Band 47

OMAR NASSAR

Innovative micro-NMR/MRI functionality utilizing flexible electronics and control systems

Omar Nassar

#### **Innovative micro-NMR/MRI functionality utilizing flexible electronics and control systems**

#### Schriften des Instituts für Mikrostrukturtechnik am Karlsruher Institut für Technologie (KIT) Band 47

Hrsg. Institut für Mikrostrukturtechnik

Eine Übersicht aller bisher in dieser Schriftenreihe erschienenen Bände finden Sie am Ende des Buchs.

## **Innovative micro-NMR/MRI functionality utilizing flexible electronics and control systems**

by Omar Nassar

Karlsruher Institut für Technologie Institut für Mikrostrukturtechnik

Innovative micro-NMR/MRI functionality utilizing flexible electronics and control systems

Zur Erlangung des akademischen Grades eines Doktors der Ingenieurwissenschaften von der KIT-Fakultät für Maschinenbau des Karlsruher Instituts für Technologie (KIT) genehmigte Dissertation

von Omar Nassar, M.Sc.

Tag der mündlichen Prüfung: 27. Juli 2021 Hauptreferent: Prof. Dr. Jan G. Korvink Korreferent: Prof. Dr. Michael Bock

**Impressum**

Karlsruher Institut für Technologie (KIT) KIT Scientific Publishing Straße am Forum 2 D-76131 Karlsruhe

KIT Scientific Publishing is a registered trademark of Karlsruhe Institute of Technology. Reprint using the book cover is not allowed.

www.ksp.kit.edu

*This document – excluding parts marked otherwise, the cover, pictures and graphs – is licensed under a Creative Commons Attribution-Share Alike 4.0 International License (CC BY-SA 4.0): https://creativecommons.org/licenses/by-sa/4.0/deed.en*

*The cover page is licensed under a Creative Commons Attribution-No Derivatives 4.0 International License (CC BY-ND 4.0): https://creativecommons.org/licenses/by-nd/4.0/deed.en*

Print on Demand 2023 – Gedruckt auf FSC-zertifiziertem Papier

ISSN 1869-5183 ISBN 978-3-7315-1176-2 DOI 10.5445/KSP/1000143838

#### **Abstract**

The central theme of this work is developing and integrating exible electronics for micro Magnetic Resonance (MR) applications. Two crucial applications have been covered within this dissertation; one application in the eld of Magnetic Resonance Imaging (MRI) and the other in the eld of Nuclear Magnetic Resonance (NMR). The MRI application focus on solving the safety and reliability issues involved with MR-catheters. The NMR application introduces a novel approach for enhancing the NMR spectroscopy throughput.

The rst part of the dissertation covers the various developed technologies to fabricate exible electronics on the micro-scale. The covered MR applications required the fabrication of inductors, capacitors, and diodes on exible substrates. Hence, the rst covered technology within the frame of micro-fabrication is depositing of conductive seed-layers onto exible substrates. Several techniques have been tested and compared. The developed technology enables multilayer fabrication for the conductive structure on a exible substrate (≈ 50µm thickness) suitable for wrapping around the slender tube (> 0.5mm diameter). The second covered method is inkjet printing of high-density low loss coecient capacitors. Two custom polymer-based dielectric inks were synthesized, by the dispersion of TiO<sup>2</sup> and BaTiO<sup>3</sup> in benzocyclobutene (BCB) polymer. The inkjet-printed capacitors show a relatively high capacitance per unit area up to 69 pFmm<sup>−</sup><sup>2</sup> while achieving quality factor (*Q*) up to 100. Moreover, a technique was developed for an inkjet-printed rectifying Schottky diode. The last covered technology is the electroplating of the conductive seed-layers. Electroplating is a well-explored technology and a very crucial process in the eld of micro-fabrication. However, it is highly dependent on the operator's level of experience. In addition, precision control of the electroplating power is required, particularly when creating small structures, where pulse electroplating has proven to provide a high degree of control over the deposited structure. In this context, a highly versatile current source was developed based on a microcontroller unit to bring accuracy into the generation of optimal plating recipes. The current source was built based on a Modied Howland Current Source (MHCS) conguration using a high-power operational amplier. As a result, the current source was validated and veried, and its high capability was demonstrated by conducting some unique applications, most important of which is enhancing the adhesion of the inkjet-printed seed-layer on exible substrates.

The second part of the dissertation addresses interventional MRI using MR-Catheter. MR-Catheters potentially have a signicant impact on the eld of minimally invasive medical operations. Implanted elongated transmission conductors and detector coils act as an antenna, that couple to the MR radio frequency (RF) transmit eld, thereby making the catheter visible during use in an MRI scanner. However, this coupling potentially heats the conductors, causing possible tissue damage, and has prevented this technology from nding widespread use. An alternative approach is to couple a resonator on the catheter tip inductively coupling to a wired surface coil outside of the body. However, The micro-resonator on the catheter's tip could heat during the excitation phase. Moreover, as the catheter axial orientation changes during motion, the visibility of the catheter tip changes and can be lost if the wireless resonator and external coil's magnetic elds are orthogonal. In this contribution, the tuning capacitance of the catheter's micro detector is controlled wirelessly via pulse sequence control sent to an RF tuning circuit integrated with a detector coil. The integrated circuit generates direct current (DC) from the transmitted RF for controlling capacitance remotely, creating an intelligent embedded tunable detector on the catheter tip. During RF transmission, decoupling is done by adjusting the micro-detector operating frequency away from the Larmor frequency. Additionally, a novel detector design is introduced based on two perpendicularly oriented micro-saddle coils that maintain constant visibility of the catheter for the entire range of axial orientations with no dead angle. The system was experimentally veried in a 1 T MR scanner.

The third part of the dissertation addresses NMR spectroscopy throughput. Flow-based NMR is a promising technique for improving NMR throughput. However, a common challenge is the relatively large dead volume in the ow tube feeding the NMR detector. In this contribution, a novel approach is introduced for fully automated high throughput NMR spectroscopy with enhanced mass sensitivity. The developed approach is accomplished by taking advantage of microuidic technologies combined with thin-lm micro-NMR detectors. A customized NMR probe head is developed with a microuidic system featuring: i) a micro-saddle detector for NMR spectroscopy and ii) a pair of ow sensors anking the NMR detector and interfaced to a microcontroller unit. A microuidic tube is used to transport a sample series through the probe head, comprising an unlimited number of dierent sample plugs separated by an immiscible uid. The developed system tracks the samples' position and velocity in this dual-phase ow and synchronizes the NMR acquisition. The developed custom build probe head is plugand-play ready with market-available NMR systems. The system was successfully utilized for automating ow-based NMR measurement in a 500 MHz NMR system. The developed micro-NMR detector enables high-resolution spectroscopy with NMR sensitivity of <sup>2</sup>.18 nmol s1/<sup>2</sup> with the ow sensors in operation. The ow sensors featured high sensitivity to an absolute dierence of 0.2 in relative permittivity, enabling dierentiation between most common solvents. It was illustrated that a fully automated NMR spectroscopy of nine dierent 120 µL samples could be achieved within <sup>3</sup>.6 min or eectively <sup>15</sup>.3 s per sample.

### **Zusammenfassung**

Das zentrale Thema dieser Arbeit ist die Entwicklung und Integration von exibler Elektronik für Mikro-Magnetresonanz (MR)-Anwendungen. Zwei wichtige Anwendungen wurden in der Dissertation behandelt; eine Anwendung auf dem Gebiet der Magnetresonanztomographie (MRI) und die andere auf dem Gebiet der Kernspinresonanz (NMR). Die MRI-Anwendung konzentriert sich auf die Lösung der Sicherheits- und Zuverlässigkeitsaspekte von MR-Kathetern. Die NMR-Anwendung stellt einen neuartigen Ansatz zur Steigerung des Durchsatzes bei der NMR-Spektroskopie vor.

Der erste Teil der Dissertation behandelt die verschiedenen Technologien die zur Herstellung exibler Elektronik auf der Mikroskala entwickelt wurden. Die behandelten MR-Anwendungen erfordern die Herstellung von Induktoren, Kondensatoren und Dioden auf exiblen Substraten. Die erste Technologie, die im Rahmen der Mikrofabrikation behandelt wird, ist das Aufbringen einer leitfähigen Startschicht auf exiblen Substraten. Es wurden verschiedene Techniken getestet und verglichen. Die entwickelte Technologie ermöglicht die Herstellung einer mehrschichtigen leitfähigen Struktur auf einem exiblen Substrat (50µm Dicke), die sich zum Umwickeln eines schlanken Rohres (> 0.5mm Durchmesser) eignet. Die zweite Methode ist der Tintenstrahldruck von Kondensatoren mit hoher Dichte und niedrigem Verlustkoezienten. Zwei dielektrische Tinten auf Polymerbasis wurden synthetisiert, durch die Dispersion von TiO<sup>2</sup> und BaTiO<sup>3</sup> in Benzocyclobuten (BCB) Polymer. Die im Tintenstrahldruckverfahren hergestellten Kondensatoren zeigen eine relativ hohe Kapazität pro Flächeneinheit von bis zu 69 pFmm<sup>−</sup><sup>2</sup> und erreichen dabei einen Qualitätsfaktor (*Q*) von etwa 100. Außerdem wurde eine Technik für eine tintenstrahlgedruckte gleichrichtende Schottky-Diode entwickelt. Die letzte behandelte Technologie ist die Galvanisierung der leitenden Startschichten. Die Galvanik ist eine gut erforschte Technologie und ein sehr wichtiger Prozess auf dem Gebiet der Mikrofabrikation. Sie ist jedoch in hohem Maße von der Erfahrung des Bedieners abhängig. Darüber hinaus ist eine präzise Steuerung der Galvanikleistung erforderlich, insbesondere bei der Herstellung kleiner Strukturen, wobei sich die Pulsgalvanik als ein Verfahren erwiesen hat, das ein hohes Maß an Kontrolle über die abgeschiedene Struktur bietet. In diesem Zusammenhang wurde eine hochexible Stromquelle auf Basis einer Mikrocontroller-Einheit entwickelt, um Genauigkeit in die Erstellung optimaler Galvanikrezepte zu bringen. Die Stromquelle wurde auf Basis einer modizierten Howland-Stromquelle (MHCS) unter Verwendung eines Hochleistungs-Operationsverstärkers (OPAMP) aufgebaut. Die Stromquelle wurde validiert und veriziert, und ihre hohe Leistungsfähigkeit wurde durch die Durchführung einiger schwieriger Anwendungen demonstriert, von denen die wichtigste die Verbesserung der Haftung der im Tintenstrahldruckverfahren gedruckten Startschicht auf exiblen Substraten ist.

Der zweite Teil der Dissertation befasst sich mit interventioneller MRT mittels MR-Katheter. MR-Katheter haben potenziell einen erheblichen Einuss auf den Bereich der minimalinvasiven medizinischen Eingrie. Implantierte längliche Übertragungsleiter und Detektorspulen wirken wie eine Antenne und koppeln sich an das MR-Hochfrequenz (HF)-Sendefeld an und machen so den Katheter während des Einsatzes in einem MRT-Scanner sichtbar. Durch diese Kopplung können sich die Leiter jedoch erhitzen, was zu einer gefährlichen Erwärmung des Gewebes führt und eine breite Anwendung dieser Technologie bisher verhindert hat. Ein alternativer Ansatz besteht darin, einen Resonator an der Katheterspitze induktive mit einer Ober ächenspule außerhalb des Körpers zu koppeln. Allerdings könnte sich auch dieser Mikroresonator an der Katheterspitze während der Anregungsphase erwärmen. Außerdem ändert sich die Sichtbarkeit der Katheterspitze, wenn sich die axiale Ausrichtung des Katheters während der Bewegung ändert, und kann verloren gehen, wenn die Magnetfelder des drahtlosen Resonators und der externen Spule orthogonal sind. In diesem Beitrag wird die Abstimmkapazität des Mikrodetektors des Katheters drahtlos über eine Impulsfolgensteuerung gesteuert, die an einen HF-Abstimmkreis gesendet wird, der in eine Detektorspule integriert ist. Der integrierte Schaltkreis erzeugt Gleichstrom aus dem übertragenen HF Signal zur Steuerung der Kapazität aus der Ferne, wodurch ein intelligenter eingebetteter abstimmbarer Detektor an der Katheterspitze entsteht. Während der HF-Übertragung erfolgt die Entkopplung durch eine Feinabstimmung der Detektorbetriebsfrequenz weg von der Larmor-Frequenz. Zusätzlich wird ein neuartiges Detektordesign eingeführt, das auf zwei senkrecht ausgerichteten Mikro-Saddle-Spulen basiert, die eine konstante Sichtbarkeit des Katheters für den gesamten Bereich der axialen Ausrichtungen ohne toten Winkel gewährleisten. Das System wurde experimentell in einem 1 T MRT-Scanner veriziert.

Der dritte Teil der Dissertation befasst sich mit dem Durchsatz von NMR-Spektroskopie. Flussbasierte NMR ist eine vielversprechende Technik zur Verbesserung des NMR-Durchsatzes. Eine häuge Herausforderung ist jedoch das relativ große Totvolumen im Schlauch, der den NMR-Detektor speist. In diesem Beitrag wird ein neuartiger Ansatz für vollautomatische NMR-Spektroskopie mit hohem Durchsatz und verbesserter Massensensitivität vorgestellt. Der entwickelte Ansatz wird durch die Nutzung von Mikrouidik-Technologien in Kombination mit Dünnlm-Mikro-NMR-Detektoren verwirklicht. Es wurde ein passender NMR-Sensor mit einem mikrouidischen System entwickelt, das Folgendes umfasst: i) einen Mikro-Sattel-Detektor für die NMR-Spektroskopie und ii) ein Paar Durchusssensoren, die den NMR-Detektor ankieren und an eine Mikrocontrollereinheit angeschlossen sind. Ein mikrouidischer Schlauch wird verwendet, um eine Probenserie durch den Sondenkopf zu transportieren, die einzelnen Probenbereiche sind durch eine nicht mischbare Flüssigkeit getrennt, das System erlaubt im Prinzip eine unbegrenzte Anzahl an Proben. Das entwickelte System verfolgt die Position und Geschwindigkeit der Proben in diesem zweiphasigen Fluss und synchronisiert die NMR-Akquisition. Der entwickelte kundenspezische Sondenkopf ist Plug-and-Play-fähig mit marktüblichen NMR-Systemen. Das System wurde erfolgreich zur Automatisierung von ussbasierten NMR-Messungen in einem 500 MHz NMR-System eingesetzt. Der entwickelte Mikro-NMR-Detektor ermöglicht hochauösende Spektroskopie mit einer NMR-Empndlichkeit von <sup>2</sup>.18 nmol s1/<sup>2</sup> bei Betrieb der Durchusssensoren. Die Durchusssensoren wiesen eine

hohe Empndlichkeit bis zu einem absoluten Unterschied von 0.2 in der relativen Permittivität auf, was eine Dierenzierung zwischen den meisten gängigen Lösungsmitteln ermöglichte. Es wurde gezeigt, dass eine vollautomatische NMR-Spektroskopie von neun verschiedenen <sup>120</sup> <sup>µ</sup><sup>L</sup> Proben innerhalb von <sup>3</sup>.6 min oder eektiv <sup>15</sup>.3 s pro Probe erreicht werden konnte.

## **Nomenclature**

#### **Abbreviations**




#### **Symbols**


#### **Constants**


## **Contents**







#### **1 Introduction**

#### **1.1 Motivation**

Since the magnetic resonance (MR) concept was rst introduced, it led to the development of two versatile technologies: magnetic resonance imaging (MRI) and magnetic resonance spectroscopy (MRS). MRI is a non-invasive medical imaging technique used extensively in hospitals and clinics to investigate and diagnose tumors, internal bleeding, and infections. What distinguishes MRI as an imaging technique is that it does not involve any harmful ionizing radiation. On the other hand, MRS is an information-rich spectroscopic technique widely used in chemistry and biology to characterize materials by determining and quantifying molecular structure. In addition, MRS oers broad chemical coverage and is a non-destructive technique, allowing further upstream analytics.

In its very primitive structure, besides the interface electronics, the MRI or MRS hardware consists of three main components: 1) a large permanent magnet responsible for the polarizing static magnetic eld (*B*0), 2) a radio frequency (RF) transmitter coil responsible for the excitation of the alternating magnetic eld (*B*1), and 3) an RF receiver to acquire the MR signals, which is commonly done using the RF transmitter coil itself. Employing micro-MR probes is a well-established approach to enhance the signal-to-noise ratio (SNR) of the MR experiment if only a small amount of sample is available. Integrating microuidic technologies and micro-embedded systems with micro-MR detectors has continued to evolve since the rst NMR-detector reports. These integrations lead to a wide range of novel applications such as biological sample monitoring [1–3], chemical reaction tracking [4–8], sensitivity enhancement techniques [9–12], and electrochemical functionality [13].

An example of the contribution of micro-MR detectors in the eld of MRI is the development of MR-catheter for minimally invasive medical operations. Locating a micro-MR detector on a catheter's tip acts as an antenna and couples to the magnetic resonance RF transmit eld, thereby making the catheter visible during use in an MR scanner. This approach provides surgeons with an accurate geometrical orientation of the catheter's tip relative to the patient's body. Catheter-based intervention is among the primary methods used in the treatment diagnosis of heart diseases. Over the past two decades, cardiovascular diseases have been reported as the leading cause of death globally, with more than seventeen million deaths in 2017 surpassing all other medical causes combine [14]. However, during the excitation stage, the coupling between the catheter detector and the RF magnetic eld can heat the conductors, causing hazardous tissue heating, which has kept this technology from being widely used. Furthermore, as the catheter's axial direction alters during interventions, the visibility of the catheter tip changes in the MR image and can be lost if the magnetic elds of the wireless resonator and external coil are orthogonal.

An example of the contribution of micro-MR detectors in the eld of NMR is the development of ow-based NMR spectroscopy. Flow-based NMR is a promising technique for transporting the samples through the NMR detector in a microuidic tube without the need for hardware disassembling and assembling. However, a common challenge is the relatively large dead volume in the ow tube feeding the NMR detector that results in the fact that NMR is a low throughput method. Therefore, there is a disconnect between applications that require large parameter spaces screening and the opportunity to take advantage of this information-dense method. Addressing the NMR throughput challenge could lead to a signicant push to diverse elds, including drug discovery screening, clinical analysis, chemical process monitoring, bio uid metabolomics, and food analysis. Consider the eld of drug discovery as an example; NMR can resolve the binding of drugs to pathogens and can be utilized for screening compound libraries to investigate the molecular interactions with target proteins [15, 16]. Pharmaceutical companies have molecular libraries containing millions of compounds, which is ever-increasing as new molecules are being synthesized utilizing combinatorial methods. For example, Pzer, Inc. maintains 4 million compounds inside their molecular library that can be used for analyzing potential medicines, as reported in 2017 [17]. Hence, the development of high throughput NMR methods is a must to enhance the screening eciency of these massive and ever-increasing molecular libraries.

#### **1.2 Objective**

To tackle the challenges facing the MR-catheters and NMR spectroscopy throughput, the present work aims to address three main hypotheses:

#### • **Interventional MRI using MR-catheters**

Integrating micro-MR detector with micro-embedded systems to wirelessly decouple the micro-resonator on the catheter's tip to circumvent the heating hazard. The developed system needs to be controlled wirelessly to avoid using conductive wires inside the patient, as they are the main cause of dangerous heating. The developed embedded system should have an antenna to pick up the transmitted RF control signal operating at a frequency range away from the Larmor frequency and rectify it to control the NMR detector's variable capacitance of. The whole system should have a thickness below 80 µm with a high degree of exibility to be rolled around the slender catheter tube (5 fr). Moreover, the detector has to be designed in a way to keep coupling with the surface coil through the whole axial orientations to avoid losing the catheter's visibility at any orientation.

#### • **Automated high throughput NMR spectroscopy**

Integrating micro-NMR detector with microuidic technologies to build a customized probe head capable of online synchronization of NMR acquisition for two-phase follow. The dead volume in the ow tube is eliminated by loading the tube with plugs of dierent samples separated by immiscible uid. The customized probe head should possess a sensing system for real-time monitoring of the position and velocity of the uid interface. The sensing system should be sensitive to dierent NMR samples and integrates a low-cost embedded controller to synchronize the NMR acquisition. In addition, the system should be compatible with market-available NMR systems.

#### • **Micro-fabrication of exible electronics**

To achieve the previously mentioned objectives, the fabrication of thin, exible electronics needs to be addressed by focusing on: 1) enhancing the thin-lm fabrication technology for depositing conductive layers onto exible substrates, 2) developing thin lm exible capacitors with high capacitance density to reduce their volume to be integrated on the catheter tip, and 3) developing an electroplating method for enhancing the adhesion of the conductive structure on the exible substrates.

## **1.3 Thesis outline**

This thesis consists of ten chapters structured in three main parts, accompanied by two introductory chapters and the conclusion. The thesis is organized as follow:

#### • **Introductory chapters**

	- 3. Chapter 3 presents the dierent developed processes for depositing conductive seed layers onto exible substrates and provides a comparison between these processes. It introduces the developed micro-fabrication process for fabricating thin-lm capacitors and diodes.
	- 4. Chapter 4 presents the development of a exible, versatile current source for electroplating. It covers the dierent novel plating applications conducted using the developed plating system.
	- 5. Chapter 5 is an overview of state-of-the-art MR-catheters focusing on the dierent developed techniques for solving the heating issue.
	- 6. Chapter 6 presents a thorough description of the design and development of the introduced micro embedded system for the interventional MRI application.
	- 7. Chapter 7 is an overview of state-of-the-art high throughput NMR spectroscopy. The chapter also presents a brief overview of the state-of-the-art micro-NMR detectors and ow sensors since they are employed for building the introduced high throughput NMR approach.
	- 8. Chapter 8 presents a thorough description of the design, characterization, and fabrication of the fully automated high throughput NMR system.
	- 9. Chapter 9 presents various NMR ow experiments demonstrating the the versatility of the developed technology for realizing automated high throughput NMR spectroscopy with high mass sensitivity.
	- 10. Chapter 10 concludes the entire thesis and gives an outlook for future work.

## **1.4 Main results**

The research conducted during the PhD work has led to four main publications with contribution as the rst author (three published and one in preparation) and three other publications with contribution as coauthor (two published and one in preparation).

#### **Micro-fabrication of exible electronics**

#### 1. **Load sensitive stable current source for complex precision pulsed electroplating**

Published in: Review of Scientic Instruments [ON1] Authorship status: Lead author

Content: This paper presents a highly exible current source based on a microcontroller unit used for the electroplating process. The functionality and stability of the developed system under dierent plating parameters were described. The developed system was used for conducting two nontrivial plating applications.

Contribution: Involved in conceiving the idea. Designed, developed, and characterized the electroplating power source. Carried out the application of electroplating of inkjet-printed tracks. Lead in writing the manuscript and preparation of gures.

2. **Polyaramid-Based Flexible Antibacterial Coatings Fabricated Using Laser-Induced Carbonization and Copper Electroplating**

Published in: ACS Applied Materials & Interfaces [ON2] Authorship status: Coauthor

Content: This paper introduced a fabrication method for developing electrical circuits on exible polyaramid substrates using laser-induced carbonization and copper electroplating.

Contribution: Performed and optimized the electroplating processes.

#### **Interventional MRI using MR-catheters**

3. **Wireless Double Micro-Resonator for Orientation Free Tracking of MR-Catheter During Interventional MRI.**

Published in:IEEE Journal of Electromagnetics, RF and Microwaves in Medicine and Biology [ON3]

Authorship status: Lead author

Content: This work proposed a novel micro-resonator based on two perpendicularly oriented saddle coils that create a homogeneous magnetic eld when wrapped around a catheter tube.

Contribution: Convinced the idea. Designed, optimized, and fabricated the introduced micro-resonator. Performed the MRI experiments. Lead in writing the manuscript and preparation of gures.

4. **Topologically optimized magnetic lens for magnetic resonance applications.**

Published in: Magnetic Resonance [ON4]

Authorship status: Coauthor

Content: This work employed the topology optimization technique to develop optimal layouts of a Lenz lens to improve the SNR of MR detection.

Contribution: Conducted the initial inductive coupling experiments.

#### 5. **Wireless tuning system towards safe interventional magnetic resonance imaging.**

Manuscript status: In preparation

Authorship status: Lead author

Content: This work introduces a novel wireless tuning system to decouple MR-catheter in an iMRI operation during RF excitation to avoid the heating hazard.

Contribution: Convinced the idea. Designed, fabricated, and characterized the wireless tuning system. Lead in writing the manuscript and preparation of gures.

#### **Automated high throughput NMR spectroscopy**

6. **Integrated impedance sensing of liquid sample plug ow enables automated high throughput NMR spectroscopy.**

Published in: Nature - Microsystems & Nanoengineering [ON5]

Authorship status: Lead author

Content: The paper introduced a novel customized NMR probe head that is plugand-play ready with market-available NMR systems for enhancing NMR spectroscopy throughput. The paper demonstrates the capability and the applications of the introduced technology.

Contribution: Involved in conceiving the idea. Designed, fabricated, and optimized the experimental setup. Conducted the system characterization, performed the NMR experiments, and conducted data analysis. Lead in writing the manuscript and preparation of gures.

7. **Highly uorinated peptide probes with enhanced in vivo stability for** <sup>19</sup>**F-MRI.**

Manuscript status: In preparation

Authorship status: Coauthor

Content: This work aimed to develop peptide-based 19F labels suited for 19F MRI at micron to milli-scale. The Highly uorinated peptides bearing up to 27 uorine nuclei possess high bio-compatibility and in vivo stability, achieving in situ labeling on BSA and glutathione.

Contribution: Developed the micro-MR saddle detector used for the MRI experiments, and participated in conducting the MRI experiments.


# **2 Theoretical background**

**N** uclear Magnetic Resonance (NMR) has a very wide range of applications in chemistry, biology, and medicine. Isidor Isaac Rabi rst introduced NMR in 1938 [18] where he was awarded the Nobel Prize in physics in 1944 for his techniques of using nuclear magnetic resonance to determine the nuclear spin and magnetic moment of atoms. Later in 1952, Felix Bloch [19] and Mills Purcell [20] shared the Nobel Prize in physics for expanding the NMR technique to include solids and liquids, leading to a new branch of science. The further research and development of NMR were highlighted with three additional Nobel Prizes; in 1991 in chemistry by Richard R. Ernst for the development of Fourier transform NMR spectroscopy, in 2002 in chemistry by Kurt Wüthrich for studying biological molecules using NMR, and in 2003 in physiology or medicine by Manseld and Lauterbur for developing magnetic resonance imaging (MRI). Paul Christian Lauterbur rst introduced MRI in 1973 [21].

In this chapter, the fundamental concepts of NMR and MRI, which are most relevant to this thesis, are introduced. The chapter is divided mainly into four sections. In Section 2.1, the fundamental concepts and terminologies of the NMR, such as nuclear spins, polarization, excitation, and relaxation, are introduced. Section 2.2 presents the basics of acquiring an NMR spectrum by explaining the concept of chemical shift and spins coupling. Section 2.3 discusses the primary technique for collecting an MR image and the eld encoding. lastly, Section 2.4 introduces the MR system's main hardware components by highlighting the dierent elements between an NMR spectrometer and an MRI scanner. Within the last subsection, some of the essential parameters for evaluating the NMR spectrum signal-to-noise ratio are discussed. The information presented in this introductory chapter were mostly collected from the following three textbooks:


## **2.1 Magnetic resonance fundamentals**

Nowadays, the concept of Nuclear Magnetic Resonance (NMR) is widely used in various applications. These applications could be categorized mainly into NMR spectroscopy and magnetic resonance imaging (MRI). NMR spectroscopy is also known as magnetic resonance spectroscopy (MRS), and it is an analytical technique to study molecules and compounds by determining and quantifying their molecular structure. Therefore, it is extensively used in medicine, chemistry, and biology. On the other hand, hence MRI is an imaging technique; its mainly used for diagnosing diseases such as tumors or internal bleeding.

### **2.1.1 Spin angular momentum**

Spin is the keyword of NMR. Every atomic nucleus consists of elementary particles called protons and neutrons, which are composite fermions particles. Fermions are particles that have half-odd integer spin and are described by the Pauli exclusion principle. Accordingly, spinning is an intrinsic property of the nuclei of the atoms. The particles' spinning originates a spin angular momentum vector (~*S*), which can be described in:

$$|S| = \hbar \sqrt{I(I+1)},\tag{2.1}$$

Where } is the Planck constant (6.62607015×10−34J s) divided by <sup>2</sup><sup>π</sup> and *<sup>I</sup>* is dened as the nuclear spin quantum number.

*I* plays a signicant role in determining the possibility of detecting a material using the NMR technique. NMR detects signals only for a non-zero nuclear spin. If a nucleus has an even number of protons and neutrons, then the nucleus has no spin (*I* = 0), and it is NMR silent. <sup>32</sup>S, <sup>16</sup>O, and <sup>12</sup>C are examples of NMR silent nuclei. On the other hand, if the number of either the proton or neutron or both is odd, then it has a non-zero nuclear spin (*I* 6= 0) and detectable by NMR. Nuclei such as <sup>1</sup>H and <sup>13</sup>C are examples of nuclei with non-zero spin.

As a consequence of the spin of protons and neutrons (which have a positive charge), a magnetic moment vector (~µ) is originated. The magnetic moment (~µ) is proportional to spin angular momentum (~*S*) by the gyromagnetic ratio γ, which is a unique intrinsic property of each nucleus,

$$
\mathfrak{A} = \mathfrak{Y} \text{ S.}\tag{2.2}
$$

The magnetic moment (~µ) and the spin angular momentum (~*S*) are vector quantities, and based on their directions, the sign of the gyromagnetic ratio (γ) is dened. Thus if the two vectors have parallel alignment, then γ has a positive sign, while the negative sign describes antiparallel alignment, as shown in Fig. 2.1. <sup>13</sup>*C* is an example of a positive gyromagnetic ratio

Figure 2.1: Determination of the gyromagnetic ratio's (γ) sign according to the alignment of the magnetic moment (~µ) with respect to the magnetic moment (~*S*).

(γ > 0), while <sup>15</sup>*N* has a negative gyromagnetic ratio (γ < 0).

### **2.1.2 Spin polarization and Larmor precession**

The spin quantum number describes the spin's possible orientation states, where 2*I* + 1 can determine the number of the spin's possible orientations. The hydrogen atom (1H) has one proton, which means that *I* = 1 2 . Therefore there are two possible orientations spin for the hydrogen proton. In the absence of an external energy source, such as a static magnetic eld, both orientations have the same energy level. On the other hand, with the presence of a magnetic eld, the two orientations have two quantized energy levels described by Zeeman splitting with magnetic the quantum number *m* (*m* = ± 1 2 for a proton), as shown in Fig.2.2a. The rst possible orientation is the lower energy level, known as α spin state, and it's parallel to the magnetic eld. The other orientation has higher energy, known as β spin state, and opposes the magnetic eld. The total energy dierence is (∆*E*) proportional to the strength of the applied magnetic eld (*B*~0), as shown in Fig.2.2b based on:

$$
\Delta E = -\sharp \mathbb{1} \cdot \mathcal{S}\_0. \tag{2.3}
$$

Conventionally the static magnetic eld is in the z-direction (*B<sup>Z</sup>* = *B*~0 <sup>=</sup> *<sup>B</sup>*<sup>0</sup> and *<sup>B</sup><sup>X</sup>* <sup>=</sup> *<sup>B</sup><sup>Y</sup>* <sup>=</sup> <sup>0</sup>) and the magnetic moment in the z-direction is µ*<sup>Z</sup>* = γ*m*}. Hence, the total energy dierence (∆*E*) for proton is,

$$
\Delta E = E \underset{m = -\frac{1}{2}}{\quad \text{ $1 \to \infty$ }} \frac{1 - E}{m + \frac{1}{2}} = -(-\frac{1}{2})\hbar \gamma \mathcal{B}\_0 + (\frac{1}{2})\hbar \gamma \mathcal{B}\_0 = \hbar \gamma \mathcal{B}\_0. \tag{2.4}
$$

With the absence of the external magnetic eld (*B*<sup>0</sup> = 0) and the assumption that there is

Figure 2.2: Zeeman splitting. **(a)** Zeeman splitting of <sup>1</sup>H nuclei with the application of an external magnetic eld (*B*~0). **(b)**The energy dierence between the two quantized energy levels is proportional to the external magnetic eld's strength (*B*0).

no spin to spin interaction, the magnetic moments have random orientations and degenerated energy levels. If, in contrast, a homogeneous static magnetic eld *B*~<sup>0</sup> is applied, conventionally in z-direction, and due to the spin vectors' energetic preference for parallel alignment with respect to the external magnetic eld, more spins prefer the parallel alignment than the antiparallel states. Consequently, a bulk macroscopic magnetization *M*~<sup>0</sup> originates, due to nuclear magnetism, parallel to the external magnetic eld and proportional to the quantity dierence of the parallel and anti-parallel states, as shown in Fig 2.3. This process of generating the net magnetization (*M*~0) of a sample is known as polarisation. The magnetic moment rotates around the external magnetic eld vector with a specic frequency called Larmor frequency (ω0), and this rotation is known as precession, shown in Fig.2.4a. According to the Plank-Einstein relation, the proton's energy (*E*) is proportional to the frequency(ν) (*E* = *h*ν = *h* ω 2π = }ω), then Larmor frequency can be formulated using Equation 2.4,

$$
\alpha\_0 = \frac{E}{\hbar} = \gamma B\_0. \tag{2.5}
$$

Equation 2.5 reveals the linear relation between the Larmor frequency and the applied static magnetic eld (*B*0), as shown in Fig. 2.4b. Hence, a homogeneous *B*<sup>0</sup> magnetic eld is required to achieve a proper spectral distribution without losing the sensitivity and selectivity of the NMR. Additionally, the radio frequency (RF) circuits in the NMR should be correctly designed to operate at the Larmor frequency.

The spin's probability distribution of occupying one of the two available energy states is described by Boltzmann distribution (*e*−*E*/*KB<sup>T</sup>* ). Hence, the ratio between the populations of the occupied high and low Zeeman states is,

$$\frac{N\_{high}}{N\_{lower}} = \frac{N\_{-1/2}}{N\_{+1/2}} = \frac{\frac{-E\_{-1/2}}{K\_B T}}{\frac{-E\_{+1/2}}{e}} = e \frac{-\Delta E}{K\_B T} \,, \tag{2.6}$$

Where *K<sup>B</sup>* is the Boltzmann constant (1.38×10−23JK−<sup>1</sup> ) and T is the temperature in K. Substituting in Equation 2.6 by Equation 2.4 with considering an approximation,

$$\frac{N\_{high}}{N\_{lower}} \approx 1 - \frac{\hbar \chi B\_0}{K\_B T} \,\text{.}\tag{2.7}$$

The net magnetization can be be expressed in terms of total spin (*N*0) using the population dierence (∆*N*),

$$
\vec{M}\_0 = \Delta N \vec{\gamma S} = \frac{B\_0 N\_0 \overleftrightarrow{\gamma}^2 \hbar^2}{4K\_B T}. \tag{2.8}
$$

Figure 2.3: Eect of applying an external magnetic eld on spin orientations. **(a)** Random spin orientations with the absence of an external magnetic eld. **(b)** Applying an external magnetic eld (*B*~0), align the spins parallel and anti-parallel to the applied eld building the two energy levels α and β. **(c)** The majority of the spins align in parallel to the external magnetic eld generating the net magnetization (*M*0).

### **2.1.3 Excitation and relaxation**

To conduct an NMR measurement, rst, the target sample is polarized by applying the external magnetic eld (*B*0) to establish the longitudinal net magnetization vector *M*<sup>0</sup> parallel to *B*<sup>0</sup> (conventionally in the z-direction). An excitation process is then done by applying a time-varying magnetic eld pulse (*B*1), subsequently evaluating the response of the nuclear magnetic moments (spins).

To realize the resonance condition, the time-varying excitation magnetic eld (*B*1) should operate at Larmor frequency (ω0) using an RF inductor. The *B*<sup>1</sup> eld should be superimposed perpendicular to the external *B*<sup>0</sup> eld to forces the net magnetization vector to ip toward the transversal XY plane and produce a ip angle (α*f a*) with respect to the Z-axis, as shown in Fig.2.5,

$$
\alpha\_{fa} = \gamma B\_{\mathsf{I}} \pi. \tag{2.9}
$$

Equation 2.9 reveals the direct proportionality of the ip angle to the *B*<sup>1</sup> eld strength and the pulse duration (τ). Soft pulses, such as weak elds and long pulse durations, are mainly used to select narrow bandwidth frequencies. On the other hand, hard pulses, with a strong eld

Figure 2.4: Spin precession and Larmor frequency. **(a)** The spin polarization axis's precession around the static magnetic eld vector (*B*~0) with the Larmor frequency (ω0). **(b)** The linearity between the Larmor frequency and the applied static magnetic eld.

and short pulse duration, excite a wide bandwidth. Therefore, the *B*<sup>1</sup> eld's pulse duration is precisely controlled to achieve a ip angle of 90°.

Before applying the RF excitation pulse, the net magnetization vector entirely points in the direction of the polarizing static *B*<sup>0</sup> eld, in this case, *M<sup>Z</sup>* = *M*0, and *M<sup>X</sup>* = *M<sup>Y</sup>* = 0. The excitation processes result in a tilted magnetization vector with three components *MX*, *M<sup>Y</sup>* , and *MZ*. During the excitation process, the magnetization vector *M*~ rotates in the space around the Z-axis within the XY-plane, forming a spherical pattern due to the torque *T* (~*T* = *d*~*s dt* ) produced by the *B*<sup>1</sup> eld, as shown in Fig. 2.5. Accordingly, the longitudinal magnetization component decreases, and the transversal component increases till the ip angle reaches 90°, at which *M<sup>Z</sup>* = 0, *M<sup>X</sup>* = *M*<sup>0</sup> cos(ω0*t*), and *M<sup>Y</sup>* = *M*<sup>0</sup> sin(ω0*t*)). Felix Bloch described the nuclear induction phenomenon as a mathematical analysis using Bloch equations:

$$\vec{M}(t) = \begin{bmatrix} M\_X(t) \\ M\_Y(t) \\ M\_Z(t) \end{bmatrix} = \begin{bmatrix} M\_0 \cos\left(a\mathfrak{u}t\right) \sin\mathfrak{a} \ e^{-t/T\_2} \\ M\_0 \sin\left(a\mathfrak{u}t\right) \sin\mathfrak{a} \ e^{-t/T\_2} \\ M\_0 \cos\mathfrak{a} + \left(1 - e^{-t/T\_1}\right) \left(M\_0 - M\_0 \cos\mathfrak{a}\right) \end{bmatrix}. \tag{2.10}$$

Where *T*<sup>1</sup> and *T*<sup>2</sup> are relaxation time constants describing the net magnetization vector's motion within the relaxation stage to reestablish thermal equilibrium. The relaxation process occurs by terminating the RF excitation pulse resulting in an alternating electric eld that could be detected using an RF coil; this signal is known as the NMR signal.

During the relaxation stage, the transverse magnetization *M<sup>T</sup>* (*M<sup>T</sup>* = *M<sup>X</sup>* + *jM<sup>Y</sup>* ) decays exponentially (*M<sup>T</sup>* ∝ *e*−*t*/*T*<sup>2</sup> ) following the *T*<sup>2</sup> relaxation time constant (spin to spin relaxation), which describes the dephasing of spins. This decay is called free induction decay (*F ID*), resulting from the uctuations of the frequency due to the inter-nuclear actions, such as the chemical interactions between the spins because of anisotropies. After starting the relaxation process by time *T*2, the transversal net magnetization decays to 37% of its maximum value *M*0.

Figure 2.5: Impact of a time-varying excitation magnetic eld (*B*1) on the magnetization vector (*M*0) of a polarized sample. **(a)** With the absence of the excitation eld (*B*1), The magnetization vector (*M*0) points in the direction of the polarizing static (*B*0) eld (*M<sup>Z</sup>* = *M*<sup>0</sup> and *M<sup>X</sup>* = *M<sup>Y</sup>* = 0). **(b)** Applying a soft pulse with a duration *t*<sup>1</sup> results in the magnetization vector's inclination with a ip angle of <sup>α</sup>*f a*<sup>1</sup> (α*f a*<sup>1</sup> <sup>&</sup>lt; <sup>90</sup>°). **(c)** <sup>A</sup> <sup>90</sup>° excitation RF pulse is applied, resulting in a total tilt of the magnetization vector to the XY plane (*M<sup>Z</sup>* = 0, *M<sup>X</sup>* = *M*<sup>0</sup> cos(ω0*t*), and *M<sup>Y</sup>* = *M*<sup>0</sup> sin(ω0*t*)).

Simultaneously, the longitudinal net magnetization *M<sup>Z</sup>* increases exponentially (*M<sup>Z</sup>* ∝ (1 − *e*−*t*/*T*<sup>1</sup> )) to restore its full longitudinal magnetization. *T*<sup>1</sup> relaxation is described as spin to lattice relaxation. The system emits the excess energy out to the environment as photons to reach equilibrium and achieve the preferred minimum energy situation. After starting the relaxation process by time *T*1, the longitudinal net magnetization restores 63% of its maximum value *M*0.

Typically, the transversal magnetization (*M<sup>T</sup>* ) vanishes before the longitudinal magnetization (*MZ*) is fully restored as *T*<sup>2</sup> is shorter than *T*1. Thus, a spiral pattern describes the magnetization vector's rotation during the relaxation stage, as shown in Fig. 2.6.

### **2.1.4 Free Induction Decay (FID)**

As explained in the previous subsection, the total transversal magnetization (*M<sup>T</sup>* ) is a complex sum of the magnetizations in both coordinates, X and Y,

$$M\mathbf{r} = M\mathbf{x} + jM\mathbf{y} = M\_0 \cos\left(a\mathbf{a}\_0 t\right) e^{-t/T\_2} + jM\_0 \sin\left(a\mathbf{a}\_0 t\right) e^{-t/T\_2} = M\_0 e^{j\mathbf{a}\_0 t} e^{-t/T\_2}.\tag{2.11}$$

*M<sup>T</sup>* decays exponentially by the decay constant *T*2, so during relaxation, the magnetization vector rotates around the Z-axis, and simultaneously its amplitude shrinks, as shown in Fig. 2.7a. The real and imaginary parts of *M<sup>T</sup>* decay are shown separately in Fig. 2.7b. The real and imaginary signals are sinusoidal with exponentially decaying amplitudes, and they have dierent phases. A Fourier transform is applied to represent the signals in the frequency domain instead of the time domain, as shown in Fig. 2.7c. The Fourier transform (FT) is a mathematical transformation by integrating the product of a sinusoidal signal and the MR time-domain signal,

$$M(\mathfrak{a}) = \int\_{\mathfrak{a}}^{-\mathfrak{a}} M(t) \, e^{-j\mathfrak{a}t} \, dt. \tag{2.12}$$

Figure 2.6: Magnetization vector's behavior during the relaxation stage after being excited with a 90 degree ip angle. **(a)** Re-alignment of the magnetization vector (*M*~0) with the polarizing eld (*B*~0). The trajectory of *M*<sup>0</sup> follows an envelope shape function due to the ratio between the time constants *T*<sup>1</sup> and *T*2. **(b)** Longitudinal recovery of *M<sup>Z</sup>* and transverse decay of *MXY* over time.

By applying Fourier transform for the transversal magnetization vector (*M<sup>T</sup>* ):

$$\begin{split} M(\mathfrak{a}) &= \int\_{0}^{\omega} M\_{0} e^{j\mathfrak{a}\_{0}t} e^{-t/T\_{2}} e^{-j\mathfrak{a}t} dt = \int\_{0}^{\omega} M\_{0} e^{[j(\mathfrak{a} - \mathfrak{a}\_{0}) - 1/T\_{2}]t} dt \\ &= M\_{0} \frac{e[-j(\mathfrak{a} - \mathfrak{a}\_{0}) - 1/T\_{2}]t}{-j(\mathfrak{a} - \mathfrak{a}\_{0}) - 1/T\_{2}} \Big|\_{0}^{\omega} \\ &= \frac{M\_{0}}{j(\mathfrak{a} - \mathfrak{a}\_{0}) + 1/T\_{2}} \times \frac{-j(\mathfrak{a} - \mathfrak{a}\_{0}) + 1/T\_{2}}{-j(\mathfrak{a} - \mathfrak{a}\_{0}) + 1/T\_{2}} \\ \frac{M(\mathfrak{a})}{M\_{0}} &= \frac{1/T\_{2}}{(\mathfrak{a} - \mathfrak{a}\_{0})^{2} + 1/T\_{2}^{2}} - j \frac{(\mathfrak{a} - \mathfrak{a}\_{0})}{(\mathfrak{a} - \mathfrak{a}\_{0})^{2} + 1/T\_{2}^{2}}. \end{split} \tag{2.13}$$

The line-shapes of the real and imaginary parts in Equation 2.13 are called absorption and dispersion mode Lorentzian line-shapes. At the Larmor frequency (ω = ω0), the real part reaches it's maximum amplitude of *T*2, while the imaginary part is eliminated, as shown in Fig.2.8a. To calculate the full width at half maximum (*FW HM*) of the absorption, the real part is equated to *T*2/2:

$$Re(\frac{M(\mathfrak{a})}{M\_0}) = \frac{1/T\_2}{(\mathfrak{a}\mathfrak{o}\_{1/2} - \mathfrak{o}\mathfrak{o}\_0)^2 + 1/T\_2^2} = \frac{T\_2}{2}$$

$$\therefore \ (\mathfrak{o}\_{1/2} - \mathfrak{o}\mathfrak{o}\_0)^2 = \frac{1}{T\_2^2} \quad \rightarrow \quad \mathfrak{o}\mathfrak{o}\_{1/2} = \mathfrak{a}\mathfrak{o}\_0 + \frac{1}{T\_2} \text{ or } \mathfrak{a}\mathfrak{o}\_0 - \frac{1}{T\_2}$$

$$\therefore \ FWHM = \Delta\mathfrak{o} = \mathfrak{a}\mathfrak{o} + \frac{1}{T\_2} - (\mathfrak{a}\mathfrak{o}\_0 - \frac{1}{T\_2})$$

$$= \frac{2}{T\_2} \text{ rad s}^{-1}. \tag{2.14}$$

Accordingly, the absorption's maximum amplitude and FWHM are directly dependent on the relaxation time constant *T*2. Practically, due to the arbitrary initial phase shift (ϕ) between the real and imaginary parts of the *M<sup>T</sup>* signal in the time domain, the frequency domain's line-

Figure 2.7: Generation of the NMR signal during the relaxation stage. **(a)** During relaxation, *M*~<sup>0</sup> rotates around *B*~<sup>0</sup> while precessing with the Larmor frequency to re-align with *B*~0. **(b)** The precessing magnetic moment generated a magnetic eld, which can be picked up using an RF coil. The generated signal in the time domain is an exponentially decaying sinusoidal with Larmor frequency. **(c)** Fourier transform is applied to transfer the signal from the time domain to the frequency domain to acquire the NMR spectrum.

shapes may neither be absorption nor dispersion, as shown in Fig.2.8. Consequently, after acquiring the NMR spectrum, a phase correction for the real part is required to phase the spectrum in absorption mode. This process is called phasing the NMR spectrum.

An RF coil with electrical resistance (*Rl*) and inductance (*L*) is placed close to the target sample to obtain the NMR signal. According to Faraday's law, based on the change in the net magnetization vector with time, an electromotive force (ε) is induced in the RF coil,

$$\varepsilon(t) = -\frac{d\phi\_B}{dt} = -\frac{d(\vec{B}\cdot\vec{A})}{dt} = -L\frac{dI\_l}{dt}.\tag{2.15}$$

Where φ*<sup>B</sup>* is the magnetic ux, equal to the dot product of the magnetic ux density ~*B* and the perpendicular vector area ~*A*. *I<sup>l</sup>* is the induced current in the RF coil as a result of the change in the transversal magnetization *M*~*<sup>T</sup>* based on the external applied excitation magnetic eld *B*~1. Consequently, the induced electromotive force (ε) is the NMR signal with direct proportionality to *M*~*<sup>T</sup>* . The evolution of this signal is described by the free induction decay (FID), as shown in Fig. 2.7.

Figure 2.8: Eect of the arbitrary initial phase shift on the line-shape in the frequency domain. **(a)** The magnetization vector (*M*0) has no initial phase shift (ϕ = 0°); hence the absorption line doesn't need phase correction in the frequency domain. **(b)** and **(c)** *M*<sup>0</sup> has an initial phase shift (ϕ) of 90° and 180°, resulting in an out of phase absorption peak in the frequency domain that requires phase correction.

## **2.2 Nuclear Magnetic Resonance (NMR) spectroscopy**

Chemists massively utilize NMR spectroscopy to analyze compounds and determine their structure and molecular identity. NMR spectroscopy is based on two essential phenomena, chemical shift and J-coupling, to have the ability to dierentiate the molecules. The fundamental steps of the NMR spectroscopy could be summarized in few steps, as shown in Fig. 2.9:


### **2.2.1 Chemical shift**

Chemical shift is a fundamental phenomenon of the NMR. The atoms' bonds result in a slight change in the energy gap between α and β spin states. Consequently, a slight shift in the range of a few Hz could be observed on the precession frequency. However, NMR mainly focuses on the nucleus, where the atom's bonds and the electrons distribution aect the output signal. As described previously, the protons' dynamics are based on the absorbed energy after being exposed to the external magnetic eld. The electron distribution around the nucleus plays

Figure 2.9: Basic procedures of NMR spectroscopy. Place the target sample in a static magnetic eld (*B*0) for polarization, then apply an alternating RF magnetic eld (*B*1) at Larmor frequency for excitation and nally switch o the *B*<sup>1</sup> eld for relaxation and picking up the nuclear magnetic eld.

a signicant role as it could shield the external magnetic eld from reaching the nucleus, as shown in Fig. 2.10. In the case of high electron density (strong shield), the external static eld has a lower eect on the nucleus, and the NMR spectrum appears at a lower frequency (upeld) and vice versa. The primary source changing the electrons distribution in an atom is the bonds with its neighboring atoms.

Figure 2.11 shows the typical NMR spectra of methyl chloride (CH3Cl) and methyl uoride (CH3F) as an example of the electron shield's eect on the NMR spectrum. For methyl chloride, the carbon atom has a polar bond with the chlorine atom, which contains 17 electrons around its nucleus. On the other hand, for methyl uoride, the carbon atom has a polar bond with the uorine atom, which contains only 9 electrons. Due to the higher electron density in the chlorine atom, the electron shield in methyl chloride is higher than that in methyl uoride. Therefore, the methyl uoride's chemical shift (towards downeld) is higher than that for methyl chloride (towards upeld).

$$\delta(\text{ppm}) = \frac{\Delta \text{\textnu(Hz)}}{\text{\textnu(MHz)}} \times 10^6 = \frac{(\text{\textnu(Hz)} - \text{\textnu}\_{\text{ref}} \text{ef}(\text{Hz}))}{\text{\textnu(MHz)}} \times 10^6,\tag{2.16}$$

where <sup>ν</sup> is the measured frequency, and <sup>ν</sup>*re f* is a known frequency for a used reference sample for calibrating the output signals such as 3-(trimethylsilyl)- propionic-2,2,3,3-d4 acid (TSP). The chemical shift is calculated relative to the NMR operating frequency to maintain the same chemical shift for each compound independent of the NMR operating frequency, leading to an independent eld measure.

### **2.2.2 Spin-Spin coupling (***J***-coupling)**

The NMR spectra consists of singlets peaks with chemical shifts in between corresponding to the dierent protons. On the other hand, within a particular molecule, the spin-spin coupling aects the resulting spectrum. There are mainly two dierent spin-spin coupling mechanisms: the direct dipole-dipole coupling and the indirect dipole–dipole coupling. The rapid tumbling

Figure 2.10: The electrons shield the nucleus from the external magnetic eld. The nucleus on the left experience a stronger magnetic eld compared to the nucleus on the right due to the lower electron density.

Figure 2.11: NMR spectra of methyl chloride (blue) and methyl uoride (red). The dierent chemical shifts (δ) are due to the dierent electron densities around the chlorine and uorine nuclei, which shield the external magnetic eld dierently.

of the molecule suppresses the direct dipole-dipole coupling.

The indirect dipole-dipole coupling, also known as *J*-coupling, is the coupling of the nuclear spins through the assistance of the electrons resulting in the observed multiplet structures. The magnetic eld generated by a spinning proton magnetizes the electrons, aecting other neighboring protons; therefore, it's known as indirect coupling. Figure 2.12 shows dierent conditions for the *J*-coupling. For example, in a molecule containing one <sup>1</sup>H and one <sup>13</sup>C, <sup>1</sup>H can adopt one of the two possible alignments (aligned with or opposed to *B*0). Consequently, the NMR spectrum of <sup>13</sup>C in *CH* group is split into two peaks of equal intensity due to the proton's eect. In case <sup>13</sup>C is coupled to two <sup>1</sup>H in a *CH*<sup>2</sup> group, then the peak is split twice, one for each bond. Each of two hydrogen protons has two possible alignments with the *B*<sup>0</sup> eld resulting in four possible alignments combinations. The combinations result in three magnetic eects. Due to the ratio between the alignments combinations, the peaks' intensities have a ratio of 1 : 2 : 1. In case <sup>13</sup>C is coupled to three <sup>1</sup>H in a *CH*<sup>3</sup> group, the peak is split into a quartet pattern with an intensity ratio of 1 : 3 : 3 : 1. The frequency separation between the peaks resulted from the J-coupling is specied in Hz. For example, the *J*-coupling frequency separation between the <sup>1</sup>H and <sup>13</sup>C spins in the ethanol's *CH*<sup>3</sup> group is <sup>124</sup>.9 Hz.

## **2.3 Magnetic Resonance Imaging (MRI)**

Magnetic Resonance Imaging (MRI) is another unique application based on the MR principle. MRI is a medical imaging technique that is used to create images of the internal organs without interventions. MRI is widely used to diagnose tumors, internal bleeding, developmental anomalies, and other internal disorders. Another medical technique for internal imaging organs is computerized tomography (CT) which is based on sending X-rays through the body and analyzing the receptor waves, then the physical location is reconstructed numerically. CT

Figure 2.12: Resulted multiplet structure due to the indirect dipole-dipole coupling (*J*-coupling). A <sup>13</sup>C doublet, triplet, quartet resulted from coupling to one, two, or three <sup>1</sup>H protons, respectively.

suers from involving harmful radiation and, in most cases, it demands contrast agents.

In contrast to CT, MRI doesn't include radiation exposure. The location in 3D space is reconstructed by using magnetic elds to manipulate the atoms' magnetic properties based on particular frequency and phase. The pure MR output signal shows the net magnetization and doesn't provide information about the spatial distribution. Hence, the keyword for MRI is the spatial encoding by using gradient coils to encode the MR FID signal, as shown in Fig. 2.13.

### **2.3.1 MRI spatial encoding**

The three linear gradient magnetic eld coils shown in Fig. 2.13b are used for encoding the magnetic eld through the three primary spatial directions by performing slice selection (SS), frequency encoding (FE), and phase encoding (PE). The part that needs to be imaged is in 3D shape, so in the beginning, the target part should be divided into slices, then processing each slice in 2D. Slice selection is based on the fact that only the exciting spins could generate nuclear magnetic signals. Slicing is done by applying linear-gradient (*GZ*) in the z-direction with a proper excitation prole to achieve the desired slicing, as shown in Fig. 2.14. Simultaneously, an RF excitation pulse is applied with an appropriate prole, corresponding to a rectangular waveform in the frequency domain and a Sinc waveform in the time domain, with excitation bandwidth ∆*fexc*. The slice thickness (∆*Z*) is proportional to excitation bandwidth and slope of the linear-gradient, and it represents the depth of the voxel:

$$B\_Z(z) = B\_0 + G\_Z Z,\tag{2.17}$$

$$
\Delta Z = \frac{\Delta f\_{\rm exc}}{\gamma G\_{\rm Z}}.\tag{2.18}
$$

Figure 2.13: The MR free induction decay (FID) signals (a) without and (b) with spatial encoding. Encoding the FID is done by employing gradient coils [top right] for creating gradient magnetic elds in three spatial directions; X, Y, and Z.

The slice thickness is determined by the gradient eld's steepness in the z-direction and the RF pulse bandwidth, as shown in Fig. 2.15. The slice thickness can be reduced by increasing the *G<sup>Z</sup>* gradient steepness or narrowing the Rf pulse's bandwidth.

Subsequently, another spatial dimension is encoded using phase coding (PE) to dierentiate the pixel within the desired slice. A linear-gradient *G<sup>P</sup>* is applied in the Y-direction for a short

Figure 2.14: Slice selection (SS) is done by applying linear-gradient eld (*GZ*) in the z-direction and RF excitation pulse with a prole corresponding to a rectangular waveform in the frequency domain represented by a Sinc waveform in the time domain.

time between the excitation and acquisition stages. As a result of this gradient, each pixel's phase within the slice changes with respect to its position on the Y-axis. The eld of view (*FOVP*), which represents the size of the area that the phase encoding covers, is proportional to the phase gradient and the number of points *N* with respect to the total duration *TP*,

$$FOV\_P = \frac{N}{\gamma G\_P T\_P} = \frac{1}{\gamma G\_P \pi},\tag{2.19}$$

Where τ is the pulse duration corresponding to the total duration divided by the number of points.

Lastly, during the readout stage, the third spatial dimension is considered for frequency encoding (*FE*) by applying a linear gradient *G<sup>f</sup>* in the X-direction. Similarly, its *FOV<sup>f</sup>* is proportional to the frequency gradient and digitized bandwidth ∆*f* , which is the receiver bandwidth ∆*facq* divided by total numbers of points *M*.

$$FOV\_f = \frac{\Delta f\_{acq}}{\gamma G\_f M} = \frac{\Delta f}{\gamma G\_f}.\tag{2.20}$$

Consequently, in the end, spatial information for each voxel within the target part is available to construct the image. The combination of the RF excitation pulses with the three dierent gradient magnetic elds to generate the MR image is known as the MR imaging sequence. Each imaging sequence has a dierent order of applying the slice selection (SS), the phase encoding (PE), and the frequency encoding (FE) with respect to the RF excitation signal. There are various MR imaging sequences to acquire voxels' information of a target part. Two of the standard MRI sequences are the spin-echo and the gradient-echo sequences. The timing diagram of both MRI sequences is shown in Fig. 2.16.

In a typical spin-echo sequence, a 90° excitation RF pulse is applied simultaneously with the slice selecting gradient in the z-direction. Consequently, the desired slice is selected, and its spins are excited. After that, a dened phase encoding pulse is applied in Y-direction. Another

Figure 2.15: Eect of changing the gradient's strength in the z-direction and the RF pulse bandwidth on the slice thickness. **(a)** The strength of the gradient eld in the z-direction is inversely proportional to the slice thickness. On the other hand, **(b)** The RF pulse bandwidth is directly proportional to the slice thickness.

Figure 2.16: The MRI sequence timing diagrams of a standard spin-echo imaging sequence and standard gradientecho imaging sequence.

180° excitation RF pulse is then applied along with the z-direction gradient to enhance the magnetic moments after dephasing to increase the net magnetization before recording the MR signals. Applying the 180° RF pulse is the main drawback of this imaging sequence because of the high RF power demand. Finally, the frequency encoding pulse is applied while the MR echo signal is recorded. These steps are repeated N times based on phase encoding steps. The time required to acquire one echo signal is called the echo time (TE) and is dened by the time between the centers of the excitation pulse and the echo signal. The time between two consecutive excitation pulses for the various phase encoding steps is known as repetition time (TR).

The gradient-echo MRI sequence is almost similar to the spin-echo sequence without the 180° excitation RF pulse. At rst, a 90° excitation pulse is applied along with the slice selection gradient as in the spin-echo imaging sequence. Simultaneously, with applying the phase encoding pulse, a negative frequency encoding pulse is applied to compensate for the expected magnetic moment's dephasing during the readout gradient. This negative frequency encoding technique is used to refocus the magnetic moment instead of using the 180° excitation RF pulse.

### **2.3.2 K-space matrix and image computation**

For obtaining a computational image, all the acquired encoded MR signals are recorded in a matrix known as the k-space matrix with size (*M* ×*N*). The k-space matrix represents an array of spatial frequencies. The MR signal (or FID) is sampled and lled into the k-space matrix during the data acquisition. There are various lling schemes, and one of the standard schemes is the sequential acquisition or Cartesian lling, where the k-space points are determined as follows:

1. In the case of the frequency encoding (in the X-direction),

$$k\_f = m\Delta k\_f = m\frac{1}{FOV\_f},\tag{2.21}$$

where *m* ∈ 1,2,...,*M*. Substituting by Equation 2.20 in Equation 2.21 and ∆*f* = 1/∆*t* where ∆*t* is the period of the sampling pulse,

$$\text{h.f.}\,k\_f = m\gamma G\_f \Delta t.\tag{2.22}$$

2. Similarly, in the case of the phase encoding (in the y-direction) and by using Equation 2.19,

$$k\_P = n\Delta k\_P = n\gamma \Delta G\_P \tau.\tag{2.23}$$

After lling the k-space matrix, a 2D discrete Fourier transform is performed to the matrix to convert the spatial frequencies (*k <sup>f</sup>* , *kP*) to the positions (X,Y) representing spin density distribution to compute the nal image, as shown in Fig. 2.17

## **2.4 Magnetic resonance hardware**

The NMR spectrometer and the MRI scanner have the same hardware units except for the gradient coils, present only in the MRI scanner. Nevertheless, some NMR experiments might require gradient elds,i.g., voxel-based or Self-Diusion (SD) NMR spectroscopy. Figure 2.18 show the main building unit of MRI scanners which consists of;


Figure 2.17: Obtaining an MRI image by applying the three axes gradient elds. **(a)** A typical MRI spin-echo pulse sequence for encoding the magnetic eld. **(b)** First, slice selection (SS) is achieved by applying a gradient eld in the z-direction and an RF pulse, then phase encoding (PE) and frequency encoding (FE) are applied to encode the selected slice. PE is achieved by applying gradient eld in the Y-direction, leading to phase change of the magnetic moments with respect to their Y-axis position. FE is achieved by applying a gradient eld in the X-direction, magnetic moments are subjected to a dierent frequency with respect to their X-axis position. **(c)** The sampled MR echo signals are used to ll the k-space matrix . **(d)** Finally, the MRI image is acquired by applying 2D Fourier transform to the k-space matrix.

### **2.4.1 Static magnet and shim eld**

The magnet is responsible for producing the static *B*<sup>0</sup> eld for polarizing the sample by aligning the magnetic moments of the sample's spinning nuclei. For human MRI applications, hospitals usually have standard MRI scanners with <sup>1</sup>.<sup>5</sup> or 3 T. The approved limit for the human MRI scanner is 7 T. Researchers are trying to push the human MRI limits for higher elds where a <sup>10</sup>.5 T magnet was used for human imaging [25]. For an NMR spectroscopy, superconducting magnets are commercially available that can provide highly homogeneous magnetic elds of up to <sup>25</sup>.9 T [26].

The static magnetic eld needs to be highly homogeneous within 1 ppb (part-per-billion) for high-resolution NMR spectroscopy and to prevent geometrical distortions in the MR imaging. Consequently, shim coils are used to generate spherical harmonic magnetic elds, which are superimposed to the main magnet *B*<sup>0</sup> static eld for inhomogeneity compensation. The inhomogeneities result from internal ferromagnetic materials, sample permeability, probe material,

Figure 2.18: Overview of the main hardware components of a typical magnetic resonance imaging (MRI) scanner.

and imperfection of the main magnet.

### **2.4.2 Gradient coils**

The gradient coils are used for performing spatial encoding. There are three gradient coils aligned in the three space directions X, Y, and Z. For each gradient coil, linear gradients are produced based on two variable spatial encoding parameters; i) The rst parameter corresponds to the gradient strength at the position Tm−<sup>1</sup> , ii) The second parameter is the gradient slope which represents the eld's change with respect to time ms. The gradient coils are connected to electronics circuits consisting of current ampliers, gradient pulse programmers, and digital to analog converters (DAC). The user controls the pulses' width, amplitude, and shape using the gradient pulse programmer through a computer user interface. The signals are then converted to analog data using DAC. Finally, the current amplier amplies the pulse signals to drive the gradient coils.

### **2.4.3 Radio frequency coils**

The MR (FID) signal is generated by changing the orientation of the net magnetization vector by a certain ip angle. This excitation procedure is performed using an alternating RF eld generated by an RF coil. In general, Two RF coils are required; i) the rst coil is used for excitation by producing the RF magnetic eld (*B*1). The generated magnetic eld can be manipulated to establish a particular ip angel,e.g., 180° ip angles in the spin-echo imaging sequence, ii) The second the RF coil is employed to acquires the MR signal (FID) during the relaxation process. Two separate RF coils could be used as transmitter and receiver, or one RF coil could be used as a transceiver based on the reciprocity principle.

The RF coils are attached to an electronic circuit consisting of a pulse programmer, an RF signal source, and power amplier. The RF coil should be tuned at the corresponding Larmor frequency to excite the spins or acquire the MR signal either as transmitter or receiver. As a transmitter, the pulse duration (τ) can be adjusted through the pulse programmer via a fast switch called a pulse gate. The adjusted pulse duration corresponds to a certain ip angle of the net magnetization vector based on the application. As a receiver, the coil should be placed near the sample, where the released energy from the sample during the relaxation induces an electromotive force in the RF coil.

For an RF coil to be utilized for MR measurement it requires an adjusting interface circuit called matching and tuning circuit. The matching procedure is performed to adjust the RF coils' internal impedance to match the connecting cable's impedance to avoid losses resulting from power reections. The tuning procedure is essential to tune the resonance frequency of the RF coil to the Larmor frequency to excite the spins during the excitation process and pick up the MR signal during the relaxation process. The standard matching and tuning circuit consists of a variable tuning capacitor (trimmer) attached in parallel with the RF coil. In contrast, a variable matching capacitor is shunted in series with the coil, as shown in Fig. 2.19a. The equivalent electrical circuit of an RF coil consists of inductance *LCoil*, intrinsic resistance *RCoil*, and parasitic internal capacitance *CCoil*.

The static magnetic eld *B*<sup>0</sup> has a much stronger eld than the generated RF magnetic eld *B*<sup>1</sup> by a factor of 10<sup>6</sup> . Hence, the RF coil should resonate at Larmor frequency to minimize the required energy to excite the spins and produce the ip angle. For an RF coil to excite the spins, the generated energy should cover the Zeeman states' energy dierence.

The upper limit of the resonance frequency of an RF coil is dened by its internal components. For any RF coils, the resonance frequency (ω*r*) is proportional to the coil's inductance and capacitance (ω*<sup>r</sup>* = 1/ √ *LCoil CCoil*). Hence, for utilizing an RF coil in MR measurements, it should be designed to have a self resonance frequency (SRF) higher than the Larmor frequency. Consequently, the coil's resonance frequency can be shifted down toward the Larmor frequency by increasing either the inductance or equivalent capacitance. For MR applications, the equivalent capacitance is increased by adding the tuning capacitor (*C<sup>T</sup>* ) in parallel to the coil. The resonator's equivalent capacitance is the summation of the coil's internal capacitance and the tuning capacitance (*CTotal* = *CCoil* +*C<sup>T</sup>* ). Therefore, the resonance frequency of the MR resonator could be adjusted by increasing the tuning capacitor to lower the frequency till it is tuned to the Larmor frequency as follow:

$$\alpha\_{\mathcal{V}} = \alpha\_0 = \frac{1}{\sqrt{L\_{Coll}(C\_{Coll} + C\_T)}}.\tag{2.24}$$

Figure 2.19: Typical matching and tuning network circuit and simulated behavior. **(a)** Schematic of the equivalent circuit of a non-ideal coil attached to matching and tuning capacitors. **(b)** and **(c)** Simulation results using Advanced Design System (ADS) software for the equivalent circuit in (a). The modeled coil (*LCoil* <sup>=</sup> 30 nH, *<sup>R</sup>Coil* <sup>=</sup> <sup>2</sup>Ω) was tuned to a resonance frequency of 500 MHz using *<sup>C</sup><sup>T</sup>* <sup>=</sup> <sup>2</sup>.71 pF and matched to <sup>50</sup> <sup>Ω</sup> using *<sup>C</sup><sup>M</sup>* <sup>=</sup> <sup>0</sup>.68 pF. (b) the results of sweeping *<sup>C</sup><sup>T</sup>* with keeping *<sup>C</sup><sup>M</sup>* constant. (c) The results of sweeping *<sup>C</sup><sup>M</sup>* with keeping *C<sup>T</sup>* constant.

Due to the non-ideality of the RF coil, it has an internal resistance modeled in series with its inductance resulting in a complex impedance of the coil (*ZCoil*), shown in Equation 2.25. The coil's impedance is the summation of the coil resistance *RCoil* and the equivalent reactance (*XCoil*) of the inductance and capacitance. For connecting two circuits, such as the RF coil and the power amplier, their impedance should be matched to eciently transfer signals and avoid power reections. The internal amplier impedance and the resistance of the connecting coaxial cables are 50 Ω, so the coil impedance should also be matched to 50 Ω at the resonance frequency (*Zcoil* = 50Ω and *XCoil* = 0). Adding a matching capacitor (*CM*) in series with the Coils gives control over the coil impedance's reactance. Hence, *C<sup>T</sup>* is used to tune the coil to the Larmor frequency and set the real part of the coil's impedance to 50 Ω, while *C<sup>M</sup>* is used to eliminate the residual imaginary part, using Equation 2.26.

$$Z\_{coll} = \frac{\frac{1}{j\rho oC\_{Coll}} (R\_{coll} + j\rho oL\_{Col})}{\frac{1}{j\rho oC\_{Col}} + (R\_{coll} + j\rho oL\_{Col})},\tag{2.25}$$

$$\Re[\frac{Z\_{Coll}}{Z\_{Coll} + \frac{1}{jaoCr}}] = \Im 0 \quad , \quad \Im[\frac{Z\_{Coll}}{Z\_{Coll} + \frac{1}{jaoCr}} + \frac{1}{jaoC\_M}] = 0. \tag{2.26}$$

A numerical simulation was done using Advanced Design System (ADS) software for the matching and tuning circuit shown in Fig. 2.26a. The modeled inductor has an inductance of *<sup>L</sup>Coil* <sup>=</sup> 30 nH and an intrinsic resistance of *<sup>R</sup>Coil* <sup>=</sup> <sup>2</sup>Ω. It was tuned to a resonance frequency of <sup>ω</sup><sup>0</sup> <sup>=</sup> 500MHz using *<sup>C</sup><sup>T</sup>* <sup>=</sup> <sup>2</sup>.71 pF and matched to <sup>50</sup> <sup>Ω</sup> using *<sup>C</sup><sup>M</sup>* <sup>=</sup> <sup>0</sup>.68 pF. The tuning capacitor (*C<sup>T</sup>* ) was then swept from <sup>1</sup>.71 pF to <sup>3</sup>.71 pF with a step of <sup>0</sup>.5 pF with keeping the matching capacitor at a constant value of <sup>0</sup>.68 pF. The eect of changing *<sup>C</sup><sup>T</sup>* on the circuit impedance and reection coecient (*S*11) is presented in Fig. 2.19b. *C<sup>M</sup>* was then swept from <sup>0</sup>.08 pF to <sup>1</sup>.28 pF with a step of <sup>0</sup>.3 pF with keeping *<sup>C</sup><sup>T</sup>* at a constant value of <sup>2</sup>.71 pF. The eect of changing*C<sup>M</sup>* on the circuit impedance and reection coecient (*S*11) is presented in Fig. 2.19c. Changing*C<sup>T</sup>* aects both the real and imaginary parts of the circuit's impedance, while *C<sup>M</sup>* only aects the impedance's imaginary part. Therefore, *C<sup>T</sup>* is adjusted rst to tune the circuit, and *C<sup>M</sup>* is then used to eliminate the imaginary term.

The matching and tuning circuit conguration shown in Fig.2.19a is known as the unbalanced matching scheme. The balanced matching scheme can be achieved using two matching capacitors instead of one capacitor, one at each terminal of the RF coil, as shown in Fig. 2.20a. The balanced scheme reduces the coil to ground losses as long as the voltage amplitude is the same at the two terminals of the circuit with inverted phase. In NMR spectroscopy, the RF coil is placed onto the probe-head inside the magnet where it is dicult to reach. So the matching and tuning capacitors (*CT*,*Ph* and 2×*CM*,*Ph*) could be used to pre-adjust the RF coil before connecting it to the NMR probe. Additionally, two more variable capacitors (*CT*,*Pb* and *CM*,*Pb*) are built-in the probe and can be mechanically controlled from outside the magnet using rods for ne-tuning, as shown in Fig.2.20a.

The probe's trimmers (*CT*,*Pb* and *<sup>C</sup>MP<sup>b</sup>*) are connected in series with the RF resonator. Hence, both trimmers will provide the matching capacitor's functionality, which does not oer full control of the RF coil. Therefore, a shunt capacitor (*CP*) is added to set *CT*,*Pb* in parallel with

Figure 2.20: Schematic circuit and simulated behavior of the balanced matching and tuning network attached to a coil on a typical NMR probe. **(a)** Schematic of the equivalent circuit of an RF coil on an NMR probe. The non-ideal coil is connected to a balanced matching and tuning circuit on the probe-head (*CT*,*Ph* and 2×*CM*,*Ph*) for pre-tuning and the probe's trimmers (*CT*,*Pb* and*CM*,*Pb*) for ne-tuning. **(b)-(e)** Simulation results using Advanced Design System (ADS) software for the equivalent circuit in (a). The modeled coil (*LCoil* <sup>=</sup> 30 nH, *<sup>R</sup>Coil* <sup>=</sup> <sup>2</sup>Ω) was tuned to a resonance frequency of 500 MHz and matched to <sup>50</sup> <sup>Ω</sup> using *<sup>C</sup>T*,*ph* <sup>=</sup> 2 pF, *<sup>C</sup>M*,*Ph* <sup>=</sup> 10 pF, and *<sup>C</sup>T*,*Pb* <sup>=</sup>*CM*,*Pb* <sup>=</sup> <sup>3</sup>.5 pF. The simulation results show the change in the circuit equivalent impedance by sweeping: (b) *CT*,*Pb* and *CM*,*Pb* with *C<sup>P</sup>* as an open circuit, (c) *C<sup>P</sup>* while keeping the rest of the capacitors constant, (d) sweeping *<sup>C</sup>T*,*Pb* with *<sup>C</sup><sup>P</sup>* <sup>=</sup> 1 pF, and (e) *<sup>C</sup>M*,*Pb* with *<sup>C</sup><sup>P</sup>* <sup>=</sup> 1 pF.

the RF resonator to give it the tuning functionality. Based on the lumped-model presented in Fig. 2.20a, the equivalent impedance (*ZTotal*) was driven with and without (*CP*). In Equations 2.27 and 2.28, the terms highlighted in blue set the tuning frequency, and the terms highlighted in red aect only the imaginary part of the impedance resulting in a matching behavior. Equations 2.27 and 2.28 reveal the role of *C<sup>P</sup>* in providing *CT*,*Pb* the tuning functionality.

$$Z\_{Total\\_\text{\textquotedblleft}D\text{\textquotedblright}C\_P\text{\textquotedblright}} = \frac{Z\_{coll} \cdot 1/j \text{joC}\_{T,Ph}}{Z\_{coll} + 1/j \text{joC}\_{T,Ph}} + \frac{1}{jjo} (\frac{2}{C\_{M,Ph}} + \frac{1}{C\_{M,Pb}} + \frac{1}{C\_{T,Pb}})\,,\tag{2.27}$$

$$Z\_{Total\\_\text{\textquotedblleft}Wb} \, C\_P \, = \left[ (\frac{Z\_{coll} \cdot 1/j \text{joC}\_{T,Ph}}{Z\_{coll} + 1/j \text{joC}\_{T,Ph}}) + (\frac{1}{jjo} (\frac{2}{C\_{M,Ph}} + \frac{1}{C\_{T,Pb}})) \right] \, \parallel \, C\_P + \frac{1}{jjo \text{C}\_M} \,\tag{2.28}$$

A numerical simulation was done using ADS software for the balanced matching and tuning circuit shown in Fig. 2.20a. The modeled inductor has an inductance of *<sup>L</sup>Coil* <sup>=</sup> 50 nH, an intrinsic resistance of *RCoil* = 2Ω, and the internal capacitance is neglected *CCoil* = 0. The coil is tuned to a resonance frequency of 500 MHz with selecting the capacitance values of*CT*,*Ph* <sup>=</sup> 2 pF, *<sup>C</sup>M*,*Pb* <sup>=</sup> 10 pF, and *<sup>C</sup>T*,*Pb* <sup>=</sup> *<sup>C</sup>M*,*Pb* <sup>=</sup> <sup>3</sup>.5 pF. First, each of *<sup>C</sup>T*,*Pb* and *<sup>C</sup>M*,*Pb* were swept separately from 1 pF to 4 pF with a step 1 pF, with holding the other capacitor at a constant value of <sup>3</sup>.5 pF and without *<sup>C</sup>P*. The result is presented in Fig. 2.20b, which demonstrates that both *CT*,*Pb* and *CM*,*Pb* have a matching eect by changing only the imaginary part of the impedance. Next, *<sup>C</sup><sup>P</sup>* is added, and its value was swept from <sup>0</sup>.1 pF to 4 pF with a step 1 pF. The eect of sweeping *<sup>C</sup><sup>P</sup>* on the impedance is plotted in Fig. 2.20c. With *<sup>C</sup><sup>P</sup>* <sup>=</sup> 1 pF, each of *<sup>C</sup>T*,*Ph* and *<sup>C</sup>M*,*Ph* were swept again (from 1 pF to 4 pF with a step 1 pF), with holding the other capacitor at a constant value of <sup>3</sup>.5 pF. The eect of sweeping*CT*,*Pb* with the presence of*C<sup>P</sup>* on the impedance is shown in Fig. 2.20d. *C<sup>P</sup>* sets *CT*,*Pb* in parallel with the RF resonator, aecting both the real and imaginary parts of the impedance giving it the tuning functionality. On the other hand, since *CM*,*Pb* stays in series with the RF resonator, it keeps its matching eect as shown in Fig 2.20e.

### **2.4.4 MR characterization**

The quality of the acquired MR signal and the hardware components' functionality can be evaluated by devaluing the Signal-to-Noise (SNR) Ratio of the resulted MR data. The SNR is a common concept in signal processing. According to conventional signal theory, SNR is dened as the square root of the ratio between the signal intensity and the noise intensity. However, in the NMR community, SNR is the fraction between the signal amplitude of the NMR spectrum and the root-mean-square of the noise as described by Hoult and Richards [27]. A reformulated

closed-form version of the SNR was introduced by Peck et al. [28] as follow:

$$SNR = \frac{\text{peak signal}}{\text{RMS Noise}} = \frac{k\_0 (B\_1/i) V\_{\text{sample}} N \gamma \hbar^2 I (I+1) \frac{a a\_0^2}{3\sqrt{2} k\_B T}}{\sqrt{4 k\_B T R\_{\text{noise}} \Delta f}},\tag{2.29}$$

where *k*<sup>0</sup> is a scaling factor related to the homogeneity of the alternating *B*<sup>1</sup> eld, *i* is the current owing through the RF coil, *Rnoise* is the AC electrical resistance of the coil comprising the proximity and skin eects, *Vsample* is the active sample volume, *N* is the spin density, and ∆*f* is the spectral bandwidth.

Equation 2.29 indicates that besides the temperature and the *B*<sup>0</sup> eld intensity, the RF coil is the fundamental unit aecting the SNR since it has to pick up the weak MR signal. Hence, the design and optimization of the RF coil should be done with great attention to reduce the noise it generates, especially while receiving the MR signal. The SNR denominator represents the MR coil's noise which is mainly thermal noise developed in the coil and sample. This noise is commonly known as Johnson noise [29] and is described by Nyquist formula:

$$V\_{noise} = \sqrt{4k\_B T R\_{noise} \Delta f}.\tag{2.30}$$

The main parameters to characterize the MR coil performance are the *B*1 homogeneity, probe eciency (η*P*), and Quality factor (*QP*) to enhance its performance and thus the SNR. The mathematical and practical methods for evaluating these parameters are summarized in Table 2.1.

The lling factor (α*F*) is another critical parameter that dierentiates one MR coil topology from another. The lling factor is the ratio between the sample volume observed by the *B*<sup>1</sup> eld to the whole *B*<sup>1</sup> eld volume. α*<sup>F</sup>* is evaluated by the weighted moment,

$$d\mu\_{\overline{\mathbb{F}}} = \frac{\int B\_1^2 \rho(r)dV}{\int B\_1^2 dV},\tag{2.36}$$

where the parameter ρ(*r*) is unity in the locations where the sample is aected by the *B*<sup>1</sup> eld and zero elsewhere. For example, a solenoid coil with the internal volume lled with the sample has a lling factor of α*<sup>F</sup>* = 1/2; other typologies usually have a lling factor much lower than that. Selecting a proper lling factor is a matter of optimization as increasing the lling factor increases the NMR signal. On the other hand, the sample should also be kept far from the coil's material where the *B*<sup>0</sup> eld is distorted due to the susceptibility mismatch, which leads to low spectral resolution. The NMR spectral line-shape is considered a benchmark of *B*<sup>0</sup> distortion, with signals showing a sharp narrow Lorentzian reecting *B*<sup>0</sup> eld with high homogeneity. Increasing the *B*<sup>0</sup> eld strength or reducing the temperature boosts the spin polarization, enhancing the signal amplitude and SNR. The minimum detectable quantity of a sample for a given system is dened as the limited detection (LOD), which is inversely pro-


Table 2.1: Methods for evaluating the MR coil's characterization parameters

<sup>a</sup> The nutation spectrum is the magnetization progression with raising the excitation pulse length (τ) at a constant supplied power (*PCoil*).

portional to the SNR. The normalized limit of detection (nLOD) is dened for an SNR of 3 and is given by [30], √

$$nLOD\_{\text{d0}} = \frac{3n\_s \sqrt{\Delta t}}{SMR\_{\text{d0}}},\tag{2.37}$$

Where ∆*t* is the single scan acquisition time, *n<sup>s</sup>* is the number of spins presented in the sample, and *SNR*<sup>ω</sup> is the calculated SNR in the frequency domain.

# Part I.

# Micro-Fabrication of Flexible Electronics

# **3 Flexible electronics fabrication**

**F** lexible electronics have drawn signicant attention since they rst appeared in the 1960s because of their magnicent potential in several emerging applications [31]. The global market size of exible electronics was valued at \$23.64 billion in 2019 and is expected to rise rapidly to reach a value of \$42.48 billion by 2027 [32]. Flexible electronics have found their way in many devices such as electronic components [33], displays [34], solar cells [35], and radio frequency identication devices (RFID) [36]. Flexible electronics provide a signicant push for developing wide range of technologies as wearable electronics [37], health monitoring devices [38], and intelligent textile [39] as they are more ecient than conventional electronics.

The rst part of the thesis (Chapters 3 and 4) discusses the dierent fabrication technologies developed for building exible electronic devices that are used in the second part (MRI application) and the third part (NMR application) of the thesis. A exible electronic device is generally built out of a thin passive exible substrate with active electronic components on one or both sides of a substrate, enabling rolls, bends, folds, and stretches without hindering its performance.

The micro MR-detector consists mainly of inductors and capacitors. The dierent inductors' typologies can be achieved by structuring a conductive layer on a dielectric substrate. On the other hand, capacitors require structuring a dielectric layer on top of the conductive layer to realize parallel plate capacitors. Besides these two passive electronic components, fabricating diodes was needed for the MRI application. Hence, this chapter is divided into three sections. Section 3.1 discusses the dierent technologies for depositing a conductive seed-layer on a exible substrate with focusing on the inkjet-printing technique. Section 3.2 introduces a novel insulator ink compatible with the inkjet-printing technique to realize a high dielectric high quality factor exible capacitor. Section 3.3 explains a fabrication process for building exible diodes using inkjet-printing.

## **3.1 Conductor seed-layer patterning**

Substrate selection is an important step in the fabrication of exible electronic devices [40]. The substrate's mechanical properties should meet the application specications regarding exibility, and it should be rigid enough to sustain mechanical stresses during the fabrication process. The electrical properties are essential to restrain leakage currents in electrical devices. The thermal properties play an essential role during the device operation for controlling the generated heat dissipation. Moreover, the substrate's characteristics, such as heat resistance, chemical resistance, and surface smoothness, determine the substrate's compliance with dierent manufacturing processes.

Several material categories have been used as a base substrate for exible electronic devices; the most common ones are papers, textiles, and polymers. *Paper* substrates are very common in the eld of exible electronics [41]. They are mainly made out of compressed cellulose bers into thin sheets. The cellulose mesh network and type of ller material dene the paper's properties, structure, and thickness [42]. Paper-based substrates have a set of unique properties such as mechanical strength, exibility, and customizability. However, paper-based substrates suer from poor thermal performance due to the large pore size and the presence of moisture in the paper [43]. Thus, paper substrates are not compatible with fabrication processes that include high curing temperature, i.g., limiting the deposition processes that can be used on paper. *Textile* substrates [39] have been recently considered for use in exible electronic devices, with most applications related to wearable electronics [44]. Due to textiles' exibility and stretch-ability, most of the coating processes used with them are rapid prototyping techniques, such as screen printing, dip coating, inkjet-printing, or hand stitching. *Polymer* substrates are one of the most attractive choices of materials for exible electronics [45]. Polymer-based substrates can be divided into: 1) semi-crystalline polymers such as polyetheretherketone (PEEK), polyethylene terephthalate (PET), and polyethylene naphthalene (PEN); and 2) amorphous polymers such as polycarbonate (PC), polyethersulphone (PES), and polyimide (PI). Due to the dierent structural organization levels and dierent chemical compositions, these polymer substrates feature a wide range of mechanical, chemical, and thermal properties. PIs are of the most promising polymer substrates for exible electronic devices due to their excellent mechanical, physical, thermal, and dielectric properties [46].

Through this thesis, Kapton polymer foil has been used as the base substrate to manufacture MRI and NMR exible electronic devices. Kapton has several advantages which makes it a perfect candidate for our applications, such as [47]: 1) thermal stability over a wide range of temperatures (from −269 °C to 400 °C) makes it compatible with most of the micro-fabrication technologies; 2) perfect electrical insulation with good dielectric quality; 3) chemical stability, compatible with most solvents and solutions involved in the micro-fabrication process as well as in the NMR/MRI experiments; and 4) availability as ultra-thin sheets with thickness down to a few micrometers. Moreover, the Kapton polymer was chosen for the NMR/MRI applications due to its magnetic susceptibility, which is close to that of water [48].

Conductive electrodes are the primary building block of the exible electronic devices. Depositing conductive electrodes with a specic structure, morphology, and pattern is an essential step in the electronic systems' fabrication process. In conventional micro-fabrication, conductive materials are deposited by employing physical vapor deposition (PVD) or chemical vapor deposition (CVD), which is commonly used in the semiconductor industry. Patterning the deposited conductive layer is conventional done with the mean of the photolithography process. On the one hand, this process results in uniform conductive tracks of a high degree of material purity with precise resolution and thickness. On the other hand, this microfabrication process involves using expensive instrumentation, which requires a clean-room environment adding to the process cost, complexity, and duration. Therefore, inkjet-printing, an alternative rapid prototyping approach, has been considered for depositing conductive tracks with the desired pattern. Inkjet-printing is done using a bench-top laboratory printer without the need for further special instrumentation, making it a relatively cheap and fast deposition method.

This section briey describes the dierent fabrication processes tested for depositing conductive structures onto exible Kapton lm to produce NMR/MRI compatible sensors. Two categories of fabrication processes have been considered: 1) photolithography processes 3.1.1 that can be categorized to "mask and deposit methods" or "mask and etch method"; and ii) rapid prototyping process 3.1.2 where inkjet-printing is used (mask-free technique). By the end of the section, all processes have been compared with highlighting the processes used further for fabricating the NMR/MRI sensors.

#### **3.1.1 Photolithography**

Photolithography is the key fabrication technology in the semiconductor industry and mass production of silicon-based devices. Photolithography allows fabricating devices with resolutions down to several nanometers and is the leading operator behind the miniaturization trend known as Moore's law. Photolithography is an established micro-fabrication method that uses mostly UV-light for micro-structuring, hence also known as UV-lithography [49]. In its simplest form, a layer of photosensitive chemical is deposited on a base substrate, and a photon beam is used to transfer a pattern written on a photo-mask to the photosensitive layer [50]. A specic series of chemical treatments are then applied either for depositing another material in the exposure pattern onto the base substrate or etching the photo-mask pattern into the based substrate. Four dierent variants of the photolithography process have been tested to deposit structured conductive tracks on the Kapton foil: 1) lift-o process; 2) molding process ; 3) liquid photo-sensitive polyimide; and 4) wet etching process. The four tested approaches of photolithography are discussed below.

#### **Lift-o micro-technology**

Figure 3.1 summarizes the fabrication ow for the lift-o photolithography process. A 25 µm Kapton foil was utilized as the base substrate for the photolithography process. With a thickness of 25 µm, the base lm is exible enough to be rolled around the slender capillaries (< 2mm), satisfying the NMR/MRI applications, and simultaneously it can be mechanically handled through the fabrication process. To facilitate the Kapton foil handling during the lithography fabrication process, it was cut into the shape of a 100 mm wafer using an infrared laser cutter (VLS3.50, Universal Laser System). Then, it is laminated to a silicon wafer using a 2 µm ma-N 1420 negative photoresist layer (micro resist technology GmbH) to promote adhesion, spin-coated at 3000 rpm. The wafer with the laminated Kapton lm was soft-backed

at <sup>80</sup> °<sup>C</sup> for 1 min. Tiny air bubbles were observed between the silicon wafer and Kapton foil. However, the binding between the foil and substrate was enough to proceed with the subsequent processing steps. After rinsing the top side of the Kapton foil with acetone and DI water, a 2 µm layer of the negative photoresist ma-N 1420 was spin-coated on top of the foil. The wafer was then soft-baked on a hotplate at <sup>100</sup> °<sup>C</sup> for 2 min. After cooling down naturally, the coated negative photoresist was then structured by UV exposure at a dose of 680 mJ cm<sup>−</sup><sup>2</sup> . The photoresist was developed for 2 min using ma-D 533 s developer. The wafer was then cleaned with DI water and soft-backed at <sup>80</sup> °<sup>C</sup> for 1 min. 20 nm chromium and 60 nm gold layers were deposited on the wafer using the physical vapor deposition process (PVD, Univex 450) at deposition rates of 5Ås−<sup>1</sup> and 8Ås−<sup>1</sup> , respectively. The chromium layer is used to promote the adhesion between the foil and the gold layer. The patterned area on top of the sacricial photoresist was then lifted o with acetone inside an ultrasonic bath for 5 min, followed by rinsing in IPA and DI water. Then the patterned region on top of the sacricial photoresist was lifted o with acetone in an ultrasonic bath, then rinsed with DI water.

The deposited seed-layer (chromium and gold) using PVD results in a thin conductor layer with a thickness below 100 nm, which is not sucient for RF applications. RF applications require a minimum thickness of the conductor; otherwise, the electric power suers from high resistance. The required conductor thickness is calculated according to the operating frequency,i.g., the <sup>1</sup>H frequency at <sup>11</sup>.7 T NMR spectrometer is 500 MHz, thus, the copper tracks must be at least 6 µm thick. Hence, an electroplating process is employed to reach the required conductor height. However, electroplating results in peeling o most of the conductor layer from the Kapton foil. The solution for the electroplating problem is reported in Chapter 4. After electroplating, The lm was nally peeled o from the silicon wafer and rinsed from both sides with acetone and DI water.

The lift-o is considered to be a relatively simple process as no etching is included. No aggressive chemical or plasma treatment is required that could aect the polyimide foil. However,

Figure 3.1: Schematic illustration of the lift-o micro-fabrication process. **(1)** Lamination of Kapton foil on a silicon wafer. **(2)** Patterning of a negative photoresist (ma-N 1420) layer using UV lithography. **(3)** Photoresist development and seed-layer deposition (Cr and Au). **(4)** Lifting o the sacricial photoresist layer. **(5)** Peeling o the Kapton lm from the silicon wafer and rinsing with acetone.

the lift-o process suers from some drawbacks. During the lift-o step, some unwanted parts of the deposited metal remain on the wafer as the resist below them was not removed. Moreover, some of the removed metal redeposited in random locations on the wafer and cannot be etched away easily.

The patterned structures on the wafer must all be connected for the electroplating process, as no common seed-layer connects the structures (i.e., during electroplating, the power source is connected to one point on the wafer where all structures need to be connected). Electroplating results in increasing the width of the tracks due to the absence of a mold.

#### **Molding micro-technology**

Figure 3.2 summarizes the fabrication ow for the molding photolithography process. A 25 µm Kapton foil was utilized as the base substrate for the photolithography process. The same lamination step is repeated as in the lift-o process to facilitate sample handling. The wafer was then soft-backed at <sup>80</sup> °<sup>C</sup> for 1 min. 20 nm chromium and 60 nm gold layers were deposited on the wafer using PVD. A 25 µm layer of the negative photoresist SU-8 3025 (micro resist technology GmbH) was spin-coated at 4000 rpm for 30 s on top of the gold layer to act as the electroplating mold. The wafer was then soft-baked on a hot plate at <sup>95</sup> °<sup>C</sup> for 13 min. After cooling down naturally, the coated negative photoresist was then structured by UV exposure at a dose of 400 mJ cm<sup>−</sup><sup>2</sup> . Then post-exposure bake was done at <sup>95</sup> °<sup>C</sup> for 4 min. The photoresist was developed for 30 min using a PGMEA developer. The wafer was then rinsed by immersion in isopropyl alcohol (IPA) for 30 min and then left to dry for 2 h at room tempera-

Figure 3.2: Schematic illustration of the molding micro-fabrication process. **(1)** Lamination of Kapton foil on a silicon wafer. **(2)** Seed-layer deposition (Cr and Au). **(3)** Patterning of a negative photoresist (SU-8 3025) layer using UV lithography. **(4)** Developing the negative photoresist to create molds for electroplating. **(5)** Electroplating process. **(6)** Stripping the negative photoresist and etching the seed-layer. **(7)** Peeling o the Kapton lm from the silicon wafer and rinsing with acetone.

tures. Electroplating process was then conducted to reach the required conductor thickness. After electroplating, the SU-8 photoresist was stripped using R3T plasma etching. Then Au and Cr etching was done to remove the unused seed-layer, and the lm was nally peeled o from the silicon wafer and rinsed from both sides with acetone and DI water.

The main advantage of the molding process is the presence of molds for the electroplating process. Hence, electroplating results in uniform tracks other than the lift-o process, where plating also occurs on the tracks' edges and sides. However, on the one hand, a stable photoresist is required to establish the mold to survive the electroplating bath. On the other hand, stripping this photoresist after electroplating usually requires aggressive plasma treatment, resulting in burning the polyimide foil.

#### **Photosensitive polyimide**

Other than the result of the studied process, a photosensitive liquid polyimide (LTC9310, FU-JIFILM electronic materials) was used instead of foil polyimide. LTC9300 is a series of negative photoresist polyimide, where LTC stands for "Low Temperature Cure", it can be cured (from liquid to foil) by heating at 200 °C other than standard PIs, which are cured with temperatures > 350°C.

Figure 3.3 summarizes the fabrication ow using liquid photosensitive polyimide. On the top of a silicon wafer, a 60 nm gold layer was deposited using PVD at deposition a rate of <sup>8</sup>Ås−<sup>1</sup> . The gold layer was deposited in a way to keep the wafer border free from gold to promote adhesion between the silicon wafer and the polyimide that will be spin-coated in the next step. No chromium was added between the gold and the silicon wafer to facilitate the nal step of peeling o the polyimide lm with the gold layer.

On the top of the gold layer, a <sup>30</sup> <sup>µ</sup><sup>m</sup> layer of LTC9310 was spin-coated at 1200 rpm for 30 s. After soft-baking at <sup>100</sup> °<sup>C</sup> for 4 min, LTC9310 was structured by UV exposure at a dose of

Figure 3.3: Schematic illustration of the liquid polyimide micro-fabrication process. **(1)** Seed-layer deposition (Au) with keeping the wafer border free to promote photosensitive polyimide adhesion. **(2)** Patterning of a negative photosensitive polyimide (LTC 9310) layer using UV lithography. **(3)** Developing the polyimide photoresist and cure in a vacuum oven. **(4)** Electroplating process and peeling o the cured polyimide lm from the silicon wafer by cutting the border. **(5)** Etching the gold seed-layer from the polyimide lm's backside.

550 mJ cm<sup>−</sup><sup>2</sup> . The wafer was then continuously sprayed by QZ3512 (FUJIFILM electronic materials) for development and sprayed by QZ3501 for rinsing. The wafer was then transferred to a vacuum oven set to <sup>250</sup> °<sup>C</sup> for 2 h, to cure the liquid polyimide into a exible lm with a thickness of around 20 µm. Electroplating process was then conducted to reach the required conductor thickness. Finally, the polyimide was peeled o by cutting the lm around the wafer border. The polyimide lm comes out with the gold layer from the backside, which can be easily etched away.

Using photosensitive liquid polyimide provides control over the polyimide lm's thickness, which can be in a few µm range. Moreover, the structures can be embedded within the polyimide lm, as in the process described here. In Chapter 6, this process was used to realize multilayer structures where vias were created within the polyimide lm.

One of the diculties facing this process is to predict the exact thickness of the nal polyimide lm since after curing, the thickness of the deposit polyimide reduces. This fabrication process's overall yield is not high as curing/electroplating can remove part of the polyimide from the silicon wafer.

#### **Wet etching micro-technology**

Figure 3.4 summarizes the fabrication ow for the wet-etching photolithography process. A 25 µm Kapton lm covered from a single side with a 9 µm copper layer (AkaFlex KCL 2-9/25 HT, Krempel GmbH) was utilized as the base substrate for the photolithography process. The same lamination step is repeated as in the lift-o process except for the adhesive layer between the foil and the silicon wafer where a 3.3 µm AZ4533 photoresist layer (Micro-Chemicals GmbH) was used, spin-coated at 4000 rpm. The wafer was then soft-backed at <sup>95</sup> °<sup>C</sup> for 1 min. No air bubbles were observed between the Kapton foil and the silicon wafer, unlike the ma-N 1420 photoresist. To promote photoresist adhesion on the copper layer, the lm was cleaned by etching the top layer of the copper surface by dipping the wafer into a sodium persulfate

Figure 3.4: Schematic illustration of the wet etching micro-fabrication process. **(1)** Lamination of Kapton foil (covered with a copper layer) on a silicon wafer. **(2)** Patterning of a positive photoresist (AZ 4533) layer using UV lithography. **(3)** Developing the positive photoresist. **(4)** Copper layer wet etching. **(5)** Peeling o the Kapton lm from the silicon wafer and rinsing with acetone and DI water.

solution (Na2S2O8, 240 gL<sup>−</sup><sup>1</sup> ) for 15 s. After the etching process, the wafer was rinsed with DI water and dried with a pressurized nitrogen stream. Directly after the rinsing process, a 3.3 µm layer of the positive photoresist AZ4533 was spin-coated on top of the copper layer to prevent oxidation. The wafer was then soft-baked on a hotplate at <sup>95</sup> °<sup>C</sup> for 1 min. After cooling down naturally, the coated positive photoresist was then structured by UV exposure at a dose of 235 mJ cm<sup>−</sup><sup>2</sup> . The photoresist was developed for 3 min using a KOH-based AZ400K developer, prepared at 1:4 dilution. The wafer was then cleaned with DI water and soft-backed at 115 °C for <sup>2</sup>.5 min. The wafer was then immersed in the sodium persulfate solution for etching the exposed copper layer. The lm was nally peeled o from the silicon wafer and rinsed from both sides with acetone and DI water.

This fabrication process's overall yield is very high (nearly 100%), owing to the standard processes used. Over etching reduces the track width, thus, the sample should be carefully observed during etching.

#### **3.1.2 Inkjet-printing**

Employing an additive manufacturing process in exible electronics fabrication, such as inkjetprinting of conductive polymers on an unstructured xable substrate, plays an essential role in research as well as in industry [51]. The global market size of printed electronics was valued at \$7.8 billion in 2020 and is expected to have a bright future with an annual growth rate of 21.5% to reach a value of \$20.7 billion by 2025 [52]. The global demand for low-cost, energyecient, thin, and exible customize electronics is the main driving power for the printed electronics market's growth. Many versatile electronic components have been realized by printing, ranging from simple components such as resistors, capacitors, and indicators [53] to more complex devices such as diodes, transistors [54, 55], displays [56], batteries [57], and microelectromechanical systems (MEMS) [58]. The inkjet-printing method is compatible with roll to roll process [59, 60], which is used to achieve relatively high production rates.

Inkjet-printing of conductive polymer on polyimide substrates (few microns thickness) enables low-cost, highly exible, lightweight, and yet complex electronic components. Via inkjetprinting, the conductive material can be precisely positioned using computer-aided tools without the need for a mask or subsequent etching processes, reducing the processing time and consumed materials [61–65], making the method both economically and ecologically attractive. The conductive material can be inkjet-printed as nanoparticles (NP) [66, 67] or metalloorganic decomposition (MOD) [68] inks. Silver nanoparticle inks are commonly used with the inkjet-printing method, enabling conductive tracks down to ≈ 30µm in width. A 25 µm wide silver tracks have been successfully inkjet-printed onto an unstructured polyimide substrate [67].

Following printing, the conductive ink requires to be sintered to get rid of the solvent content,

which typically demands high temperatures (<sup>&</sup>gt; <sup>150</sup> °C) for a duration of more than 30 min, limiting the materials that may be used as a substrate. Alternative low-temperature sintering techniques have also been considered [65] but have not yet become mainstream. Since the inks contains a relatively large percentage of solvent that evaporates, inkjet-printed and sintered conductors have thicknesses in the range of hundreds of nanometers. Hence, electroplating is an essential step for RF applications.

Throughout the whole thesis, Fujilm Dimatix DMC-2800 printer was used for printing, loaded with a 10 pL print head-cartridge (DMC-11610, Fujilm Dimatix) lled with silver nanoparticle ink. The used ink was 30–35 wt.% suspension of silver nanoparticles (Sigma-Aldrich®) with diameters<sup>≤</sup> 50 nm dispersed in triethylene glycol monomethyl ether. The silver ink is rated with a surface tension of <sup>35</sup>-40 mNm<sup>−</sup><sup>1</sup> and a viscosity of <sup>10</sup>-18 mPa s at room temperature.

The print head-cartridge posses 16 printing nozzles with an integrated heater. The control pulse supplied for all nozzles has the same waveform, but voltage levels can be controlled for each nozzle individually. The ink should have a viscosity of <sup>10</sup>-12 mPa s and surface tension of <sup>28</sup>-33 mNm<sup>−</sup><sup>1</sup> at jetting temperature to achieve optimum printing performance. For high viscosity inks, the ink printability can be tuned by adjusting viscosity using the integrated heaters, which can heat the ink at the nozzle outlet up to 70 °C.

The dot spacing is set based on the droplet size, where every two successive droplets should half overlap to result in a uniform printed lines. Otherwise, if the dot spacing was small will result in the accumulation of the ink, and in case the dot spacing is wide will result in a wavy line. Figure 3.5 shows separately printed droplets of the silver ink. The average droplet diameter was measured to be 50 µm, therefore, the dot spacing was set to 25 µm, resulting in printable minimum line width of ≈ 60µm. The droplet size can be controlled by optimizing the jetting waveform supplied to the printing nozzles. The printing nozzle contains a piezo-

Figure 3.5: Printed droplets of silver ink. The droplet diameter was measured to be 51 µm, therefore, the dot spacing was set to 25 µm.

Figure 3.6: Inkjet-printing typical jetting waveform. The jetting waveform divided mainly into four consecutive phases: **(1)** Bending of the PZT element outward to draw ink; **(2)** Bending the PZT element inward to eject the ink; **(3)** Reducing the PZT element bending to release the droplet; **(4)** Resetting to the initial position to prepare for pulling the ink for the next droplet.

electric element (PZT) whose deection is controlled by the jetting waveform, which in turn controls the droplet formation. Figure 3.6 demonstrates the typical jetting waveform supplied to a printing nozzle with illustrating a PZT element in operation. The jetting waveform is divided mainly into four consecutive segments (draw, eject, release, reset). The used time and voltage parameters for silver ink printing were *T*<sup>1</sup> = 3.264µs, *T*<sup>2</sup> = 3.84µs, *T*<sup>3</sup> = 3.392µs, *T*<sup>4</sup> = 0.832µs, *L*<sup>1</sup> = 0%, *L*<sup>2</sup> = 100%, *L*3 = 60%, and *L*<sup>4</sup> = 27%.

Figure 3.7a summarizes the inkjet-printing process, demonstrating a printed saddle coil in Fig. 3.7b. During printing, the substrate plate and the printing head were maintained at room temperature. Before printing, the Kapton lm was rinsed with acetone and DI water. After

Figure 3.7: Inkjet-printing fabrication process ow and result. **(a)** Schematic illustration of the inkjet-printing process: (1) Silver ink printing; (2) Sintering the silver ink by heating; (3) Electroplating by a layer of copper to increase the printed track thickness. **(b)** Saddle coil fabricated using inkjet-printing. The coil is shown after (1) Printing and (2) Electroplating. The SEM images of the printed silver and electroplated copper are demonstrated.

printing, the lm with the printed tracks was transferred to a hot plate for drying at 80 °C for 15 min, followed by sintering in an oven at <sup>150</sup> °<sup>C</sup> for 1 h. Subsequently, electroplating was carried out.

### **3.1.3 Micro-fabrication processes comparison**

According to the extensive study of the previously mentioned dierent fabrication process:


Table 3.1 summarizes the main advantages and disadvantages of each of the studied microfabrication processes. The comparison focuses on the observed adhesion strength between the deposited seed-layer and the polyimide lm before electroplating, as it was one of the main selecting criteria for the fabrication process.

## **3.2 Metal-insulator-metal (MIM) exible capacitor**

#### **3.2.1 Introduction**

Capacitors are one of the most essential and extensively used electronic passive components in modern electronics, with an estimated global market size of \$18.8 Billion in 2020 and expected to reach \$23.5 Billion by 2027 [69]. Thin-lm exible capacitors are a fundamental unit in various applications such as wearable devices, bendable devices, rollable displays, and interventional medicine [70–73].

Thin-lm capacitors could be realized using inkjet-printing as printing has been used for depositing conductive as well as semi-conductive and insulating lms [74,75]. However, printing


Table 3.1: Micro-fabrication processes comparison

the insulating layer of metal-insulator-metal (MIM) capacitors still cannot satisfy the need for high quality and high density capacitors due to the absence of printable materials with high relative permittivity (εr). Increasing the density of the printed capacitors is essential to reduce chip sizes and costs.

The typical capacitance formula of a parallel plate capacitor can be expressed as,

$$C = \frac{\varepsilon\_0 \varepsilon\_r A}{d},\tag{3.1}$$

ε<sup>0</sup> is the vacuum permittivity, ε*<sup>r</sup>* is the insulating layer permittivity (dielectric constant), *A* is the overlapped surface area between the two metal plates, and *d* is the gap between the

two metal plates (insulator thickness). Achieving high capacitance density by decreasing the insulator thickness leads to increased current leakage and low RF loss tangent [76–78]. The apparent solution for realizing high capacitance density with high performance is by integrating a high permittivity insulating layer [79–81]. Hence, developing a printable high permittivity ink is a must.

Polymer-based inks were commonly used for inkjet-printing the insulating layer such as PI [82], poly(4-vinylphenol) (PVP) [83], poly(methyl methacrylate) (PMMA) [84], and SU-8 [85]. Nonetheless, all of which oer very low dielectric constants in the range of 2 to 4.3. Increasing a polymer's dielectric constant is commonly done by adding ceramic llers with high permittivity to the polymer [86]. This technique is widely used in dierent applications but recently started getting attention in the inkjet-printing eld. Lim et al. printed a BaTiO3-resin with an <sup>ε</sup><sup>r</sup> of 70 at 1 MHz [87]. However, an inltration printing technique was followed, resulted in a thick insulator layer of 20 µm. Kang et al. dispersed BaTiO<sup>3</sup> in PVP, but because of the low ceramic content used (20 wt.%), PVP ε<sup>r</sup> was just doubled from 3.6 to 7.4 [83]. Vesico et al. used HfO<sup>2</sup> based ink, showing an <sup>ε</sup><sup>r</sup> of 12.6 at 1 MHz [88]. Mikolajek et al. dispersed barium strontium titanate (BST) in PMMA enhancing <sup>ε</sup><sup>r</sup> to 55 at 1 kHz [89].

In the following section, two ceramic/polymer composite inks were prepared for inkjet-printing high permittivity and high quality MIM insulating layer. The inks preparation and capacitor printing are described, and the MIM printed capacitors' characteristics are measured and discussed.

### **3.2.2 Material and method**

Benzocyclobutene (BCB) polymers (CYCLOTENE 3022-35, Dow Chemical) is used as the inks' base polymer for its signicant advantages as perfect adhesion to substrates, high planarization, thermal and chemical stability, and being pinhole-free [90]. More importantly, BCB has a low dissipation factor of 0.0008 [91] compared to that of SU-8 (0.015) and polyimide (0.025). BCB is commonly used in high-frequency multi-chip modules [90] and chips packaging.

TiO<sup>2</sup> and BaTiO<sup>3</sup> were chosen as examples of Class I and Class II ceramic dielectric llers, respectively, for their high chemical stability and dielectric constants. Class II ceramics have relatively high dielectrics (1,000 to >20,000) compare to Class I ceramics (5 to a few hundred), but Class I ceramics have a lower dissipation factor (<<0.01) than Class II (0.01 to 0.03) [92]. Commercially available BaTiO<sup>3</sup> (50 nm cubic powder, Sigma-Aldrich) and TiO<sup>2</sup> (Anatase, 25 nm powder, Sigma-Aldrich) NPs were used to prepare the dielectric inks. The smallest available NPs were used to get smooth surfaces.

Each of the ceramic NPs (TiO<sup>2</sup> and BaTiO3) were dispersed into the BCB polymer with two dierent concentrations (20wt.% and 40wt.%). Each ink was stirred at room temperature using a thermal shaker (Thermo shaker LLG-uniTHERMIX 1 pro) for 3 h at 1500 rpm. The viscosities

of the inks were adjusted to 11 mPa s using Mesitylene (98%, Sigma-Aldrich) solvent to meet the inkjet-printing required viscosity (10-12 mPa s). The viscosities were evaluated at room temperature using a rheometer (Thermo Scientic HAAKE RheoStress 300) with a cone plate probe of 30 mm diameter. Figure 3.8 presents the measured viscosities of the prepared inks before adding the solvent. The viscosities (η) were measured as a function of shear rate (˙γ). TiO<sup>2</sup> composites show higher viscosities than that of BaTiO3, especially at high concentrations. Figure 3.8 shows photographs of the prepared inks after adding Mesitylene solvent to adjusts the viscosity. The inks were ltered before lling the inkjet printer's head-cartridge using a 0.45 µm pore size syringe lter. It was observed that the TiO<sup>2</sup> powder diuse better than the BaTiO<sup>3</sup> powder in the BCB polymer, where less amount of TiO<sup>2</sup> remained in the lters than BaTiO3.

Printing was done directly onto a <sup>25</sup> <sup>µ</sup><sup>m</sup> Kapton lm. Piezoelectric cartridges of a 10 pL droplet volume and a total capacity of <sup>1</sup>.5 mL were used. The BCB inkjet-printed droplet diameter was measured to be ≈ 105µm (see Appendix A for a photo of printed droplets and lines of BCB), therefore, the dot spacing was set to 50 µm, and the same jetting waveform for the silver ink was used. Figure 3.9a summarizes the fabrication process ow and illustrates the dimensions of the printed MIM capacitor. Firstly, the bottom layer was printed using Ag ink with . Then the sample was dried on a hot plate at <sup>90</sup> °<sup>C</sup> for 15 min before printing the insulator to prevent layers dissolution. Before printing any of the dielectric inks, they were shaken for 30 min using a vibration shaker (Vortex-5 Shaker) to avoid sedimentation. The dielectric ink was then printed on the bottom layer and soft-backed using a hot plate at <sup>90</sup> °<sup>C</sup> for 15 min. The top conductive layer was then printed onto the top of the insulating layer and dried on a hot plate at <sup>90</sup> °<sup>C</sup> for 15 min. Three printed layers were required for the top layer to achieve a hole-free sheet. Finally, the device was hard-backed inside a vacuum oven at <sup>200</sup> °<sup>C</sup> for 1 h to remove any solvent content.

Figure 3.8: Measured viscosities of the prepared dielectric inks. The graph [left] presents the measured viscosities of the prepared inks before adding the solvent, as well as the viscosities of pure BCB (16.5 mPa s) and Mesitylene solvent (0.7 mPa s). The photographs [right] are of the prepared inks after adding Mesitylene solvent to adjusts the viscosity. TiO<sup>2</sup> powder was observed to diuse better than BaTiO<sup>3</sup> powder in the BCB polymer.

Figure 3.9: Inkjet-printing process of a metal-insulator-metal (MIM) capacitor. **(a)** Schematic illustration of the inkjet-printing process: (1) Bottom layer printing using silver ink; (2) Sintering of the silver ink by heating; (3) Insulating layer printing and sintering by heating; (4) Top layer printing using silver ink; (5) Curing of the device in a vacuum oven. **(b)** 2D schematic of the printed MIM capacitor. The printed parallel plate capacitor has an area of 4 mm<sup>2</sup> .

The thickness, smoothness, and micro-structures of the printed layers were investigated using a chromatic white light interferometer (Bruker ContourGT-K 3D Optical Microscope) and a scanning electron microscope (Carl Zeiss AG—SUPRA 60VP SEM). The insulating layers' dielectric properties and the RF performance of the MIM capacitors were characterized using a network analyzer (E5071C, Keysight) connected to a probe station (MPS150, FormFactor Thiendorf, Germany) employing a Z-probe (Z0-20-K3N-GS-50, FormFactor Thiendorf, Germany). Measurements were done in ambient laboratory conditions.

### **3.2.3 Results and discussion**

The roughness of the deposited insulating layer plays an essential role in dening the performance of the printed MIM capacitors; as the roughness of the deposited layer increases, the quality and the reproducibility of the capacitor decrease. Figure 3.10 presents SEM images and WLI measurements (2D and 3D) of the printed bottom silver and pure BCB insulating layers. The printed silver layer had a thickness and roughness of 470 nm and 73 nm, respectively. Printing one layer of the dielectric ink was not sucient to prevent pinholes in the insulating lm resulting in capacitors with a short circuit. Hence, at least three printed layers were required to avoid pinholes in the insulating lm, resulting in a pure BCB layer with a thickness of 1.56 µm. The printed BCB layer was observed to be smooth with measured roughness of 69 nm. Dispersing ceramic llers in the BCB polymer increases the deposited layer roughness, which reduces the quality factor of the printed capacitor. The minimum achievable thickness of the insulating layer is proportional to the percentage of the ceramic llers' contents. The printed insulating layers using TiO<sup>2</sup> composite inks have rough surfaces (900-1250 nm) and a

Figure 3.10: Surface morphology of inkjet-printed pure BCB insulating lm and conductive silver layer. **(a)** and **(c)** Scanning electron microscope (SEM) images of the printed BCB and bottom silver BCB layers, respectively, with a magnication of ×20K. **(b)** and **(d)** White light interferometry (WLI) measurements with a focusing area of <sup>131</sup> <sup>µ</sup><sup>m</sup> <sup>×</sup> <sup>174</sup> <sup>µ</sup>m. The roughnesses of the printed BCB and silver layer were 69 nm and 73 nm, respectively.

minimum lm thickness in the range of 1.9-2.9 µm. On the other hand, the achieved insulating layers using BaTiO<sup>3</sup> composite inks have roughness in the range of <sup>800</sup>-1000 nm and a thick lm with thicknesses in the range of 4.5-5 µm. Table 3.2 summarizes the roughness, minimum thickness, and loss tangent (tan δ = 1/Q-factor); see Appendix A for the SEM image and WLI measurement of the insulating lm resulted from printing each of the composite inks.

The eect of printing multi-layers of the insulating BCB ink on the capacitance, self-resonance frequency (SRF), and quality factor of the capacitors was studied. The measured impedance (*<sup>Z</sup>* <sup>=</sup> *<sup>R</sup><sup>C</sup>* <sup>+</sup> *jXC*) was used to calculate the capacitance (*<sup>C</sup>* <sup>=</sup> <sup>1</sup>/2<sup>π</sup> *f XC*) and the quality factor (*<sup>Q</sup>* <sup>=</sup> <sup>−</sup>*XC*/*RC*) of the capacitors over a frequency range of 100 kHz to <sup>2</sup>.2 GHz. Figures 3.11 and 3.12 show the measured capacitance and quality factor of MIM capacitors with dierent numbers of printed insulating layers. Besides inkjet-printing, two capacitors were prepared by drop-casting (DC) of BCB inks, one with modied viscosity of 11 mPa s and the other with the viscosity provided by the supplier of 16 mPa s. Increasing the number of printed insulating layers results in: i) reducing the capacitance as it is inversely proportional to the distance between the capacitor's two conductive plates, as explained in Equation 3.1; 2) increasing of the capacitor's SRF as it is inversely proportional to the capacitance ( *f* = 1/ √ *LC*); and iii) enhancing the capacitor's quality factor, as increasing the gap between the parallel plates avoid any current leakages or pinholes. The eect of adding the ceramic llers (BaTiO<sup>3</sup> and TiO2) with dierent concentrations on the relative permittivity and quality factor was investigated. The relative permittivity was calculated by substituting Equation 3.1 with the measured capacitance, measured insulating lm thickness using WLI, and the capacitor dimensions. The dimensions of the capacitors' bottom and upper conductive layers are <sup>2</sup>.2×2.2 mm and <sup>2</sup>×2 mm,


Table 3.2: The measured parameters of the dielectric layers printed using developed composite inks - comparison with state of the art.

<sup>c</sup> Calculated using insulating layer's thickness and dielectric coecient

<sup>b</sup> The Measuring frequency isn't mentioned

<sup>c</sup> @ 0.1 MHz

respectively, resulting in a capacitor area of 4 mm<sup>2</sup> .

In the case of composite inks with 40wt.% concentration of BaTiO<sup>3</sup> and TiO2, the relative permittivity of the BCB was enhanced from 2.3 to 37 and 20, respectively. The addition of the ceramic llers leads to a decrease in the capacitor's quality factor since they increase the surface roughness; moreover, ceramic powders have higher dissipation factors than that of the BCB polymer. Dispersing of BaTiO<sup>3</sup> powder in the BCB polymers leads to less quality than adding TiO2; however, BaTiO<sup>2</sup> leads to better enhancement in the relative permittivity. These results agree with the fact that TiO<sup>3</sup> belongs to Class I of the ceramic dielectrics, and BaTiO<sup>2</sup>

Figure 3.11: Eect of multi-layers printing and drop-casting (DC) of the insulating layer on the capacitance of a 4 mm<sup>2</sup> MIM capacitor. The capacitor's SRF increases as the gap between the two parallel plated increases since SRF is inversely proportional to the capacitance ( *f* = 1/ √ *LC*). the Focused view [Left] show the capacitance at relatively low frequencies (100 kHz-80 MHz).

Figure 3.12: Eect of multi-layers printing and drop-casting (DC) of the insulating layer on the q-factor of a 4 mm<sup>2</sup> MIM capacitor. The quality factor increases at low frequencies as the gap between the capacitor's parallel plates increases. However, at wide gaps, the quality factor drops dramatically as the frequency increases.

belongs to class II [92]. Table 3.2 summarizes the measured parameters of the developed composite inks and composite inks from literature for comparison.

## **3.3 Metal-semiconductor (MS) exible diode**

#### **3.3.1 Introduction**

Organic electronics in general and organic diodes, in particular, have attracted signicant consideration for next-generation electronics due to their uncountable advantages as exibility, scalability, light-weight, and applicability in numerous elds [93]. Organic diodes have been intensively investigated over the past decades, driven by commercial demands and improvement potential. Organic electronic components show their promising potential to be fabricated using rapid and cost-eective techniques that push toward an era of disposable electronics [94, 95].

Diodes are the primary building block in near eld communication and wireless power transmission applications; they are employed as rectifying switches where high switching speeds between forward and reverse bias are required [96]. Schottky diodes are considered one of the simplest electronic components in the semiconductor industry; it consists of two parallel electrodes of dierent metals separated by a single layer of a p-type or n-type semiconductor, as shown in Fig. 3.13a. One of the metal/semiconductor (MS) interfaces presents low resistance current conduction, creating an ohmic contact. In contrast, the other MS interface is responsible for rectication by allowing charge injection in one bias direction and prevents it in the other direction creating a Schottky barrier (φ <sup>B</sup>). In the case of a p-type semiconductor, the rectifying properties originate by bringing it in contact with a metal having a working function (WF) lower than the semiconductor's Fermi level (EF). The MS junction forces the two energy levels (metal's WF and semiconductor's EF) to align, leading to band bending and creating the Schottky barrier, as shown in Fig. 3.13c and d. The MS barrier height in Schottky diodes is lower than that of junction barriers in p–n junction diodes, resulting in a lower voltage droplet in the forward bias while allowing a higher forward current density. Moreover, the absence of a charge carrier depletion region in Schottky diodes eliminates the required reverse recovery time for switching the diode state from forward bias to reverse bias, leading to a higher switching frequency.

Schottky diodes based on vertical structures are the most compatible diode design with inkjetprinting [97]; vertical structures tend to reduce the lateral resolution requirement. Printed silver electrodes are always expected to create ohmic contact with a typical p-type organic semiconductor as silver features high WF (φ*<sup>M</sup>* ≈ −5.2 eV). However, printing a metal electrode with low WF to generate the Schottky barrier with organic semiconductors is still challenging. A low WF metal such as copper or aluminum is required for the Schottky diode's cathode electrode. Aluminum suers from a high degree of oxidation, which leads to a forward voltage

Figure 3.13: Schematic illustration of the working principle of the metal-semiconductor diode (Schottky diode). **(a)** Construction and symbol of the Schottky diode. The MS diode is constructed from a low WF metal, high WF metal, and p-type or n-type semiconductor. Based on the type of the semiconductor, one of the junctions creates ohmic contact (allow charge ow in both directions), and the other junction creates Schottky contact (has the rectication behavior. **(b)** VI-Characteristics of the Schottky diode. **(c)** [Right] coss-section view of the printed MS diode's structure. [Left] Energy band diagram showing the relative WF for copper (Cu), PTAA (p-type semiconductor), and silver (Ag) [Left]. **(d)** Energy band diagram of a Cu-PTAA-Ag diode under no, forward, and reverse biases. The arrows illustrate the holes' movement.

oset. On the other hand, copper does not suer from the same degree of oxidation as aluminum. Still, the available copper-based inks need aggressive post-treatments, making them incompatible with polymeric substrates and organic semiconductors [98, 99].

The following section demonstrates the fabrication of an inkjet-printed Schottky diode on an unstructured exible polyimide substrate.

### **3.3.2 Method and results**

The Schottky diode was printed directly onto an unstructured 25 µm Kapton lm. Silver NPs ink was used for printing the conductive electrodes. The low WF copper electrode was achieved by electroplating the printed silver electrode. The roughness of the plated copper layer can be controlled by employing pulse electroplating, as going to be discussed in Chapter 4.

Poly(triaryl amin) semiconductor (PTAA, Sigma-Aldrich®) was chosen as a p-type organic semiconducting with hole mobilities of 10−<sup>3</sup> up to 10−<sup>2</sup> cm<sup>2</sup> V <sup>−</sup><sup>1</sup> s−<sup>1</sup> , resulting in high carrier mobility. PTAA was chosen for its excellent ambient air stability, not restricted to be used in an inert atmosphere like most other semiconductors. The PTAA's highest occupied molecular orbital (HOMO) and lowest unoccupied molecular orbital (LUMO) levels lie at <sup>−</sup>5.<sup>0</sup> ∼ −5.1 eV and <sup>−</sup>1.<sup>8</sup> ∼ −2.0 eV, respectively. Hence, PTAA is expected to create ohmic contact with printed silver (φ*<sup>M</sup>* ≈ −5.2 eV) and Schottky contact with copper (φ*<sup>M</sup>* ≈ −4.7 eV). A homogeneous solution of the PTAA was prepared by dispersing 2.0wt% of the polymer in toluene at room temperature. As the percentage of the PTAA to the solvent increases the operating frequency increases, due to the higher carrier mobility concentration. However, increasing the PTAA concentration leads to a cracked deposited layer. Hence, 2wt.% was found to achieve a hole-free layer (check Appendix A for spin-coated PTAA layers with 2.0wt.% and 5.0wt.% concentrations).

Figure 3.14 summarizes the Schottky diode fabrication process ow. At rst, the bottom layer was inkjet-printed directly on the Kapton lm using silver ink. Then the sample was sintered using an oven at <sup>150</sup> °<sup>C</sup> for 1 h. The bottom silver layer was then electroplated with a <sup>3</sup> <sup>µ</sup><sup>m</sup> copper layer. Uni-polar pulse plating was implemented using a home-built power source (Section 4.3) to enhance the surface roughness. The plating waveforms are explained in detail in Chapter 4. On top of the copper electrode, a single layer of PTAA was spin-coated at 1000 rpm for 1 min and dried on a hot plate at <sup>100</sup> °<sup>C</sup> for 5 min. The top silver electrode was printed on the PTAA layer and dried on a hot plate at <sup>90</sup> °<sup>C</sup> for 15 min. For the top silver layer, at least three printed layers were required to achieve a hole-free sheet. Finally, the whole device was hard-backed in an vacuum oven <sup>200</sup> °<sup>C</sup> for 1 h to remove any solvent content. Vacuum ovens are generally recommended for hard-backing to prevent oxidation and help with gases extraction. Figure 3.15a shows a microscopic photo of the fabricated MS diode. The SEM im-

Figure 3.14: Inkjet-printing process of a metal-semiconductor (MS) diode. **(a)** Schematic illustration of the inkjetprinting process: (1) Bottom layer printing using silver ink; (2) Sintering of the silver ink by heating; (3) Electroplating of the bottom layer with copper; (4) Semiconductor layer spin-coating and sintering by heating; (5) Top layer printing using silver ink; (6) Curing of the device in a vacuum oven. **(b)** 2D schematic of the printed MS diode. The printed diode has an area of 4 mm<sup>2</sup> .

ages and WLI measurements of the spin-coated PTAA semiconductor layers and the bottom copper layer are shown in Fig. 3.15b and c. The spin-coated PTAA layer had a thickness and roughness of 410 nm and 52 nm, respectively.

Figure 3.15: Fabricated MS diode using inkjet-printing. **(a)** Microscopic image of the fabricated MS diode. **(b)** SEM image and WLI measurement of the PTAA layer. The thickness and roughness of the PTAA layer were 410 nm and 52 nm. **(c)** SEM image and WLI measurement of the bottom copper layer. The thickness and roughness of the copper layer were <sup>10</sup> <sup>µ</sup><sup>m</sup> and 72 nm.

#### **4 Electroplating**

**E** lectroplating is a common technology in the area of micro-fabrication. Deposited seedlayers typically have thicknesses within the range of tens to hundreds of nanometers. Consequently, it results in high resistivity for applications that demand high electrical conductivity in DC to RF ranges. The high resistance generates Joule heat losses and aects the transferred electrical signal by introducing thermal noises. The high resistance of slender conducting tracks arises from the distribution of an AC signal throughout the conductor, where the current density is highest at the conductor's surface and drops exponentially with depth; this phenomenon is referred to as the skin eect. The current ows through the conductor's skin, which lies between the conductor's outside surface and a level known as the skin depth. The requisite minimum skin depth (δ) of the conductive tracks can be calculated using: δ = p ρ/π *f* µ, where *f* is the electrical signal's frequency, and µ and ρ are the permeability and resistivity of the conductive layer's material.

Electroplating has consistently shown its ability to coat dierent substrates with various materials eciently. Pure metals, metal composites, alloys, conductive polymers, semiconductors, and oxides can be deposited via electroplating [100]. Due to the diverse applications of electroplating and its impact on the eld of material science, it has been under investigation for the last centuries. Electroplating has been involved in multiple areas starting from the conventional ones as erosion, corrosion, and wear resistance [101], up to highly advanced applications such as micro and nano fabrication [61], integrated circuits [102], and thick-walled electroformed products (above millimeter) [100]. Electroplating is widely used in industry as well as in research due to its multiple benets. Besides being low cost, rapid, and able to coat dierent substrates as well as controlled alloy composition, it results in pure free from pores structures with the range of mm down to nm [103,104]. Manipulating the parameters involved in the electroplating process can control the deposited lm features. Operating parameters such as applied current density, electrolyte bath temperature and composition, pH value, and additives provide control over the deposited lm crystallite size, texture, and embedded lm stresses [100].

This chapter focuses on controlling the deposited lm features by controlling the waveform of the applied current density. Through the chapter, we also show the signicant advantages of controlling the deposited lm's micro-structures on the adhesion of inkjet-printed tracks onto an unstructured exible substrate.

Generating current pulses for pulsed electroplating, especially with complex waveforms, typically requires complicated and expensive equipment compared to the conventional DC tools [105–107]. Hence, applications with more complex pulsed electroplating still face some di-

This chapter's text and gures were reproduced and adapted from the author's published work [ON1] with permission of the American Institute of Physics (AIP) Publishing (CC BY 4.0, 2019)

culties and are seldom found in practice. In this chapter, we present an electroplating power source with high functional capability 4.3. The developed current source can be used for standard DC as well as complex pulsed electroplating. It can also be employed for galvanostatic and potentiostatic electroplating processes. The current source has been used for dierent complex applications, as demonstrated in the applications Sections 4.4 and 4.5.

The fundamentals of the electroplating process presented in this chapter are mainly derived from the following two textbooks:


## **4.1 Electroplating working principle**

By dipping a metal in an aqueous solution containing ions of that metal, ions exchange happens between the metal and solution. Some ions move from the crystal lattice structure of the metal electrode to the solution, and in return, some ions move from the solution to the metal's crystal lattice structure. Since the two ways of exchange happen at dierent rates, this leads to a potential dierence. This potential interface dierence for metal can be calculated according to the Nernst equation:

$$E = E^0 + \frac{RT}{zF} \ln a(M^{\odot+}),\tag{4.1}$$

*E* 0 is the relative standard electrode potential of the metal, *R* is the gas constant, *T* is the absolute temperature, *z* is the number of electrons in the reaction, *F* is Faraday constant, and *a*(*M*z+) is the activity of the ion.

The relative standard electrode potential of metal can be measured by immersing the metal in a solution of its ions and connecting a reference electrode creating an electrochemical cell. Standard hydrogen electrode (SHE) is commonly used as the reference electrode since it has zero potential. The relative standard potential for electrodes of dierent materials is reported in the literature [109].

Electroplating a target conductive substrate with a particular ion can be achieved by immersing the substrate in a solution containing the ions of the desired metal and the plating metal electrode. A power source is used to force the ions to deposit from the plating electrode to the target substrate. The target substrate is connected to the cathode, and the plating electrode is connected to the anode, as shown in Fig. 4.1. Electrodeposition can be done in two manners, galvanostatic or potentiostatic. In the potentiostatic, the applied voltage is set to a constant level, leading to a current changing with time. In contrast, in the galvanostatic, the electric current is the controlled parameter leading to a continuous rate of deposition [106]. One of the common simple electroplating processes is copper electroplating, where at the anode cop-

Figure 4.1: Schematic diagram of copper electroplating setup.

per is dissolved in copper sulfate (CuSO4) electrolyte and at the cathode copper is deposited according to chemical formula [110]:

$$\begin{aligned} \text{At anode Cu(dissolved)} &\longrightarrow \text{Cu}^{2+} + 2\text{ e}^-\\ \text{At cathode Cu}^{2+} + 2\text{ e}^- &\longrightarrow \text{Cu}\_{\text{(deposited)}} \end{aligned}$$

According to Faraday's law, the amount of electrochemical reactions within an electrochemical cell is proportional to the quantity of electric charge *Q* through the substance. For the accumulated weight *w* at the electrode in electrolysis, Faraday's law states the relationship:

$$\mathbf{w} = \mathbf{Z} \cdot \boldsymbol{\mathcal{Q}}, \quad \text{with} \quad \mathbf{Z} = \frac{\mathbf{M}}{z\_l F} \quad , \text{and} \quad \boldsymbol{\mathcal{Q}} = \int i \, d\mathbf{t} \,. \tag{4.2}$$

*Z* is the constant of proportionality and is known as the electrochemical equivalent, *M* is the molar mass of the substance, *z<sup>i</sup>* is the valency of the ion, and *F* is the Faraday constant. The deposition rate can be within a wide range of 0.1-10 µm/min. The deposit mass and thickness are calculated through Faraday's equations:

$$mass = \frac{\alpha ItM}{nF},\tag{4.3}$$

α is the deposition eciency, *I* the applied current, *t* the electroplating time, *M* Molar mass, *n* the charge state, and *F* Faraday's constant.

$$Thickness = \frac{M\_at}{Fn\rho} \varepsilon\_f i,\tag{4.4}$$

*i* is the applied current density, *M<sup>a</sup>* atomic weight, ρ metal density, *n* number of electron

transfer in electroplating reaction, ε<sup>f</sup> Faraday eciency.

Three forms of current distributions impact metal deposition during the electroplating process: primary, secondary, and tertiary. The primary current distribution is inuenced by the electroplating system's geometric aspects, especially the plated structures. Higher current densities are built at edges, resulting in more deposited metal around these areas leading to the well-known phenomena (dog boning). The secondary current distribution is the electron transfer associated with metal dissolving from the anode or depositing on the cathode. The tertiary current distribution is the ion's transportation within the electrolyte between the electrodes [106].

The deposition of metal can be modeled as a capacitor parallel to resistance, as shown in Fig. 4.2a. The capacitor represents the layer between the electroplated electrode (cathode) and the electrolyte, known as the electric double layer (EDL). The resistance represents the electrochemical reaction. When the power source is connected to the electrodes, the capacitive current *I<sup>C</sup>* is used for charging the EDL. After the EDL is charged, the faradic current *I<sup>F</sup>* deposits the metal. The deposition doesn't occur until the EDL is charged. Turning o the power leads to the discharge of the EDL. The time is taken by *I<sup>F</sup>* to reach 99% of the applied current (*IP*) is the charging time (*TC*). While the time taken by *I<sup>F</sup>* to reach 1% of *I<sup>P</sup>* is the discharging time (*Td*) [111]. As a rule of thumb, the charging of the EDL is seven times faster than the discharging and can be roughly calculated by [112]

$$T\_c = \frac{17}{I\_P}, T\_d = \frac{120}{I\_P}.\tag{4.5}$$

Figure 4.2: Equivalent electrical circuit and ion transportation of the electroplating process. **(a)** Schematic diagram of the electrode's equivalent electrical circuit [106, 113]. **(b)** Schematic diagram illustrating the dierent layers formed by the mass transportation of the ions from the anode toward the cathode within the electrolyte [114, 115].

## **4.2 Pulsed electroplating**

Pulsed electroplating has drawn signicant attention in research as well as in industrial fabrications. One of the rst signicant applications of pulsed electroplating was depositing gold toward switching crossover electronic circuits performed by Western Electric in the 1970s [116]. A reduction in the electroplating time by a factor of 2 was achieved when implementing pulsed electroplating compared to the conventional direct current electroplating. The timing parameter of pulsed electroplating of this application (on/o time is 0.1/0.9 s) is still used till today for soft gold deposition. Pulsed electroplating has been considered in other elds as well, e.g., in the eld of optics, highly anti-reective surfaces were achieved via the implementation of bipolar pulsed electroplating [117].

Modifying the current distribution with the electroplating bath results in controlling the deposited lm's internal structure as well as its physical and mechanical properties. The current distributions within the electroplating bath can be modied either by adding chemical additives or employing pulsed electroplating [100]. However, it's preferable to limit the usage of the chemical additives for a more reliable electroplating process [118]. On the other hand, pulsed electroplating always shows its ability to control the deposited lm properties without involving chemical additives. Furthermore, implementing reverse pulsed electroplating can replace the additives and eliminate the drawbacks resulting from using them [100].

Via pulsed electroplating, several features of the deposited material can be controlled, such as: the atomic micro-scale structure [118], the grain size [ON1], the mechanical and physical properties of the deposited lm [119], the surface roughness [ON1], and the hydrogen content [120].

Pulsed electroplating can be used to modify the diusion layer, as shown in Fig. 4.3. In direct electroplating, a single diusion layer is created. In contrast, pulsed electroplating leads to two diusion layers, increasing uniformity in mass transportation, leading to a more uniform metal deposition and a better lling of holes and vias. The concentration of ions is constant with respect to the distance from the electrodes beyond the diusion regions (bulk concentration) [106].

A passive negatively charged layer is built around the cathode during the DC electroplating. The formed layer hinders the ions from transferring uniformly toward the cathode. pulsed electroplating leads to a uniform transfer of the ions by discharging this built layer during the switch-o time. Moreover, applying the opposite polarity (reversed pulsed electroplating) reduces the discharging time [100, 105].

As discussed in the previous Section 4.1, the current density is not uniformly distributed within the electroplating bath, which leads to higher consumption of the ions in the spaces with high current density than that with lower current density resulting in a non-uniform deposition. Pulsating the applied current during the electroplating process allows the ions to transfer

Figure 4.3: Diusion layer within the electrolyte in case of direct and pulsed electroplating [106,121]. **(a)** A single diusion layer is formed in the case of direct electroplating. **(b)** The diusion layer is divided into two layers by the pulsed electroplating, which results in a more uniform mass transportation and thus a more uniform metal deposition.

from high concentration to low concentration spaces leads to a uniform distribution of the ions within the bath hence a uniform deposited lm.

### **4.2.1 Waveforms for pulsed electroplating**

In pulsed electroplating, the applied current density is always alternating between dierent set-points, as shown in Fig. 4.4b and c. Other than the direct current electroplating, where the applied current density is set to a xed value through the whole process as shown in Fig. 4.4a. The material's deposition rate and thickness in pulsed electroplating are equivalent to a DC electroplating operating at the average current density of the pulsed electroplating.

The pulsed electroplating processes are generally split into two categorize:

1. Uni-polar pulsed electroplating, also referred to as ordinary pulsed electroplating. In this category, all pulses have the same polarity and the applied current frequently oscillates between a positive state (*IPeak*) and the o state, producing a sequence of pulses

Figure 4.4: Main categories of the applied current densities Waveforms for electroplating. **(a)** Direct current pulsed electroplating. **(b)** Ordinary uni-polar pulsed electroplating. **(c)** Bipolar pulsed electroplating.

with similar duration and amplitude as shown in Fig. 4.4b. The duration of applied current is denoted as the on-time (*TON*), and the duration of the o-state is denoted as the o-time (*TOFF*). The duty cycle of the pulse (γ) is the proportion of the on-time to the total pulse time (*TPulse*),

$$\gamma = \frac{T\_{On}}{T\_{On} + T\_{off}}.\tag{4.6}$$

The average current (*IAvg*) is the actual applied value that is used for calculating the material depositions' rate and thickness. It can be calculated based on the peak current and the pulse duty cycle,

$$I\_{\text{Avg}} = I\_{\text{Peak}} \mathcal{Y}.\tag{4.7}$$

2. Bipolar pulsed electroplating; also referred to as reverse pulsed electroplating. In this category, the applied current oscillates between values with dierent polarities. The applied waveform has a total time of *TPulse*. The cathodic part of the applied pulse with the positive polarity current is referred to as *I<sup>C</sup>* operating for time *T<sup>C</sup>* and the anodic part of the pulse has a negative polarity current referred to as *I<sup>A</sup>* operating for time *TA*, as shown in Fig. 4.4c. For bipolar pulsed electroplating, the duty cycle and the average current (*I* 0 *Avg*) of the pulse can be calculated by:

$$\gamma' = \frac{T\_\mathrm{A}}{T\_\mathrm{A} + T\_\mathrm{C}},\tag{4.8}$$

$$I\_{\rm Avg} \,' = \frac{I\_A T\_A - I\_C T\_C}{T\_A + T\_C} \,. \tag{4.9}$$

Besides the ordinary pulses, more complex waveforms have been investigated in the past [105, 107,122], some of which are presented in Fig. 4.5 and Fig. 4.6. These waveforms provide higher control over the deposited materials. For uni-polar pulsed electroplating, the other candidates were:


For bipolar pulsed electroplating, the other introduced waveforms were:

1. Reverse pulse with o period, where the bipolar pulses are separated with o time.


For pulsed electroplating, typically, the pulse duration are in the range of milliseconds with a duty cycle > 5% to be enough for the EDL to charge and the deposition happens. For any of the mentioned complex waveforms in Fig. 4.5 and Fig. 4.6, the average current density can be determined by dividing the integration of the applied current through a single full pulse over the total pulse time,

$$I\_{\rm Avg} = \frac{1}{T\_{\rm Pules}} \int\_0^{T\_{\rm Puure}} I(t)dt. \tag{4.10}$$

Figure 4.5: Dierent complex waveforms introduced for uni-polar pulsed electroplating. The candidates for unipolar pulsed electroplating are **(a)** Superimposed pulse, **(b)** Duplex pulse, **(c)** Pulsed pulse, and **(d)** Pulsed on pulse. For all curves, the Y-axis is the current density (J), and the x-axis is the time (T).

Figure 4.6: Dierent complex waveforms introduced for bipolar pulsed electroplating. The candidates for bipolar pulsed electroplating are **(a)** Reverse pulse with o-time, **(b)** Reverse duplex pulse, **(c)** Reverse pulsed pulse, and **(d)** Reverse pulsed on pulse. For all curves, the Y-axis is the current density (J), and the x-axis is the time (T).

## **4.3 Electroplating tool: current source**

Generating the complex waveforms mentioned in the previous section for the electroplating process is challenging. Numerous instruments and devices have been patented that are used for generating the required waveforms for pulsed electroplating and pulse anodization. The technologies range from manually switching the applied current for periods of a few minutes to a programmable power source using a microprocessor [123]. Silicon-based pulse rectiers have been extensively considered. Nevertheless, they suer from relatively slow falling and rising times and cannot provide high pulse rates; they have grown into disuse. The review by Raj et al. [122] covers the various previously developed devices for pulse anodizing having similar specications to those needed for pulsed electroplating.

In this section, a low-cost electroplating power source with high functional capability is introduced. The system can be used to perform complex galvanostatic and potentiostatic electroplating processes. The developed system was used to solve the diculties of electroplating inkjet-printed conductive tracks on unstructured exible substrates, as explained in Section 4.4. The system has also been used for performing other miscellaneous applications, as demonstrated in the applications Section 4.5.

### **4.3.1 Current source design**

The developed power source precisely controls the waveform and the amplitude of the current pumped through the electroplating bath. The circuit possesses a feedback system for online monitor and recording the output current for documentation and diagnostic purposes. The circuit is controlled wirelessly over the internet connection via a developed graphical user interface (GUI).

A BeagleBone Black Wireless®(BBB) is used as a microcontroller unit (MCU) of the system. This microcontroller board posses a built-in wireless module that allows online access to the circuit over an internet connection. The remote access to the circuit gives the benet of remotely setting the electroplating process parameter. Moreover, the vital benet is observing the output voltage, which is used for monitoring the change in the impedance of the connected load. The BeagleBone is a Linux-based microcontroller. It includes an SD card as a storage memory, which provides enough space to record every electroplating process with dierent values. A real-time clock (RTC) integrated circuit (IC)(BQ32000, Texas Instruments®) is included in the circuit and connected to the MCU to keep track of the executed processes in real-time.

The current source was built mainly based on three modules and a microcontroller: (I) an isolated power supply for supplying the circuit with the necessary powers; (II) a feedback module to sense the output signal and send it to the microcontroller; and (III) a voltage-controlled current circuit (VCCC) to convert a voltage signal provided by the microcontroller to a controlled output current.

*Module (I).* A DC power supply is used to provide the circuit with a DC power of 12 V and 2 A. The input power provided by the supply is branched into four routes:


4. The last rout is connected directly to a lter and used to supply the circuit with 12 V used to power up a cooling fan.

*Module (II).* A feedback signal from the output port is shunted to a 16-bit analog to digital converter (*ADC*) (ADS1112, Texas Instruments®), which sends the converted signal to the microcontroller through the inter-integrated circuit (*I* <sup>2</sup>*C*) protocol. For protecting the microcontroller board, an *I* <sup>2</sup>*C* isolator chip (ADuM1250, Analog Devices) is located between the ADC and the microcontroller, as shown in Fig. 4.7.

*Module (III).* The microcontroller sets the control voltage (VControl) through the serial peripheral interface (*SPI*) by a 16-bit digital to analog converter (*DAC*) (DAC8831, Texas Instruments®). The used DAC can generate a bipolar output signal in the range of ±VREF when working with an external buer. The reference voltage (VREF) was set to <sup>2</sup>.5 V. Since a bipolar *DAC* is used, the most signicant bit (*MSB*) of the 16 bits acts as a sign bit, while the signal is dened by the rest of the acquired digital word. Hence, the resolution can be calculated as follow:

$$Resolution = \frac{V\_{\text{ref}}}{2^{n-1}}.\tag{4.11}$$

For protecting the microcontroller board, an isolated dc-to-dc converter chip (ADuM6401, Analog Devices) is located between the DAC and the microcontroller, as shown in Fig. 4.7. The control signal (VControl) is then provided to a VCCC, which is the most critical element of the whole circuit as it denes its performance. Numerous approaches and designs of the voltage-controlled current source have been introduced before and utilized in dierent applications [124–130]. The primary dening characteristic of a VCCC is the nominal value and stability of the output impedance while changing the connected load impedance or the out-

Figure 4.7: Schematic diagram illustrating the interface circuits connected to the microcontroller. The interface electronics are used for generating the input control signal (VControl) to the VCCC and measuring the feedback signal (VFeedback).

put frequency. A current source derived from the modied Howland current source (MHCS) was developed, characterized, and subsequently used in our electroplating experiments. The (MHCS) was a perfect candidate as a VCCC for the electroplating application due to its simple conguration, stability and reliability in the performance under dierent surrounding conditions, and voltage compliance.

### **4.3.2 Voltage control current circuit (VCCC)**

The modied Howland current source (MHCS) is used for controlling the amplitude and waveform of the output current as a function of the input control voltage. The MHCS is famous for the high stability of the output current over a broad frequency band. The MHCS con guration can sink or source bipolar current with a high output impedance, preventing the connected load from aecting the circuit's performance. Moreover, It can operate with oating or grounded loads [131]. The MHCS is appropriate for applications that require stability in the controlled current with dierent waveforms over a wide range of frequencies. Due to its reliable performance, The MHCS has been widely considered in numerous critical applications in dierent elds, e.g., detecting breast cancer using the electrical impedance tomography (EIT) [132] and in functional neural stimulation (FNS) [133, 134].

The basic Howland current pump (HCS) was rst introduced by Prof. Bradford Howland of MIT for controlling current via voltage in 1962 [135]. However, in the initial proposed design, the output voltage did not swing close to the rail voltage supplied to the circuit leading to complications regarding the voltage compliance [136]. Hence, the basic conguration was inecient with the loads that draw high voltages leading to waste in power. Consequently, the modied version was introduced where the non-inverting port of the OPAMP was connected to the output port via a passive resistor excluding the loading eects or the common-mode voltages.

A schematic gure of the MHCS is shown in Fig. 4.8. The MHCS consists mainly of a set of passive electronic components (resistors and capacitors) and one active electronics component (operational amplier (OPAMP)). Through this subsection, the MHCS is modeled, then the circuit's dierent components (resistors, capacitors, and OPAMP) are discussed.

#### *MHCS circuit modeling:*

The transfer function between the input control signal (*VControl*) and the output current (*I*OUT) of a MHCS can be driven using a simple circuit analysis assuming an ideal OPAMP as shown in Equation 4.12 (see Appendix B for derivation). The transfer function of the MHCS reveals a simple linear relation between the input control voltage (*VControl*) and output current (*IOUT* ).

$$I\_{\rm OUT} = -V\_{\rm control} \frac{R\_2}{R\_1 r} + V\_{\rm OUT} \frac{R\_1 (R\_4 + r) - R\_2 R\_3}{R\_1 r (R\_3 + R\_4)}.\tag{4.12}$$

For a high precision current source, the circuit's output impedance must be much higher than the load through which the current is pumped, so that the load impedance doesn't aect the value of the output current. The output impedance (*ZOUT* ) of the MHCS can be derived by *VOUT* /*IOUT* while connecting the control signal (*VControl*) to the ground as stated in Equation 4.13. In general, *ZOUT* should be in the range of a few MΩ for single-digit currents and voltages.

$$Z\_{OUT} = \frac{V\_{OUT}}{I\_{OUT}} = \frac{R\_1 r (R\_3 + R\_4)}{R\_1 (R\_4 + r) - R\_2 R\_3}.\tag{4.13}$$

Because of the OPAMP's non-ideal performance, it has a nite common mode rejection ratio (CMRR), which contributes to the output signal with a part known as common-mode noise. The CMRR of a dierential amplier is the ratio between the dierential gain (*Ad*) and the common-mode gain (*ACM*).

$$V\_{out} = V\_d A\_d + V\_{CM} A\_{CM}, \quad CMRR = \frac{A\_d}{A\_{CM}}.\tag{4.14}$$

The close loop CMRR as function in the resistors can be derived by assuming a non-ideal OPAMP with nite gain (A). The output is calculated in the presence of inputs on the positive and negative terminal of the amplier [137].

$$CMRR = 1 + 2A(\frac{R\_1}{R\_1 + R\_2} + \frac{R\_4}{r}).\tag{4.15}$$

As the opamp is the key component of the MHCS, the circuit's performance can be strongly aected as the opamp's temperature increases. Hence, for the application where high power is required for a long time, it's essential to provide an excellent cooling system. The power

Figure 4.8: A schematic gure of the modied Howland current source (MHCS).

dissipation (*PD*) through the OPAMP can be estimated by the following,

$$PD = P\_{\text{Supply}} - P\_{\text{Load}} = V\_{\text{S}}I - I^2 R\_{\text{Load}} \,\text{.}\tag{4.16}$$

Equation 4.16 shows that the dissipated power in the OPAMP is the remaining of the supplied power (*PSupply*) after deducting the power consumed by the load (*PLoad*). For resistive load, the power consumption is directly proportional to the square of the output voltage. Hence, the maximum power dissipation (*PDmax*) occurs at half of the maximum supplied voltage (*I* = *VSmax*/2*RLoad*) [138], assuming that the output voltage can swing to the voltage's extreme. Hence, the maximum dissipated power in the OPAMP can be expressed by,

$$PD\_{\text{max}} = \frac{V\_{\text{Smax}}^2}{4R\_{Load}}.\tag{4.17}$$

The power dissipation in the OPAMP can be used for estimating the junction temperature *T<sup>J</sup>* by:

$$T\_J = PD \,\,\Theta\_{J\Lambda} + T\_{\Lambda},\tag{4.18}$$

where θ*JA* is the junction to ambient thermal resistance and *T<sup>A</sup>* is the ambient temperature. θ*JA* (°C/W) is function in the heat-sink's area attached to the OPAMP; this relation is usually provided in the OPAMP's technical data-sheet. Hence, the heat-sink's required area can be estimated to keep the OPAMP temperature below the safety limits.

#### *Components selection:*

The Howland current source's performance deviates from the theoretical predictions due to ideality assumptions. Hence, each of the passive and active components of the MHCS must be chosen carefully to achieve the high performance of stable and precise output current with broad bandwidth. The main aecting factors on the circuit's performance are the degree of non-ideality of the OPAMP and the tolerance in the values of the resistors. Several comprehensive studies were done to study the performance of the MHCS under dierent conditions, for example:


• The practical problems involved with the dierent congurations of the Howland current sources were discussed by Texas instrument [136].

*I. Resistors selection.* The resistors' values should be selected in a way for the circuit to be symmetrical to improve the stability. By achieving *R*<sup>1</sup> = *R*<sup>3</sup> and *R*<sup>2</sup> = *R*<sup>4</sup> + *r*, so that *R*1/*R*<sup>2</sup> match *R*3/(*R*<sup>4</sup> +*r*), achieves the goal. So that the total output current of MHCS (*IOUT* ) in Equation 4.12 will depend only on the control signal (*VControl*) and is not aected by the output voltage (*VOUT* ), then it can be expressed as:

$$I\_{OUT} = -V\_{Control} \frac{R\_2}{R\_1 r}.\tag{4.19}$$

Equation 4.19 reveals the simple linear relation between the input control signal and output current, where *r* sets the gain and *R*1/*R*<sup>2</sup> scales it. The resistor *r* is selected to have low resistance in the range of few to hundred ohms. The remainder of the circuit's resistors are in the range of kilo-ohms to mega-ohms. Consequently, *R*<sup>2</sup> *r* so the output current direction can be assumed to ow through *r* towards the load based on the current divider rule. Thus, a high power resistor is employed for *r*. A small value of *r* reduces *ZOUT* in Equation 4.13, but at the same time, it improves the CMRR in Equation 4.15.

Resistance tolerance plays an essential role in the MHCS behavior. Any small miss-match between *R*2*R*<sup>3</sup> and *R*1(*R*<sup>4</sup> + *r*) will permit the connected load to inuence the output current. On the other hand, *ZOUT* and CMRR are both inversely proportional to the resistance tolerance [137, 140]. The resistors' miss-match results in poor CMRR no matter how good the OPAMP is. Consequently, as the used resistances in the MHCS have wide ranges of tolerance, the circuit's performance is strongly reduced. It's preferable to increase the close loop CMRR to overcome the common-mode input signal.

Trimming the resistances is essential to increase the CMRR and overcome the output current dependency on the load impedance leading to improved performance of the MHCS. Several approaches have been introduced to accurately set the resistance values, summarized in the technical report provided by Texas Instruments [136]. Trimming can be achieved by using trimmer resistors or using highly accurate resistors with low error percentages, which are relatively expensive. A potentiometer could be connected in series with R3 to balance the gain. Consequently, the output current can be controlled eciently with a resolution in the µA range, with output voltages up to few volts.

*II. OPAMP selection.* In general, any OPAMP can be utilized for building a MHCS. However, it's recommended to choose the OPAMP based on the application. In the case of applications where a wide range of output voltages are required, *OPAMPs* with relatively high supply voltages (e.g., ±15V) should be used. Additionally, It's recommended to use rail to rail OPAMP where the output voltage (dynamic range) can swing between the supply voltages' extremes. In the case of low output voltage applications, low voltage CMOS ampliers can be utilized.

For achieving a fast switching current pumps, current feedback ampliers can be used. However, current feedback OPAMP usually operates at low impedance and has insucient CMRR, even below 60 dB. Hence, excellent trimming of the resistors is required to balance the poor CMRR.

*III. Capacitors selection.* Two essential capacitors need to be added to the MHCS. The added capacitors don't aect the amplitude of the output signal; however, they are necessary to suppress the noise that can inuence the circuit's output.


$$f\_{\text{CutOff}} = \frac{1}{2\pi C\_2 \mathcal{R}\_2}.\tag{4.20}$$

The negative feedback capacitor's value should be high in the range of a few nF to narrow the bandwidth and thereby enhances the output stability. On the other hand, a high capacitance will slow down the output. For the applications that require a borad bandwidth, the feedback capacitor's value can be reduced down to few pF.

The virtual capacitance of the MHCS (*COUT* ) can be assessed using Equation 4.21 to estimate the circuit's dynamic behavior (see Appendix B for derivation).

$$C\_{OUT} = \frac{I\_O}{dV\_O/dt} = \frac{R\_3(R\_1 + R\_2)^2}{2\pi r R\_1^2 (R\_3 + R\_4) G BWP}. \tag{4.21}$$

Reducing the virtual capacitance of the circuit results in increasing the output speed. The virtual capacitance can be reduced either by increasing r or selecting an OPAMP with high gain-bandwidth product (GBWP).

### **4.3.3 Electroplating setup**

Table 4.1 reports the actual components of the MHCS used for the construction of the electroplating current source shown in Fig. 4.9. Considering the output current will be owing


Table 4.1: The components used in constructing the modied Howland current source.

through *<sup>r</sup>*, a thick lm power resistor with a rating power of 25 W is used for *<sup>r</sup>*. The maximum current that the circuit can provide is below 370 mA. Accordingly, the resistor's maximum dissipated power is estimated to be ≈ 1.4W, which is way below the resistor's rating power. Thus, the resistor's temperature is ensured not to increase during the operation, keeping *r* at a stable value. For the rest of the resistors, low power resistors with excellent tolerances of ±1% were used; hence no trimming was required. The current source in isolation can operate up to a frequency of more than 1 MHz based on Equation 4.20, upon the used value for the negative feedback capacitor *C*<sup>2</sup> and resistor *R*2.

A high-speed power OPAMP is used with dual supply voltages of ±3.5−±7.5V and can drive output current up to <sup>1</sup>.2 A. The amplier possesses a thermal shutdown safety option. Additionally, it provides a notication signal through a status pin (E/S) in case of thermal shutdown. The thermal shutdown option is essential for the electroplating application as processes can run continuously for hours. A sucient cooling system should be provided to ensure the per-

Figure 4.9: Photograph of the developed electroplating power supply. The power source features an isolated power supply, bipolar *DAC* for generating the control voltage signal, cooling fan, and the voltage-controlled current source (VCCC) based on the modied Howland current source (MHCS). The circuit is attached on the top of the BeagleBone black wireless (BBB) board.

formance's stability and avoid the OPAMP's shutdown condition, mainly when operating a long electroplating process with relatively high current densities.

A cooling fan is used, as shown in Fig. 4.9. Additionally, a cooling pad with holes created in the *PCB* under the amplier following the data-sheet recommendation. Under the operating parameters of *V<sup>S</sup>* = ±5V and *r* = 10Ω, the opamp's maximum dissipation power (*PDmax*) is estimated to be below 1 W, using Equation 4.17. Consequently, the amplier junction temperature in case of maximum power dissipation is below 70 °C at room temperature, which is suciently below the maximum junction temperature of the OPAMP of 150 °C.

A photograph of the full electroplating setup is shown in Fig. 4.10. A self-built copper electrolyte bath (Cuprostar® LP-1, Enthone-OMI GmbH) is used with a composition of 2 M concentration Cu2SO4, <sup>0</sup>.7 mM <sup>H</sup>2SO4, and <sup>0</sup>.009 mM HCL. The bath was continuously stirred and ltered using an electric pump. Schematic of each of the current source modules, as well as the whole circuit's schematic and the electrolyte composition, are available as supplementary material found in Appendix. B.

### **4.3.4 Graphical user interface (GUI)**

A graphical user interface (GUI) is created based on a python script, as shown in Fig. 4.11. The GUI is pre-loaded on the MCU (Beaglebone) and can be accessed using a personal computer. The script is pre-programmed with all the previously mentioned electroplating waveforms in Subsection 4.2.1. The GUI allows the user to dene the surface area required to be plated and the process total time. Then the user can choose from a drop-down list the required type of

Figure 4.10: Photograph of the whole electroplating setup. The setup features copper electrolyte, electric pump for steering and ltering, the developed power source, and connecting wires.


Figure 4.11: Graphical user interface (GUI) for controlling the electroplating current source.

the waveform, which activates a query with the parameters that need to be lled by the user. The user lls the required electroplating current densities (A dm<sup>−</sup><sup>2</sup> ), and the script calculates the required current (mA) based on the specied area and current densities. The developed device gives the advantages of autonomously running a series of dierent electroplating waveforms, allowing depositing several layers with dierent micro-structures, as discussed later in Section 4.4. The total electroplating time, the remaining time, the currently running waveform, and the measured output are displayed on the GUI with a real-time update.The python codes for controlling the current source and the BBB shell conguration le can be found in Appendix B.

### **4.3.5 Current source characterization**

This subsection presents the experimental results of characterizing the circuit's output currents and voltages under dierent load resistances. Additionally, the time response of the developed current source was evaluated during switching between dierent current levels. Furthermore, the circuit was utilized to generate the dierent electroplating waveforms discussed previously in Section 4.2.1. The waveforms were generated with relatively high current densities while being connected to an electroplating setup.

For studying the eect of the load resistance on the circuit's output current, the current source was attached to a resistor kit, and an oscilloscope was connected in parallel with the load resistance to calculate the output current *IOUT* . Figure 4.12 presents the experimental result of investigating the impact of the connected load resistance on the circuit's output current. The relationship between the programmed current (*IIN*; the value inserted to the controller) and the actual measured output current when varying the connected load resistance is demonstrated in Fig. 4.12a. The reference line (black line) in the graph represents the current ideal performance, where the deviation from the reference line is the error value. The error increases proportionally with the circuit's output current value *IOUT* . The measured error values are in the range of 0%–2% until the maximum driving voltage is reached, where the output current reaches a saturation state, after which it does not increase. Figure 4.12b presents the measured maximum output current *IOUT* at each load resistance value. Increasing the load resistance limits the maximum current that the circuit can deliver. Electroplating setups typically have an electric resistance of only a few ohms. Hence, from the experimental characterization, the full range of output currents can be employed for the electroplating process.

Practically all current sources have a saturation value of the output voltage (*VSaturation*). Below *VSaturation*, the circuit's output current is constant, and above *VSaturation*, the circuit's output current follows a decaying curve, as shown in Fig. 4.13b. This behavior is typical for most of the current sources where at low load resistance, the output current is constant and stable. In contrast, as the connected load increases, the output current starts dropping and deviates from the desired behavior. For the developed MHCS circuit, *Vstauration* was measured to be ±4.3 V, which is an excellent performance for an MHCS with an OPAMP supplied with voltages of <sup>±</sup>5 V. This good performance mainly results from carefully designing and choosing the electronics components to satisfy the points mentioned in the component selection Subsection 4.3.2.

During the pulsed electroplating process, the applied current continuously oscillates between dierent values; therefore, the settling time, overshoot, and nal error of the output current

Figure 4.12: Experimental data showing the impact of the connected load resistance (*RLoad*) on limiting the circuit's output current (*IOUT* ). **(a)** The programmed output current (*IIN*) versus its measured actual output current (*IOUT* ) value at dierent load resistances. **(b)** The maximum output current that can be pumped at dierent load resistances.

are of interest. Figure 4.13a illustrates the alterations in these three parameters with respect to the dierence between the output current levels while changing the connected load resistance. Three families of curves are shown, representing the dierent output current levels. Within each group of curves, the connected load resistance is swept. When the output current alternates between relatively close current levels, the circuit takes around 30 µs to settle to its nal value. The settling time decreases as the gap between the output current levels increases. On the one hand, increasing the connected load resistance increases the circuit's settling time as well. On the other hand, a higher load resistance dampens the overshoot value. The overshoot value increases at higher output currents. For the pulsed electroplating process, the pulse width is in the range of ms, and hence, a settling time in the range of µs is acceptable.

The developed current source capability for electroplating was tested by generating the waveforms discussed in Subsection 4.2.1, Fig.4.5, and Fig. 4.6. The current source cathode was connected to a copper electrode, and the anode was connected to a metalized silicon substrate of rectangular shape, with dimensions of 4 cm×1.5 cm. The silicon substrate was covered with 20 nm chromium and 60 nm gold layers. The chromium layer is used to promote the adhesion between the silicon and the gold layer. The chromium and gold layers were deposited using the physical vapor deposition process (PVD) (Univex 450) at deposition rates of 5Ås−<sup>1</sup> and 8Ås−<sup>1</sup> , respectively. The copper electrode and metalized silicon substrate were immersed in a self-made sulfuric acid-based electrolyte bath (Cuprostar® LP-1, Enthone-OMI GmbH). The electrolyte was constantly stirred and ltered by an electric pump setup, as discussed previously in Fig. 4.10.

Figures 4.14 and 4.15 show samples of the waveforms that were successfully generated by the current source when connected to the electroplating bath. During the electroplating process, the pulse width should be enough for the EDL to be charged; otherwise, electrode-position will not happen. The negative pulse width should also be enough for the interface to discharge, as

Figure 4.13: Dynamic behavior and saturation states of the developed current source. **(a)** The experimentally measured settling time of the circuit's output current *IOUT* at dierent current levels and load resistances. **(b)** The experimentally acquired saturation curves, illustrating the output current drop at dierent load resistance.

Figure 4.14: Dierent complex uni-polar waveforms generated by the current source when connected to an electroplating path. The generated uni-polar waveforms are: **(a)** Superimposed pulse; **(b)** Duplex pulse; **(c)** Pulsed pulse; and **(d)** Pulsed on pulse.

discussed previously in Section 4.1. Hence, The pulse widths were chosen in the range of ms to t with the electroplating application. The ripples and oscillations on the measured waveforms are barely observed because of the time scale. The purpose of these experiments was to demonstrate the capability of the developed device under extreme conditions. Therefore, the generated waveforms had relatively high current densities of 6 A dm<sup>−</sup><sup>2</sup> . Moreover, each of the presented waveforms was kept running for more than 2 h. The circuit operated in a stable and robust manner in generating all indicated electroplating waveforms. It was also successful in generating the same waveforms with dierent time intervals and current densities.

Figure 4.15: Dierent complex bipolar waveforms generated by the current source when connected to an electroplating path. The generated bipolar waveforms are: **(a)** Reverse pulse; **(b)** Reverse pulse with o-time; **(c)** Reverse pulsed pulse; and **(d)** Reverse pulsed on pulse.

## **4.4 Electroplating of inkjet-printed exible electronics**

Inkjet-printing a conductive seed-layer onto an unstructured polymer substrate is an essential process step in the eld of exible electronics [51, 65], as discussed in the previous Chapter 3.1.2. The thickness of the printed layer is usually in the range of a few hundred nanometers which is very small for RF applications. Increasing the printed conductive layer's thickness can be realized either by printing multilayers [141] or by electroplating, which is more eective. Moreover, electroplating allows increasing the thickness of the conductive layer to several micrometers [61, 142, 143]. However, the adhesion of printed metallic lms on the exible substrate is particularly challenging, especially after electroplating. During electroplating the printed tracks, a charged layer is created around the electroplated tracks [105]. This charged layer results in high current concentrations nodes along the tracks' corners and edges, resulting in plating stress that leads to peeling o of the structures, as shown in Fig. 4.17. Using pulsed electroplating for manipulating this charged layer is a remedy and one of the main reasons for employing dierentiated electroplating voltage waveforms.

In the following subsections, the developed current source is utilized to solve the adhesion problem accompanying the electroplating of printed conductive tracks on exible substrates.

### **4.4.1 Materials and methods**

Dierent electroplating waveforms were tested to study their impact on the adhesion of inkjetprinted tracks on a exible substrate, as well as on the metallic micro-structure of the deposited layer. Silver rectangular tracks with 10 mm <sup>×</sup> 13 mm dimensions were inkjet-printed on a 125 µm Kapton sheet, as shown in Fig. 4.16. The ink used for inkjet-printing was 30–35 wt.% suspension of silver nanoparticles (Sigma-Aldrich®) with diameters<sup>≤</sup> 50 nm dispersed in triethylene glycol monomethyl ether. The silver ink is rated with a surface tension of 35- 40 mNm<sup>−</sup><sup>1</sup> and a viscosity of <sup>10</sup>-18 mPa s at room temperature. A Fujilm Dimatix DMC-2800 printer was used for printing. A 10 pL print head-cartridge (DMC-11610, Fujilm Dimatix) was lled with the ink and loaded to the printer.

During printing, the substrate plate and the printing head were maintained at a temperature of 25 °C. The printing dot spacing was set to 25 µm, and the printing frequency was set to 5 kHz using one of the 16 nozzles for printing, which resulted in uniformly printed tracks. Before printing, the Kapton foil was rinsed with acetone, isopropyl alcohol (IPA), and deionized water (DI water). After printing, the Kapton foil with the printed tracks was transferred to a hot plate for drying at <sup>80</sup> °<sup>C</sup> for 15 min, followed by sintering in an oven at <sup>150</sup> °<sup>C</sup> for 1 h. Subsequently, electroplating was carried out in the previously discussed setup in Subsection 4.3.3 and Fig. 4.10. Electroplating was performed without adding special additives or heating the bath, as the interest was to control the electroplating outcome by controlling the waveform only. After electroplating, the sample was rinsed using DI water, followed by drying using an

Figure 4.16: Inkjet-printed silver ink on a 125 µm Kapton lm. The printed track is presented for three levels of detail. The inner focused view is a scanning electron micro-graph (SEM).

air gun and placing the sample in a vacuum for 20 min to restrain oxidation.

#### **4.4.2 Results**

First, samples were electroplated by applying DC waveforms with relatively low, medium, and high current densities. The applied current densities were <sup>0</sup>.5 A dm<sup>−</sup><sup>2</sup> for 90 min, 1.5 A dm<sup>−</sup><sup>2</sup> for 30 min and 3 A dm<sup>−</sup><sup>2</sup> for 15 min, leading to deposited copper layers of thickness of <sup>≈</sup> <sup>10</sup>µm. For all samples electroplated using DC waveform, the tracks peeled o from the Kapton substrate, as shown in Fig. 4.17. It was observed that as the applied current density increases, the structures start to roll up due to the internal plating stress.

Then, samples were electroplated by applying pulsed electroplating. Uni-polar and bipolar with o-time waveforms were selected. The uni-polar waveform is dened with three parameters; *IPeak*, *TON*, and *TOFF*, as shown in Fig. 4.4b. The selected parameters of the uni-polar pulsed electroplating were *<sup>I</sup>Peak* <sup>=</sup> <sup>0</sup>.6 A dm<sup>−</sup><sup>2</sup> , *<sup>T</sup>ON* <sup>=</sup> 250 ms, *<sup>T</sup>OFF* <sup>=</sup> 50 ms. The bipolar with o-time waveform is dened with six parameters; *IC*, *TC*, *IA*, *TA*, and the o-times between the pulses (*TOFF* and *T* '*OFF*), as shown in Fig. 4.6a. The selected parameters of the bipolar pulse were *<sup>I</sup><sup>C</sup>* <sup>=</sup> <sup>0</sup>.8 A dm<sup>−</sup><sup>2</sup> , *T<sup>C</sup>* = 300 ms, *I<sup>A</sup>* = 1.5 A dm<sup>−</sup><sup>2</sup> , *T<sup>A</sup>* = 50 ms, *TOFF* = 10 ms, *T* '*OFF* <sup>=</sup> 20 ms. Both types of waveforms have an average current density of <sup>0</sup>.5 A dm<sup>−</sup><sup>2</sup> calculated using Equation 4.10, resulting in the same deposition rate. The pulsed electroplating processes

Figure 4.17: DC electroplating of inkjet-printed silver tracks on a 125 µm Kapton substrate using dierent current densities. The current densities applied are **(a)** low, **(b)** medium, and **(c)** high current. The plating stresses cause the curling of the electroplated lms, which increases with the applied current densities as seen in (b) and (c).

were run for 90 min, achieving deposited copper layers of thickness <sup>≈</sup> <sup>10</sup>µm.

The lm deposited by the bipolar waveform shows a perfect adhesion to the Kapton substrate. On the other hand, the lm deposited by the uni-polar waveform show adhesion better than the DC waveform but not as good as the bipolar plating. The eect of the two waveforms on the metallic micro-structure of the deposited copper layer was investigated with respect to the surface topography, roughness, and grain size, employing three characterization techniques:


Figure 4.18: Pulsed electroplating of inkjet-printed silver tracks on a 125 µm Kapton substrate. The waveform used are **(a)** uni-polar and **(b)** bipolar pulsed electroplating. For each electroplated lm, SEM images are presented.

Figure 4.19: X-ray diractograms of electroplated copper sheets with uni-polar and bipolar waveforms.

*K*λ/(β *cos*θ), where τ is the ordered crystallite's mean size, *K* is the shape factor, λ is the X-ray wavelength (λ = 1.54Å), β is the selected reection's full width at halfmaximum (FWHM), and θ is Bragg's angle. The crystallite sizes for the uni-polar and bipolar plated samples were 30 nm and 307 nm, respectively.

The impact of varying the pulsed electroplating parameters on the metallic micro-structure was investigated. For both uni-polar and bipolar electroplating, samples were plated with relatively low, medium, and high average current densities. The applied average current densities were <sup>0</sup>.5 A dm<sup>−</sup><sup>2</sup> for 90 min, 1.5 A dm<sup>−</sup><sup>2</sup> for 30 min and 3 A dm<sup>−</sup><sup>2</sup> for 15 min, leading to deposited copper layers of thicknesses of ≈ 10µm. The used current densities ratings are equivalent to that used for the DC electroplating shown in Fig. 4.17.

Tables 4.2 and 4.3 summarize the plating printed silver layer results using uni-polar and bipolar waveforms with dierent current densities (mentioned in the footnote of the tables). In uni-polar pulsed electroplating, the SEM images show a very similar micro-structure with increasing the applied current density. Additionally, the WLI measurements show that the surface roughness decreases as the applied current density increases. In the bipolar pulsed electroplating. The SEM shows dierent micro-structures with increasing the applied current density, where high current density leads to conglomerates formation. Moreover, the surface roughness was observed to be increasing proportionally with the applied current density. In conclusion, bipolar electroplating was observed to promote the printed conductive layer's adhesion to the exible substrate compared to the uni-polar plating. However, the uni-polar electroplating results in a better surface nish with roughness below 80 nm other than the bipolar plating, which results in roughness in the µm scale.

The developed power source provides the exibility to run a sequence of electroplating processes where each process has dierent waveforms. With this sequence of the plating process, a lm can be deposited with several layers where each has individually determined microstructures. A series of bipolar and uni-polar pulsed electroplating was performed for an inkjetprinted silver track on a Kapton lm, as indicated in Fig.4.20. This plating sequence combines the advantage of the bipolar plating on promoting the adhesion to the substrate and the advantage of the uni-polar waveform toward a better surface nish. The parameter of the conducted


Table 4.2: The eect of changing the uni-polar pulsed electroplating parameters on the metallic micro-structure of plated inkjet-printed silver tracks on Kapton foil.

c *<sup>I</sup>*Peak <sup>=</sup> <sup>3</sup>.6A dm<sup>−</sup><sup>2</sup> , *T* Total = 15min.

a, b, c *<sup>T</sup>* ON <sup>=</sup> 250ms, *<sup>T</sup>* OFF <sup>=</sup> 50ms.

Table 4.3: The eect of changing the bipolar pulsed electroplating parameters on the metallic micro-structure of plated inkjet-printed silver tracks on Kapton foil.


a *I*<sup>C</sup> = 0.8A dm<sup>−</sup><sup>2</sup> , *I*<sup>A</sup> = 1.5A dm<sup>−</sup><sup>2</sup> , *T* Total = 90min.

b *I*<sup>C</sup> = 2.2A dm<sup>−</sup><sup>2</sup> , *I*<sup>A</sup> = 3.3A dm<sup>−</sup><sup>2</sup> , *T* Total = 30min.

c *I*<sup>C</sup> = 4A dm<sup>−</sup><sup>2</sup> , *I*<sup>A</sup> = 4.2A dm<sup>−</sup><sup>2</sup> , *T* Total = 15min.

a, b, c *<sup>T</sup>* <sup>C</sup> <sup>=</sup> 300ms, *<sup>T</sup>*<sup>A</sup> <sup>=</sup> 50ms, *<sup>T</sup>* O <sup>=</sup> 10ms, *<sup>T</sup>* 'O = 20ms.

Figure 4.20: Double-layer electroplating waveform.

electroplating process is as follow: (1) the bipolar waveform: *<sup>I</sup><sup>C</sup>* <sup>=</sup> <sup>1</sup>.25 A dm<sup>−</sup><sup>2</sup> , *T<sup>C</sup>* = 250 ms, *I<sup>A</sup>* = 1 A dm<sup>−</sup><sup>2</sup> , *T<sup>A</sup>* = 100 ms, *TOFF* = 25 ms, *T* 'OFF <sup>=</sup> 50 ms, and total process time (*TBipolar*) = 15 min; and (2) the uni-polar waveform: *<sup>I</sup>Peak* <sup>=</sup> <sup>0</sup>.8 A dm<sup>−</sup><sup>2</sup> , *<sup>T</sup>ON* <sup>=</sup> 250 ms, *<sup>T</sup>OFF* <sup>=</sup> 150 ms, and the total process time (*TUni*−*polar*) = 75 min.

Both pulses had an applied average current density of <sup>0</sup>.5 A dm<sup>−</sup><sup>2</sup> and the total plating time was 90 min, resulting in a copper lm with a total thickness of <sup>≈</sup> <sup>10</sup>µm. Figure 4.21 shows the SEM images of the printed silver layer as well as both of the deposited layers. SEM reveals both layers' micro-structures, where the rst layer resulted from the bipolar waveform featured a relatively rough surface but resulted in good adhesion of the printed track to the Kapton substrate. In contrast, the upper layer resulted from the uni-polar waveform featured a smoother surface. The roughness of the nal surface was measured using WLI to be 328 nm.

## **4.5 Various applications for the developed current source**

The developed electroplating current source was distributed in our research group and has been used for dierent applications. To show its applicability, three examples are selected and

Figure 4.21: Double-layer electroplating of a printed silver track on a 125 µm Kapton lm. The SEM micrographs are for: **(a)** Printed silver layer; **(b)** Bipolar electroplated layer, which promotes the adhesion of the conductive layer on the exible substrate but results in a rough surface; and **(c)** Uni-polar electroplated layer with a relatively smooth surface for better surface nish.

demonstrated in this section. The experiments shown here were not entirely conducted by myself: laser carbonization of Nomex in Subsection 4.5.1 was done by *Emil R. Mamleyev*, 3D printing of carbon micro-lattice architectures in Subsection 4.5.2 was done by *Monsur Islam*, and the setup presented in Subsection 4.5.3 was designed and built by *Markus V. Meissner*.

#### **4.5.1 Electroplating of a laser carbonized polyaramid-based lm.**<sup>1</sup>

Electroplating substrates with a high electrical resistivity as carbon is challenging compared to electroplating materials with a high conductivity as gold or silver. The challenge arises from the electroplating process nature, where the electrical current should be pumped through the material that needs to be plated. Thus for electroplating material with high electrical resistivity, a powerful current source is required to pump the current through the resistive substrate. Within this subsection, the developed current source was successfully utilized for electroplating carbon tracks on polyaramid-based exible substrate. The deposited copper layer was then used to generate customized exible antibacterial coatings.

A 130±13 µm thick Poly(1,3-phenylene isophthalamide) is used as the base substrate, it's commercially known as Nomex®Type 410 insulation paper (DuPont™; supplied by RS Components). Figure 4.22 summarizes the fabrication process. The Nomex lm was rst rinsed with isopropanol and dried for 10 min at room temperature. The lm was then glued on a glass wafer with an acrylic adhesive sheet (3M-467MP) to prevent the lm from bending during fabrication. Laser-induced carbonisation was performed using a CO2-laser (ULS Versa Laser 3.50, wavelength: 10.6 µm, lens 2.0', beam diameter 120 µm at focus).

Performing carbonization with a single irradiation process results in carbon tracks with very high electrical resistivity. Thus, two dierent carbonization recipes with multiple irradiations were adopted for the manufacturing of the carbon layers. The rst carbonization recipe

Figure 4.22: Schematic illustration of the fabrication ow of electroplating carbonized Nomex. **(1)** Laser irradiation of a Nomex sheet xed on a glass substrate using LC2 and LC3 recipes. **(2)** Copper electroplating of the resulted carbonized tracks on the Nomex sheet. **(3)** Resulted exible copper tracks on the Nomex sheet.

<sup>1</sup>This subsection's text and gures were partially adapted from the author's published work [ON2] with permission of American Chemical Society. Laser Carbonization of Nomex was done by *Emil R. Mamleyev*.

was conducted with a double irradiation procedure, and the resulted samples are denoted as LC2. During the rst irradiation scan, parameters were adjusted to a scanning speed of 3 W, 13 cm s<sup>−</sup><sup>1</sup> , a distance between pulses in a line at ∼65 µm, an interline distance of ∼100 µm, and a focal distance of <sup>2</sup>.5 mm. During a second irradiation scan, the scanning speed was raised to 32 cm s<sup>−</sup><sup>1</sup> , and the distance between pulses was lowered to ∼25 µm. The second carbonization recipe was performed with a triple irradiation procedure, and the resulted samples are denoted as LC3. During the rst irradiation scan, parameters were adjusted to a scanning speed of 3 W, 13 cm s<sup>−</sup><sup>1</sup> , a distance between pulses in a line at ∼25 µm, an interline distance of ∼25 µm, and a focal distance of <sup>2</sup>.5 mm. Subsequently, the same procedures as LC<sup>2</sup> were repeated except for the pulse distance, which was kept at 25 µm for all scans. Subsequently, after carbonization, the LC<sup>2</sup> and LC<sup>3</sup> samples were electroplated using the developed electroplating setup discussed in the electroplating setup Subsection 4.3.3. A DC electroplating was applied with a current density of 1 A dm<sup>−</sup><sup>2</sup> for 5 min. After electroplating, the samples were rinsed with DI water and dried on a hot plate at <sup>95</sup> °<sup>C</sup> for 5 min to prevent copper oxidation by evaporating the residual moisture. Both samples LC<sup>2</sup> and LC<sup>3</sup> were successfully plated with copper. The samples maintained their electrical resistance and remained xed to the Nomex substrate without damage even after bending. Figure 4.23 demonstrates the surface morphology of the carbon layer as well as the plated copper layer. The deposited copper particles have crystallike shapes with lengths over 3.2 µm and 3 µm and widths over 1.13 µm and 1.15 µm for LC2, and LC3, respectively. In the LC<sup>3</sup> sample, most of the deposited crystals were established on a central active site and consistently interconnected with neighboring crystals, providing better electrical stability than LC2. The thickness of the deposited copper layer was determined to be slightly above 3.5 µm.

Figure 4.23: Surface morphology after carbonization and electroplating of Nomex. SEM images of the carbon layer generated by double LC<sup>2</sup> **(a)** and triple LC<sup>3</sup> **(b)** irradiation procedure. SEM images of the copper layer after electroplating of the LC<sup>2</sup> **(c, d)** and LC<sup>3</sup> **(e, f)** samples.

#### **4.5.2 Electroplating of 3D printed carbon micro-lattice architectures**<sup>2</sup>

This subsection introduces another application of the developed electroplating current source for plating 3D carbon structures. The 3D carbon architectures were fabricated by printing 3D structures out of an epoxy resin using stereo-lithography and then carbonized in a nitrogen environment [145]. SLA-3500 laser stereo-lithography system (3D Systems, USA) was used for printing the epoxy resin WaterShed XC11122 (DSM Desotech BV, Netherlands) to realize the 3D architectures. The stereo-lithography system works layer-by-layer, enabling the polymerization of the resin and produce 3D solid complex components. After printing the 3D prototype using stereo-lithography, it was rinsed by immersion in acetone for 2 min and post-cured in a UV oven for around 20 min to improve mechanical properties. The pyrolysis process for obtaining carbon scaolds was conducted in a horizontal tube furnace (Carbolite Giro, Germany). The prototype obtained from the stereolithography process was heat-treated at 900 °C. The steps of the heat-treatment process were [146–149]: 1) heating from room temperature to <sup>900</sup> °<sup>C</sup> under a heating rate of <sup>2</sup>.<sup>5</sup> °C/min; 2) dwell for 1 h at <sup>900</sup> °C; and 3) cooling the sample down to room temperature by turning o the heater of the furnace. The whole heat-treatment process was conducted in an inert atmosphere achieved by a constant nitrogen ow with a ow rate of <sup>0</sup>.8 L/min.

Before electroplating, the 3D carbon micro-lattice architectures were rinsed by immersion in DI water. Electroplating was performed using the developed electroplating setup. A DC

Figure 4.24: Electroplating of 3D printed carbon micro-lattice architectures. **(a)** Photograph of the 3D carbon architectures before and after electroplating with copper. The 3D carbon architectures are fabricated by stereolithography printing then carbonization in an inert environment. **(b-d)** SEM images of the deposited copper layer.

<sup>2</sup>3D laser stereolithography printing of the carbon micro-lattice architectures was done by *Monsur Islam*.

electroplating was employed with a current density of <sup>1</sup>.5 A dm<sup>−</sup><sup>2</sup> for 2 min. After plating, the sample was rinsed using DI water, followed by drying using an air gun and placing the sample in a vacuum for 30 min to restrain oxidation.

Figure 4.24a shows the 3D carbon structure before and after electroplating. The achieved 3D copper cube had a total surface area of ≈ 100mm<sup>2</sup> . Figures 4.24b-d present SEM images of the deposited copper. The deposited copper layer thickness was measured to be ≈ 2µm.

### **4.5.3 Parallel electroplating**<sup>3</sup>

The developed electroplating power source is compatible with the parallelization concept. A PBC was developed possessing eight current channels where each being separately and realtime controllable, as shown in Fig. 4.25. A self-constructed electroplating setup was prepared for parallel electroplating. The setup was build-out of four 3 L glass beakers, each equipped with a single copper anode and two wafers holders connected to the cathodes (power source channels). Holders for the electrical connections, thermometers, anodes, and cathodes (wafers) were cut of PMMA (Poly(methyl methacrylate)) sheets using a laser cutter (VLS2.30, Universal Laser Systems Inc.). The electrolyte allowed plating all of the eight wafers at room temperature at relatively high current densities of up to 4 A dm<sup>−</sup><sup>2</sup> .

Figure 4.25: Electroplating setup for plating eight 100 mm wafers in parallel. The power supply with eight channels is presented in the foreground left, and a PMMA wafer holder is shown to its right. The gure is generated by *Markus V. Meissner*.

<sup>3</sup>The electroplating setup with eight channels shown here was designed and built by *Markus V. Meissner*.

# Part II.

# MRI Application - Interventional Magnetic Resonance Imaging (iMRI)

# **5 State-of-the-art: interventional Magnetic Resonance Imaging (iMRI)**

**C** atheter-based intervention is one of the eective medical methods used in treating and diagnosing cardiovascular diseases. It plays an essential role in the eld of minimally invasive medicine [150, 151]. Over the past two decades, cardiovascular diseases have been reported as the leading cause of death globally, with more than seventeen million fatalities in 2017 surpassing all other medical causes combine [14]. Patients suering from large vessel occlusion should be treated fast as it leads to death or long-term disability. Catheter-based endovascular embolectomy is an eective treatment where the doctors use a catheter under the visualized tracking approach to control its path through the vessels [152]. Hence, providing a surgeon with a real-time precise location of the catheter's tip within the patient's body is a prerequisite for further advancement of minimally invasive medical operations. Tracking the catheter can be achieved using real-time X-ray uoroscopy. However, X-rays aect the patient and the medical sta with doses of potentially harmful ionizing radiation [153]. Moreover, tracking catheters using X-ray suers from poor contrast of soft tissues, making the method unsuitable for complex interventional procedures [150].

Hence, tracking a catheter's tip within a magnetic resonance environment has been a topic of great attention during the past decade and has been considered a promising approach for interventional diagnosing and immediate treatment [154–156]. Several approaches have been developed to track catheters in an MR environment, categorized into passive and active techniques. The passive approaches depend on the magnetic susceptibility dierence between the catheter and the body tissues, which results in a visualized location of the catheter [157]. Therefore, it requires either susceptibility artifacts or contrast agents [158]. Although passive approaches are easier to implement and cost-eective, they are less robust and lack reliability than other approaches [159]. The second approach is electrically active, where additional hardware such as resonators (coil || capacitor) [160,161] or RF antennas [162,163] are placed on the catheter's tip resulting in a traceable visible spot in the MR image. The active approaches are divided into active and semi-active methods. In the active methods, the micro-resonators on the catheter tips are connected to the MRI scanner through a conductive wire for data transfer. On the other hand, the semi-active method eliminates conductive wires where the catheter's resonator provides a resonant inductive coupling eect, resulting in a wireless local signal amplication of the *B*<sup>1</sup> eld.

However, installing an RF receiver coil of the MRI to the catheter still faces some unsolved issues concerning patient safety and catheter visualization, which has prevented this tech-

This chapter's text and gures were reproduced and adapted with permission from the author's published works [ON3] (CC BY 4.0, 2020 IEEE) and [ON6] ( Copyright © 2019, IEEE)

nology from nding widespread use. The excessive RF heating and high induced currents in the micro-resonator during transmission of the RF eld can harm the patient by dangerously heating the surrounding tissues [164, 165] and may deliver an electric shock to the patient. The implanted conductive coil and transmission line inside the human body acts as an electromagnetic antenna [166], which is sensitive to both electric and magnetic elds, although the MRI detector should be sensitive only to the nuclear magnetic eld.

## **5.1 Heating problem**

Heating conductive electrodes or implants inside the human body during MRI procedures may lead to immediate and severe damage to the surrounding tissues. When subjected to a temperature above 44 °C, tissue injury will happen due to the disintegration of proteins. The alternating *B*<sup>1</sup> eld is the electromagnetic eld in MRI that contributes the most to the heating of conductive implants. The *B*<sup>1</sup> magnetic eld generates an electric vortex eld which induces eddy currents inside the conductive parts. This eddy current leads to direct heating of the conductive subject. The frequency-dependent skin eect denes the eective resistance, thus the conductor's thickness determines the heating generated by the eddy current.

When it comes to long conductive wires, the scattered electric eld has a more signicant impact. The *B*<sup>1</sup> elds induce surface currents into the conductive wire, resulting in a scattered electric eld. The magnitude of the scattered electric eld is dened by the length of the conductive wire, diameter, insulation material and thickness, and tip geometry [167]. The magnitude of the scattered electric eld is magnied at the tip of the conductive wires, resulting in the maximum harmful heating at the wire's tip [167, 168]. This eect of the scattered electric eld becomes very obvious when the wire length (*Lwire*) is in the range of the wavelength λ of the alternating *B*<sup>1</sup> eld; *Lwire* = λ/4,λ/2,λ,3λ/2,...etc. Park et al. [168] showed mathematically and experimentally that the resonant length (length at which maximum heating occurs) of a bare wire in a <sup>1</sup>.5 T MRI scanner is 20 cm, while, for insulated wire, the resonant length exceeds 40 cm. Konings et al. [165] tested dierent lengths of conductive wire with the same conguration in a <sup>1</sup>.5 T MRI scanner. The temperature rises from <sup>37</sup> °<sup>C</sup> to <sup>72</sup> °<sup>C</sup> by increasing the wire length from 45 cm to 85 cm.

A conductive material in an MR environment absorbs a certain quantity of the transmission power, leading to an increase in its temperature. The specic absorption rate (SAR) denes the quantity of the absorbed power per mass. The SAR distribution can be derived from the peak amplitude of the electric eld (E) by [24]

$$SAR = \frac{\sigma |E|^2}{2\rho},\tag{5.1}$$

where σ is the conductivity and ρ is the mass density. Equation 5.1 can be rewritten as a

function of the parameters of the MR sequence (this simplied form is adapted from "From picture to proton, ch. 20.2 [24]")

$$SAR = \frac{\sigma \pi^2 r^2 f^2 B\_1^2}{2\rho} D,\tag{5.2}$$

Where *f* is Larmor frequency which is in turn function in *B*0, *D* is the duty cycle which is the fraction of the total scan time with RF eld operating, and *r* stands for the sample size. The previous equation shows that SAR increases with the square of *B*<sup>0</sup> and *B*<sup>1</sup> magnetic elds. Moreover, SAR increases linearly with the number of RF pulses per cycle (duty cycle).

## **5.2 Active MR-catheter markers**

This section presents an overview of the state-of-the-art MR-catheters, focusing on the introduced approaches for eliminating the long conductive transmission wires in MR-catheters to avoid the heating hazard. Ladd and Quick [169] used RF coaxial chokes for reducing the RF heating. An RF choke is a type of inductor that blocks high-frequency signals while it allows only low frequency or DC signals to pass. Several techniques could be employed to implement the chokes in a transmission cable, including winding loops around the cable, which add self-inductance to the system. Another possibility is adding discrete passive components like capacitors into the cable. However, windings and discrete passive components add challenges for interventional catheters due to their geometry constraints. An alternative is using a coaxial sleeve with a specic length (λ/4). Several coaxial chokes can be added within the transmission cable, where each choke participates in increasing the impedance and reducing the amplitude of the RF standing wave. Figure 5.1d shows a schematic diagram of two coaxial chokes in series integrated around the transmission line. In this design, a triaxial cable is used, where the chock is prepared by soldering the secondary shield to the inner shield and then a small segment of the secondary shield is removed after a length of λ/4. The other end of the secondary shield (chock) is left open. The coaxial chokes increase the impedance for the highfrequency currents owing through the outside shield layer of the coaxial cable, eliminating unwanted currents. On the other hand, chokes do not aect the impedance within the conductor layer for low-frequency currents. After integrating a choke in the cable, the spot of the maximum electric eld is shifted near to the choke, resulting in high impedance and reducing the standing wave. Consequently, the change in temperature is spatially distributed based on the chokes locations to avoid spots with high temperatures. In this approach, the catheter needs to be well insulated as it can deliver electric shock at the open end of the chocks.

Weiss et al. [170] reduced the electrical length of the conductive transmission wire by integrating miniaturized transformers within the wire, as shown in Fig. 5.1a. The conductive wire was divided into small segments, and the signal transfer between the segment was done using miniaturized transformers. The reduction of the conductive wire length restricts the induced electric eld in the wire, thus reduces the generated heating. The miniaturized transformers were single loop transformers fabricated on a multi-layer PCB with a 127 µm gap between the two transformer coils. The results showed a reduction of the heat to a safe level. However, a loss in the transmission signal occurs but still sucient for the active marker tracking purpose.

Syms et al. [171,172] used thin-lm magneto-inductive (MI) waveguides to transmit the signal instead of conductive transmission wire. The used MI waveguide had a periodic array of lowfrequency resonators (LC circuits), as shown in Fig 5.1c. The LC circuits were structured on both sides of a thin-lm substrate alternately, xed around the catheter tube by the mean of heat-shrink tubing. The length of the elements was chosen to be short enough to protect the catheter against the external alternating B1 eld and the induced electric elds. The system was developed to overcome the heat and orientation problems. The orientation problem is the loss of coupling of the catheter receiver during its movement because of bending or axial orientation.

Fandrey et al. [173] used laser light and optical bers to transmit the MR signal instead of the conductive wire to tackle the heating problem. However, the optical approach requires placing bulky components on the catheter's tip, which added another challenge to the iMRI system. The system consists of a resonant receiver circuit for acquiring the MR signal, an optical modulator on the catheter's tip, and the optical power supply. Figure 5.1b shows a schematic gure for the optical system for MR-catheter. The received MR signal is fed directly to the optical modulator that modulates the current via a laser diode. The power for the optical modulator on the catheter's tip is transmitted through another optical bers. The bers are located around

Figure 5.1: Examples of state-of-the-art signal transfer in MR-catheter. **(a)** Integrating miniaturized transformers to divide the conductive transmission wire into small segments, *reprinted from [170] with permission of John Wiley & Sons, Inc. (Copyright © 2005 Wiley-Liss, Inc.)*. **(b)** Employing an optical system to transmit the MR signal instead of using conductive wire, *reprinted from [173] with permission of John Wiley & Sons, Inc. (Copyright © 2011 Wiley-Liss, Inc.)*. **(c)** Utilizing thin-lm magneto-inductive (MI) waveguides to transmit the MR signal, *reprinted from [172] with permission of IEEE (Copyright © 2013 IEEE)*. **(d)** Using RF coaxial chokes to block the high-frequency currents, *reprinted from [169] with permission of John Wiley & Sons, Inc. (Copyright © 2000 Wiley-Liss, Inc.)*.

the catheter tube. The resonator is passively decoupled using a pair of diodes on the catheter's tip. This technique ensured reliable tracking of the tip of the catheter and high contrast visualization. However, the optical method adds handling diculties because of the mechanical connection between the scanner and the catheter using the optical ber, where bers manipulation, bending, or rotation should be avoided. Sarioglu et al. [174] tackled the drawback of the bulky components of the optical system by integrating metal-oxide-semiconductor photovoltaics. In addition, the interface circuit between the optical and electronic components was developed as an o-chip MEMS-based resonator array to reduce power consumption. A photovoltaics circuit as a power supply unit is also integrated to convert the input optical power to drive the CMOS chip. The CMOS chip is responsible for detecting the input signals then creates a PWM electrical current.

### **5.2.1 Inductive coupling**

An alternative approach for tracking the catheter is to wirelessly couple the resonator on the catheter's tip through inductive coupling to an RF surface coil outside the patient [175– 179]. The surface coil is conventionally connected to the MRI scanner through a matching and tuning circuit, as shown in Fig. 5.2. A wireless coupling is implemented between the catheter's resonator and the surface coil to avoid any long conductive cables inside the magnet to overcome the heating problem. Tracking the catheter using inductive coupling has proven to be reliable, delivering high contrast visualization.

The micro-resonator on the catheter's tip is tuned to the Larmor frequency, so the exciting

Figure 5.2: interventional Magnetic Resonance Imaging (iMRI) through inductive coupling for safe minimally invasive medical operation.

magnetic eld results in high induced current into the micro-resonator. This current generates a local magnetic eld whose ux depends on the coil geometry. The local magnetic eld amplies the MR signal and enhances the tracking image where the micro-coil is shown as a bright spot in the MR image. Consequently, the heat problem of the long wire connection between the catheter and the scanner is eliminated. The only source of heat is the induced current in the resonator.

Multiple surface coils (coils array) can enhance the acquired MR image, which can be postprocessed using image processing techniques. The magnetic *B*<sup>1</sup> elds of the catheter's coil and the surface coil should be aligned for maximum coupling. Otherwise, the coupling decreases as the angle between the *B*<sup>1</sup> eld increases, resulting in losing visibility in orthogonal *B*<sup>1</sup> elds. Recently, various micro-coil geometries have been adapted as a catheter's micro-resonator, including a surface loop coil, at spiral coils, two opposing solenoids, a double helical inductors, and a surface loop coil. Figure 5.3 shows an overview of the previously implemented coils for MR-catheter. All proposed designs have xed capacitance except for the design introduced by Ellersiek et al. [177], where a exible resonant circuit was developed and fabricated, including a at spiral coil and tunable capacitor. The tunable capacitor consists of a main capacitor and detachable capacitors. Calibration is performed manually by cutting parts of the detachable capacitor using a scalpel. After calibration, the resonant circuit is coated by a bio-compatible layer of Parylene-C then attached to the catheter.

### **5.2.2 Marker designs**

The developed markers (resonators) for the iMRI application should feature several functionalities, including bio-compatibility, sterilization, reliability and validity for tracking, and patient safety. The MRI active marker should be tuned at Larmor frequency [180]. Then during the RF excitation phase, the nuclear spins deect from their equilibrium position with a ip angle proportional to the alternating magnetic eld *B*1. The MRI marker (resonant circuit on the catheter's tip) is coupled with the transmit coil, and RF current is induced through the microresonator coil. This current generates a magnetic eld around the micro-coil that enhances the nuclear spins, enhancing the local magnetic eld and the ip angle.

The MRI marker design is inuenced by the main *B*<sup>0</sup> eld orientation and amplitude which denes the Larmor frequency and the required orientation of the resonator's *B*<sup>1</sup> eld; *B*<sup>0</sup> and *B*<sup>1</sup> elds should be perpendicular. Dierent shapes of resonant MRI markers are shown in Fig 5.4. For example, in the case of closed MR scanners, the *B*<sup>0</sup> eld is parallel to the interventional instrument, so the saddle, spiral, and single loop coils are preferred. On the other hand, for open MR scanners, cylindrical loops are preferred as the *B*<sup>0</sup> eld is perpendicular to the interventional instrument.

Additionally, for each application, the orientation, geometries, surface material, and stiness

Figure 5.3: Examples of state-of-the-art MR-catheter using wireless inductive coupling. **(a)** A planar rectangular spiral coil attached to a tuning non-magnetic SMD capacitor xed around the catheter using medical-grade heatshrink tubing, *reprinted from [179] with permission of Elsevier (Copyright © 2014 Elsevier Inc.)*. **(b)** A single loop coil attached to the tuning capacitor is placed around the catheter and coupled to the RF surface coil, *reprinted from [178] with permission of John Wiley & Sons, Inc. (Copyright © 2005 Wiley-Liss, Inc.)*. **(c)** A spiral square coil with the capacitor fabricated on thin-lm polymer and rolled around the catheter (the 2D structure of the resonator is shown in Fig. 5.5a), *reprinted from [177] with permission of Elsevier (Copyright © 2009 Elsevier B.V.)*. **(d)** Two perpendicularly oriented saddle coils were deposited on a polyimide lm and rolled around the catheter, *reprinted with permission from [ON3] (CC BY 4.0, 2020 IEEE)*. **(e)** A double helix inductor on a polymer catheter achieved using aerosol jet deposition. *reprinted from [175] with permission of IEEE (Copyright © 2019 IEEE)*. **(f)** Two separate resonators rolled at a 45° angle with the catheter's axis to decouple them. The resonator can be positioned either adjacent or on top of each other, *reprinted from [176] with permission of John Wiley & Sons, Inc. (Copyright © 2003 Wiley-Liss, Inc.)*.

of the interventional instrument dene the coil design, fabrication technology, and the required adhesion material. For example, for soft intervention into the body, on the one hand, the catheter end should be soft to avoid damaging the blood vessels. On the other hand, it

Figure 5.4: Dierent shapes of resonant MRI markers used for MR-catheters [180]. **(a)** Cylindrical coil. **(b)** Singleloop coil. **(c)** Spiral coil. **(d)** Saddle coil.

should be sti enough to support push ability. Hence, the size of the micro-coil is important to support the catheter properties. Moreover, optimizing the coil's geometry denes its self-resonance frequency and quality factor.

The basic conguration of a resonant circuit consists of a coil and capacitor in parallel to tune the resonance frequency ( *f* = 1/ √ *LC*). Using SMD capacitors is not preferred as they are bulky and not bio-compatible. Other approaches have been to integrate the capacitor, including a cylindrical or roll-up capacitor [177]. Multi-layer capacitors can also be fabricated along with the coil to increase the capacitor aspect ratio to be suitable for integrating on an interventional instrument [72], as shown in Fig. 5.5. Another good approach is using small capacitors with detachable connections [177], allowing editable capacitance for ne-tuning purposes.

### **5.2.3 Marker fabrications**

Dierent fabrication technologies have been considered for developing resonance markers around the catheter tube, including wire-winding, thin-lm technology, inkjet-printing, hot embossing technology, and aerosol deposition process. Kaiser et al. have compared the dierent fabrication technologies for iMRI in their review paper [180]. The wire-winding technology is a simple 3D manufacture process where a wire is wound around a tube with the required geometry then soldered to an SMD capacitor. However, the wire-winding process faces many issues, including poor reproducibility, instrumentation diameter enlargement, restriction to cylindrical coils, and good xation techniques.

Thin-lm technology includes the patterning of the target 2D structure of a conductive layer on a carrier insulating lm, then roll it around a tube to achieve the required 3D structure. This technology includes micro-fabrication methods such as spin coating, photoresist patterning, masked UV light exposure, etching, galvanization, and cleaning processes. On the one hand, thin-lm technology provides accurate structures, exible substrate, and high reproducibility. However, on the other hand, it is costly and requires expensive equipment.

The hot embossing fabrication process [181] is used to achieve a conducting structure onto a thermoplastic instrument using negative embossing stamps under suitable temperature and pressure. First, the stamp is pressed against a metallic foil that is mounted around the instru-

Figure 5.5: Examples of state-of-the-art fabricated capacitors for MR-catheter. **(a)** A manually editable tuning capacitor fabricated on a polymer substrate, *reprinted from [177] with permission of Elsevier (Copyright © 2009 Elsevier B.V.)*. **(b)** A fabricated xed multi-layer tuning capacitor, *reprinted from [72] with permission of Elsevier (Copyright © 2015 Elsevier B.V.)*.

ment. Then the stamp punches out the unneeded parts of the metallic foil, leaving behind the required structure. This technique provides high reproducibility of mechanically stable structures with low cost. However, the base instruments should be made of thermoplastic, and 3D structures can be challenging to be fabricated.

The aerosol-deposition [175] process is used to deposit catalytic nano ink to fabricate 3D structures directly on the instrument. This process can print conductive, insulating, and adhesive layers. Aerosol Jet deposition technology can deposit structures independent of the base substrate shape and material. However, the resulted structures have low mechanical stability, and the process requires high-cost materials.

# **6 Novel wireless decoupling system**

**C** atheters can be visualized in the MR environment during minimally invasive surgery by locating a micro-resonator on the tip of the catheter. By tuning the resonator to the Larmor frequency of abundant nuclei and exposing it to the RF magnetic eld, its motion can be tracked in real-time under direct image guidance. However, the conductive wires that connect the resonator to the MRI spectrometer, for signal transfer, tuning and matching, is known to cause dangerous tissue heating. A promising approach is to track the resonator wirelessly using a wired resonator outside of the body through inductive coupling. The matching and tuning network is then connected to the external resonator. The inductive coupling approaches face two challenges: i) during the RF excitation stage, a strong electric eld is induced inside the MR-catheter resonator, leading to dangerous heating. However, the resonator should be sensitive only to the nuclear magnetic eld during the receiving (sensing) stage. ii) As the catheter axial orientation changes during motion, the coupling between the MR-detector and the surface coil will be aected, which results in a dierence in the catheter visibility. Visibility can be lost if the resonator and external coil elds *B*<sup>1</sup> are orthogonal. In this contribution, a novel wireless decoupling system is introduced where the resonance frequency of the resonator on the catheter tip can be tuned wirelessly. Consequently, the resonator's resonance frequency is controlled via a pulse sequence sent to an RF microelectronic system integrated with the detector coil to couple it during the acquisition stage. Moreover, a novel detector design is proposed based on two perpendicular miniaturized saddle coils that maintain constant coupling between the external coil and the MR-catheter detector at all axial orientations.

This chapter is divided into three main sections. The rst section 6.1 explains the working concept of the wireless decoupling system. The second section 6.2 presents the technical details of the dierent electronic building blocks of the system proposed for decoupling the MRcatheter resonator wirelessly. Within this section, the fabrication and characterization of the developed system are also discussed. The third section 6.3 presents a novel double saddle coil for wireless tracking of the MR-catheter during interventions with orientation independence. This section discusses the modeling and design of the proposed double detector. Moreover, the performance of the detector is characterized using numerical simulations and conrmed with experiments.

## **6.1 System description**

Figure 6.1 demonstrates an overall schematic diagram of a minimally invasive medical operation under iMRI through inductive coupling. It also shows a detailed illustration of the

This chapter's text and gures were reproduced and adapted with permission from the author's published works [ON3] (CC BY 4.0, 2020 IEEE) and [ON6] ( Copyright © 2019, IEEE)

Figure 6.1: Overall schematic diagram of a minimally invasive medical operation under iMRI through inductive coupling with implementing wireless detuning system. [Bottom] The adapted MRI pulse sequence and RF connections. [Top] A detailed illustration for the developed micro-electronic system. Two surface coils are used, one coil to transmit and receive the MR signal and the other coil for transmitting a control signal. Two microresonators are positioned on the catheter tip, one resonator for collecting the nuclear magnetic signal to be visualized in the MR image. The other micro-resonator pickups the control signal to tune the rst micro-resonator.

developed microelectronics system as well as the used MRI pulse sequence. The developed system for tracking the catheter tip consists of four resonators; two relatively big surface coil outside the body and two micro-resonators on the catheter tip inside the body. Each of the four resonators has a specic functionality:


signal fed to a variable capacitor (varicap) to control the tuning frequency of the other MR-catheter resonator. This coil is referred to as the *"control signal receiver"*.

4. The other micro-resonator on the catheter tip is tuned to a frequency close to the Larmor frequency and is connected in parallel to the varicap. Hence, the DC control signal fed to the varicap can tune it to the Larmor frequency to be visible in the MR image. This coil is referred to as the *"MR-catheter detector"*.

Initially, the MR-catheter detector is not tuned to the Larmor frequency for considering failsafe condition; therefore, no electric eld is generated during the excitation phase, preventing the heating hazard. Then during the relaxation stage, a TTL pulse is sent by the MRI console, which switches on the control signal transmitter coil to transmit the control signal. Consequently, the control signal is picked up by the control signal receiver on the catheter tip. A half-wave rectier generates a DC signal from the received control signal, which is then fed to a varicap to tune the MR-catheter detector to the Larmor frequency. Hence the MR-catheter detector becomes sensitive only to the nuclear magnetic eld and could be visually detected in the MR image with eliminating the heating eect.

## **6.2 Wireless decoupling system**

Figure 6.2 presents an equivalent schematic circuit of the developed wireless decoupling system. Three electric circuits represent the whole MR-catheter tracking system with the integrated wireless decoupling feature:

1. The equivalent circuit of the MRI surface coil (LS1), as shown in Fig. 6.2 (shaded in blue). The coil is connected to the MRI scanner through the matching and tuning network (CM1 and CT1). For connecting the coil to the MRI scanner, its impedance should match the transmitting cable impedance (50 Ω). Hence CT1 is used to tune the surface coil

Figure 6.2: Equivalent schematic circuit of the wireless decoupling system for safe MR-catheter tracking. The equivalent circuits of MRI surface coil resonator [Blue], control signal transmitting resonator [Green], and wireless tunable detector on the catheter tip **[**Red].

( *fLS*<sup>1</sup> = 1/ √ *LS*1*CT*1) and set the impedance real part to 50 Ω, while CM1 eliminates the residual imaginary part.


$$f\_{\text{Court}} = \text{F}\_{\text{S}2} = 1/\sqrt{L\_{\text{S}2} \text{Cr}\_2} = \text{F}\_{\text{C}2} = 1/\sqrt{L\_{\text{C}2} (D\_2 \text{C}\_2/D\_2 + C\_2)}.\tag{6.1}$$

The micro-resonator (LC1) is tuned using D<sup>1</sup> and C<sup>1</sup> to a frequency close to the Larmor frequency to be coupled to the MRI surface coil during the acquisition stage to appear in the MR image;

$$f\_{Larmor} = F\_{L\_{\odot 1}} = 1/\sqrt{L\_{\odot 1}C\_{\text{T}}} = F\_{\text{Cl}\_{\text{Cl}}} = 1/\sqrt{L\_{\text{Cl}}(D\_{\text{l}}C\_{\text{l}}/D\_{\text{l}} + C\_{\text{l}})}.\tag{6.2}$$

The D<sup>2</sup> and C<sup>2</sup> conguration create a half-wave rectier. Hence it generates a DC signal from the received control signal, which in turn is fed to D<sup>2</sup> (varicap) to change the resonance frequency of the MR detector (LC1). The resistance (r2) is added to the ground line to isolate the electric elds of the two micro-resonators so that they do not interfere, i.e., r<sup>2</sup> should be in the MΩ range. The resistance (r1) is added in parallel to D2, so once the control signal is switched o, D<sup>2</sup> can dissipate the provided dc signal to the ground line and returns to its initial. The value of r<sup>1</sup> should be high enough to force the DC control signal to pass through D<sup>2</sup> during the tuning phase; i.e., r<sup>1</sup> should be in the MΩ range.

### **6.2.1 Fabrication of the wireless decoupling device**

The whole microelectronics circuit on the catheter tip was fabricated using the thin-lm fabrication processes described in Chapters 3 and 4. The dimensions of the fabricated device with its electronic components are dened mainly by the operating frequencies of the MR signal and control signal. The MR signal frequency is determined by the Larmor frequency. On the other hand, the control signal frequency could be adjusted to satisfy the t to the application. The control signal should operate at a frequency away from the Larmor frequency to not interfere with the MR signal. Hence, in principle, the control signal could operate above or below the Larmor frequency. Operating above Larmor frequency provides the advantage of a low tuning capacitor (smaller dimension), but rectifying thin-lm diodes are limited to low frequency. On the other hand, operating below Larmor frequency, the signal could be recti ed using a thin-lm diode, but the coil requires high capacitance for tuning. Therefore, an operating frequency of 20 MHz was chosen for the control signal to be rectied using a thin lm diode. In addition, the resonance frequency of the LC2 resonator is reduced to 20 MHz by increasing the coils' inductance (increase outer dimensions and the number of turns), as shown in Fig. 6.3a. For both resonators (MR-catheter detector and control signal receiver), a spiral surface coil was used. Table 6.1 summarizes the dimensions of both resonators with the experimentally measured inductance.

Based on the inductance of both resonators, the required capacitance was calculated to tune the resonator for the intended frequencies. The MR-catheter detector LC1 was designed to have a resonance frequency of 60 MHz (18 MHz above the Larmor frequency in a 1 T MRI scanner). Hence, considering the capacitance of D<sup>1</sup> of 62 pF, C<sup>1</sup> was designed (by setting the capacitor area and selected insulating material) to have a capacitance of 196 pF. The control signal resonator LC2 was designed to have a resonance frequency of 20 MHz (more than 20 MHz away from Larmor frequency in a 1 T MRI scanner). Hence, the capacitance of C<sup>2</sup> and D<sup>2</sup> was designed to be 379 pF and 389 pF, respectively. Table 6.1 summarizes the designed dimensions and the expected values of the fabricated capacitors and diode.

The capacitors C<sup>1</sup> and C<sup>2</sup> were fabricated using the thin-lm technology developed in Section 3.2, where an insulating lm is inkjet-printed between two conductive layers. The rectifying diode D<sup>2</sup> was fabricated using the technology developed in Section 3.3, where a ptype semiconductor is spin-coated between two conductive sheets having dierent working functions (WF). The resistors r <sup>1</sup> and r<sup>2</sup> were fabricated by drop-casting of the conductive polymer PEDOT:PSS (Poly(3,4-ethylenedioxythiophene)-poly(styrenesulfonate), Sigma-Aldrich) between two discontinuous conductive traces. PEDOT:PSS polymer was employed to build the resistors as it possesses higher resistivity than most other inorganic material, resulting in the required MΩ resistance. Moreover, it can be processed in solution with stability in

Figure 6.3: Schematic illustration of the designed electronic device for controlling the MR-catheter resonance frequency wirelessly. **(a)** Top view of the device with the bottom copper layer and top silver layer. **(b)** Isometric views of the capacitor, resistor, and diode.


Table 6.1: The geometrical dimensions and value of each electronic component included in the micro MR-catheter device.

<sup>a</sup> Measured experimentally using snier coil and network analyzer.

<sup>b</sup> Calculated according to: *Capacitance* = ε0ε*rA*/*d*.

the ambient atmosphere. For the varicap D2, a discrete component was used (BB640E6327HTSA1, Inneon). However, fabricating a varicap using thin-lm technology is under investigation in our group.

### **6.2.2 Experimental result of the wireless decoupling system**

First, the bottom layer was inkjet-printed using silver ink on a 25 µm Kapton lm. The bottom layer contains the two resonators (LC1 and LC2), the bottom layer of the capacitors (C<sup>1</sup> and C2) and diode D2, and discontinuous lines for building the resistors (r<sup>1</sup> and r2). The bottom layer was then electroplated with a 10 µm copper layer, as shown in Fig. 6.4a. The copper layer is required to reduce the RF resistance and provide a low WF material for the diode. Next, on top of the C<sup>1</sup> pad, an insulating layer is printed using BCB/TiO<sup>2</sup> composite ink (20wt.%, ε*<sup>r</sup>* = 7) developed in Section 3.2. This ink was selected as C<sup>1</sup> has a relatively low capacitance (<sup>≈</sup> 196 pF). On the other hand, for C2, the composite ink BCB/TiO<sup>2</sup> (40wt.%, <sup>ε</sup>*<sup>r</sup>* <sup>=</sup> <sup>20</sup>) was used as it has relatively high capacitance (<sup>≈</sup> 379 pF). Then, on top of the diode D<sup>2</sup> pad, a layer of the PTAA semiconductor was spin-coated (measured thickness= 410 nm, <sup>ε</sup>*<sup>r</sup>* <sup>=</sup> <sup>3</sup>). Consequently, on top of the insulating and semiconductor layers, a silver layer was inkjet-printed. Then, the resistors were achieved by drop-casting of the PEDOT:PSS polymer. Figure 6.4b presents the

Figure 6.4: Fabricated device for wirelessly tuning the MR-catheter detector resonance frequency. **(a)** Bottom layer achieved by silver inkjet-printing and copper electroplating. **(b)** Full device after depositing capacitors insulator, diode semiconductor, top silver layer, and resistors conductive polymer.

nal device before rolling around the catheter tube. Finally, the device was rolled around an 8 fr catheter tube.

The measured tuning frequency of the MR-catheter resonator and control signal receiver resonator were <sup>56</sup>.5 MHz and 23 MHz. Figure 6.5 presents an experimental setup to verify the functionality of the fabricated device for wireless tuning. First, a macro resonator (mimic LS2 in Fig. 6.2) tuned to 23 MHz and connected to a function generator placed 1 cm away from the control signal receiver coil LC1. Moreover, a snier coil was positioned facing the MRcatheter detector LC1 and connected to a network analyzer (Agilent E5071C) for measuring the detector resonance frequency. The snier coil enables detecting the resonance frequency of a resonator without any direct wire connections [182, 183]; a description of the snier is attached in appendix C.

The plot in Fig. 6.5 shows the measured reection coecient (S11) of the snier coil. Each peak represents the resonance frequency of the MR-catheter detector at dierent voltage levels supplied to the control signal transmitter coil via the function generator. Thus, the developed system successfully tuned the resonance frequency of the MR-detector wirelessly, which proves that it can decouple the resonator inside the patient wirelessly. The resonance frequency could be tuned within a 15 MHz range equivalent to a 7 V range of the control signal amplitude. Increasing the control signal amplitude over 7 V did not further change the resonance frequency, indicating that the varicap reached its limit. It was observed that as the amplitude of the control signal increases, the resonance frequency decreases, which indicates that the varicap D<sup>1</sup> capacitance increases with the control signal. The increase in the varicap capacitance implies that it is supplied with a negative voltage. This is because the D<sup>2</sup> conguration (silver-PTAAcopper) cuts the positive part of the control signal and recties the negative part of the signal.

Figure 6.5: Experimental results of tuning the MR-catheter detector wirelessly. **(a)** Reection coecient of the snier coil, where each peak represents the MR-catheter detector resonance frequency at dierent amplitudes of the control signal. **(b)** Photo for the testing setup.

Nevertheless, if it is required that the resonance frequency increase with the control signal, the diode conguration could also be inverted (copper-PTAA-silver). However, this was done to simplify the fabrication process; otherwise, more than two printed conductive layers will be required. Moreover, the working principle of the system is still valid.

The introduced fabrication process's overall yield is relatively low, out of a batch of 20 devices, only four of them work as expected. This is due to the complexity of the device and the involvement of several materials (conductor, semiconductor, insulator) in the fabrication process. The main problem with inkjet-printing that each device is fabricated separately, and the operators skills signicantly aect the result. Hence, one solution would be utilizing a dierent process compatible with mass production where the operators skill is marginalized, such as screen printing or roll-to-roll processes.

## **6.3 Double-saddle micro-resonator**

In the introduced system, the catheter tip is being tracked inside the patient by wirelessly coupling a micro-resonator on the catheter tip with an outside surface coil through inductive coupling. The surface coil in this system is connected to the MRI scanner through a matching and tuning network. The maximum coupling between the micro-resonator and the surface coil occurs when their *B*<sup>1</sup> magnetic elds are aligned. At this arrangement, with both *B*<sup>1</sup> elds aligned, the maximum visibility of the catheter tip in the MRI scanner is achieved. However, as the catheter axial orientation changes during the medical operation, the resonator's visibility varies and can be lost when the *B*<sup>1</sup> elds of the catheter micro-resonator and the surface coil are approximately orthogonal [176,178]. Additionally, as the magnetic coupling changes with the catheter motion and orientation, the entire system demands to be re-matched and tuned.

In the following subsections, the design and fabrication of a micro-resonator based on a double saddle coil is proposed. The developed micro resonator maintains almost constant maximum coupling between the *B*<sup>1</sup> elds of the micro-resonator on the catheter tip and the external surface coil throughout the entire range of axial orientations.

### **6.3.1 Modeling and simulation of inductive coupling for iMRI**

An RF resonator mainly consists of an inductor connected in parallel to a capacitor. For a lumped-parameter RF resonator, the angular resonance frequency ω<sup>0</sup> depends on its capacitance *C* and inductance *L* and is given by ω<sup>0</sup> = 1/ √ *LC*. When the coils of two RF resonators are in close proximity to each other, where the magnetic eld generated by one coil cuts the loops of the other coil, they turn to be inductively coupled and share a mutual inductance *M*12, as shown in Fig. 6.6. Depending on the coupling strength, the resonance frequency of each of the two inductively coupled resonators is shifted away from its initial value. If both of the RF resonators are tuned to the same initial resonance frequency (ω<sup>0</sup> = ω<sup>01</sup> = ω02), then coupling will split this frequency, resulting in two frequencies shifted ω1,<sup>2</sup> to the left and right of the initial value,

$$
\omega\_{1,2} = \frac{a\_0}{\sqrt{1 \pm K}},\tag{6.3}
$$

where *K* is the inductive coupling coecient, which depends on the mutual inductance between the two coils and is given by *K* = *M*12/ √ *L*1*L*2. Applying voltage to one of the two inductively coupled resonators (VS) induces a voltage in the other resonator (VOUT), as shown in Fig.6.6:

$$V\_{\rm Out} = V\_S \frac{Z\_2 jo\mathcal{M}\_{12}}{Z\_1 Z\_2 + (jo\mathcal{M}\_{12})^2},\tag{6.4}$$

where *Z*<sup>1</sup> and *Z*<sup>2</sup> are the impedance of the two resonators. Equations 6.3 and 6.4 are driven in appendix C. Agilent's advanced design system (ADS) software was used to simulate the frequency splitting for a pair of inductively couple RF resonators. The two simulated resonators were identical (R= <sup>1</sup> <sup>Ω</sup>, L= 50 nH, C= 125 pF) with an angular resonance frequency of ω<sup>0</sup> = 63.6MHz, a photo of the ADS simulation environment is attached in appendix C. The voltage supplied to one of the two resonators was set to unity, and the coupling coecient (*K*) was swept. Figure 6.7b shows a comparison between the simulation results and the mathematical model. Figure 6.3b illustrates the shape of the induced voltage in the other resonator at dierent coupling condence using Equation 6.4. The gure shows the induced voltage reduction as the coupling decreases.

Tracking catheter using a single coil is not reliable due to the impact of the catheter axial orientation on the catheter visibility. The coupling coecient changes as the catheter's axial orientation varies with respect to the static surface receiver coil during interventions. Hence, the entire system needs to be re-tuned and matched with the catheter motion and orientation. Moreover, the visibility can be lost if the *B*<sup>1</sup> elds of the catheter-detector and surface coil become orthogonal. Therefore, a novel micro-detector is introduced based on two orthogonal saddle coils, as shown in Fig. 6.8a. The proposed detector maintains near-constant coupling with the surface coil under the entire range of the catheter axial orientations. The two saddle

Figure 6.6: Equivalent schematic circuit of two inductively coupled RF resonators.

Figure 6.7: Resonance frequency splitting due to inductive coupling. **(a)** Degree of the resonance frequency splitting versus coupling coecient (*K*), comparing the mathematical model is compared to the ADS simulation. **(b)** The induced output voltage at dierent *K*, labeled in the gure.

coils are arranged to have their *B*<sup>1</sup> elds orthogonal, which eectively decouples them. The two coils are connected in series and tuned by a single capacitor to the Larmor frequency, as illustrates in the equivalent schematic circuit in Fig 6.8b.

The nite element analysis software COMSOL multiphysics was used to gain insight into the *B*<sup>1</sup> eld distribution generated by micro-detector in the space around the catheter. Additionally, the simulation was utilized to evaluate the angular dependence of the coupling between the micro detector and the surface coil. Two simulation scenarios were considered:

1. In the rst simulation model, a micro-saddle detector was excited using a 50 Ω port, and the magnetic ux density was plotted around the detector. Two micro-detector designs were simulated based on single and double saddle coil. In each case, the perimeter gap (see Fig. 6.8 for description) between the saddle's two square loops was swept to achieve the best magnetic eld distribution.

Figure 6.8: Double saddle coil for MR-catheter tracking through inductive coupling. **(a)** CAD-drawing illustrating the design of the double saddle coil. **(b)** Equivalent schematic circuit for a tuned double saddle detector inductively coupled to a surface pickup coil.

2. In the second simulation model, the excitation port was connected to a stationary surface coil, and the micro-detector was connected to a port for measuring the induced voltage, as shown in Fig. 6.10. Single and double saddle micro-detectors were considered, and in each case, the three possible relative orientations between the surface coil and the micro-detector (roll, pitch, and yaw) were swept from 0° to 360°.

The COMSOL radio frequency module was used for both simulation models, and simulations were run at a 63 MHz excitation frequency. As part of the simulation environment, the microdetector was positioned inside a box-shaped domain to represent the surrounding sample (water), which was then located inside a bigger box-shaped domain to mimic the ambient air. A radiation boundary condition was employed to act as an absorbing layer to prevent backscattering. The free tetrahedral mesh was rened to resolve the modeled structures correctly, as shown in Fig. 6.10a.

Figure 6.9 summarizes the results of the rst simulation scenario. The results revealed that the micro-detector with a double-saddle provides more magnetic ux sources around the detector than the single-saddle detector. The distribution of the generated magnetic eld around the detector can be manipulated by varying the perimeter gap between the saddle coil's two square loops. By setting the detector dimensions to the values listed in Table. 6.2 and sweeping the

Figure 6.9: The simulated magnetic eld generated the micro-saddle detectors excited at 63 MHz. The top row demonstrates the magnetic eld distribution generated by a single saddle coil, while the bottom row presents the magnetic eld generated by a double saddle detector. For each case, the magnetic eld direction is presented at two dierent cross-section views, one perpendicular to the coil's axis (XY-plane) and the other along its axis (XZ-plane). In addition, the magnetic eld direction is plotted at dierent perimeter gaps to deduce the change in the magnetic eld distribution with the gap.

Figure 6.10: Simulation results of the voltage induced in a catheter micro-saddle coil due to inductive coupling with a surface coil at dierent relative orientations. **(a)** The simulation model of a micro-saddle coil positioned within a stationary single loop surface coil. The diagram illustrates the rened tetrahedral mesh applied for the numerical model. The plots illustrate the voltage induced in the micro-saddle coil caused by the surface coil as a function of the three orthogonal relative rotations: **(b)** Relative axial orientation of between the saddle and surface coil (saddle rotation around the Z-axis), **(c)** Alignment of the axes of micro-saddle and surface coil (saddle rotation around the X-axis), and **(d)** Rotation of the saddle coil in the plane of the surface coil ( saddle rotation around Y-axis).

gap, it was found that the best homogeneity for the magnetic eld was obtained at a gap of 450 µm. Hence, a gap value of 450 µm was used for further simulations and fabrication.

Figure 6.10b-d summarize the results of the second simulation scenario. The variation in the induced voltage in the micro detector with the change in the relative axial orientation is shown in Fig. 6.10b. In the case of a single saddle coil, the coupling between the micro-coil and the surface coil was maximum when their *B*<sup>1</sup> elds were aligned, and as the relative angle between the *B*<sup>1</sup> elds increases, the coupling decreases. Accordingly, no coupling was observed (i.e., no


Table 6.2: Geometrical dimensions of the double saddle coil used for MR-catheter tracking.

voltage induced in the micro-coil) when an orthogonal position was taken. On the other hand, for a double saddle coil, the coupling remained constant over the entire range of relative axial orientations, with values alternating above the single saddle coil's maximum value. This is because the total coupling of the whole double saddle detector is the summation of the coupling of each of the single saddle coils within the double saddle detector with the surface coil. Thus, a coupling reduction of a single saddle element is accompanied by a coupling increase in the other saddle element. The variation in the induced voltage in the micro detector with the change in the relative angle between coil axes shown in Fig. 6.10c. For both double and single saddles, the maximum coupling occurs when the surface coil axis and the saddle coil axis are orthogonal, and minimum coupling occurs when the two axes coincided. The change in the coupling as the micro detector orient in the 2D plane is shown in Fig. 6.10d. The orientation of the micro-detector in the 2D plane does not aect the coupling.

During invasive medical procedures using MR catheter, it is expected that the catheter is intensively subjected to rotations along its axis and turns in the surface coil's plane (along the length of a laying body, and across its width). On the other hand, the angle between the micro-coils and the surface coil axes rarely varies with wide angles (into the depth of the body). Therefore, the proposed micro-double saddle coil is considered to be a highly suitable detector for the MR-catheter application.

### **6.3.2 Fabrication of the double saddle resonator**

Building a micro-double saddle detector around a slender catheter tube is challenging as it demands a micro-fabrication process for creating multi-layer structures in the 3D. The fabrication is done by 2D patterning of conductive layers separated by a thin exible polymer lm. Then the polymer lm with the conductive layers is rolled around the catheter tube to achieve the 3D structure. The conductive layers 2D patterning has been performed utilizing two dierent techniques: inkjet-printing and photolithography process. Direct inkjet-printing onto an unstructured substrate is an ecient and fast prototyping process but has a low resolution (above 50 µm). On the other hand, the photolithography process required expensive instrumentation but oers a high resolution and suitable for mass production.

#### **Photolithography process**

Fabricating the double saddle coil using the photolithography process was conducted in three main steps, as explained in Fig. 6.11a. The rst step was done by chemical wet etching to pattern the bottom saddle coil on a 25 µm Kapton lm (Fig. 6.11a(i)). Subsequently, the second step was done by structuring a liquid photosensitive polyimide on top of the bottom saddle to create the separation layer, including via that connects the top and bottom metal layers (Fig. 6.11a(ii)). Lastly, the third step was done by lift-o process to deposit the top saddle coil on top of the separation layer (Fig. 6.11a(iii)). The three performed processes (wet etching, photosensitive polyimide, and lift-o) are explained in detail in the photolithography Subsection 3.1.1 in the micro-fabrication Chapter 3.

#### **Inkjet-printing process**

The inkjet-printing of silver metal onto unstructured Kapton lm is explained in detail in the inkjet-printing Subsection 3.1.2 in the micro-fabrication Chapter 3. Kapton lm with a thickness of 25 µm was used. Printing of the double saddle was done in two steps, as shown in Fig.6.11b. The rst silver layer was printed, then the Kapton substrate dried at <sup>90</sup> °<sup>C</sup> for 15 min using a hot plate. Subsequently, the Kapton substrate was ipped over, and after printing the second layer, it was dried again at <sup>90</sup> °<sup>C</sup> for 15 min. Lastly, the device was sintered in an oven at 150 °C for 1 h.

The thicknesses of the conductive layers deposited using either inkjet-printing or lift-o process are within a few hundreds of nm. However, for RF applications, a minimum thickness of the conductive track is required according to the operating frequency to provide low RF resistance. Therefore, the thickness of the tracks was increased by electroplating to ≈ 10µm using the home-built electroplating setup discussed in Subsection 4.3.3 in the electroplating Chapter 4. Finally, the Kapton lm with the double saddle coil was rolled around a 5 fr catheter tube. Figure 6.12 shows microscopic images of the 2D patterned micro double-saddle coil achieved by the photolithography process and inkjet printing.

Figure 6.11: Schematic illustration of the two micro-fabrication techniques used for developing a micro-double saddle coil. **(a)** Photolithography process consists of three main fabrication steps: (1) Wet etching for patterning the bottom saddle coil, (2) Structuring photosensitive liquid polyimide for creating a separation layer possessing via, and (3) Lift-o process for depositing the top saddle coil. **(b)** Inkjet-printing of silver ink on both sides of a Kapton lm.

### **6.3.3 Experimental results of the double saddle resonator**

The enhancement in MR-catheter tracking provided by the proposed micro-double saddle coil was veried by conducting two experiments. The rst experiment was done to characterize the inductive coupling between the micro-saddle coil and the surface coil at various relative axial orientations. The second experiment was done to map the *B*<sup>1</sup> eld of the micro-saddle coil within an MRI scanner.

An experimental setup was created from laser-cut PMMA sheets to characterize the inductive transmission coecient between a micro-detector immobilized on a catheter tip and a primary stationary surface resonator, as demonstrated in Fig. 6.13. The primary surface coil (radius = 2 cm) was connected to a network analyzer through a matching and tuning network ((*LCoil* k *CM*) −*C<sup>T</sup>* ) containing variable capacitors. The catheter tip with the microdetector was adjusted to face the center of the primary surface coil at a distance of 2 cm. The catheter axial orientation was actuated utilizing a stepper motor actuator (step motor 28byj-48) controlled using a microcontroller (Arduino-Uno board) through a motor driver board (ULN2003APG).

The experiment was carried out twice, the rst run with a single saddle coil and subsequently the second run with a double saddle coil. The primary stationary surface coil was tuned to 63 MHz resonance frequency and matched to <sup>50</sup> <sup>Ω</sup>. The single saddle coil had a measured inductance of 16 nH and was tuned to a resonance frequency of 63 MHz by a 390 pF capacitor (SMD capacitor 0402). In contrast, the double saddle coil (two single coils connected in series, *<sup>L</sup>double* <sup>=</sup> <sup>2</sup> <sup>×</sup> *<sup>L</sup>single*) was tuned to the same resonance frequency by a 190 pF capacitor. The capacitance required for tuning the double saddle is almost half of the capacitance used for tuning the single saddle; this is because the double coil's inductance is twice that of the single coil ( *f* = 1/ √ *LC*). The primary coil's reection coecient (S11) was recorded at dierent relative axial orientations (swept from 0° to 90° with a step of 10°) with the micro-detector immobilized on the catheter tip. The relative axial orientation range from 0° to 90° was enough to fully characterize the coupling coecient due to the symmetric structure of the saddle coils.

Figure 6.14 presents the experimental results of the inductive coupling characterization. The

Figure 6.12: Microscope images of the 2D patterned micro-double saddle coil. The coils are fabricated by **(a)** Photolithography process and **(b)** Inkjet-printing.

Figure 6.13: Experimental setup used to characterize the inductive coupling strength between a stationary scale primary coil and a rotatable micro-detector.

practical results match the COMSOL simulation results in Fig. 6.10b; The coupling between the primary surface coil and the single saddle coil decreases as the relative angle between their magnetic *B*<sup>1</sup> elds increases. Furthermore, when the *B*<sup>1</sup> elds were nearly orthogonal, the reection coecient curve showed no splitting in the resonance frequency, conrming that no inductive coupling occurred (Fig. 6.14b). In contrast, for the double saddle coil, the inductive coupling between the surface resonator and the micro-detector was maintained at a near-constant prole for the entire range of the relative axial orientations (Fig. 6.14a).

The magnetic *B*<sup>1</sup> elds of the micro-saddle coils were mapped utilizing a low eld MRI scanner (Bruker ICON, *B*<sup>0</sup> = 1T). As revealed in Fig. 6.15a, the double saddle coil generated a homo-

Figure 6.14: Measured reection coecient (S11) curves of a stationary primary resonator due to coupling with micro-saddle detector. The S<sup>11</sup> curves demonstrate the resonance frequency splitting of a stationary resonator tuned at 63 MHz due to inductive coupling with **(a)** Micro-double saddle coil and **(b)** Micro-single saddle coil both tuned at 63 MHz.

geneously distributed magnetic eld surrounding the catheter tube, conrming the simulated magnetic eld distribution in Fig. 6.9. The MR images had a slice thickness of <sup>1</sup>.25 mm, a eld of view (FoV) of 20mm×20mm, with an imaging matrix size set as 128×128 using echo time (TE) = <sup>4</sup>.5 ms, repetition time (TR) = 278 ms, a ip angle of <sup>40</sup>°, and using four averages.

Figure 6.15: Mapping the *<sup>B</sup>*<sup>1</sup> elds of **(a)** A micro-double saddle coil and **(b)** A micro-single saddle coil in a 1 T MRI scanner.

# Part III.

# NMR Application - Automated High Throughput Nuclear Magnetic Resonance

# **7 State-of-the-art: high throughput NMR spectroscopy**

**N** MR spectroscopy is an information-rich analytical method utilized for characterizing materials by determining and quantifying molecular structure. NMR is an attractive screening tool as: i) it oers broad chemical coverage; ii) it is a non-destructive technique, permitting further upstream analytics; and iii) it does not require signicant sample preparation. The NMR technique measures the response of a sample's spins to a polarizing static magnetic eld (*B*0) and an excitation RF magnetic eld (*B*1), which generates a small electrical signal in an RF resonant detector. The polarizing *B*<sup>0</sup> eld is provided by an external magnet, while the *B*<sup>1</sup> excitation eld is commonly generated by the RF detector itself, as explained in detail in Chapter 2.

A standing challenge facing the present NMR technique is that it is generally limited to single sample measurements under low throughput conditions. Therefore, there is a disconnect between applications that require large parameter spaces screening and the opportunity to take advantage of this information-rich method. By addressing the NMR throughput challenge, it could lead to a signicant push to diverse elds, including drug discovery screening, clinical analysis, chemical process monitoring, bio-uid metabolomics, and food analysis.

Consider the eld of drug discovery as an example; NMR can resolve the binding of drugs to pathogens and can be used for screening compound libraries to investigate the molecular interactions with target proteins [15, 16]. Pharmaceutical companies have molecular libraries containing millions of compounds, which is ever-increasing as new molecules are being synthesized utilizing combinatorial methods. For example, Pzer, Inc. maintains 4 million compounds inside their molecular library that can be used for analyzing potential medicines, as reported in 2017 [17]. Hence, the development of high throughput NMR methods is a must to enhance the screening eciency of these massive and ever-increasing molecular libraries.

## **7.1 High throughput NMR approaches**

For the last decades, developing high throughput NMR techniques has been addressed extensively with the goal to involve this powerful technique in screening these large compound libraries. For a fully automated NMR measurement, several steps should be controlled [184]. The NMR spectroscopy process involves sample preparation, transferring the desired sample into the magnet, adjusting the NMR instrument (tuning, matching, shimming), performing the

This chapter's text was partially reproduced and adapted from the author's published work [ON5] (CC BY 4.0, 2021 Springer Nature)

measurement, transferring the sample outside the magnet, and processing the data. Several approaches have been developed, which can be divided into three main concepts: i) automated sample changers; ii) probe with multi RF coils; and iii) ow-based NMR systems.

### **7.1.1 Automatic sample changer**

The rst approach to be discussed is the automated sample changer systems. This approach was developed to analyze multiple samples with avoiding the manual samples changing as possible, i.e., overnight or weekends. The system possesses a sample preparation holder where each sample is physically separated to avoid cross-contamination. The sample holder is connected to the NMR magnet to transfer each sample individually using a pneumatic control system. Automated sample changers provides the user with easy and safe access to the NMR magnet without the need for ladders or steps. This automation approach allows users to access and automated operation overnight, which enhances the NMR spectroscopy throughput. Besides the automation process, manual control for the sample handling system is usually possible using a push button. Therefore, automated sample changer systems have been adequate for individual NMR glass sample tubes. However, this approach automates only the sample transfer into and outside the NMR magnet. The throughput is still limited by the NMR instrument re-adjusted (tuning, matching, shimming) after each sample exchange.

Since the automated sample changer was rst introduced, it has been enhanced and developed where commercial automated changers are now available and capable of queuing up to ∼ 500 samples for the NMR measurement [184,185]. For illustrating the capability of the automated sample changers, two examples of commercially available systems are discussed below. The rst example is for a sample changer system provided by Brucker [186], a photo of the system is shown in Fig. 7.1a. The sample holder possesses 24 or 60 sample positions compatible with various NMR sample tube lengths and diameters. The system can be connected to almost all Brucker's NMR magnets. The system has the distinction of being vibration-free and has no RF interference eect on NMR coils. Two versions of the systems are provided to control the samples set-point temperature, heated and cooled version. The heated version has a temperature range from room temperature up to 125 °C; thus, liquid polymers could be analyzed. On the other hand, the cooled version has a temperature range from 4 °C up to 40 °C, which is appropriate for analyzing various samples such as protein or biouid solutions.

The second example is the ASC24 system provided by JEOL [187], a photo of the system is shown in Fig. 7.1b. Several versions of the sample holder are available that possess 30, 64, or 100 sample positions. The sample changer is at the same height as the user, so no need for ladders or steps. Besides, there is a top-loading sample changer version installed on the top of the NMR magnet to save space. A manipulator controls the sample for insertion and ejection through an NMR magnet. The automatic sample changer comes with several advantages, However, the main drawback of these systems is that it requires NMR instrument readjustment after the

Figure 7.1: Examples of the commercially available automated NMR sample changer systems. *The pictures are reprinted from the companies' websites [186, 187]*.

insertion of each sample, such as tuning and matching of the RF coils and shimming of the main magnet.

### **7.1.2 Parallel detectors**

Developing NMR probes with multiple RF coils has been considered a practical approach for enhancing the NMR spectroscopy throughput. Instead of using a single RF detector to acquire an NMR spectrum from a single sample, a probe with multiple RF detectors could be designed to excite and acquire signals from various sample volumes simultaneously. Commonly, the reported multiple-coils probes have been designed in a way that each RF coil is warped around a capillary where the sample is located.

Sample excitation and NMR signal acquisition have been conducted following two main approaches using parallel coil NMR probes. The rst approach is attaching all coils to a single resonant circuit, and a gradient eld is applied during sample excitation to provide spatial frequency dependence. Thus, one receiver is used to acquire the NMR signal from the multiple coils, and each sample's signal is separated based on the frequency shift introduced by the gradient eld. This approach provides the advantage of simple hardware construction. Mac-Namara et al. [188] introduced a four coils NMR probe, shown in Fig. 7.2a. The four coils are connected in parallel and attached to a single resonance circuit with trimmer capacitors; the circuit is located beneath the sample region. To dierentiate the acquired signal based on the sample location, a static linear eld gradient has been applied in the direction of the *B*<sup>0</sup> eld by changing the shim power supply.

The other approach is providing a separate resonant circuit for each coil to excite and detect signals separately with avoiding RF interference and coupling between the coils. Fabricating several RF receivers adds to the system hardware complexity and cost. Wang et al. [189] introduced an eight coils probe head developed with an interface to four receiver channels

spectrometer, shown in Fig. 7.2b. The eight coils have been divided into two groups (four coils each) where data from one group are acquired at a time. For designing an eight coils probe to t into the NMR magnet homogeneous region, the impedance matching circuits have been designed to be away from the coils. This design reduces the eciency of the RF coils, but a compromise for enhancing the lling factor was considered. Additionally, sucient grounding and proper shielding between the coils' circuits were required to avoid RF cross-coupling and interference as much as possible.

An application for the multiple coils principle is NMR detection for continuous-ow capillary electrophoresis (CE) [190]. Capillary electrophoresis is a powerful separation method for a mixture of various samples by exposing them to a transient electric eld. Although no physical separation, NMR measurement can be conducted based on the dierent electrophoretic mobility of the dierent samples [191]. Wolters et al. [190] introduced a two coils probe that has been used with performing CE NMR spectroscopy. Figure 7.2c shows a schematic illustration of the CE/NMR system with multiple coils. The system possesses a single separation capillary connected to multiple outlets, where each outlet has a coil with its detecting circuit. The coils' circuits are attached to an electromagnetic relay switch to determine which coil is

Figure 7.2: Examples of state-of-the-art multiple parallel coil NMR probes. **(a)** NMR probe head with four parallel ow-through coils [Right]. Spatial frequency encoding is considered for using single resonance circuit [Left], *reprinted from [188] with permission of Elsevier (Copyright © 1999 Elsevier Science B.V.)*. **(b)** NMR probe head with eight parallel RF coils. Each coil has its own resonant circuit, *reprinted from [189] with permission of Elsevier (Copyright © 2004 Elsevier Inc.)*. **(c)** Schematic illustration of the CE/NMR system with two micro-NMR RF coils [Right]. The results [Left] show that system with multiple coils and CE ow could separate the spectra, as shown in (A) and (B). On the other hand, without CE, the spectra are overlapped, as shown in (C), *reprinted with permission from [190], Copyright ©2002 American Chemical Society*.

connected to the transmit/receive channel, i.e., one coil is active at a time. Hence, continuous CE ow sequentially direct sample ow into the various outlets, and cycled stopped-ow detection process is conducted between the multiple NMR detectors.

In principle, the number of coils in multiple coil probes can be scaled up to increase the throughput, nevertheless, a limitation arise from the small volume of the homogeneous *B*<sup>0</sup> eld inside the NMR magnet. The design of the coils and the minimum coils spacing should be considered to avoid RF cross-coupling and interference that could result in signals bleed between neighboring coils, i.e., exciting a specic coil can generate an RF signal in the adjacent detecting coil. These challenges could be addressed in several ways, such as; considering only one active coil and de-tuning the inactive neighboring coils or using dierent frequency adjustments for the various coils.

### **7.1.3 Flow-based NMR**

Flow-based NMR systems have also been considered to achieve high throughput NMR spectroscopy [192, 193]. Flow-based NMR spectroscopy is done by transferring a sample from outside the magnet through the NMR detector using uidic tubes. Several liquid handlers are available commercially (e.g., multiple probe 215 liquid handler provided by Gilson Inc [194]), or even standard syringe pumps can be used to inject the sample into the inlet of the owinjection probe. After analyzing the sample, it is transferred out of the magnet, and the following sample is injected without the need for hardware disassembling. Solvents can be used between every two successive samples for the rinsing purpose of the probe's tube to avoid cross-contamination. The main advantage of the ow-based approach is that the NMR instruments should be adjusted once (shimming, matching, and tuning) for the rst sample, and no need for re-adjustment.

Rehbein et al. [195] introduced a ow probe with a double resonance solenoidal detector having an active detection volume of 2 µL used for capillary HPLC-NMR spectroscopy coupling, a picture of the probe head and the whole system is shown in Fig. 7.3a. The NMR RF coil is warped directly around a horizontal capillary (OD/ID, <sup>1</sup>.5/0.8 mm) attached at both ends to vertically oriented ow tubes. The NMR detector is immersed in a uorinated oil (FC-43) to match the susceptibility miss-match between air and copper. By using the introduced system, NMR spectroscopy can be done in a continuous-ow or stopped-ow measurement. The advantage of the stopped-ow method is that the NMR acquisition time could be increased leading to a greatly enhanced SNR. Also, with this method, it is possible to record 2D correlation NMR spectra.

Carret et al. [10] introduced a ow-based device that is compatible with any commercial NMR probe head, a picture of the device is shown in Fig. 7.3b. The device possesses a micro-coil with an inlet and outlet for the sample ow. The NMR signal is transferred from the micro-coil to the NMR system detector through induction coupling.

Macnaughtan et al. [196] introduced a system where a multiple coil probe is combined with ow-based NMR spectroscopy, a picture of the system is shown in Fig. 7.3c. The system injects four samples simultaneously into four coils installed on a probe head using a liquid handler possessing four controllable syringe pumps. The time consumption is improved by a factor of four compared to a single ow-through system. The four micro-solenoidal detectors were connected in parallel and attached to a single resonance circuit with two variable capacitors for tuning and matching. NMR analysis was performed based on the chemical shift imaging or selective excitation between the four NMR detectors [197].

The ow-based approach has proven to be a reliable alternative to enhance the NMR spectroscopy throughput, with additional enhancements by developing probes with multiple RF

Figure 7.3: Examples of state-of-the-art ow-through NMR probes. **(a)** A probe head possessing a microsolenoidal coil for ow-based NMR spectroscopy [Left]. Schematic illustration for a capillary HPLC-NMR spectroscopy coupling system, *reprinted from [195] with permission of John Wiley & Sons, Inc. (Copyright © 2007 WILEY-VCH Verlag GmbH & Co. KGaA)*. **(b)** Schematic illustration for a ow-based NMR device that is compatible with commercial NMR probe heads. *reprinted with permission from [10], Copyright ©2018 American Chemical Society*. **(c)** Schematic illustration for a system combining multiple coil probe with ow-based NMR spectroscopy, *reprinted with permission from [196], Copyright ©2003 American Chemical Society*.

NMR detectors. However, ow-based NMR spectroscopy faces mainly two challenges:


## **7.2 Adapted approach**

The sample exchange phase during an NMR experiment is not necessarily the only timeconsuming step. NMR spectroscopy is famous for its high resolution, with spectral line widths in ppb. This is because nuclear spins have relatively long relaxation periods, typically in the term of seconds for hydrogen nuclei in the liquid state. The relatively lengthy re-equilibration delay results in restricting the number of times an NMR experiment can be replicated. This recycling delay can be eliminated under the ow-based approach as freshly polarized samples are continuously transferred into the NMR-coil detection volume, avoiding the need for re-equilibration. This is particularly signicant when the target nuclei have relatively long relaxation time constants [10]. Combining the lengthy recycle times with the insensitivity of NMR spectroscopy results in the NMR being a low throughput technique. Hence, in situations where sample exchange and recycle delay become time-limiting processes, a ow-based approach will outperform the traditional sample handling technique and enhance throughput.

Therefore, a ow-based NMR spectroscopy is adapted here to enhance the NMR throughput. Instead of exchanging individual sample containers traditionally, a ow-through NMR system can be used to deliver a series of concentrated sample plugs separated by an immiscible uid from outside the magnet through the NMR detector. This two-phase ow eliminates the dead volume but requires continuous online measurement to track the sample plug's position and velocity to ensure that the target sample lls the NMR detection volume. In this contribution, a novel ow-based NMR system is introduced. The ow system features: i) integrated capacitive sensors for sample position and velocity measurements; and ii) a micro-saddle coil for

NMR spectroscopy. This ow system enables fully automated NMR spectroscopy while taking advantage of the improved experimental time per sample due to continuous sample ow. The system is discussed in detail in the next Chapter 8.

A ow-based NMR spectroscopy is extremely attractive when target sample quantities are small; instead of dealing with poor lling factors or diluting samples using standard NMR hardware, microuidic technologies and micro-NMR techniques can be used. The following subsections provide an overview of the state-of-the-art micro-NMR detectors and ow sensors.

## **7.3 Micro-NMR detectors**

The RF coil is the core element of the NMR system, allowing interaction of measurement device with the target sample magnetization. In this part of the thesis, 'coil' refers to any device capable of generating and detecting magnetic elds. The topology and design of the RF coil have a signicant impact on the NMR measurement. This section provides an overview of various NMR coils, which can be divided into two main categories: volume and planar.

### **7.3.1 Volume coils**

There are many shapes of the volume coils, such as solenoidal and Helmholtz coils. Volume coils have better eld homogeneity than planar coils. In the case of analyzing samples that can be cast into a cylindrical shape, such as liquids or powders, volume coils are the proper choice since the probe volume can be lled optimally, resulting in a high lling factor and sensitivity.

#### **Solenoidal coil**

Solenoidal coils have the classic topology represented by helically wrapped thin conductive wire around a cylinder used as a mechanical support for the coils. A schematic of the solenoidal coil is shown in Fig. 7.4a. In the case of NMR application, the supporting cylinder is a capillary that holds the sample during the measurement. Solenoidal coils are attractive for micro-NMR applications due to their several advantages including high lling factor, high sensitivity, ease of electric decoupling, highly uniform magnetic eld within the coil's core, and easily manufactured on the micro-scale [201].

However, integrating solenoidal coils into an NMR system faces some issues. The rst drawback is the sample handling, especially in ow-based spectroscopy, since the coil should be transversely located inside the NMR magnet. Another drawback is that the NMR magnet homogeneous *B*<sup>0</sup> eld needs to penetrate through the coil's tracks before it polarizes the sample. This leads to the distortion of the *B*<sup>0</sup> eld due to the susceptibility mismatch of dierent material interfaces, resulting in broader NMR signals. There are several approaches to overcome this challenge, for example, adding susceptibility materials for matching purposes [202].

Figure 7.4: Examples of state-of-the-art volume NMR coil. **(a)**, **(d)**, and **(f)** Schematic illustration for the solenoidal, Helmholtz, and saddle coils, respectively. The schematic illustrations show the *B*<sup>0</sup> and *B*<sup>1</sup> elds' expected orientations. **(b)** and **(c)** Examples for micro-solenoidal coils. **(b)** *Reprinted from [198] with permission of AIP Publishing (Copyright © 1997 AIP.)*, and **(c)** *reprinted from [199] with permission of Elsevier (Copyright © 2009 Elsevier B.V.)*. **(e)** Example for micro-Helmholtz coil, *reprinted from [200] with permission of IOP Publishing (Copyright © 2014 IOP Publishing Ltd)*. **(g)** and **(h)** Examples for micro-saddle coils. **(g)** *Reprinted from [143] with permission of IOP Publishing (Copyright © 2017 IOP Publishing Ltd)*, and **(h)** *reproduced from the author's published work [ON5] (CC BY 4.0, 2021 Springer Nature)*.

The coil eciency (*B*1/*I*) is the coils ability to produce a magnetic eld from a unit current which also expresses the detector ability to convert the NMR signal into a voltage signal [27]. Solenoidal coil eciency depends on the number of turns and the coil diameter [28],

$$\frac{B\_1}{I} = \frac{\mu\_0 N\_C}{d\_C \sqrt{1 + (h\_C/d\_C)}},\tag{7.1}$$

where *N<sup>C</sup>* is the number of turns, *d<sup>C</sup>* is the coil diameter, and *h<sup>C</sup>* is the coil height. The previous equation shows that decreasing the coil diameter, with maintaining a constant *hC*/*d<sup>C</sup>* ratio, increases the coil eciency. Hence, solenoidal coils on the micro-scale were developed using MEMS fabrication techniques [203].

#### **Helmholtz coil**

The basic shape of a Helmholtz coil is a pair of identical circular coil loops which are placed apart with a distance equivalent to the radius of the coil loop with a common axis. A schematic of the Helmholtz coil is shown in Fig. 7.4d. The two loops are electrically connected in series so that the same current passes through them in the same direction to generate a homogeneous *B*<sup>1</sup> eld in the center.

The main advantage of Helmholtz coils over solenoidal coils is the homogeneity and direction of the *B*<sup>1</sup> eld, which aligns the coil loops with the *B<sup>O</sup>* eld. Hence, the *B*<sup>0</sup> eld does not penetrate several material interfaces as in the case of solenoidal coils, reducing *B*<sup>0</sup> eld distortion. Moreover, the Helmholtz coil orientation inside the *B*<sup>0</sup> eld results in more convenient sample handling, especially in ow-based NMR, where there is no need to remove the probe head to exchange the sample. Helmholtz coils can be fabricated using the MEMS technologies, where each of the two loops could be considered as a planar structure and then connected together [200,204]. But, the Helmholtz coils have lower sensitivity than the solenoidal coils [27].

#### **Saddle coil**

Saddle coil is a special type of a Helmholtz coil, where the coil loops take a cylindrical arrangement other than the planar arrangement in classical Helmholtz coils. A schematic of the saddle coil is shown in Fig. 7.4f. The cylindrical geometry maximizes the coil's lling factor, thus, saddle conguration for NMR is an appealing possibility for tube-based NMR experiments. For better magnetic eld homogeneity, it was shown that the angular width for each coil loop should be ≈ 120°, and the separation arc in between the loops should have angular width of ≈ 60° [205].

Saddle coils feature similar advantages as Helmholtz coils regarding *B*<sup>1</sup> eld homogeneity and direction, as previously discussed. Moreover, they provide ample sample space, so an easier sample handling that is compatible with stationary and ow-based NMR.

The main challenge facing the saddle coil is the fabrication at the micro-scale due to the 3D conguration. In 2018 Wang et al. [143] introduced a robust micro-fabrication process for developing a micro-NMR saddle coil with an active sample volume of 82 nL. The developed process was to pattern 2D conductive layer onto a thin planar polyimide lm then rolling it around a glass capillary to achieve the 3D structure. The supporting glass tube was used as a microuidic sample chamber.

### **7.3.2 Surface coils**

Surface coils are the second group of coils beside the volume coils. Surface coils have simple structures that can be eciently designed and optimized compared to the volume coils. 2D planar coils could be easily fabricated at the micro-scale using standard lithography or fast prototyping techniques such as inkjet-printing [206]. The main advantage of the surface coils is

locating the sample on the top of the probe instead of inside of it, which means more straightforward sample handling and larger available sample space. Planar spiral and striplines are wide-used arrangements in MR applications.

#### **Planar spiral coils**

Planar spiral coils have been developed to oer small NMR detectors with high sensitivity and a straightforward sample handling. A schematic of the planar spiral coil is shown in Fig. 7.5a. The magnetic eld of the planar coils is not uniformly distributed in space; the magnetic *B*<sup>1</sup> eld strength drops with distance from the coil surface. Hence, it is important to consider the sample location (volume of interest) with respect to the coil structure [209]. Moreover, a eld distortion is introduced by the electrical bridge between the outermost and innermost winding to create a closed-loop.

In the beginning, Stocker et al. [210] developed a rectangular planar micro-coil that was successfully employed to acquire NMR spectra of liquid samples. Later Dechow et al. [211] introduced micro-coils of gold tracks fabricated on gallium arsenide substrate with inner diameters ranging from 50 µm to 400 µm to serve samples with nL volumes. In 2015, Grisi et al. [212] introduced a single transceiver chip with dimensions of 1 mm<sup>2</sup> , which features an integrated planar micro-coil. Montinaro et al. [213] used this chip to design 3D micro-channels to analyze sub nL samples using NMR spectroscopy. Commercial micro planar coils have been provided by Bruker BioSpin (Billerica, MA and Fällanden, Switzerland), which were used to interface

Figure 7.5: Examples of state-of-the-art surface NMR coil. **(a)**, **(b)**, and **(c)** Schematic illustration for the spiral planar, phased array, and striplines, respectively. **(d)** Example for micro-planar NMR coils, *reprinted from [206] with permission of Elsevier (Copyright © 2002 Elsevier Science B.V.)*. **(e)** Example for phased array coils, *reprinted from [207] with permission of IEEE (Copyright ©2011 IEEE)*. **(f)** Example for stripline, *reprinted from [208] with permission of AIP Publishing (Copyright © 2008 AIP)*.

digital microuidics with NMR micro-coils to manipulate droplets [214]. Lee et al. [215] introduced a phased array micro coils conguration to increase the eld of view (FOV), schematic for phased array coils is shown in Fig. 7.5b. A phased array is a conguration of multiple planar coils with a carefully designed dened overlap where the mutual inductance between the neighboring coils is canceled [207, 216]. This conguration is valid for samples with a large area or several parallel samples with minimal interference in between. Davoodi et al. [217] developed a broadband micro spiral coil that could acquire NMR spectra over a wide range of frequencies (125–500 MHz in an <sup>11</sup>.74 T NMR magnet) without the need for a tuning interface circuit.

#### **Striplines**

Striplines or micro slot coils have a very simple structures; they are made od a single planar lines that can be easily fabricated using micro-fabrication technologies. A schematic of the stripline is shown in Fig. 7.5c. It operates as a wave-guide with geometrical dimensions that determine the resonance frequency, such as the conductive strip's geometries and the substrate's relative permittivity. Additionally, its magnetic eld *B*<sup>1</sup> is proportional to the substrate (dielectric) thickness [218]. The matching circuit interface connected to the stripline signicantly aects its quality factor because of the thermal noise arise by the lumped components. This problem was raised while developing a lumped-element planar strip array (LPSA) to enhance parallel MRI imaging and decrease the scan time [219].

Several approaches have been developed to benet from the simple stripline structure and high sensitivity. The volume microstrip approach has been developed as a microstrips transmission line (MTL) to develop MR imaging of the human head [220]. In addition, this volume approach was also applied by using double microstrips for achieving a uniform magnetic eld within the volume of interest and better SNR. Moreover, they were able to analyze at samples such as cell layers [221]. Another approach to enhance the stripline magnetic eld homogeneity is adding two metal planes to force the eld line to be parallel to the planes, which enlarge the volume with a homogeneous magnetic eld [222]. Striplines have been utilized for analyzing living cells like cancer cells, where a microuidic device was developed to hold the sample then placed on the top of a micro slot detector to perform the NMR analysis [1].

## **7.4 Flow sensors**

Developing microuidic-compatible NMR devices has drawn signicant attention in recent years. Microuidic devices have been used in high-eld NMR spectroscopy (1H resonant frequency > 200 MHz), often integrating ow for biological sample monitoring [1–3], chemical reaction tracking [4–8], sensitivity enhancement techniques [9–12], and most recently, electrochemical functionality [13]. Except for the digital microuidic approach [8,223], the target sample in these examples has always been transferred from outside the magnet without considering the sample handling situation. For ow-based NMR spectroscopy, sample loading is done from outside the NMR magnet. Thus, eliminating dead volumes in lling transfer lines (often a few meters in length) before the sample reaches the NMR detector is the challenge. The feading dead volume can be converted into a multi-sample queue using two-phase ow, achieving high throughput by taking advantage of microuidic technologies. Hence, online tracking of the sample position and velocity must be addressed to ensure each NMR signal acquisition is synchronized to the target sample.

Fluids in microuidic applications are commonly detected by employing either optical or electrical sensing systems, examples of state-of-the-art ow sensor are shown in Fig. 7.6 . Electrical sensing is appealing for two-phase ow since samples are usually optically similar. Moreover, aqueous samples have a dielectric constant at least an order of magnitude dierent from standard immiscible oil. Hence, electrical sensing is an excellent candidate for two-phase plugs ow detection.

### **7.4.1 Optical sensors**

An optical sensing system consists mainly of an optical source, optical detector, and waveguides to guide the light beam. Several optical sources have been used, e.g., semiconductors laser sources, based on amplifying the stimulated emissions. Each laser has specic parameters

Figure 7.6: Examples of state-of-the-art ow sensor. **(a)** Liquid droplets detection using an optical system. [Top] Schematic illustration of the system, and [Bottom] microuidic device with the ow optical detecting system, *reprinted from [224] with permission of Elsevier (Copyright © 2005 Elsevier B.V.)*. **(b)** Multiplexed arrays (4 × 4) of capacitive and resistive sensors for sensing the ow of liquid plugs. The resistance of the resistive sensors (thin-lm serpentine resistors) varies according to the thermal conductivity of the liquid in contact. [Top] Schematic illustration of the multiplexed sensors, and [Bottom] microscopic image of the resistive sensors, *reprinted from [225] with permission of Elsevier (Copyright © 2008 Elsevier B.V.)*. **(c)** Microuidic device integrating capacitive sensors (coplanar electrodes) positioned beneath a ow channel with a thin insulation layer to prevent direct contact with the target uids. The capacitance of the sensor changes according to the permittivity of the owing uid. [Top] Schematic illustration of the microuidic device with an array of the capacitive sensors, and [bottom] microscopic image of the coplanar capacitive sensor, *reprinted from [226] with permission of Elsevier (Copyright © 2011 Elsevier B.V.)*.

such as the operating wavelength, single or multiple modes, and beam quality; the parameter of the laser is selected based on the application. Several optical detectors have been utilized, such as photo-diodes, photo-multiplier tubes (PMT), and charged coupled devices (CCD). The optical detector is used to receive the optical signal and detect the change in the signal intensity corresponding to a parameter of the sample. The detector converts the change in the light intensity to an electrical signal to be analyzed.

Optical sensing systems are widely used for medical, cosmetic, and food industries. Nguyen et al. [224] developed an optical sensing system to control droplet size in a microuidic dropletbased system. The laser source was located above the micro-channel and a photo-diode detector on the other side of the channel. The laser light intensity varies as a droplet crosses the light beam due to diraction and absorption. As a result, the size and the shape of the droplets can be recorded. Camer et al. [227] developed an optical sensing system for counting formed droplets in a dripping system. The optical system was built around a transparent rectangularshaped ow channel with a width of <sup>2</sup>.5 mm. A 760 nm laser source is located on one side of the channel, and a detector is located opposite it. The detector senses the interruption of the laser beam whenever a drop crosses it and transmits electrical data with a frequency of 200 kHz. Additionally, a CCD camera is used to monitor the drop breakup during dripping.

The optical methods have the advantage of being a non-invasive sensing technique, extending the exibility in sample handling. However, the main challenge is that the target samples are often optically similar. Other than the optical sensing system, the sensing element of the electric sensors can be easily fabricated using the standard lithography process within microuidic devices.

### **7.4.2 Electrical sensors**

Flow electrical sensors are mainly based on changing the impedance of the sensor corresponding to a change in the physical properties of the material in their vicinity. The commonly used ow electrical sensors could be categorized into resistive and capacitive sensors. Capacitive sensors provide an output signal based on the target sample dielectric constant, while resistive sensors sense the samples based on their thermal conductivity.

#### **Resistive sensors**

Resistive sensors are thermosensitive sensors, where the resistance changes with the change of the temperature. By applying a constant potential dierence across the resistive sensor, its temperature settles at constant temperature based on its temperature coecient of resistance (TCR) and the surroundings thermal conductivity. When samples with dierent thermal conductivity, or samples of the same liquid owing at dierent velocities, ow over the resistive sensor, the sensor's temperature varies correspondingly, changing the sensor's resistance and thus the output current.

Several studies have employed resistive sensors in droplet detection [225, 228]. However, The main limitation of the resistive sensor is that the target uid should come in direct contact with the sensor to change its resistance. This direct physical contact between the sensor and the sample could lead to sample's temperature uctuation and cross-contamination.

#### **Capacitive sensors**

The capacitive sensor is a suitable choice for most microuidic applications. In general, the capacitance of a capacitive sensor changes according to the change in the insulator dielectric constant or the capacitance dimensions. Capacitive sensors are suitable for various applications such as position, angle, pressure, and ll level measurements. The sensor conguration is optimized according to the application, e.g., in the case of ll-level capacitive sensors, the sensor conguration depends on whether the uid is conductive or non-conductive [229]. For conductive uid, the uid is considered as counter electrode, and an insulated electrode is immersed into the liquid with a single conductive electrode. On the other hand, for nonconductive uids, the uid is considered the dielectric insulator between the two inserted conductive electrodes. Soenksen et al. [230] developed a capacitive sensor to monitor and control a uid height without direct contact with the uid. Elbuken et al. [226] developed a planar interdigitated capacitive sensor for measuring droplets' size and speed. Ghafar-Zadeh and Sawan [231] developed a hybrid microuidic/CMOS capacitive sensor. The design of the sensing part is based on interdigitated electrodes (IDC), where the system output signal varies according to the dielectric constant of the material in contact.

Capacitive sensors feature several advantages including high sensitivity, no physical contact or interference with the sample, small temperature inuence, and good stability and reproducibility. In contrast, they feature some drawbacks such as the electrical connection inuences the output reading (i.e., interfacing electronic should be located near it), sensitivity to humidity and dust, and sometimes shielding is required to avoid susceptibility for interfering signals.

# **8 Automated high throughput NMR system**

**T** his chapter presents a thorough description of the design, characterization, and fabrication of a fully automated high throughput NMR system. Automating the NMR spectroscopy is achieved by combining microuidic technologies with thin-lm micro-NMR detectors. The system is based on the concept of ow-based NMR spectroscopy, where plugs of the target samples are loaded into a exible microuidic tube separated by an immiscible uid. The exible tube is fed through a glass capillary with a micro-NMR detector, and ow sensors are rolled around it and placed in the isocenter of an NMR magnet. The arrangement of successive plugs of samples and immiscible uid is kept under continuous ow. The ow sensors detect the aqueous-oil interfaces and synchronize the NMR spectroscopy acquisition via a microcontroller. The system possesses a pair of ow sensors that allows the system to operate regardless of the ow velocity and the plug volume.

This chapter is divided into four sections. Section 8.1 explains the technical details for the different components of the automated NMR system. Within this section, the control system's hardware connections and software algorithms are presented. Section 8.2 discusses the design, optimization, and characterization of the micro-NMR detector. Section 8.3 discusses the design, optimization, instrumentation, and validation of the ow sensing system. Section 8.4 introduces a robust thin-lm micro-fabrication process used for fabricating the micro-detector and ow sensors.

## **8.1 System description**

Figure 8.1a presents an overall schematic of the hardware components of the actively synchronized NMR system, with a detailed illustration of a custom-built NMR probe head in Fig. 8.1b. The primary part of the automated NMR system is a custom-built NMR probe head. The probe head is built from a 3D printed holder (see Appendix D for the holder's working drawing) supporting a PCB that possess three substantial modules:

1. **A micro-NMR detector** attached to pre-tuning and matching capacitors.

The pre-tuning and matching network is built from three non-magnetic capacitors (S111DUE, Johanson Technology) soldered directly on the PCB, having a balanced conguration of *CM*,*PCB* − (*L* k *CT*,*PCB*) −*CM*,*PCB*. The capacitor values can be modied to adjust the resonance frequency of the micro-NMR detector so the NMR spectroscopy could, in principle, be performed on any NMR-sensitive nucleus. This tuning and matching conguration is discussed previously in detail in Subsection 2.4.3, where

This chapter's text and gures were reproduced and adapted from the author's published work [ON5] (CC BY 4.0, 2021 Springer Nature)

Figure 8.1: Overall diagram of the fundamental elements of the novel high throughput NMR experimental setup. **(a)** A microuidic Teon tube (OD/ID, <sup>1</sup>.6/1 mm) is utilized to transport a sequence of plugs of various samples separated by an immiscible uid through a custom-built NMR probe installed in an <sup>11</sup>.7 T NMR magnet. A syringe pump is used to regulate the ow from outside the NMR magnet. A microcontroller is used to drive a pair of ow sensors integrated on the probe head to determine the samples' velocity and location and synchronize the NMR data acquisition. **(b)** Photograph of a customized probe head installed on the NMR probe. The probe head consists of a PCB supported by a 3D printed holder. The PCB possesses a micro-NMR coil with a pretuning and matching network, two ow sensors with their interface electronics, and a glass capillary (OD/ID, <sup>2</sup>.15/1.72 mm) through which the exible sample ow tube is fed. Right: microscopic images for the micro-NMR detector, the IDC sensor, and the insertions of the Teon tube inside the glass capillary. Left: schematic diagram of the equivalent electronic circuit in the NMR probe.

*CT*,*PCB* is used to tune the NMR detector to <sup>1</sup>H NMR frequency, and *CM*,*PCB* is used to match the circuit to 50 Ω to guarantee signal transfer and prevent signal reection.

2. **Two capacitive ow sensors**, before and after the micro-NMR detector, electrically connected to a capacitance-to-digital converter (CDC) microchip to convert the sensors' reading to digital data to be transferred outside of the magnet.

Two sensors were used to enable velocity calculation, where the total separation distance between them is set to 30 mm. The two sensors are anking the NMR detector with a separation of 10 mm w.r.t. the detector so that the NMR automated spectroscopy is independent of the ow direction. The capacitive sensors' readings were transferred to a microcontroller unit (MCU) located outside the NMR magnet via the CDC integrated circuit (IC). In addition, a digital pin of the MCU was connected to the triggering signal input-output unit (TTL I/O) of the NMR spectrometer console to synchronize NMR experiment acquisition.

3. **A glass capillary** with the micro-NMR and the ow sensors rolled around to contain the exible microuidic tube holding the ow.

The exible microuidic tube (OD/ID, <sup>1</sup>.6/1 mm) holding the sequence of samples was

fed from outside of the NMR magnet through a customized NMR probe and the glass capillary (OD/ID, <sup>2</sup>.15/1.72 mm). This arrangement of tting the ow tube inside the glass capillary provides several advantages:


The probe head's polylactic acid (PLA) material and the FR-4 of the PCB were eliminated next to the micro-NMR detector as they are a source of background signal and aect the homogeneity of the magnetic eld. The probe head was designed to be compatible with market available micro-imaging NMR probe and a triple-axis micro-gradient sleeve (Bruker). Hence, the system is compatible with pulsed-eld gradient (PFG) sequences for the magnetic resonance imaging (MRI) experiments.

The probe head is connected to the <sup>1</sup>H channel of the custom-built NMR probe. The design of the NMR probe is adopted from commercial probes. The main addition in the developed probe is four built-in pipes inside the probe, allowing uidic and electrical access to the probe head. Like other commercial probes, the probe posses two trimmers (0.<sup>5</sup> - 10 pF, TG 092 ROHS, EXXELIA) connected in series to the probe head for ne-tuning and matching. The trimmers can be mechanically adjusted using glass ber rods from outside the magnet. Figure 8.2 shows the conguration of the tuning and matching network in the probe and the probe head. The xed capacitor *C<sup>P</sup>* is added to the probe's circuit to set one of the trimmers in parallel with the probe head circuit to enable the tuning functionality; this electrical conguration is discussed in detail in Subsection 2.4.3. The probe was designed to locate the micro-NMR detector in the NMR magnet isocenter (the region with the magnet's highest homogeneity ). Locating the micro-detector in the isocenter enables the spectrometer shim system to correct the static *B*<sup>0</sup> eld's minor deviations within the detection volume. A 3D CAD model of the probe is available in Appendix D.

Figure 8.2: Electrical RF tuning and matching network in the NMR probe. The saddle coil's three-terminals are connected in a way to keep the electrical current owing in the opposite direction in the two loops so that their magnetic eld constructively interfere in the coil center. The probe head has xed capacitors for pre-tuning and matching, and the probe has trimmers for ne-tuning and matching.

The train of samples was kept underow using a syringe pump positioned outside of the NMR magnet. Loading the microuidic tube with the samples was done manually and automatically, as discussed in Subsection 8.1.2.

### **8.1.1 Control system**

Figure 8.3 presents a schematic diagram of the hardware components of the automated NMR control system. As the train of aqueous plugs and oil plugs ow through the capillary tube with the ow sensors, a dierence in the sensors' capacitance is detected. The change in the capacitance results from the dierent eective dielectric interacting with the sensor's electric fringe eld. The changes in the capacitive sensor's values are then sensed by the MCU, which in turn synchronizes the NMR acquisition to collect the spectrum for each plug separately. The MCU sends the digital triggering signal (TTL) via a DIO pin (high = 5 V, Low = 0 V) to the TTL I/O unit of the NMR console. Hence, during the automatically synchronized NMR experiment, two software algorithms are executed in parallel. One algorithm on the MCU responsible for sending the TTL signal based on the ow interface. The other algorithm is the pulse sequence program on the NMR acquisition software (TopSpin 3.5pl2, Bruker BioSpin) to receive the TTL signal and increment the FIDs to save the spectrum from each plug on a dierent memory location.

Figure 8.4 presents a ow diagram for both algorithms, the source codes of the two algorithms are attached as supplementary material in Appendix D. On the NMR pulse program side, the acquisition software starts collecting the NMR scans of the rst FID following the system start. Following each collected scan, the NMR acquisition software checks the triggering input signal value sent from the MCU on the TTL signal pin in the NMR consol TTL I/O unit. If the TTL signal is low (0 volts), the spectrometer accumulates the collected scan to the current FID. In case the TTL signal is high (5 volts), the current FID is saved to the disk after adding the last NMR scan and then initiates the next FID measurement. Meanwhile, on the MCU side, the program monitors the capacitance values of the two sensors simultaneously. Once the measured capacitance value of either of the two sensors overrides a user-set threshold value,

Figure 8.3: Schematic diagram demonstrating the main hardware elements of the control system. The NMR data acquisition is synchronized by the microcontroller, which sends the TTL signal to the NMR spectrometer console's TTL I/O unit. The microcontroller generates the TTL pulse according to the values of the ow sensors.

the current time is recorded (*t*1). Additionally, a TTL signal is sent from the MCU'S DIO pin to the TTL pin in the TTL I/O unit of the NMR console after a delay (*d*). Subsequently, the MCU monitors the other sensor's capacitance value until it overrides the threshold value, and then

Figure 8.4: Flow chart illustrating the executed algorithms on both the MCU and the NMR acquisition software. During an automated NMR experiment, the two algorithms run in parallel. The algorithm on the MCU triggers the NMR pulse sequence to increment the FID based on the sample presence.

it records the time (*t*2). Afterward, the MCU monitors the rst sensor's value until it becomes lower than the threshold level, after which the MCU sends a TTL signal after a delay (*d*). Consequently, for each plug, the NMR TTL I/O unit receives two triggering pulses, the rst TTL pulse when a sample plug enters the NMR micro-detector (start new NMR FID acquisition) and the second TTL pulse when it leaves the micro-detector (save the current NMR FID). Afterward, the MCU monitors both sensors' capacitance value in a threshold detection loop, waiting for the next plug. This control algorithm results in saving the NMR scans collected for each sample in a dierent FID. At the beginning of each loop, the MCU monitors both sensors' capacitance values to maintain the exibility of running the NMR experiment in a bidirectional ow.

The triggering signal is always sent from the MCU to the NMR TTL I/O unit after a time delay (*d*) to ensure that the sample plug either reached or entirely left the NMR micro-detector. The delay (*d*) is calculated base on the times (*t*<sup>1</sup> and *t*2) recorded during the ow using Equation 8.1, taking into consideration the separating distance between the ow sensor and the NMR detector of 10 mm, the total gap between the two ow sensors of 30 mm. The initial value of the time delay *d* is set to zero and updated after the MCU measures each plug's velocity. Hence, a dummy plug is usually added at the beginning to determine the time delay's rst value, which depends on the ow velocity.

$$d = \frac{t\_2 - t\_1}{3}.\tag{8.1}$$

On the one hand, to guarantee that the NMR spectrometer captures the triggering pulse, the pulse width (t*T T L*) needs to be longer than the summation of the NMR scan's acquisition time (*TAQ*) and relaxation delay (*D*1). On the other hand, the TTL pulse width must also be shorter than twice the summation of acquisition time and the relaxation delay; otherwise, it will be captured twice by the NMR spectrometer:

$$(T\_{AQ} + D1) < \mathbf{t}\_{TTL} < 2 \times (T\_{AQ} + D1). \tag{8.2}$$

### **8.1.2 Sample loading**

Filling the microuidic tube with the samples can be done manually or automatically. In both lling regimes, the same sequence was adopted as follow:


After each lling step, the tube's terminal was immersed in a pure deionized (DI) water reservoir for cleaning to avoid cross-contamination. Once all sample plugs are loaded in the tube, the ow was continuously switched on for the NMR acquisition.

In the manual lling regime, the motion of the tube's terminal between the dierent reservoirs and the synchronization of the syringe pump were both done by the user. On the other hand, in the automatic lling regime, a robotic arm was used to move the tube's terminal. A MCU was used to control the syringe pump and synchronize it with the robotic arm motion. An overview of the components of the automatic lling is illustrated in Fig. 8.5.

The automatic loading system consists of an MCU (Arduino Mega 2560 and customized Arduino shield), a robot arm (UR3e, UNIVERSAL ROBOTS), a syringe pump (Fusion 200-X Touch, KR Analytical), and a router. This shield contains a Wi-Fi module (ESP8266), a serial-RS232 converter, a switch for resetting the controller, and an LED to indicate the current controller activity. The MCU is connected to the syringe pump via a DB-9 serial connector, through which the pump receives the required commands to change pumping settings. The router is

Figure 8.5: Overview of the hardware components of the automated plugs loading system.

used as a message queuing telemetry transport (MQTT) host, through which the communication between the MCU and the UR3e robot arm is achieved wirelessly.

The user interface to control the automated loading system comprises three programs that all run on a computer and are connected to the MCU. The user interface allows the researcher to program the system's pumping sequence and know its progress. These programs include: i) an excel workbook that serves as the front-end where the user can set up the experiment sequence; ii) a python le that runs in the Windows shell, takes the input from the excel le and sends the control signal to the pump and the wi module; and iii) a program to monitor the MQTT client. A description of the user interface provided in Appendix D.

## **8.2 Micro NMR-detector**

### **8.2.1 Detector design and optimization**

The topology of an MR coil is typically dened based on the application and the target sample type. In general, volume coils are favored for NMR applications where the samples are chemical solutions or suspensions. In contrast, planar coils might be the choice for MRI applications where the samples have arbitrary shapes as tissues and animal organs. Key design considerations to maximize the performance (sensitivity and spectral resolution) for an MR detector are: i) the strength and homogeneity of the excitation magnetic *B*<sup>1</sup> eld generated by the RF coil; and ii) the impact of the MR coil conductive tracks on the homogeneity of the primary static magnetic *B*<sup>0</sup> eld. The performance of the *B*<sup>1</sup> eld has a direct impact on the measurement sensitivity. The *B*<sup>0</sup> eld homogeneity deviates within the RF region due to a magnetic susceptibility mismatch between material interfaces, directly impacting the MR signal spectral resolution. An MR coil with a slight deviation in the *B*<sup>1</sup> eld homogeneity is acceptable in MRI applications even though this will aect the image contrast. On the other hand, for NMR spectroscopy, any deviation in the coil performance means wasted signals which can be a problem with low concentrate samples. Hence for NMR spectroscopy, the RF coil needs to be carefully optimized.

For the presented ow NMR application, a saddle coil geometry has been adapted for its advantages. Saddle coil is one of the most used geometries in NMR spectroscopy. A saddle coil is a special case of a Helmholtz coil; it is built out of two square loops facing each other and arranged in a cylindrical manner, as shown in Fig. 8.6.

The homogenous *B*<sup>1</sup> eld generated by the saddle coil is in the center of the arrangement and is perpendicular to the coil structure, making it very suitable for the ow-based NMR applications. Finite element simulations were performed for the magnetic eld generated by the coil. The simulation results in Fig. 8.7 show the magnetic eld orthogonality with the coil loops. Since the *B*<sup>1</sup> eld should be orthogonal to the *B*<sup>0</sup> eld, the coil is oriented along the

Figure 8.6: Photograph (left) and schematics diagram of a micro-saddle coil. Design parameters of the saddle coil include the perimeter gap between the two square conductor loops, the length of the coil detection region (L2), and the length of the electric power leads (L1).

axis of the *B*<sup>0</sup> eld. Therefore, the ow tube can be easily fed along the magnet's axis, making the sample handling comfortable. Moreover, in this arrangement, the *B*<sup>0</sup> eld needs to cross fewer material interfaces near the RF detection region, making re-shimming of the coil more manageable. A solenoidal coil has a much higher lling factor than a saddle coil; nevertheless, its *B*<sup>1</sup> eld is along its axis, which places it perpendicular to the *B*<sup>0</sup> eld. Hence, for a solenoidal coil shimming and ow tube feeding can be a big challenge.

Finite element simulations using COMSOL Multiphysics software were performed to optimize the micro saddle coil geometrical dimensions of the saddle coil. The optimization aims to select the coil geometrical dimensions in a way to enhance the generated radio frequency eld *B*<sup>1</sup> per unit current while maintaining the static *B*<sup>0</sup> and RF *B*<sup>1</sup> elds as homogeneous as possible. In the simulation model, a micro saddle coil geometry was constructed around two concentric tubes to model a glass capillary and a Teon tube lled with water. The saddle coil with the

Figure 8.7: Simulated saddle coil model (left) and the resulted magnetic eld (right). The simulated magnetic *B*<sup>1</sup> elds of the saddle coil's two loops are pointing in the same direction, and thus they constructively interfere in the coil center (sample region).

glass and Teon tubes was located inside a cylinder-shaped domain to model the surrounding air. Two simulation scenarios were considered:


In both simulation scenarios, the coil had a track width of <sup>100</sup> <sup>µ</sup><sup>m</sup> with *<sup>L</sup>*<sup>1</sup> and *<sup>L</sup>*<sup>2</sup> of 2 mm and 4 mm, respectively (see Fig. 8.6 for description). Additionally, the gap between the saddle coil two rectangular loops (θ2) was swept to obtain the best performance (highest sensitivity with the most homogeneous *B*<sup>0</sup> and *B*<sup>1</sup> magnetic elds). The gap was swept in a way that the gap perimeter (θ<sup>2</sup> ×*r*) changed from 100 µm to 1300 µm at steps of 100 µm. The simulation results are presented in Fig. 8.8, the coil sensitivity and the static *B*<sup>0</sup> eld homogeneity are plotted versus the gap perimeter (arc distance between the rectangular loops). Furthermore, the *B*<sup>0</sup> and *B*<sup>1</sup> magnetic elds distribution are illustrated at dierent gaps.

The simulation results reveal that for a saddle coil, conductive tracks with a high degree of symmetry not only enhance the *B*<sup>0</sup> and *B*<sup>1</sup> magnetic elds homogeneity but also increase the sensitivity. According to the simulation study, a gap of 700 µm (equivalent to θ<sup>2</sup> of 37.3°) was selected for the coil fabrication. The full geometrical dimensions of the implemented saddle coil are summarized in Table 8.1. The fabrication process can be found in micro-fabrication Section 8.4.

Table 8.1: Geometrical dimensions of the integrated saddle coil in the custom-built NMR probe head.


Figure 8.8: Finite element simulation (COMSOL Multiphysics) studies the eect of changing the perimeter gap between the saddle coil's loop on the coil sensitivity and the *B*<sup>0</sup> magnetic elds' homogeneity. The graph is the simulated coil sensitivity and the simulated *B*<sup>0</sup> homogeneity versus the gap between the saddle coil's loop. The *B*<sup>0</sup> and *B*<sup>1</sup> eld top view proles at a cross-section in the center of the saddle coil are plotted for dierent gap values (Top).

### **8.2.2 Detector characterization**

The fabricated micro-saddle coil was evaluated following the dierent evaluation approaches summarized in Table 2.1 in the MR characterization Subsection 2.4.4.

With a rolled-up micro-saddle coil connected to the custom-built probe, pre-matching and tuning was accomplished by adjusting the capacitors soldered on the probe head's PCB and validated using a network analyzer (Agilent, E5071B). The reection coecient *S*<sup>11</sup> was used to check the resonance of the resonator. The coil was consecutively tuned to 500 MHz, the <sup>1</sup>H proton resonance frequency (Larmor frequency) at <sup>11</sup>.7 T, and matched to <sup>50</sup> <sup>Ω</sup>, as shown in Fig. 8.9. By knowing the bandwidth at <sup>−</sup>3 dB (∆*f*−3 dB) of the resonance frequency ( *<sup>f</sup>*0) and substituting in Equation 2.35, the loaded quality factor for a batch of fabricated coils was measured to be *Q<sup>P</sup>* = 50±2.

The probe head was characterized using NMR spectroscopy measurements conducted using an AVANCE III <sup>11</sup>.7 T NMR system (1H resonance frequency 500 MHz, Bruker BioSpin) under

Figure 8.9: Measured reection coecient S<sup>11</sup> of a micro-saddle coil. The coil is connected to the custom-built NMR probe and lled with an aqueous sample, tuned at 500 MHz and matched to <sup>50</sup> <sup>Ω</sup>. The bandwidth at <sup>−</sup>3 dB was measured to be approximately 20 MHz, resulting in a loaded quality factor of <sup>∼</sup> 50.

stationary (non-owing) conditions. The coil was lled with a test sample (30 mM TSP dissolved in deionized (DI) water) using a Teon tube with an inner diameter of <sup>1</sup>.0 mm, yielding an active sample volume of 3.14 µL. FWHM of the <sup>1</sup>H NMR water resonance was used to evaluate the spectral resolution (Fig. 8.10, red spectrum), with the micro-coil yielding a line-width

Figure 8.10: Characterization of the NMR micro-saddle coil. Measurements were performed using a micro-saddle coil at 500 MHz NMR magnet. The graph shows the NMR spectrum of water (red, lower scale) and NMR Excitation pulse nutation spectrum (blue, upper scale). The time delay between successive nutation measurements was 10 s.

of <sup>1</sup>.25 Hz. The excitation *<sup>B</sup>*<sup>1</sup> magnetic eld performance was assessed by acquiring a nutation spectrum (a series of <sup>1</sup>H NMR spectra at xed supplied power of *P* = 1W, varying the excitation pulse width from 1-105 µs, in increments of 5 µs (Fig. 8.10, blue spectra). The nutation spectrum envelop time period was measured to be *T*nut = 204µs (90° ip angle achieved at <sup>51</sup> <sup>µ</sup>s), thus the nutation frequency was calculated to be *<sup>f</sup>* nut <sup>=</sup> <sup>4</sup>.9 kHz.

The *B*<sup>1</sup> eld strength was calculated using Equation 2.33 to be *B*<sup>1</sup> ≈ 230µT ( <sup>1</sup>H gyromagnetic ratio <sup>γ</sup> <sup>1</sup><sup>H</sup> <sup>=</sup> <sup>42</sup>.756MHzT−<sup>1</sup> , angle between the *B*<sup>1</sup> and *B*<sup>0</sup> elds θ = 90°). Thus, the micro-coil sensitivity (η*P*) was calculated using Equation 2.31 to be approximately 230µT/ <sup>√</sup>W, comparable to the simulation result in Fig. 8.8.

Additional electronic components (sensors, AD7746, and MCU) were added to the NMR system for realizing a synchronized NMR spectroscopy. These hardware components come with the cost of adding noise to the NMR signal, as illustrated in Fig. 8.11. A single scan <sup>1</sup>H NMR spectroscopy was performed on a stationary (non-owing) sample (250 mM sucrose and 30 mM TSP in DI water) before and after powering up the periphery sensing components, leading to a reduction of the SNR by 55%, so that the *nLOD*<sup>ω</sup> increases from <sup>1</sup>.19 nmol s1/<sup>2</sup> to 2.18 nmol s1/<sup>2</sup> . Nevertheless, the NMR probe head sensitivity was comparable to other micro-coils, even with the added noise [30]. The measured sensitivity of <sup>1</sup>.19 nmol s1/<sup>2</sup> is at <sup>a</sup> 500 MHz static magnetic eld. To facilitate comparison with other reported micro-coils, the sensitivity needs to be scaled to 600 MHz, which is then converted to <sup>0</sup>.86 nmol s1/<sup>2</sup> . For context, Finch *et al*. showed a sensitivity of nLODω(600 MHz) <sup>1</sup>.57 nmol s1/<sup>2</sup> using their optimized stripline micro-detector [232]. Even with the noise added with the active ow-sensing elements in operation (yielding nLODω(600 MHz) <sup>1</sup>.58 nmol s1/<sup>2</sup> ), the NMR detector performance is adequate.

Figure 8.11: Comparison of a single-scan <sup>1</sup>H NMR spectra without (blue) and with (red) connecting the electrical power to the ow sensors' electronics interface. In this experiment, the sample was stationary (250 mM sucrose and 30 mM TSP dissolved in DI water). SNR was noted to decrease by <sup>∼</sup>55 % after connecting the power for the electronics interface. The regions between 5.56 - 5.28 ppm, 3.68 - 3.36 ppm, and 0.15 - -0.15 ppm are re-plotted in red boxes for illustration. The water peak's intensity (5.2 - 4.3ppm) was set manually to zero.

A test sample was prepared with 300 mM glucose, 75 gL<sup>−</sup><sup>1</sup> coloring powder, and 30 mM TSP dissolved in DI water to characterize the SNR of the ow NMR system. The characterization results are presented in the next Section 9.2. The coloring powder is added to the sample to visually separate the sample plugs from the transparent immiscible oil plugs. A single scan <sup>1</sup>H NMR spectroscopy of a stationary sample is shown in Fig. 8.12. The anomeric glucose hydrogen signal J-coupling could be resolved to <sup>∼</sup> 50 % from baseline. For comparison, Finch *et al*. demonstrated <sup>∼</sup> 85 % from baseline resolution of the same signal using the stripline micro-detector [232].

## **8.3 Capacitive Sensor**

### **8.3.1 Sensor design and optimization**

Interdigitated capacitive sensors (IDC) were used for detecting the ow, illustrated in Fig. 8.13. Interdigitated electrodes are sensitive to dierences in material permittivity, interfering with its fringe eld by modulating the sensor capacitance. Flow sensing dependent on capacitance calculation has several appealing characteristics: i) the non-invasive measurement avoids sample contamination; and ii) it requires low power minimizing temperature uctuations. Figure 8.13 presents a 2D layout of the interdigitated electrodes and the expected fringe eld lines within a rolled sensor detection region. The fringe eld is directed toward the ow by rolling the sensor around the ow capillary. Rolled IDC sensors have improved sensitivity to dierences in dielectric constants and are much more compact than conventional co-planar capacitive sensors.

Figure 8.12: Single-scan <sup>1</sup>H NMR spectra of a stationary sample (300 mM glucose, 75 gL<sup>−</sup><sup>1</sup> coloring powder, and 30 mM TSP dissolved in DI water). In this experiment, the electrical power was not connected to the ow sensors' electronic interface. The regions between 5.5 - 5.1 ppm and 4.3 - 3.2 ppm are re-plotted in red boxes for illustration. The spectral region between 3.0 - 0.5 ppm (free region) was eliminated for clarity.

Figure 8.13: 2D schematic layout of the interdigitated capacitive sensors (IDC) and the expected fringe eld generated by the sensors after rolling around the glass capillary.

The strength and uniformity of the electric eld generated by the sensor dene its sensitivity and performance. Improving the sensor sensitivity can be achieved by optimizing the interdigitated sensor geometrical dimensions to enhance the sensor generated electric eld. To optimize the ow sensor performance, the sensor total width (*WS*), and the gap between the electrodes (*G*) were explored (Fig. 8.14). Optimization was done by simulating the capacitive sensor's generated electric eld using nite element simulation. In the simulation model, the interdigitated sensor was constructed around two concentric tubes to model a glass capillary and a Teon tube lled with air and placed in an air-like environment. The simulated sensor was excited with a sinusoidal signal of 5 V at 32 kHz (mimicking the existing system parameters), and the sensor generated electric eld was plotted. The simulated sensor had a

Figure 8.14: Finite element simulation (COMSOL Multiphysics) studies the impact of varying the gap between the IDC sensor's ngers (G) and the sensor total width (WS) on the produced electric eld inside the sensor. **(a)** Gap dependence: The electric eld at the sensor center is plotted versus the gap (bottom). Top view proles of the electric eld at a cross-section in the center of the sensor are plotted for dierent gap values of 20 µm, 60 µm, and 100 µm. The electric eld generated by the sensor decreases as the gap between the ngers increases. **(b)** Width dependence: The electric eld at the sensor center is plotted versus sensor width (bottom). Top view proles of the electric eld at a cross-section in the center of the sensor are plotted for sensors with widths of 2 mm, <sup>3</sup>.5 mm, and 5 mm. The electric eld generated by the sensor increases as the sensor width increases.

total width *W<sup>S</sup>* = 3.725mm, a gap of *G* = 25µm, the number of sensing electrodes *N* = 30, each with a width of *W* = 100µm, and a length of *L<sup>S</sup>* = 6.6mm (slightly less than the outside circumference of the glass capillary <sup>6</sup>.75 mm).

Figure 8.14a shows the eect of the gap width between the electrodes (*G*) on the generated electric eld. The gap width was swept from 10 µm to 100 µm at steps of 10 µm. Subsequently, the electric eld strength at the sensor center was plotted against the gap, as shown in Fig. 8.14a. Reducing the gap width increases the generated electric eld strength and hence increases the sensor's sensitivity as in the case of co-planar interdigitated capacitive sensors [233]. Figure 8.14b demonstrates the eect of the capacitive sensor overall width (WS) on the generated electric eld. The sensor width was swept from 2 mm to 5 mm with a step of <sup>0</sup>.5 mm. Subsequently, the electric eld strength at the sensor center was plotted against the sensor width, as shown in Fig. 8.14b. On the one hand, expanding the sensor sensing region by increasing the sensor's width along the glass capillary increases the generated electric eld's strength and increases sensors sensitivity. On the other hand, the sensor width should be minimized to reduce the sensor's settling time. Increasing the number of sensing electrodes (*N*) within the sensor improves the generated electric eld homogeneity and hence the smoothness of the signal [226]. According to the simulation results and considering fabrication complexity to not select dimensions below 25 µm, several design iterations were tested. The selected dimensions of the interdigitated sensor implemented in the NMR probe head are summarized in Table 8.2.

The electric eld generated by the implemented sensor is simulated. The domain inside the Teon tube was modeled using, the relevant parameters, once for water (ε*<sup>r</sup>* = 80) and again for the separating oil FC-43 (ε*<sup>r</sup>* = 1.9). The electric eld within the detection volume is lower in water than in FC-43 due to the relatively high dielectric coecient of water compared to FC-43, as shown in Fig. 8.15. This variation in the sensor electric eld leads to the change in the interdigitated sensor capacitance, which is then sensed to detect the solutions' interface. The capacitance instrumentation technique is discussed in the following Subsection 8.3.2.

### **8.3.2 Sensor instrumentation and characterization**

The IDC sensor capacitance was read-out utilizing an o-the-shelf capacitance-to-digital converter (CDC) microchip (AD7746, Analog Devices). The AD7746 microchip is a high resolution sigma-delta (Σ−∆) CDC that posses two separate single-ended measurement channels, each of 24-bit resolution. The full-scale capacitance of each of the CDC's channels is <sup>±</sup>4 pF from an internal set reference value (0-17 pF). The CDC measurement has an accuracy of 4 fF and a resolution down to 4 aF. The AD7746 chip provides two built-in excitation sources (up to 5 V at 16 o<sup>r</sup> 32 kHz) and amplifying/ltering circuits. The sampling rate of the CDC can be set between <sup>9</sup>.<sup>1</sup> <sup>−</sup>90.9 Hz resulting in root-mean-square (RMS) noise levels of <sup>4</sup>.<sup>2</sup> <sup>−</sup>40 aF.

The AD7746 chip has a 16-lead TSSOP (Thin Shrink Small Outline Package) conguration


Table 8.2: Geometrical dimensions of the integrated ow IDC sensor in the custom-built NMR probe head.

(4.<sup>5</sup> <sup>×</sup>5 mm), enabling integration on the NMR probe head's PCB. Each of the two capacitive ow sensors was attached to a separate measurement channel of the CDC, via a positive capacitance input pin (CIN 1/2 (+)) and an excitation pin (EXC A/B) leveraging the single-end measurement. The excitation signal was set to 5 V at 32 kHz, and the CDC's sampling rate was set to <sup>9</sup>.1 Hz. The PCB's conductive tracks connecting the capacitive sensors to the CDC chip were kept as far as possible from each other to limit cross-coupling between the two sensors. The CDC's digitized capacitance measurements were transmitted via I2C serial communication to a microcontroller outside the NMR magnet, as shown in Fig. 8.16. An Arduino board was used as an o-the-shelf MCU. Two Arduino shields have been developed compatible with Arduino Uno and Micro modules. The schematic designs and photographs of each of the two Arduino shields are provided in Appendix D. The Arduino shield was used to connect the I<sup>2</sup>C wires from the CDC to the Arduino board; it additionally featured LEDs for visual feedback and push-buttons for triggering the NMR experiment manually. The performance of the two microcontrollers (Arduino Uno and Micro) was observed to be equivalent.

The instrumentation system's functionality to detect permittivity dierences was validated by sensing the capacitance read-outs in the presence of various solvents, summarized in Fig. 8.17. The developed system was eciently able to dierentiate between toluene (ε<sup>r</sup> = 2.38) and the uorinated oil FC-43(ε<sup>r</sup> = 1.9) used for samples separation. Also, a dierence in capacitance

Figure 8.15: Finite element simulation (COMSOL Multiphysics) illustrating the electric eld prole inside the IDC sensor (water and FC-43 oil as examples). Left: an overview of the whole sensor, right: focusing on the target sensing region inside the Teon ow tube.

Figure 8.16: Electronic hardware components implemented for realizing the automatically triggered NMR experiment. Each of the two IDC sensors is connected to an input channel of a CDC microchip on the NMR probe head. The digitized sensors' readings are transferred to an MCU outside the NMR magnet over an I2C communication. An Arduino shield is developed comprising I2C interface, push-buttons for manual triggering, LEDs for visual feedback, and TTL signal interface.

measurement was observed between air and FC-43, opening the possibility of air-bubble detection. The measurement system's sensitivity was independent of the magnetic eld, with only a constant oset of +0.3 pF observed with the sensor located in the NMR magnetic eld. This capacitance oset could be compensated by re-calibrating the CDC IC, but since the o set does not aect the working principle of the system, this re-calibration was not performed. The measurement system's sensitivity to a dierence in relative permittivity was determined to be ∆ε*<sup>r</sup>* = 0.2.

Figure 8.17: IDC sensor measured capacitance when lled with dierent solutions. The error in the capacitance measurement arises from the CDC chip's noise (<sup>±</sup> <sup>0</sup>.03 pF). The system is eciently capable of dierentiating toluene (ε*<sup>r</sup>* = 2.38) from FC-43 (ε*<sup>r</sup>* = 1.9) (inset).

The developed measurement system features two ow sensors to enable ow velocity measurement and bi-directional NMR ow synchronization. The system accuracy to measure the velocity was characterized by loading the line with plugs of colored aqueous samples separated with the uorinated oil FC-43 and recording the ow using a camera (60 fps). The plug velocities were extracted from the recorded video data using motion analysis software (Kinovea), which were then compared to the velocities measured by the developed system. The velocity measured by the system was within 4% of the camera results. The velocity characterization experiment was done outside of the NMR magnet to enable video recording.

Validating the sensors capability to synchronize NMR data acquisition was achieved by loading the ow tube with a line of three aqueous sample plugs separated by FC-43 oil. The three sample plugs are of dierent volumes and glucose concentrations, but all owing at 4 mm s<sup>−</sup><sup>1</sup> . Figure 8.18 demonstrates the sensors reading from the three aqueous sample plugs. The results reveal that a capacitance change above 100 fF is generated by the aqueous solutions separated by FC-43 plugs. Therefore, the capacitive sensor SNR is satisfactory even when the CDC microchip operates at the highest sampling rate of 90 Hz (RMS noise of 40 aF). As can be observed from the sensors measurements, the capacitance reading between the two ow sensors was not identical. The measured capacitance variation between the aqueous samples and the FC-43 oil for the 1st sensor was <sup>∼</sup> 190 fF (closer to the CDC IC) and <sup>∼</sup> 120 fF for the

Figure 8.18: Output readings of the ow sensors during a triggered NMR experiment. The experiment was performed on three plugs of dierent volumes (40 µL, 80 µL, and 120 µL) and dierent glucose concentrations (0 M, 1 M, and 2 M) separated by plugs of FC-43 (<sup>80</sup> <sup>µ</sup>L). In this experiment, the plugs were owing at a velocity of 4 mm s<sup>−</sup><sup>1</sup> . The MCU sends the TTL signals to the NMR console's TTL I/O unit after a calculated delay (*d*/3). This guarantees that a new NMR acquisition starts once the sample enters the NMR detection region. The time delay (*d*) is calculated using the sample's velocity measured by the capacitive sensors (top).

2 nd sensor. The dierence between the two capacitive sensors can be attributed to the fabrication tolerances and the variation in proximity between each sensor and the CDC microchip, where long tracks build up parasitic capacitance. Nevertheless, the system working principle is not aected by the variation in the ow sensors readings since an individual threshold value could be set for each sensor.

The sensors performance was comparable to other systems utilizing the same sensing principle, even with the parasitic capacitance eect [226,234]. A sensitivity advantage was achieved by rolling the ow sensors around the ow tube compared to planar IDC sensors [226] and capacitive sensors with less lling factor [234]. The signals acquired from the CDC microchip for the two ow sensors are suciently distinct that a robust triggering algorithm could be developed, as described previously in control system Subsection 8.1.1.

#### **8.4 Micro-fabrication**

The fabrication process used for realizing the micro-saddle NMR detector and the interdigitated capacitive ow sensors around a slender glass capillary with an outer diameter of <sup>2</sup>.15 mm consist of two main steps, as shown in Fig. 8.19. The rst fabrication step was 2D patterning of a conductive layer in the form of the detector and the sensor on a exible polymer lm. The second fabrication step was rolling the 2D structured lm around the slender glass capillary, thereby realizing a 3D conductive structure around the glass capillary.

The conductive layer 2D patterning was achieved following a standard wet-etching photolithography process suitable for mass production at the wafer scale. A 25 µm Kapton lm covered from a single side with a 9 µm copper layer (AkaFlex KCL 2-9/25 HT, Krempel GmbH) was utilized as the base substrate for the photolithography process. Kapton was used due to its magnetic susceptibility, which is close to that of water [48]. A conductive copper layer of <sup>9</sup> <sup>µ</sup><sup>m</sup> provides the requisite conductor skin depth at 500 MHz, ensuring the minimal electrical resistive losses of the NMR detector. The wet-etching fabrication process is discussed in details in Section 3.1. After the 2D patterning process, the NMR saddle coil and ow sensors were cut from the Kapton lm (a photograph of the entire wafer with a batch of detectors and sensors is provided in Appendix D). After rolling the lm around the glass capillary (OD/ID, 2.15/1.72 mm, borosilicate glass, Hilgenberg GmbH), it was xed using epoxy glue (UHU-Plus). The glass capillary with the NMR detector and ow sensors was glued to the probe head PCB using the same epoxy.

The fabrication process's overall yield is very high (nearly 100%), owing to the standard processes used. After multiple uses, no failures in the glue holding the lm to the capillary were observed or in the glue holding the capillary to the PCB. If necessary, adhesion could be enhanced using a low temperature cure liquid polyimide such as LTC9300 provided by FUJI-FILM. The liquid polyimide can be applied and then hard-baked inside an oven at 200 °C for

180 min, then it will be converted into a lm, resulting in a solid xation of the structures.

Figure 8.19: Schematic illustration of micro-fabrication process of the micro-saddle coil and IDC sensor. **(a)** Five steps categorized in two processes. First process: a wet etching photolithography process: (1) Substrate preparation by laminating 25 µm Kapton lm covered with a 9 µm copper layer on a silicon wafer, (2) Patterning of a positive photoresist (AZ4533) layer using UV lithography, (3) Etching the exposed regions of the copper layer; and (4) Peeling o the Kapton lm from the silicon wafer. Second process: (5) Rolling the Kapton lm with the patterned conductive layer around the glass capillary. **(b)** Microscopic images of the 2D patterned conductive layer in the shape of (1) Saddle coil and (2) IDC sensor.

# **9 Automated ow NMR results**

**T** his chapter presents the results of the dierent experiments performed using the developed system for automating the NMR spectroscopy. This chapter is divided into four sections. Section 9.1 describes the material used for the NMR experiment as well as the sample preparation. Additionally, Within this section, NMR spectroscopy data processing is explained. Sections 9.2 and 9.3 present the experimental result for characterizing the automated ow-based NMR system where the system SNR was expressed as a function of the sample velocity, volume, and analyte concentration. Section 9.4 demonstrates the results of automating the NMR screening of various commercial beverages as a piece of evidence that the system can eciently automate NMR spectroscopy on dierent samples.

## **9.1 Sample preparation and NMR spectroscopy**

<sup>A</sup> <sup>4</sup>.5 m microuidic tube was tted through the customized probe head and used for sample ow. The used tube is made out of polytetrauoroethylene (PTFE) (OD/ID, <sup>1</sup>.6/1 mm, S1810-12, Bohlender GmbH), which provides the required exibility to be tted through the NMR system. The Fluorinert oil FC-43 ( > 99%, 3MTM Fluorinert Electronic Liquid) is used for separating the target sample plugs and was purchased from Ionic Liquids Technologies (IOLITEC) GmbH. FC-43 oil is commonly used in many electronic and chemical applications because of its technical properties [235]. It is a perfect candidate for separating the target sample plugs within a microuidic channel: i) it is chemically inert, thermally stable with low toxicity, making it compatible with the Teon tube and various target samples, ii) it is immiscible with many popular NMR solvents, iii) its viscosity is relatively low (4.7 mPa s) and therefore ows easily inside the microuidic channels, and iv) its electrical resistivity is relatively high (3.4×10<sup>15</sup> Ωcm) preventing static charges.

The test sample used for characterizing the ow system SNR presented in Section 9.2 was prepared as 300 mM glucose, 75 gL<sup>−</sup><sup>1</sup> food coloring powder, and 30 mM TSP dissolved in deionized (DI) water. The glucose concentration series from 1 mM to 1 M presented in Section 9.3 was prepared using 50:50 DI water : D2O. D-glucose, 3-(trimethylsilyl)- propionic-2,2,3,3-d4 acid sodium salt (TSP), and deuterium oxide (D2O, 99.9% deuterium, Sigma Aldrich) which were purchased from Sigma-Aldrich. A market-available food coloring powder (Brauns Heitmann) was used. The beverages used in the automated NMR experiment represented in Section 9.4 were a selection of commercial beverages that were privately donated. Beverage samples were prepared by adding 1 mL of 5 mM TSP in D2O to 3 mL of the beverage ([TSP]*final* = 1.25 mM).

This chapter's text and gures were reproduced and adapted from the author's published work [ON5] (CC BY 4.0, 2021 Springer Nature)

NMR experiments were conducted on an <sup>11</sup>.7 T wide-bore AVANCE III NMR spectrometer ( <sup>1</sup>H resonance frequency 500 MHz). The customized ow-sensing probe head was installed on a custom micro-imaging NMR probe and is therefore fully compatible with magnetic resonance imaging (MRI). The NMR pulse sequence was a modied version of a conventional single-pulse experiment, including the trigger signal control. For a typical FID, the number of collected points was 5998 over a spectral width of 10 kHz, using a recycle delay of <sup>0</sup>.1 s. For the beverage NMR experiments in Section 9.4, the number of collected points was raised to 12014. In all synchronized NMR experiments, FIDs were collected in a pseudo-2D NMR experiment, which was later separated and processed individually after the NMR experiment for all plugs was completed. Each FID was zero-lled by a factor of 2 prior to the Fourier transformation and multiplied by an exponential function equivalent to a line broadening of <sup>0</sup>.3 Hz. After separating each spectrum, it was referenced to TSP (0 ppm), and the phase and baseline were corrected.

## **9.2 NMR characterization - varying sample volume and ow rate**

Characterizing the NMR performance was done by observing the acquired SNR using the automatic and manual acquisition of NMR data. Two series of experiments were done using the same test sample:


For manual triggering, a single sample plug was loaded into the ow tubing, bordered by FC-43. NMR acquisition was started and stopped manually when the sample entered and exited the probe. As a metric of NMR performance, the SNR of the TSP signal (0 ppm) was considered under each test condition. The normalized limit of detection (nLODω) was then calculated, as explained previously in Section 2.4.4, for the spectrum of each case based on the SNR of the TSP signal. The acquisition time for a single scan NMR experiment was set to be <sup>0</sup>.299 s with the NMR detector sensitive volume of 3.14 µL.

The experimental results of the rst series of the NMR experiments, where the sample ow rate was varied, are presented in Fig 9.1a and b. The developed triggering system acquires and stores the time domain NMR signal (FID) separately, as shown in Fig. 9.1a. The ow rate was incremented after the NMR spectroscopy for an oil-sample plug pair was completed. The velocity values reported in the results are the values calculated by the MCU using the acquired capacitive sensor signals. The system performance proves that the developed system could robustly dierentiate the oil from aqueous signals. The TSP signal SNR was calculated in the frequency domain after applying the Fourier transformation to the FID.

The SNR at the dierent velocities for various volumes is plotted in Fig. 9.1b and d, and selected spectra are plotted in Fig. 9.3a. As the results reveal, the NMR spectrum's SNR decreases

Figure 9.1: Performance of the automated ow-based NMR experiments under varying sample volumes and ow velocities. **(a)** Measured FID of seven sample plugs (80 µL) separated by FC-43 oil plugs (80 µL) with varying the ow rate. **(b)** The relationship between TSP peak SNR (0 ppm) and the sample velocity measured by the dual IDC sensors. The sub-gure shows the nLOD<sup>ω</sup> at each sample volume (mean <sup>±</sup> standard deviation over all tested ow velocities). **(c)** Measured FID of seven sample plugs separates by FC-43 oil plugs with varying the sample volume. The ow velocity was kept constant at 8 mm s<sup>−</sup><sup>1</sup> , and the oil plugs volume held constant at 80 µL. **(d)** The relationship between TSP peak SNR (0 ppm) and the sample volume measured by the dual IDC sensors. The sub-gure shows the nLOD<sup>ω</sup> at each sample velocity (mean <sup>±</sup> standard deviation over all tested sample volumes). In (b) and (d), each data point represents the mean ± standard deviation of the NMR experiment performed in quadruplicate. The data were t using the damped least-squares method to a 2nd order polynomial.

as a function of the sample ow rate, as expected, since increasing the ow rate reduces the sample residence time in the NMR detector's detection region. The SNR was continuously enhanced using the developed automated synchronized system as compared to manual triggering. Consequently, approximately improvements by factors of 2.5, 3.0, and 4.0 in the spectra SNR was observed for sample plug volumes of 120 µL, 80 µL, and 40 µL, regardless of the sample ow velocity. The nLOD<sup>ω</sup> calculated from the SNR is plotted in the inset of Fig. 9.1b. The performance under the automated triggering system is four-time better than the previous micro-saddle coil results [143], with a value of 4-5 nmol s1/<sup>2</sup> .

As the results reveal, the nLOD<sup>ω</sup> under manual triggering conditions increases as the sample volume decreases due to the increased contribution of the FC-43 oil signal (essentially noise) to the FID. The experimental results of the second series of the NMR experiments, where the sample plug volume was varied, are presented in Fig 9.1c and d. The FID signals of seven oil-sample FID pairs are shown in Fig. 9.1c, the volume of the sample plugs was incremented while the volume of the FC-43 oil plugs was held constant at 80 µL. The expected enhancement in the SNR of the TSP with increasing the sample volume is illustrated in Fig. 9.1d (selected spectra are plotted in Fig. 9.3a). The triggering system, once again, yielded a systematic SNR enhancement in comparison to the manual triggering condition, approximately by factors of 2.5, 3.0, and 4.0 for plugs velocities of 4 mm s<sup>−</sup><sup>1</sup> , 8 mm s<sup>−</sup><sup>1</sup> , and 12 mm s<sup>−</sup><sup>1</sup> regardless of sample plug volume. The enhancement in the NMR sensitivity when implementing the automated triggering system, as well as the micro-NMR detector performance, is further conrmed by the calculated nLOD<sup>ω</sup> for each ow rate (inset, Fig. 9.1d). The calculated nLOD<sup>ω</sup> and at each ow velocity are plotted in Fig. 9.2b.

Figure 9.2: NMR system's sensitivity under varying sample ow rates and volumes. **(a)** The change in the normalized limit of detection (nLOD<sup>ω</sup> ) with the sample velocity. Under automatic triggering conditions, The nLOD<sup>ω</sup> alters within a small range over the enter range of tested velocities for a given sample volume. Under manual triggering conditions, the nLOD<sup>ω</sup> increases as the volume of the sample decreased due to the increased signal contribution from the FC-43 oil signal (noise). **(b)** The change in the normalized limit of detection (nLOD<sup>ω</sup> ) with the sample volume. Under automatic triggering conditions, the nLOD<sup>ω</sup> is minimally enhanced by increasing the sample volume. Under manual triggering conditions, the nLOD<sup>ω</sup> is enhanced by increasing the volume of the sample as the contribution of the oil signal is reduced. The mean ± standard deviation of the data presented here are plotted in Fig. 9.1.

Figure 9.3: Selected <sup>1</sup>H NMR spectra resulted from the ow NMR spectroscopy experiments on a test sample (300 mM glucose, 75 gL<sup>−</sup><sup>1</sup> coloring powder, and 30 mM TSP dissolved in DI water) with varying the sample volumes and ow rates (SNR of all extracted data are plotted in Fig. 9.1). **(a)** The SNR of TSP peak (0 ppm) for each spectrum is noted next to the signal. A focused view of the glucose anomeric proton (5.2 ppm) is located next to each spectrum in a gray box for spectral resolution comparison. The water signal (5.1-4.3 ppm) was eliminated for clarity. The baseline distortion in the spectra with a manual triggering system results from signal accumulation from the FC-43 oil phase (noise). **(b)** Representative full <sup>1</sup>H NMR spectra of a 120 µL sample plug owing at velocity of 4 mm s<sup>−</sup><sup>1</sup> . The blue spectrum is obtained with the auto triggering system. In contrast, the red spectrum is obtained with manual triggering. The automatic triggering system eliminates the noise contribution from the oil phase.

A balance should be realized between the sample volume, ow rate, SNR, and spectral resolution to enhance the NMR throughput. Increasing the samples ow rate with keeping all other parameters constants result in the degradation of SNR and spectral resolution, as illustrated in Fig. 9.3. On the other hand, increasing the sample volume enhances SNR and spectral resolution. Increasing the ow rate reduces the NMR measurement time, leading to less noise

averaging over the accumulated signals. Simultaneously, increasing the ow rates allows a bigger fraction of the excited sample to leave the NMR detection region during NMR signal acquisition. Consequently, the FID is articially shortened and manifesting as a rise in the apparent relaxation time T2, results in broadening the spectral line as shown in Fig. 9.3. At extremely high ow rates, the whole excited part of the sample leaves the detection volume before NMR signal acquisition is completed, therefore adding more noise into the collected signal.

## **9.3 NMR characterization - concentration limit of detection**

The dependence of the developed automated NMR spectroscopy system on the sample concentration was estimated by loading the ow tube with a series of nine aqueous samples spanning a glucose concentration range from 1 mM to 1 M. The aqueous sample plugs with dierent glucose concentrations had a volume of 120 µL and were separated by FC-43 oil plugs with a volume of 80 µL. The automated ow NMR experiment was done at two dierent ow velocities 6 mm s<sup>−</sup><sup>1</sup> and 10 mm s<sup>−</sup><sup>1</sup> .

The results of the concentration experiment are summarized in Fig. 9.4. The automated NMR triggering system, as with previous experiments, could eciently separate signals from aqueous plugs and oil plugs, as shown in the FID signals in Fig. 9.4a. The relation between the SNR of the glucose anomeric hydrogen signal (5.1-5.3 ppm) and the glucose concentration is plotted in Fig. 9.4b for the two ow rates tested. Selected NMR spectra for these auto synchronized experiments are plotted in Fig. 9.4c. The NMR signal SNR has a linear relation with the sample concentration (R<sup>2</sup> <sup>&</sup>gt; 0.99). The SNR is greater at a lower ow velocity of 6 mm s<sup>−</sup><sup>1</sup> versus the higher ow velocity of 10 mm s<sup>−</sup><sup>1</sup> , considering the samples at lower velocity have a longer residence time in the NMR coil detection region, allowing greater signal accumulation. The total experimental time required by the automated ow-based NMR system to acquire the NMR signal of all oil and aqueous plugs (18 measurement in total) was <sup>6</sup>.1 min and <sup>3</sup>.6 min, or eectively <sup>25</sup>.5 s and <sup>15</sup>.3 s per aqueous sample at ow velocities of 6 mm s<sup>−</sup><sup>1</sup> and 10 mm s<sup>−</sup><sup>1</sup> .

Obtaining the relation between SNR and sample concentration is suitable for examining the sample plugs' cross-contamination. After loading the line with the nine aqueous samples, the measurements were performed in four consecutive runs after reversing the ow direction (i.e., all plugs signal were acquired in one ow direction followed by reversing the sample ow direction and repeating the measurements). Hence, the linear plot between the SNR and concentration after repeating the measurement four-time conrms that there was no crosscontamination of the Teon tube or between the dierent sample plugs.

Employing a plug ow approach will always raise the issue of cross-contamination. In the

case reported here of aqueous samples and FC-43 plugs owing in a Teon tube, cross contamination was not recognized at the NMR detection limit. However, in situations where cross contamination may be problematic, one could introduce additional solvent plugs to clean the ow line.

Figure 9.4: Performance of the automated ow-based NMR experiments under varying sample concentrations. **(a)** Automatically measured FID signals of nine aqueous sample plugs (120 µL) separated by FC-43 oil plugs (80 µL). owing at 10 mm s<sup>−</sup><sup>1</sup> . The aqueous samples contained glucose concentrations ranging from <sup>0</sup>.001 M to 1 M. **(b)** The SNR (calculated using the glucose anomeric hydrogen signal (5.3-5.1 ppm)) versus glucose concentration under two ow velocities. Each data point is the mean ± standard deviation of the NMR experiment performed in quadruplicate. The Linear data t (dashed lines) had R<sup>2</sup> > 0.99 in both cases. **(c)** Selected <sup>1</sup>H NMR spectra from the automated ow NMR spectroscopy experiments with varying the glucose concentrations dissolved in DI water. The water signal (5-4 ppm) was eliminated for clarity. The glucose anomeric proton's SNR for each spectrum is noted next to the signal with a focused view of the peak (5.18-5.14 ppm) in a gray box for spectral resolution comparison.

## **9.4 Automated sample throughput - beverage application**

The developed system functionality was validated on a set of samples beyond simple test solutions. The ow tube was loaded with seven samples of dierent beverages and separated by FC-43 oil plugs. Each beverage sample plug had a volume of 120 µL, and the FC-43 oil plugs had a volume of <sup>80</sup> <sup>µ</sup>L, and the NMR spectroscopy was done at a ow rate of 4 mm s<sup>−</sup><sup>1</sup> .

The total experimental time of the NMR signal acquisition under these conditions was 7 min with eectively <sup>38</sup>.3 s acquisition time per beverage sample. Figures 9.5a and b demonstrate the resulting FID and spectra of all samples. TSP was added to the samples with a target concentration of <sup>1</sup>.25 mM to facilitate referencing the resulted spectrum. However, it was observed from the results that the solubility of the TSP within the beverages was sampledependent, and therefore the TSP signal could not be utilized for quantication (only for chemical shift referencing).

#### **9.5 Conclusions**

As the results reveal in all previous experiments: i) the developed automatic triggering system was able to separate the sample and oil signals robustly; ii) the resulted NMR signals for the dierent tested sample were of excellent spectral quality (both in terms of resolution and SNR); and iii) there were no additional signals recognized in the oil NMR spectra, ruling out possible diusion into the oil plugs from the aqueous samples (at the NMR detection limit). Assuming a <sup>1</sup>.0 mm I.D. PTFE tube, <sup>100</sup> <sup>µ</sup><sup>L</sup> sample plugs, and <sup>100</sup> <sup>µ</sup><sup>L</sup> oil plugs, then in a 100 m long PTFE tube, approximately 400 individual samples can be loaded. The number of the NMR visible plugs could be doubled by loading the immiscible phase with measurable molecules. Loading a 100 m PTFE tube is a time-limiting process, although this was also automated, as

Figure 9.5: Automated NMR acquisition of beverage sample plugs. **(a)** Automatically measured FID signals of seven beverage and FC-43 oil plugs. The sample volume was 120 µL, and the FC-43 oil volume was 80 µL, the experiment ran at a ow velocity of 4 mm s<sup>−</sup><sup>1</sup> . **(b)** The resulted NMR spectra of the beverage samples.

discussed in the loading Section 8.1.2. Using the developed automated measuring system with selecting a reasonable ow rate of 4 mm s<sup>−</sup><sup>1</sup> (minimum tested ow rate), considering an acquisition time of 40 s per sample (as was done in the beverage example in Section 9.4), and both plug phases contain NMR measurable content, then these 800 samples would require <sup>∼</sup> 9 h of measurement time. If the immiscible oil phase serves only to separate samples, then 400 samples would be screened in <sup>∼</sup> 9 h. This could be enhanced by incorporating the ow pump into the control loop and adjusting the ow rate so that the oil plugs ow through the detection volume more quickly. In contrast, 300 samples require a measurement time of 120 h when done manually following the conventional procedure (<sup>500</sup> <sup>µ</sup><sup>L</sup> per sample, 5 mm sample tubes, 7 min acquisition time per sample neglecting instrument adjustment time after each sample exchange, without implementing automatic sample exchanger) [236].

Conclusions and Outlook

# **10 Conclusion and outlook**

Through the introduction chapter of this work, three main objectives were raised. The rst objective was enhancing the thin-lm micro-fabrication technology to address micro-MR applications. The second objective was utilizing the developed micro-fabrication technologies to solve MR-catheter issues regarding visual reliability and heating hazard. The third objective was employing the improved micro-fabrication technologies in tackling the NMR spectroscopy drawback of being a low throughput technique. This chapter concludes the key achievements in each objective and provides potential possibilities for future work.

## **10.1 Micro-fabrication of exible electronics**

An extensive study has been presented for the dierent depositing approaches for conductive seed-layers onto a exible polymer substrate. The study covers various photolithography methods as well as fast prototyping techniques. It focuses on the seed-layer adhesion onto the exible substrate as it is a challenging topic. Moreover, a benecial comparison between the dierent approaches has been given, covering each depositing technique's advantages, disadvantages, resolution, and reproducibility.

Two novel composite insulating inks have been developed for fully inkjet-printed high quality MIM capacitors with high capacitance density. Benzocyclobutene (BCB) was used as the base polymer for the developed inks due to its low dissipation factor. In addition, BaTiO<sup>3</sup> and TiO<sup>2</sup> were dispersed in the BCB polymer to increase its dielectric constant, reducing the capacitor size to be compatible with microelectronic applications. Dierent weight ratios of the ceramic powder to BCB polymer have been analyzed. The dielectric constant, roughness and thickness of the deposited lm, and quality factor have been characterized for each ink. The TiO2/BCB inks provide surface roughness of <sup>970</sup>-1220 nm, a loss tangent of 0.037-0.04, a dielectric constant of 7-20, and capacitance density <sup>32</sup>-64 pFmm<sup>−</sup><sup>2</sup> . The BaTiO3/BCB inks have surface roughnesses of <sup>830</sup>-1000 nm, loss tangents of 0.045-0.052, dielectric constants of 10-37, and capacitance densities of <sup>19</sup>-67 pFmm<sup>−</sup><sup>2</sup> . A comparison between the produced capacitors and state of art has been given. The developed printing process shows no blending between the conductive layer and the insulating layers. Moreover, the process is compatible with the fabrication of multi-layer components.

An organic Cu-PTAA-Ag Schottky diode has been developed by using inkjet-printing and the electroplating process. The low working function electrode (copper) of the diode was formed by printing silver electrode and then electroplating with copper. PTAA p-type semiconductors have been used due to their stability in the ambient atmosphere. The developed diode was successfully used to build a half-wave rectier, which provided a DC output voltage up to 3 V at 1 MHz.

A versatile and programmable current source based on MHCS was developed for the pulsed electroplating process. The current source have precise and stable output power even when operating under relatively high current densities (<sup>≈</sup> 6A dm<sup>−</sup><sup>2</sup> ) and various frequencies (DC up to few kHz). The developed system oers extensive freedom to set the pulsed electroplating parameters, opening further exploration of complex pulsed electroplating waveforms. The developed plating setup has been used in controlling the deposited layer's micro-structure and has been employed in several crucial applications: i) solving the delamination issue when electroplating inkjet-printed conductive traces on unstructured exible substrates, ii) electroplating of a laser carbonized polyaramid-based lm, iii) electroplating of carbon micro-lattice architecture achieved by 3D stereo-lithography printing. The current source is compatible with the parallelization concept, where the output current channel can be duplicated. A developed system with eight channels was successfully utilized for plating eight 4 inch wafer in parallel, which saved materials, time, and eort. The setup is cost-eective and straightforward to be copied and used laboratory which maintains moderate electronic skills.

## **10.2 Interventional MRI using MR-catheters**

A novel wireless decoupling approach has been introduced to address the MR-catheter heating hazard. Moreover, an innovative double-saddle coil is developed to maintain catheter visibility in MR images through the entire range of axial orientations during interventions. The system tracks a micro-resonator placed on the catheter tip wirelessly through induction coupling. The catheter micro-resonator is connected to a variable capacitor controlled via a DC signal provided by a half-wave rectier. The input power to the half-wave rectier is wirelessly transmitted through induction coupling between two coils tuned at a frequency far from the Larmor frequency. The MR-resonator on the catheter tip along with the decoupling system has been fabricated using inkjet-printing on a 25 µm Kapton lm. The developed system was successfully used for wirelessly controlling the resonance frequency of the catheter microresonator within a range of 15 MHz.

The proposed double saddle detector oers wireless tracking of MR-catheters with orientation independence during interventional MRI operations. A comparison between the performance of the proposed double saddle detector and that of a traditional single saddle detector was done using numerical simulations and veried by experiments. The generated magnetic elds *B*<sup>1</sup> by the micro detectors were numerically simulated using COMSOL multiphysics and were experimentally mapped in a low eld 1 T MRI scanner. The double saddle detector generated a homogeneously distributed magnetic eld surrounding the catheter tube. The induction coupling eciency between a surface resonator and the micro detectors was numerically simulated using COMSOL multiphysics and experimentally characterized using a home-built setup. The double saddle detector maintains near-constant coupling through the entire range of axial orientations. On the other hand, the coupling between the surface coil and the conventional

single detector decreases as the angle between their *B*<sup>1</sup> elds increases, and no coupling was observed at orthogonal orientation of their elds. A novel multi-layer fabrication processes for rolled MEMS were developed for fabricating the double saddle detector. One method was based on standard lithography micro-fabrication, and the other employed inkjet-printing. Both processes were successful in fabricating double saddle detectors, which were then rolled around <sup>a</sup> 5 fr catheter tube.

Within this work, the developed coupling system is fabricated with inkjet-printing except for the varicap. As a future plan, the varicap could also be fabricated via inkjet-printing of p-type and n-type semiconductors. Hence, a wireless tunable MR-resonator could be fully printed to be compatible with roll to roll inkjet-printing fabrication technique, enabling mass production and disposable electronics. The main focus of this work was developing and characterizing the hardware. However, the introduced wireless tuning methodology oers a large room for further enhancements and applications in micro MRI and NMR applications where the MRdetector has limited access. Therefore, the developed MR-catheter with a wireless decoupling system is ready for the next step to be tested in vivo.

## **10.3 Automated high throughput NMR spectroscopy**

A novel plug ow-based approach has been introduced to address the NMR spectroscopy drawback of low sample throughput. The proposed technology enhances NMR throughput by a factor of 40 (when operating at the minimum eciency) and reduces the NMR sample volume at the same time to under a hundred micro-liters. The introduced system consists of a custom-built NMR probe head that takes advantage of thin-lm micro-NMR detectors combined with microuidic technologies. The key concept integrates thin-lm ow capacitive sensors interfaced to a microcontroller unit. Together with the NMR sensor, they enable realtime detection of sample plugs in a ow-based sample handling system with high sensitivity to dierences in relative permittivity. The NMR signal acquisition system gains a degree of automation by implementing an independent measuring mode for ow positing and velocity, further improving throughput. A detailed description of the system design, fabrication, and characterization has been given in Chapter 8. The micro-NMR detector enables highresolution NMR spectroscopy with a sensitivity of <sup>2</sup>.18 nmol s1/<sup>2</sup> with the ow sensors in operation. The spectral resolution under stationary conditions, quantied by resonance line width, was found to be <sup>1</sup>.25 Hz for the water signal. The ow sensors had a high sensitivity to a relative permittivity detection dierence of 0.2, allowing them to distinguish between most typical solvents. The modeling, simulation, and characterization of the NMR-detector and ow sensor are included. The probe head is plug-and-play ready with market-available NMR systems. After installing the developed probe head into the NMR system, a microuidic tube is fed through the probe head loaded with a sample series, comprising of an unlimited number of dierent sample plugs, separated by an immiscible uid. The system was successfully used to automate ow-based NMR measurement in a 500 MHz NMR system. A complete characterization of the automated ow-based NMR system has been given in Chapter 9. It was demonstrated that fully automated NMR spectroscopy of nine dierent 120 µL samples could be done in <sup>3</sup>.6 min or eectively <sup>15</sup>.3 s per sample.

The NMR ow-sensing approach, described in this thesis, has opened up a new concept toward automated high throughput NMR spectroscopy by splitting a large sample into small droplets. The NMR spectrum is individually acquired for each droplet with high SNR. Furthermore, the developed NMR probe head is an additional step towards increasing the NMR system functionality, allowing for greater experimental exibility and complexity or improved SNR compared to standard equipment. The platform is compatible with the parallelization concept described in Chapter 7 and, as recently reported for low-eld NMR spectroscopy [237], potentially further increasing NMR sample throughput scaled by the number of parallel NMR channels. Additionally, the application space could be extended by reducing the system dimensions to enable particle sensing within a carrying medium towards, for example, NMR measurement of cells.

### **Bibliography**


thrombectomy eligible patients indicates signicant potential for growth of endovascular stroke therapy in the usa," *Journal of neurointerventional surgery*, vol. 9, no. 8, pp. 722–726, 2017.


# **A Appendix to chapter 3**

**Inkjet-printed BCB insulating ink**

Figure A.1: Inkjet-printed droplets and straight lines of the BCB insulating ink. The measured droplets diameter was ≈ 105µm, therefore, the drop spacing was set to 50 µm.

#### **Inkjet-printed composite insulating ink**


Table A.1: Surface morphology of the inkjet-printed insulating lm using composite inks.

#### **Spin-coated PTAA semiconductor of MS diode**

Figure A.2: Spin-coated PTAA semiconductor of MS diode with dierent weight concentrations. PTAA powder was dissolved in toluene solvent with concentrations of **(a)** 5wt.% and **(b)** 2wt.%. Both solutions were spin-coated at 1000 rpm for 1 min. High PTAA concentrations lead to a cracked semiconductor layer which results in a short circuit.

# **B Appendix to chapter 4**

## **2.1 Modied Howland Current Source (MHCS) modeling**

**Transfer function**

Figure B.1: Equivelant schematic circuit of the Modied Howland Current Source (MHCS).

The output current (*IOUT* ) of the *MHCS* can be expressed in terms of the input control signal (*VControl*) by applying Kirchho's Current Law (*KCL*) (nodal analysis) at the nodes from 1-4,

$$\text{(@ node 1: } \quad \frac{V^- - V\_{Control}}{R\_1} + \frac{V^- - V\_O}{R\_2} + \frac{V^- - V^+}{Z\_{in}} = 0,\tag{B.1a}$$

$$\text{Q0 node 2:} \quad \frac{V^{+}}{R\_{3}} + \frac{V^{+} - V\_{OUT}}{R\_{4}} + \frac{V^{+} - V^{-}}{Z\_{\text{in}}} = 0,\tag{\text{B.1b}}$$

$$\text{@ node 3:} \quad \frac{V\_O - V\_{OUT}}{r} + \frac{V\_O - V^-}{R\_2} + \frac{V\_O - (V^+ - V^-) \cdot A}{Z\_O} = 0,\tag{\text{B.1c}}$$

$$\text{Q} \text{ node 4:} \quad I\_{OUT} = \frac{V^{+} - V\_{OUT}}{R\_4} + \frac{V\_O - V\_{OUT}}{r}. \tag{8.1d}$$

An ideal operational amplier (*OPAMP*) is assumed to have innite open-loop gain, innite input impedance (*Zin*), and zero output impedance (*ZO*). Hence, no electric current ow between the input terminals of the opamp leading to *V* <sup>+</sup> =*V* <sup>−</sup>. Solving the three equations B.1a - B.1c together, *V* <sup>+</sup>, *V* <sup>−</sup> and *V<sup>O</sup>* can be expressed in terms of *VControl* and *VOUT* ,

$$V^{+} = V\_{OUT} \frac{R\_3}{R\_3 + R\_4},\tag{B.2a}$$

$$V\_O = V\_{OUT} \frac{R\_3(R\_2 + R\_1)}{R\_1(R\_3 + R\_4)} - V\_{Courlol} \frac{R\_2}{R\_1} \,. \tag{B.2b}$$

Substituting by B.2a and B.2b in B.1d, the output current (*IOUT* ) can be driven as follow,

$$I\_{\rm OUT} = -V\_{\rm control} \frac{R\_2}{R\_1 r} + V\_{\rm OUT} \frac{R\_1 (R\_4 + r) - R\_2 R\_3}{R\_1 r (R\_3 + R\_4)} \,. \tag{B.3}$$

#### **Virtual capacitance**

Following the technical report provided by Texas Instrument [136], the virtual output capacitance of the *MHCS* (*COUT* ) can be derived by calculating the *OPAMP*'s output current (*IO*) in the presence of a time-varying signal at the output node of the circuit (*dV<sup>T</sup>* /*dt*) and a very small error input signal (*Verror*) across the *OPAMP*'s inputs. The applied voltage at the output node of the *MHCS* is assumed to be a ramp signal moving from a negative voltage in the positive direction. The system is solved at the moment when the signal passes zero volts, (*V<sup>T</sup>* = 0+*dV<sup>T</sup>* /*dt*). The calculated current, in that case *IO*, is known as the capacitive current since it's related only to *dV<sup>T</sup>* /*dt*.

The rate of change of an *OPAMP*'s output voltage is function in the input signal across it's inputs and the Gain Bandwidth Product (*GBMP*):

$$\frac{d\text{V}\_O}{dt} = V\_{error} \, 2\pi GBWP,\tag{B.4a}$$

$$
\ldots \, V\_{error} = \frac{dVo}{dt} \, \frac{1}{2\pi G BWP}.\tag{B.4b}
$$

When applying a time-changing signal at the output node of the *MHCS* (*dV<sup>T</sup>* /*dt*), the rate of change of the voltage at the positive terminal of the *OPAMP* can be derived using potential divider rule:

$$\frac{dV^{+}}{dt} = \frac{R\_{3}}{R\_{3} + R\_{4}} \frac{dV\_{T}}{dt},\tag{B.5a}$$

Taking into consideration the applied error signal (*Verror* and assuming an ideal *OPAMP*, the the rate of change of the voltage at the negative terminal of the *OPAMP* can be derived:

$$\frac{dV^{-}}{dt} = \frac{R\_3}{R\_3 + R\_4} \frac{dV\_T}{dt} + V\_{error},\tag{B.5b}$$

hence, the rate of change of the voltage at the *OPAMP*'s output is the same as the change at

the negative input magnied by *R*<sup>1</sup> +*R*2/*R*1,

$$\frac{d\,\delta O}{dt} = \frac{R\_3}{R\_3 + R\_4} \,\frac{R\_1 + R\_2}{R\_1} \,\frac{d\,V\_T}{dt} - V\_{error} \,\frac{R\_1 + R\_2}{R\_1} \,. \tag{B.5c}$$

By solving Equations B.4b and B.5c together:

$$V\_{error} = \frac{dV\_T}{dt} \frac{1}{2\pi GBWP} \frac{R\_3}{R\_3 + R\_4} \frac{R\_1 + R\_2}{R\_1} - V\_{error} \frac{1}{2\pi GBWP} \frac{R\_1 + R\_2}{R\_1} \tag{B.6}$$

The momentary when *V<sup>T</sup>* = 0+*dV<sup>T</sup>* /*dt*, the value of the *OPAMP*'s output voltage at is:

$$V\_O = \frac{R\_1 + R\_2}{R\_1} \ V\_{error},\tag{B.7a}$$

$$\text{Let } V\_O = \frac{dV\_T}{dt} \frac{1}{2\pi G B W P} \frac{R\_3}{R\_3 + R\_4} \left[\frac{R\_1 + R\_2}{R\_1}\right]^2 - V\_{error} \frac{1}{2\pi G B W P} \left[\frac{R\_1 + R\_2}{R\_1}\right]^2. \tag{B.7b}$$

The momentary when*V<sup>T</sup>* = 0+*dV<sup>T</sup>* /*dt*, no electric current is going through *R*<sup>4</sup> and the current going through *r* is,

$$I\_O = I\_r = V\_O/r,\tag{B.8a}$$

$$\text{i.: } lo = \frac{dV\_T}{dt} \, \frac{1}{2\pi rGBWP} \, \frac{R\_3}{R\_3 + R\_4} \, [\frac{R\_1 + R\_2}{R\_1}]^2 - V\_{error} \, \frac{1}{2\pi rGBWP} \, [\frac{R\_1 + R\_2}{R\_1}]^2. \tag{B.8b}$$

This capacitive current is equivalent to *I<sup>O</sup>* = *C<sup>O</sup> dV<sup>T</sup>* /*dt*,

$$\left| C\_O = \frac{\mathcal{R}\_3 (\mathcal{R}\_1 + \mathcal{R}\_2)^2}{2\pi r \mathcal{R}\_1^2 (\mathcal{R}\_3 + \mathcal{R}\_4) G B W P}. \right| \tag{8.9}$$

## **2.2 Electronic schematics of the electroplating current source**

Figure B.2: Schematic of the module used for generating the voltage control signal.

Figure B.3: Schematic of the MHCS.

Figure B.4: Schematic of the power supply module.

Figure B.5: Schematic of the feedback module module.

Figure B.6: Schematic of the real-time clock (RTC) module.

Figure B.7: Full schematic without the power supply module.

Figure B.8: Schematic of the fan control module.

## **2.3 Source codes**

After connecting the developed current source circuit to the BeagleBone Black (BBB) wireless board, the following steps need to be followed for the BBB software setup:


#### **BeagleBone board conguration code**

The conguration code (in Shell Script) needs to run once after booting the BeagleBone board to congure the SPI and I2C pins :

```
1 #!/bin/bash
2 config-pin P9_17 spi_cs
3 config-pin P9-18 spi
4 config-pin P9-21 spi
5 config-pin P9-22 spi_sclk
6 config-pin P9-28 spi_cs
7 config-pin P9-29 spi
8 config-pin P9-30 spi
9 config-pin P9-31 spi_sclk
10 config-pin -q P9_17
11 config-pin -q P9_18
```
 config-pin -q P9\_21 config-pin -q P9\_22 config-pin -q P9\_28 config-pin -q P9\_29 config-pin -q P9\_30 config-pin -q P9\_31 config-pin P9\_19 i2c config-pin P9\_20 i2c config-pin -q P9\_19 config-pin -q p9\_20

#### **Electroplating power source code**

The following python code generate uni-polar waveform:

```
1 import Adafruit_BBIO.GPIO as GPIO #library controlling GPIOs
2 from Adafruit_BBIO.SPI import SPI #library controlling SPIs
3 import time #BBB time library
4
5 #function for calulating SPI 16bit word for a given current
6 def word(I):
7 Rone = 10000 #Resistors values of the MHCS
8 Rtwo = 6800
9 Rfive= 10
10 Vin = ((I)*Rtwo*Rfive/(Rone+Rfive))/1000 #MHCS transfer function
11 volt_b = int((abs(Vin)*32768/2.5)+32768)
12 if I > 0:
13 volt_b = volt_b ^ 0xffff
14 a = (volt_b >> 8) & 0xff
15 b = volt_b & 0xff
16 return a, b
17
18 #Setup GPIOs
19 GPIO.setup("P8_27",GPIO.OUT) #led
20 GPIO.setup("P8_29",GPIO.OUT) #led
21 GPIO.setup("P8_8",GPIO.OUT) #power supply
22 GPIO.setup("P8_7",GPIO.OUT) #fan
23 GPIO.setup("P9_25",GPIO.OUT) #LDAC
24 GPIO.setup("P9_27",GPIO.OUT) #CS
25
26 #Control GPIOs
27 GPIO.output("P8_27",GPIO.LOW) #led off
```

```
28 GPIO.output("P8_29",GPIO.LOW) #led off
29 GPIO.output("P8_8",GPIO.LOW) #power supply off
30 GPIO.output("P8_7",GPIO.LOW) #fan off
31 GPIO.output("P9_25",GPIO.LOW) #LDAC is tied low
32 GPIO.output("P9_27",GPIO.HIGH)
33
34 time.sleep(0.05)
35
36 # Send the SPI word from the BBB to the ADC
37 spi = SPI(2,1)
38 a, b = word(0+2.7)
39 GPIO.output("P9_27",GPIO.LOW)
40 print spi.xfer2([a,b])
41 GPIO.output("P9_27",GPIO.HIGH)
42 GPIO.output("P8_8",GPIO.HIGH)
43
44 #A loop for generating a uni-polar waveform with Ip=20mA, TON=200ms, and
      TOFF=100ms.
45 while 1:
46 a, b = word(20)
47 GPIO.output("P9_27",GPIO.LOW)
48 print spi.xfer2([a,b])
49 GPIO.output("P9_27",GPIO.HIGH)
50 time.sleep(0.2)
51 a, b = word(0+2.9)
52 GPIO.output("P9_27",GPIO.LOW)
53 print spi.xfer2([a,b])
54 GPIO.output("P9_27",GPIO.HIGH)
55 time.sleep(0.1)
```
The previous Python code is used for generating any other waveform by replacing the part of the code inside the while loop (from line number 46) with the required waveform. Below is a python code template for explaining how to generate any waveform:

```
1 #This following block with five lines is used to generate a single pulse
    :
2
3 #The following code line is used to take the required current's value (I
    ) and convert it to a 16-bit word using the previously defined
    function.
4 #The value of I is in mA and can be positive, negative, or zero.
```

```
5 a, b = word(I)
6
7 #The following three code lines are used to transfer the 16-bit word
    from the BBB to the ADC.
8 GPIO.output("P9_27",GPIO.LOW)
9 print spi.xfer2([a,b])
10 GPIO.output("P9_27",GPIO.HIGH)
11
12 #The final line of the code is used to set the pulse-width by inserting
    T in ms.
13 time.sleep(T)
14
15 #The previous code block is repeated each time the current level change
    within a single waveform cycle.
```
Below are the python codes for the plating waveforms used through the thesis:

Bipolar waveform: *I*<sup>C</sup> = 20mA, *I*<sup>A</sup> = −20mA, *T* <sup>C</sup> = 200ms, and *T*<sup>A</sup> = 50ms.

```
1 # Positive pulse
2 a, b = word(20)
3 GPIO.output("P9_27",GPIO.LOW)
4 print spi.xfer2([a,b])
5 GPIO.output("P9_27",GPIO.HIGH)
6 time.sleep(0.2)
7 # Negative pulse
8 a, b = word(-20)
9 GPIO.output("P9_27",GPIO.LOW)
10 print spi.xfer2([a,b])
11 GPIO.output("P9_27",GPIO.HIGH)
```

```
12 time.sleep(0.05)
```
Bipolar with o time: *I*<sup>C</sup> = 20mA, *I*<sup>A</sup> = −20mA, *T* <sup>C</sup> = 200ms, *T*<sup>A</sup> = 50ms, *T* O = 10ms, and *T* 'O = 20ms.

```
1 #Positive pulse
2 a, b = word(20)
3 GPIO.output("P9_27",GPIO.LOW)
4 print spi.xfer2([a,b])
5 GPIO.output("P9_27",GPIO.HIGH)
6 time.sleep(0.2)
7 #Off
8 a, b = word(0)
```
GPIO.output("P9\_27",GPIO.LOW)

```
10 print spi.xfer2([a,b])
```

```
13 #Negative pulse
```

```
14 a, b = word(-20)
```

```
15 GPIO.output("P9_27",GPIO.LOW)
```

```
16 print spi.xfer2([a,b])
```

```
17 GPIO.output("P9_27",GPIO.HIGH)
```

```
18 time.sleep(0.05)
```

```
19 #Off'
```

```
20 a, b = word(0)
```

```
21 GPIO.output("P9_27",GPIO.LOW)
```

```
22 print spi.xfer2([a,b])
```

```
23 GPIO.output("P9_27",GPIO.HIGH)
```

```
24 time.sleep(0.02)
```
# **C Appendix to chapter 6**

**Frequency splitting modeling**

Figure C.1: Equivalent schematic of inductively coupled resonators. **(a)** Magnetic ux linkage between two indicators. **(b)** Equivalent schematic circuit of a pair of inductively coupled resonators.

As a pair of inductors approach each other, the electric current (I) variation in one of the inductors developed a magnetic eld around the inductor that induces an electromotive force (emf) in the other inductor, as shown Fig. C.1a. The magnetic coupling linking inductors is described by Faraday's law of induction, where the emf is the variation of the magnetic ux with time. The magnetic ux, in turn, is dened by the magnetic ux density (B) and the enclosed area (A),

$$\text{emf} = \frac{d\phi\_B}{dt}, \quad \phi\_B = \int\_{\Sigma} B \, dA. \tag{C.1}$$

The electric currents owing in both inductors (*I*<sup>1</sup> and *I*2) contribute to the ux leakage of each inductor (λ<sup>1</sup> and λ2),

$$
\mathcal{A}\_{\parallel} = L\_{\parallel}I\_{\parallel} + M\_{\parallel 2}I\_{\perp}, \tag{C.2a}
$$

$$
\lambda \mathfrak{z} = L \mathfrak{z} I \mathfrak{z} + M \mathfrak{z} I \mathfrak{z}, \tag{\text{C.2b}}
$$

where *L*<sup>1</sup> and *L*<sup>2</sup> are the self-inductance of each inductor, and *M* is the mutual induction connecting the two inductors. The magnitude of the magnetic coupling can be expressed by calculating the ratio between the ux linkage of the two inductors at two special cases: 1) in the one case, the rst inductor is left open (*I*<sup>1</sup> = 0); and 2) in the other case, the second inductor is left open (*I*<sup>2</sup> = 0). The previous calculation result in a ratio between the mutual inductance and the mean of the self-inductance, which is known as coupling coecient (K),

$$\boxed{K = \frac{M\_{12}}{\sqrt{L\_1 L\_2}}}\,\,\,\,\tag{C.3}$$

An electrical resonator is an inductor connected in parallel to a capacitor, as shown in Fig. C.1b. Practically, the conductive wire forming the inductor has a resistance, thus the total impedance (*Z*) of a resonator (*RLC*) is,

$$Z = j(a\alpha L - \frac{1}{a\alpha C}) + R(a\alpha). \tag{C.4}$$

The resonance frequency (ω0) of a resonator is the frequency at which the electrical inductance and capacitance compensate each other. At the resonance frequency, the energy on the resonator oscillates between its two forms; electrical and magnetic. From Equation C.4, the resonator's resonance frequency can be derived:

$$
\rho\_0 = \frac{1}{\sqrt{LC}}.\tag{C.5}
$$

Equation C.2 reveals that mutual inductance, as for self-inductance, relates magnetic ux and electric current. Hence, mutual inductance contributes to the resonator's impedance and consequently inuences its resonance frequency. According to Kirchho's law, the loop equations of a pair of coupled resonators equivalent to zero [238]:

$$Z\_1 I\_1 + j \alpha M\_{12} I\_2 = 0,\tag{C.6a}$$

$$Z\_2 I\_2 + j\alpha M\_{12} I\_1 = 0.\tag{C.6b}$$

Solving these two loop equations together results in two resonance frequencies (ω<sup>1</sup> and ω2). This means that magnetic coupling between a pair of resonator introduces an additional degree-of-freedom (current path), as shown:

$$a\_{1,2}^2 = (a\_{01}^2 + a\_{02}^2) \frac{-1 \pm \sqrt{1 - \frac{4(1 - K^2)}{\left(\alpha\_{01}/\alpha\_{02}\right)^2 + \left(\alpha\_{02}/\alpha\_{01}\right)^2 + 2}}}{2(1 - K^2)}.\tag{C.7}$$

This Equation shows the two resulting resonance frequencies (ω1,2) from the magnetic coupling depends on the individual resonators' initial resonance frequencies(ω<sup>01</sup> and ω02). In case both of the inductively coupled resonators are tuned to the same resonance frequency (ω<sup>01</sup> = ω<sup>02</sup> = ω0), a splitting of this initial resonance frequency occurs, resulting in two resonance frequencies shifted to the right and left of the initial value:

$$
\omega\_{1,2} = \frac{a\_0}{\sqrt{1 \pm K}}.\tag{C.8}
$$

Applying a potential dierence to one of the two coupled resonators (*VS*) induces an output voltage in the other resonator (*VOUT* ), as shown in Fig. C.1b. By solving the two loop equations in (C.6), we obtain:

$$\boxed{V\_{\text{Out}} = V\_S \frac{Z\_2 ja\bullet M\_{12}}{Z\_1 Z\_2 + (ja\bullet M\_{12})^2}} \,\,\,\,\,\tag{C.9}$$

Figure C.2: ADS simulation environment of inductively coupled resonators.

#### **Snier coil**

A *snier coil* is a probe coil used for the experimental characterization of the resonators, e.g., determining the resonance frequency of the resonators [183,239]. Snier coil is made of a 50 Ω semi-rigid coaxial cable with a particular conguration as a loop antenna. The main advantage of the snier coil is that it is sensible only to magnetic elds and suppresses the eect of electric elds on the inner conductor of the cable. The snier coil is constructed by stripping o the heat shrink tubing and insulation materials of one end of the coaxial cable, connecting the ground and signal wires to an SMA connector. At the other end, the ground wire, heat shrink tubing, and the insulation materials are stripped o. Consequently, the signal wire is rounded in the form of a small loop and soldered to the ground to achieve a short circuit, as shown in Fig.5a. Based on Faraday's law, if a magnetic eld passes through a coil loop, a voltage is induced proportional to the magnetic ux's rate of change,

$$
\varepsilon = -N \frac{\Delta \phi}{\Delta t},
\tag{C.10}
$$

where ε is the induce voltage, *N* is the number of turns, and ∆ϕ/∆*t* is the rate of change of

Figure C.3: **(a)** Photo of a snier coil. **(b)** Expected reection coecient (S11) of a snier coil.

the magnetic ux. A resonator's resonance frequency can be determined by placing the snier coil in its vicinity and measuring the reection coecient (S11) of the snier coil over a wide range of frequencies. The whole signal going through the snier coil will be reected except for the resonator's resonance frequency, as the magnetic eld generated by the snier coil at this frequency will be converted into an induced voltage in the resonator. Hence, a dip in the snier coil S<sup>11</sup> curve will be observed at the resonator's resonance frequency as it absorbed the snier coil magnetic eld, as illustrated in Fig.5b.

# **D Appendix to chapter 8**

## **4.1 Technical working drawing**

Figure D.1: Technical working drawing of the 3D printed probe head holder. The insert was printed using an Ultimaker 2<sup>+</sup> printer out of polylactic acid (PLA). All dimensions are in mm, and the gure scale is 1.5:1.

Figure D.2: NMR probe compatible with ow-based spectroscopy. The probe was designed and built by *Erwin Fuhrer*.

## **4.2 Source codes**

#### **NMR acquisition software pulse program**

The pulse program was written for software: TopSpin 3.5pl2, Bruker BioSpin.

```
1 #include <Avance.incl>
2 "d11=1m" ;Delay for writing to disc
3 "l2 = 1"
4 ze ;Reset the scan counter
5 1 d1
6 2 p1 ph1
7 0.1m
8 if trigpl2
9 {
10 if "l2 != 1"
11 {
12 goto 3 ;l2 is greater than 1
13 }
14 else
15 {
16 0.2u iu2 ;l2 is less than 1
17 go=1 ph31
18 }
19 }
20 else
21 {
22 0.2u iu2 ;Trigger is not on
23 go=1 ph31
24 }
25 3 d11 wr #0 if #0 ;FID is written
26 0.1u ru2
27 lo to 1 times td1
28 exit
29 ph1=0 2 2 0 1 3 3 1
30 ph31=0 2 2 0 1 3 3 1
```
#### **Micro-controller unite source code**

The MCU source code was written for Arduino Uno board using Arduino IDE software.

```
1 #include <Wire.h> //I2C library
2
3 ////Define Variables////
4 int flag_1 = 0;//flag
5 int flag_2 = 0;//flag
6 unsigned long t;//hold current time
7 unsigned long tt;//hold current time
8 unsigned long d = 0;//Delay before TTL signal
9 unsigned long sensor_1_th = 3500000;//1st IDC sensor threshold
10 unsigned long sensor_2_th = 5000000;//2nd IDC sensor threshold
11 int TTL_width = 500;//The TTL pulse width in ms
12 int button_state = 0 ;//hold the manual triggering pushbutton state
13 byte shift = 88;//Variable to shift the capacitance input range
14 int TTL = 2;//Trigger signal Digital PIN
15 int TTL_LED = 3;//Trigger signal LED state Digital PIN
16 int Sample_LED_1 = 6;//Sample LED state 1 Digital PIN
17 int Sample_LED_2 = 7;//Sample LED state 2 Digital PIN
18 float sensor_1_value;//hold 1st IDC sensor value
19 float sensor_2_value;//hold 2nd IDC sensor value
20
21 ////AD7746 registers' addresses////
22 #define RGTR_Status 0x00
23 #define RGTR_Cap_Data 0x01
24 #define RGTR_Cap_Setup 0x07
25 #define RGTR_Exc_Setup 0x09
26 #define RGTR_Configuration 0x0A
27 #define RGTR_Cap_DAC_A 0x0B
28 #define RGTR_Cap_Offset 0x0D
29 #define I2C_ID 0x48
30 #define Reset_ID 0xBF
31 #define Manual_TTL A0//Pushbutton for manual triggering
32
33 ////Set the IC (AD7746) settings////
34 void setup()
35 {
36 Wire.begin();//Start I2C connection
37 Serial.begin(9600);//Serial communication's baud rate
```

```
38 pinMode(TTL,OUTPUT);//Set TTL pin as an output
39 pinMode(TTL_LED,OUTPUT);//Set TTL LED state pin as an output
40 pinMode(Sample_LED_1,OUTPUT);//Sample LED state 1 pin as an output
41 pinMode(Sample_LED_2,OUTPUT);//Sample LED state 2 pin as an output
42 digitalWrite(TTL,LOW);//Start with TTL switched off
43 digitalWrite(TTL_LED,LOW);//Start with TTL LED switched off
44 digitalWrite(Sample_LED_1,LOW);//Start with sample LED 1 switched off
45 digitalWrite(Sample_LED_2,LOW);//Start with sample LED 2 switched off
46 Wire.beginTransmission(I2C_ID);//start I2C communication channel
47 Wire.write(Reset_ID);//Reset the IC (Ad7756) settings
48 Wire.endTransmission();//Terminate I2C communication channel
49 delay(50);//Delay 50 ms
50 updatergtr(RGTR_Exc_Setup,_BV(5)|_BV(3)|_BV(1)|_BV(0));//Enable EXC_A
    and EXC_B with a voltage level of Vdd
51 updatergtr(RGTR_Cap_Setup,_BV(7));// Enable capacitive channel
52 updatergtr(RGTR_Configuration ,_BV(7)|_BV(6)|_BV(5)|_BV(4)|_BV(3)|_BV
    (0));//Set Conversion times and mode
53 shift &=0x7f;//AND with 1111111 to make sure shift is 7bit
54 updatergtr(RGTR_Cap_DAC_A,_BV(7)|shift);//shift the capacitance input
    range
55 }
56
57 ////Main program////
58 void loop()
59 {
60 button_state = analogRead(Manual_TTL);// Read the manual triggering push
     button's state
61 if (button_state > 0)//In case the pushbutton is pressed, send a TTL
    signal
62 {digitalWrite(TTL,HIGH);
63 digitalWrite(TTL_LED,HIGH);
64 delay(TTL_width);
65 digitalWrite(TTL,LOW);
66 digitalWrite(TTL_LED,LOW);}
67
68 updatergtr(RGTR_Cap_Setup, 0b10000000);//Switch to the first capacitive
    input
69 sensor_1_value = readcap();//Get the value of the first sensor
70 while(sensor_1_value > sensor_1_th)
71 {t = millis();
```

```
72 if (flag_1==0)
73 {delay(d);
74 digitalWrite(TTL,HIGH);
75 digitalWrite(TTL_LED,HIGH);
76 delay(TTL_width);
77 digitalWrite(TTL,LOW);
78 digitalWrite(TTL_LED,LOW);
79 digitalWrite(Sample_LED_1,HIGH);
80 flag_1 = 1;}
81 updatergtr(RGTR_Cap_Setup,0b11000000);//Switch to the second
    capacitive input
82 sensor_2_value = readcap();//Get the value of the first sensor
83 updatergtr(RGTR_Cap_Setup, 0b10000000);//Switch to the first
    capacitive input
84 sensor_1_value = readcap();//Get the value of the first sensor
85 while (sensor_2_value > sensor_2_th)
86 {tt = millis();
87 d = (tt-t)/3;
88 digitalWrite(Sample_LED_2,HIGH);
89 updatergtr(RGTR_Cap_Setup, 0b10000000);//Switch to the
    first capacitive input
90 sensor_1_value = readcap();//Get the value of the first
    sensor
91 updatergtr(RGTR_Cap_Setup,0b11000000);//Switch to the
    second capacitive input
92 sensor_2_value = readcap();//Get the value of the first
    sensor
93 if (sensor_1_value < sensor_1_th && flag_2==0)
94 {delay(d);
95 digitalWrite(TTL,HIGH);
96 digitalWrite(TTL_LED,HIGH);
97 delay(TTL_width);
98 digitalWrite(TTL,LOW);
99 digitalWrite(TTL_LED,LOW);
100 digitalWrite(Sample_LED_1,LOW);
101 flag_2 = 1;}
102 }
103 digitalWrite(Sample_LED_1,LOW);
104 flag_1 = 0;
105 flag_2 = 0;}
```

```
106 updatergtr(RGTR_Cap_Setup,0b11000000);//Switch to the second capacitive
    input
107 sensor_2_value = readcap();//Get the value of the first sensor
108 while(sensor_2_value > sensor_2_th)
109 {t = millis();
110 if (flag_1==0)
111 {delay(d);
112 digitalWrite(TTL,HIGH);
113 digitalWrite(TTL_LED,HIGH);
114 delay(TTL_width);
115 digitalWrite(TTL,LOW);
116 digitalWrite(TTL_LED,LOW);
117 digitalWrite(Sample_LED_1,HIGH);
118 flag_1 = 1;}
119 updatergtr(RGTR_Cap_Setup, 0b10000000);//Switch to the first
    capacitive input
120 sensor_1_value = readcap();//Get the value of the first sensor
121 updatergtr(RGTR_Cap_Setup,0b11000000);//Switch to the second
    capacitive input
122 sensor_2_value = readcap();//Get the value of the first sensor
123 while (sensor_1_value > sensor_1_th)
124 {tt = millis();
125 d = (tt-t)/3;
126 digitalWrite(Sample_LED_2,HIGH);
127 updatergtr(RGTR_Cap_Setup,0b11000000);//Switch to the
    second capacitive input
128 sensor_2_value = readcap();//Get the value of the first
    sensor
129 updatergtr(RGTR_Cap_Setup, 0b10000000);//Switch to the
    first capacitive input
130 sensor_1_value = readcap();//Get the value of the first
    sensor
131 if (sensor_2_value < sensor_2_th && flag_2==0)
132 {delay(d);
133 digitalWrite(TTL,HIGH);
134 digitalWrite(TTL_LED,HIGH);
135 delay(TTL_width);
136 digitalWrite(TTL,LOW);
137 digitalWrite(TTL_LED,LOW);
138 digitalWrite(Sample_LED_1,LOW);
```

```
139 flag_2 = 1;}
140 }
141 digitalWrite(Sample_LED_1,LOW);
142 flag_1 = 0;
143 flag_2 = 0;}
144 }
145 ////End main program////
146
147 long readcap()//Function to read the input capacitance
148 { long capreturn = 0;
149 uint8_t data[3];
150 char status = 0;
151 while (!(status & (_BV(0) | _BV(2))))//wait for conversion
152 {status= readrgtr(RGTR_Status);}
153 unsigned long value = readLong(RGTR_Cap_Data);
154 value >>=8;
155 capreturn = value;
156 return capreturn;}
157
158 void updatergtr(unsigned char m, unsigned char n)//Function to write
     data to a register in AD7746
159 { Wire.beginTransmission(I2C_ID);
160 Wire.write(m);
161 Wire.write(n);
162 Wire.endTransmission();}
163
164 unsigned char readrgtr(unsigned char m)//Function to read data from a
     register in AD7746
165 { unsigned char n;//Variable to hold the received data
166 Wire.beginTransmission(I2C_ID);
167 Wire.write(m);//Address of the register to read
168 Wire.endTransmission();
169 Wire.requestFrom(I2C_ID, 1);
170 while(Wire.available()==0) {}//Wait till data is received
171 n = Wire.read();
172 return n;}
173
174 unsigned long readLong(unsigned char r)
175 { union {
176 char data[4];
```

```
177 unsigned long value;
178 }
179 byteMappedLong;
180 byteMappedLong.value = 0L;
181 Wire.beginTransmission(I2C_ID);//begin read cycle
182 Wire.write(0);//pointer to first data register
183 Wire.endTransmission();//end cycle
184 //the data pointer is reset anyway - so read from 0 on
185 Wire.requestFrom(I2C_ID,r+4);//reads 2 bytes plus all bytes before the
      register
186 while (!Wire.available()==r+4) {
187 ; //wait
188 }
189 for (int i=r+3; i>=0; i--) {
190 uint8_t c = Wire.read();
191 if (i<4) {
192 byteMappedLong.data[i]= c;
193 }
194 }
195 return byteMappedLong.value;}
```
## **4.3 Automatic loading system GUI**

The MCU controls the robot arm and the syringe pump for loading the microuidic tube with the sample plugs. The user lls a 96 well plate with the target samples and immiscible uid required to be loaded into the microuidic tube. The user sets the lling sequence of the experiment via the excel workbook le (below). Three columns are available in the "Sample Sequence" sheet of the worksheet. The rst column contains well plate locations and is limited by selecting 96 possible well coordinates for the 96 well plate, the order of the entered well coordinates dene the loading order. The second column corresponds to the name of the sample, the user is free to enter any descriptive name. The third column is used for specifying the sample volume in µL, positive values for loading the plug into the tube, and negative values to push the sample out of the tube. Once the experiment parameters are entered, the user presses the button "Program Experiment". The button will launch a back-end accompanying python script that transfers the values entered by the user to the MUC, which in turn controls the robot arm's motion with synchronizing the pump activity to complete the required lling sequence.


Figure D.3: GUI of the automatic sample loading system.

## **4.4 Electronic schematics**

1 cm

Figure D.4: Probe head PCB using CDC chip.

Figure D.5: Probe head PCB without CDC footprint to be used with Lock-in amplier.

Figure D.6: Arduino Uno NMR shield.

Figure D.7: Arduino Micro NMR shield

#### **Microfabrication results (full)**

Figure D.8: Representative microfabrication results of the ow NMR detector and sensors. **(a)** A photographic image of a fully patterned Kapton foil in the shape of a 100 mm wafer. The foil contains a batch of the micro-saddle NMR coil and the IDC ow sensor, demonstrating the developed process's compatibility with mass production. The sub-gures illustrate focused microscopic images of the 2D patterned **(b)** Micro-saddle coil and **(c)** IDC sensor.

# **List of Figures**




List of Figures



D.8 Representative microfabrication results of the ow NMR detector and sensors. 229

# **List of Tables**


### **Publications**

## **Journal articles**


#### **Proceedings**


## **Conference contributions**


## **Acknowledgement**

I would like to express my sincere gratitude to my supervisor Prof. Dr. Jan Korvink for giving me the opportunity to conduct my PhD degree in his group. Working under your supervision taught me a lot, not only on the scientic side but also on the personal and thinking side. Your support, guidance, and advice are invaluable, and I am incredibly grateful for all the knowledge you passed to me. I appreciate the vibrant and exciting research atmosphere you established in your group, which made the journey full of fun and a fantastic experience. I appreciate the freedom you oer to the students for directing their research and the time you spend with the students, constantly listening and oering out-of-the-box solutions and intelligent ideas.

I am my very thankful to Prof. Dr. Michael Bock for examining my thesis. I would like to express my gratitude for hosting me in a visit to your laboratory. It is my great pleasure to have you in my examination committee.

I owe my deepest gratitude to my group leader Dr. Dario Mager for his outstanding support and guidance. You were always willing to assist and support not only on the research side but also in professional and personal life. Thank you very much, it wouldn't be possible without your help. I would also like to express my most tremendous appreciation to Dr. Neil MacKinnon for the knowledge he gave me in the eld of NMR. I have learned a lot from you. I am very appreciative of the success we achieved together, it was my pleasure collaborating with you.

Many thanks to Prof. Dr. Uli Lemmer for spending time and eort attending my TAC meetings. I am grateful for your advice and motivation.

I would like to express special gratitude to Dr. Mazin Jouda and Dr. Hossein Davoodi for their kindness, friendship, and support. They are always willing to support and never hesitate to oer help and advice. I appreciate all the help and knowledge you provided. I would like to thank Emil Mamleyev for all the help and support he provided regarding the SEM images and the technical advice in the eld of material science. I would like to acknowledge Dr. Nurdiana Nordin for answering all my chemistry questions. I would like to present my gratitude to Shyam Adhikari for providing help and support in the cleanroom. I would like to thank Erwin Fuhrer for the fruitful discussions we had together.

It was my pleasure to share the oce with Anna Zakhurdaeva, Sagar Wadhwa, and Natheer Almtireen. Thank you for all the moments and discussions we had together for more than three years. I would like to thank the rest of the low-cost MEMS group Markus Meissner, Dr. Monsur Islam, Ahsana Sadaf. I was lucky to be part of the group.

I would like to show my most tremendous appreciation to the entire Prof. Korvink group for the professional and friendly environment they provide. I would like to thank Prof. Dr. Jürgen Brandner, Dr. Lorenzo Bordonali, Albina Julius, Mohammad Abdo, Mehrdad Alinaghian, Andrea Rivera, Julia Schulte-Hermann, Pedro Silva, Sofa Toto, Dr. Ali Moazenzadeh, and Dr. Ronald Kampmann.

I am particularly grateful for the assistance provided by the cleanroom team at IMT, Dr. Uwe Köhler, Mrs. Heike Fornasier, Mr. Marco Heiler, Mrs. Barbara Matthis, and Dr. Richard.

I would like to sincerely acknowledge the nancial support from the DAAD (German Academic Exchange Service) under the German Egyptian Research Long-Term Scholarship Program (GERLS) and the partial funding provided by the BioInterfaces International Graduate School (BIF-IGS) at the Karlsruhe Institute of Technology (KIT) and the European Union's Future and Emerging Technologies Framework (Grant H2020-FETOPEN-1-2016-2017-737043- TISuMR).

My heartfelt gratitude goes to my lovely parents Amr and Eman, my brother Alaa, my sister Sarah and my wife's family. Thank you very much for everything. It was only possible because of your motivation, support, and prayers.

Most important, I would like to thank my lovely wife Nourhan for all the support she provided throughout the entire process. I am indebted to you for everything and for taking care of our lovely son Malek, whom I am grateful for all the joy he added to my life. I am thankful for all the moments, happiness, and challenges we went through together. It was only you the reason for me to reach this point.

Schriften des Instituts für Mikrostrukturtechnik am Karlsruher Institut für Technologie (KIT)

#### ISSN 1869-5183

Band 1 Georg Obermaier Research-to-Business Beziehungen: Technologietransfer durch Kommunikation von Werten (Barrieren, Erfolgsfaktoren und Strategien). ISBN 978-3-86644-448-5


Die Bände sind unter www.ksp.kit.edu als PDF frei verfügbar oder als Druckausgabe zu bestellen.


Die Bände sind unter www.ksp.kit.edu als PDF frei verfügbar oder als Druckausgabe zu bestellen.


Die Bände sind unter www.ksp.kit.edu als PDF frei verfügbar oder als Druckausgabe zu bestellen.

Band 22 Michael Röhrig Fabrication and Analysis of Bio-Inspired Smart Surfaces. ISBN 978-3-7315-0163-3

#### Band 23 Taleieh Rajabi

Entwicklung eines mikrofluidischen Zweikammer-Chipsystems mit integrierter Sensorik für die Anwendung in der Tumorforschung. ISBN 978-3-7315-0220-3

Band 24 Frieder Märkle Laserbasierte Verfahren zur Herstellung hochdichter Peptidarrays. ISBN 978-3-7315-0222-7


Band 29 Nicole E. Steidle Micro- and Nanostructured Microfluidic Devices for Localized Protein Immobilization and Other Biomedical Applications. ISBN 978-3-7315-0297-5

Band 30 Jochen Heneka Prozessentwicklung eines industrietauglichen Verfahrens zur Fertigung von vereinzelten LIGA-Mikrobauteilen. ISBN 978-3-7315-0326-2


Band 35 Anne Bäcker Veränderliche 3D Zellgerüstträger auf Cryogelbasis zur Kultivierung von Prostatakarzinomzellen. ISBN 978-3-7315-0676-8

Band 36 Frieder Johannes Koch X-ray optics made by X-ray lithography: Process optimization and quality control. ISBN 978-3-7315-0679-9

Band 37 Tobias Jörg Schröter Vergrößerung des Sehfeldes der Röntgen-Phasenkontrast-Bildgebung für die klinische Anwendung. ISBN 978-3-7315-0731-4

Band 38 Felix Vüllers

Bioinspired Superhydrophobic Nano- and Microstructured Surfaces for Drag Reduction and Optoelectronics. ISBN 978-3-7315-0816-8

Band 39 Frederik Kotz Entwicklung neuer Materialien für die additive Fertigung und das Rapid Prototyping von Glas und Polymethylmethacrylat. ISBN 978-3-7315-0835-9

Band 40 Michael Oldenburg Photon upconversion heterostructures made from surface-anchored metal-organic frameworks. ISBN 978-3-7315-0863-2


Band 44 Andreas Striegel Werkzeug- und Prozessentwicklung des Roll-to-Roll-Hochdurchsatzverfahrens zur kontinuierlichen, großflächigen Mikrostrukturierung. ISBN 978-3-7315-1106-9

Die Bände sind unter www.ksp.kit.edu als PDF frei verfügbar oder als Druckausgabe zu bestellen.

Band 45 Marlene Kopf Langzeitstabilität der Innendrücke von Kavernen benachbarter MEMS-Sensoren auf Siliziumbasis. ISBN 978-3-7315-1121-2

Band 46 Simon Ternes

In Situ Characterization and Modelling of Drying Dynamics for Scalable Printing of Hybrid Perovskite Photovoltaics. ISBN 978-3-7315-1255-4

Band 47 Omar Nassar Innovative micro-NMR/MRI functionality utilizing flexible electronics and control systems. ISBN 978-3-7315-1176-2

Die Bände sind unter www.ksp.kit.edu als PDF frei verfügbar oder als Druckausgabe zu bestellen.

#### OMAR NASSAR

Innovative micro-NMR/MRI functionality utilizing flexible electronics and control systems

The advantages offered by the flexible electronics and control systems technologies were utilized for tackling the challenges facing two crucial Magnetic Resonance (MR) applications. The first application is in the field of interventional Magnetic Resonance Imaging (MRI), and the other application is in the field of Nuclear Magnetic Resonance spectroscopy (NMR). The first part of the book demonstrates the progress in the various flexible electronics fabrication technologies. The work focuses on developing components required by micro-scale MR applications, such as inductors, capacitors, and diodes. The second part of the book addresses the interventional MRI application using MR-catheters. A wireless tuning approach is introduced to solve the heating and reliability issues involved with MR-catheters. The micro detector loaded on the catheter is controlled wirelessly via a control pulse sequence sent to a radio-frequency (RF) tuning circuit embedded with the micro-detector. The third part of the book addresses NMR spectroscopy throughput. A novel approach is introduced for fully automated high throughput NMR spectroscopy with enhanced mass sensitivity. The developed system is accomplished by taking advantage of microfluidic technologies combined with thin-film micro-NMR detectors.

ISSN 1869-5183 ISBN 978-3-7315-1176-2