Polyaniline-based neuromorphic devices towards interfacing sensing

UNIVERSITA’ DEGLI STUDI DI PARMA DOTTORATO DI RICERCA IN “Scienza e Tecnologia dei Materiali” CICLO XXXIV Polyaniline-based neuromorphic devices towards interfacing sensing Coordinatore: Prof. Enrico Dalcanale Tutore: Dr. Salvatore Iannotta Co-Tutore: Dr. Pasquale D’Angelo Dottorando: Roman Sajapin Anni academici: 2018/19 – 2021/22 Note of thanks Before all, I would like to thank Dr. Salvatore Iannotta for the assignment of this challenging and interesting topic, the trust he placed in me to accomplish the goals that have been set, and the great opportunity to learn and grow beyond myself in the process. I am very grateful for our fruitful collaboration and admire his readiness for constructive discussions and the ability to find approaches to upcoming challenges. I learned a great deal about scientific work and his guidance throughout this period has been invaluable. I felt and enjoyed a pleasant amount of mutual appreciation for the work we have done which truly motivated me all along. A huge thank you goes to Dr. Pasquale D’Angelo for his willingness to co-supervise my thesis, the many discussions we had and the many valuable suggestions I received from him. Some of the most interesting ideas were born during our conversations, sharing thoughts and doubts alike. I will always appreciate his upbeat demeanour that was very encouraging and a needed source of optimism for me, especially during the last year of my PhD program. Dr. Tatiana Berzina deserves my biggest gratitude as she introduced me to our exciting research field and practically showed me everything there was to know about the materials and techniques we worked with. I will remember the times when we worked together as some of the most exciting ones. The work with Tatiana taught me the value of being self-critical and maintaining a healthy amount of doubt about obtained results. At the same time, it taught me to stand by my own convictions, even if they are not initially met with approval. Without a doubt, my thesis would not have been the same without her involvement. I would like to thank all the other members of our research group at IMEM and the Physics department, Dr. Victor Erokhin, Dr. Svetlana Erokhina, Dr. Silvia Battistoni, Dr. Valentina Ricci, Dr. Regina Burganova, Dr. Adelia Faizullina, Dr. Stefania Boi, Dr. Davide Vurro and Dr. Giuseppe Tarabella. It was an enrichment to work and exchange opinions with them and I have learned lots of valuable lessons over the course of these three years, not only about science, but life in general. A warm thank you to Stefania for our much appreciated, encouraging conversations. As I am sure most who worked with her would agree, the period when she joined our group felt very refreshing. Special thanks go to the administrative of IMEM and the Physics department and the doctoral office for making my work a lot more trouble-free. I also thank Dr. Davide Vurro once again for his help in operating the AJP and SEM as well as Dr. Matteo Cocuzza and Dr. Simone Marasso for providing the photolithographed electrode samples. I thank Dr. Milad Takhsha for the muchappreciated help with AFM imagery. I am grateful to Salvatore Vantaggio from the group of Prof. Antonella Parisini as well as Dr. Marco Villani, Dr. Riccardo Manfredi and Dr. Andrea Sala from IMEM for being ready to help with one or the other small thing that I needed. I thank Dr. Mirko Buttrini and Prof. Adriana Calderaro from the Department of Medicine and Surgery for the opportunity to acquire the MALDI-TOF spectra. I also thank Dr. Silvia D’Auria for the recorded IR spectra and other members of the group of Prof. Enrico Dalcanale who treated me very kindly and made me feel welcome in the Chemistry department. Special thanks are due to Prof. Dalcanale himself for generously providing a few valuable reagents and, more importantly, for his teaching and the valuable scientific advice and encouragement I received over these years. I would like to express my deepest gratitude to Dr. Tobias Gruber for the motivation and support he gave me before and during my doctorate. I am grateful for the education I received during my undergraduate studies at the University of Freiberg. Despite having missed out on one or the other lecture of his (which Tobias likes to remind me of sometimes!), I consider myself lucky to have been his student. He laid the foundation for my views on science and it is safe to say that my interest in research was awoken during my bachelor thesis under his supervision. A huge thanks is due to all my old friends and the new ones I made along the way. They have been a very welcome distraction that I needed at times and their moral support has kept my spirit up. Especially I would like to mention Igor, who at times seemed to be even more emotionally invested in my doctorate than myself. I hope one day I can make up for the times when I couldn’t be as present as I wish I was. No words can describe my gratitude for my family. I am glad you don’t expect anything to be said because I don’t quite know how to express it. Thank you for your understanding, patience, and foresight. Thank you for your unconditional support and for always being there when I needed you, even at times when we disagreed on things. Спасибо вам за всё. Finally, I am grateful to those who motivated me to take this step. It was a rich source of personal growth and probably the best decision I have ever made so far. Without you, none of this would have happened. I’m glad it did. I Abstract Memristors are electronic elements that belong to a new generation of computational systems with a great potential to contribute to the promising revival of scientific research dedicated to the hardware realization of Artificial Neural Networks (ANNs) towards Artificial Intelligence (AI) and computer-brain interfaces. A distinctive property of these devices is the dependence of their internal resistance on the electrical charge that passed through them. In other words, there is a correlation of the output characteristics of the device and the history of its use, in which it resembles biological synapses and can be considered as their artificial analogue, displaying spike time dependent plasticity (STDP). Furthermore, the ability of controlled switching between different internal resistive states makes memristors some of the most promising candidates for the implementation in memories. Contrary to the conventional Von Neumann architecture, hardwarerealized ANNs have the potential of combining the storage and processing of information, carried out by the same kind of elements, mimicking biological neurons in the brain. Parallel information processing would allow to simultaneously work with a whole array of inputs, rather than carrying out one single operation at a time. These properties pave the way towards more energyand time-efficient computing by avoiding the need to exchange information between the processor and a passive memory. The similarity to biological synapses suggests an excellent biocompatibility and the possibility to emulate some functionalities of biological systems, enabling computer-brain interfaces with a seamless transformation and processing of bioelectronic signals. The Organic Memristive Device (OMD) is a polymer-based representative of such elements. It is a two-terminal device featuring a conductive channel of polyaniline (PANI) whose resistance is modulated through electrochemically controlled transitions between the polymer’s insulating and conductive state. These transitions take place in a heterojunction of PANI and a polyethylene oxide (PEO)-based solid polyelectrolyte (SPE) doped with a source of chloride among other stability enhancing additives, promoting the reaction with a silver counter electrode. Its advantages with respect to other memristors are its low-cost fabrication and the ability of fine-tuning of the channel resistance, granted by accessing intermediary resistive states. The main goal of the present PhD thesis is to develop OMDs suitable for neuromorphic applications such as the interfacing sensing and signal processing that can be realized by means of multilayer perceptron structures. The combination of a large array of elements in one network as well as working with complex biological systems implies the occurrence of electrical noise, which poses the question how it interferes with the functioning of our device. Preliminary research has shown that the current-state OMD does not possess the necessary level of stability to reliably to carry out such sensitive experiments. Hence, a significant part of our research was devoted to the optimization of the materials employed in OMD fabrication. The goal is to advance towards devices with higher endurance, i. e., reproducible output characteristics over longer periods of time, and to examine how to design the fabrication techniques to improve biocompatibility. II A significant part of our work was focused on the stabilization of the labile, polyethylene oxide-based SPE system by optimizing its composition with particular attention to the dopant salts. This work follows and builds upon an extensive theoretical study of the role of the single dopant ion species on the operation mechanism of the OMD. It was demonstrated that the former concept which attributes a significant role in the switching process to lithium ion doping is inaccurate. Instead, it is shown that the anions are the most critical dopant species in terms of device operation. Furthermore, through a combination of theoretical and extensive practical work, novel recipes of PEO-based SPEs have been developed, providing unprecedented short-term and long-term stability with a remarkable reproducibility of I-V-characteristics, improving the endurance of the OMD by up to two orders of magnitude. A significant part of this success is due to the implementation of aluminium chloride to the SPE system, providing intrinsic acidity and enabling to avoid doping by strong, volatile acids such as HCl. This discovery was followed up by intensive research of routes to stabilize the PEObased gels and impart them with favourable properties for long-term endurance by preventing aggregation. During this research, the formerly unheeded concept of lyotropicity was introduced to OMDs, recognizing it as a major aspect for the stability of the polyelectrolyte system. Furthermore, silk fibroin solutions have been successfully employed for the first time as an alternative to PEObased SPEs, further advancing prior research towards the introduction of biocompatible materials into OMD fabrication. Another important aspect of the implementation of OMDs in complex systems for the interfacing of sensing and signal processing is the integrability with other electronic systems such as OECT-based sensors. This can be most effectively achieved by shifting the paradigm of OMD manufacturing from largely manual towards high-precision, automated fabrication. To this end, we developed materials that can be applied by means of direct-writing techniques such as Aerosol Jet Printing (AJP). Particular attention was paid to the conductive polymeric component of the device. Successful attempts have been made by developing AJP-compatible inks based on pristine PANI in a mixture of organic solvents such as NMP, xylene and isopropanol, that were successfully applied on quartz and Si/SiO2 substrates. Furthermore, a water dispersible composite material of PANI and chitosan was synthesized, further improving the quality of the printed features and the coverage of the substrates. Both printed materials have been implemented to successfully fabricate the first functioning PANI-based OMDs featuring a printed conductive channel, displaying the characteristic properties of a memristor. The combination of all the experimental results and intensive theoretical work lays the foundation and paves the way towards the implementation of OMDs in advanced, complex, biocompatible, (bio)integrable systems for bio-interfacing purposes. III List of utilized abbreviations 12C4 [12]crown-4 (crown ether compound) AJP aerosol jet printing ANN artificial neural network APS ammonium peroxydisulfate (or ammonium persulfate) AZ active zone of the OMD BHT 3,5-di-tert-butyl-4-hydroxytoluene (an antioxidant) CGF carrier gas flow (in AJP) CPA chitosan/polyaniline composite CSA camphorsulfonic acid DBSA p-dodecylbenzenesulfonic acid GndCl Guanidinium chloride (or guanidine hydrochloride) IEL electronic current (in I-V characterization) IG ionic current or gate current (in I-V characterization) ISD or ITOT total current or source-drain current (in I-V characterization) IJP inkjet printing IPA isopropanol (isopropyl alcohol) LIB lithium-ion battery LS Langmuir-Schaefer technique NEA Norland Electronic Adhesive (product name) NMP N-methylpyrrolidone (or N-methyl-2-pyrrolidinone) OMD organic memristive device PANI polyaniline (-LE, EB, ES, PS) (leucoemeraldine, emeraldine base, emeraldine salt, pernigraniline salt) PEDOT:PSS poly(3,4-ethylenedioxythiophene) doped with polystyrene sulfonate PEO poly(ethylene oxide) (or polyethylene oxide) TSA or TsOH tosylic acid (or p-toluenesulfonic acid) SCE standard calomel electrode SDS sodium dodecyl sulfate SF silk firoin ShGF sheath gas flow (in AJP) SPE solid polyelectrolyte (or solid-state polymer electrolyte) STDP spike-time-dependent plasticity IV Table of contents 1 Introduction and state of the art .................................................................................................... 1 1.1 Memristors ........................................................................................................................... 1 1.2 The Organic Memristive Device ......................................................................................... 5 1.2.1 Materials employed in OMD manufacturing ........................................................... 6 1.2.1.1 Polyaniline as the electrically conductive polymer component .............. 6 1.2.1.2 Poly(ethylene oxide) as the solid polyelectrolyte .................................. 12 1.2.2 OMD switching mechanism .................................................................................. 13 1.2.3 Electrochemical model of OMD switching ........................................................... 16 1.2.4 Experimental evidence for lithium migration ........................................................ 18 2 Materials and methods ................................................................................................................ 22 2.1 Chemicals .......................................................................................................................... 22 2.2 Instruments ........................................................................................................................ 23 2.3 Experimental ...................................................................................................................... 24 2.3.1 LS-deposition of pristine PANI ............................................................................. 24 2.3.2 Synthesis of the chitosan/PANI composite ............................................................ 24 2.3.3 Preparation of PANI and CPA inks ....................................................................... 25 2.3.4 Aerosol Jet Printing experiments ........................................................................... 25 2.3.5 Preparation of PEO gels ......................................................................................... 26 2.3.6 OMD fabrication .................................................................................................... 26 2.3.7 Electrical characterization ...................................................................................... 26 3 Organic memristive devices interfacing biological systems ....................................................... 27 3.1 Electrical noise and OMD operation ................................................................................. 28 4 PANI-based OMD with improved stability and performance .................................................... 34 4.1 The role of SPE dopant additives in device functioning ................................................... 34 4.1.1 Li+ free SPE containing other hygroscopic salts .................................................... 38 4.1.2 The role of dopant cations ..................................................................................... 41 4.1.3 SPE containing Li+ with an addition of [12]crown-4 ............................................. 43 4.1.4 The role of dopant anions ...................................................................................... 44 4.2 Improved SPE compositions for increased device performance ....................................... 48 4.2.1 Intrinsically acidic AlCl3-doped SPE with increased long-term stability .............. 48 4.2.2 The concept of kosmotropicity, chaotropicity and the Hofmeister series ............. 54 4.2.3 Optimization of intrinsically acidic SPE ............................................................... 56 6 The resistivity of the deposited PANI can be tuned in a narrow zone in the middle of the channel that is covered by the manually cast stripe of a polyethylene oxide (PEO) gel that solidifies and functions as the SPE. The silver wire is placed at a short distance above the PANI channel, submerged in the PEO gel perpendicular to the channel direction. Because the device somewhat resembles a field effect transistor (FET) schematically, the corresponding denotation has been adapted for the three electrodes. The silver electrode is named the gate (G), and the two electrodes on both ends of the PANI channel are named the source (S) and drain (D). The gate is short-circuited with the grounded source and a voltage is applied to the drain and distributed across the channel, causing electrochemical reactions at certain given potentials. The process of the resistive switching of the device as well as the underlying chemical transformations of the polyaniline are explained below in more detail. 1.2.1 Materials employed in OMD manufacturing 1.2.1.1 Polyaniline as the electrically conductive polymer component Polyaniline has a basic poly(p-phenyleneimineamine) structure, consisting of alternating benzenoid amine and quinoid imine moieties. As shown in Figure 1-6 for a minimum degree of polymerization of 8, in earlier works, five different oxidative states could be isolated and chemically interconverted through oxidation with hydrogen peroxide or reduction with TiCl3 or phenylhydrazine [53]–[55]. The polymerization of the monomer aniline (or its hydrochloride) is most commonly carried out in acidic media with of oxidizing agents such as ammonium peroxydisulfate (APS) or electrochemically, yielding the polymer in its emeraldine salt form (PANI-ES). The forms of polyaniline can be distinguished with spectroscopic methods. Figure 1-7 shows the characteristic absorption spectra of the three polymer base forms in the UV/Vis range [55],[56]. Leucoemeraldine displays only one significant absorption band at 343 nm which corresponds to the π-π* transition in the aromatic system. In emeraldine, this band is hypsochromically shifted and a new band at 637 nm appears which is attributed to a molecular exciton associated with the quinone diimine fragment. In pernigraniline, the band at 530 nm is attributed to a Peierls gap transition [57]. FTIR spectra are presented in Figure 1-8 [58],[59]. The most relevant characteristic vibrations for the Figure 1-5: Schematic representation of the Langmuir−Schaefer technique. 7 Figure 1-7: Optical absorbance spectra of the base forms of polyaniline, adapted from [56]. Figure 1-6 : a) General chemical structure of PANI and b) its oxidative states for a minimum degree of polymerization of 8 (adapted from [55]). 8 base polymer are δop(C−H) on 1,4-disubstituted rings at 823 cm-1 and δip(C−H) at 1100−1170 cm-1, ν(C−N) of the secondary amine groups with a band at 1299 cm−1, ν(C−N) of the imine groups at 1378 cm−1, ν(C=C) in benzenoid rings at 1496 cm−1, ν(C=C) in quinoid rings at 1588 cm−1 [59]. The latter two bands and their ratio can be used to distinguish between the leucoemeraldine, emeraldine and pernigraniline forms [60]. The conductive emeraldine salt state can be recognized by the appearance of bands at 1238 cm−1 and 1140 cm-1 that are assigned to the ν(C−N+•) and ν(=NH+−) vibrations of the polaron and bipolaron structures, respectively [59], the origin of which is explained below. Doping of polyaniline The emeraldine salt is the transient state with the highest conductivity and hence the most relevant one for OMD operation, reaching conductivity levels in the order of 1−5 S cm−1 [61]. All forms of polyaniline can be electrochemically transformed in one another and can exist in their respective base and salt form, depending on the pH of the environment, as demonstrated in Figure 1-9 [62]–[64]. The so-called doping of a base form to its respective salt form occurs due to the acid-base activity of the amine and imine groups of the polymer chains. For instance, the commercially available nonconductive emeraldine base (PANI-EB) can be chemically doped with an acid such as HCl to obtain the conductive emeraldine salt (PANI-ES). The process is fully reversible as treating the doped polymer with an ammonia solution deprotonates it, yielding PANI-EB [61]. This can be seen in Figure 1-8 as the spectra of PANI-ES are largely equal after and prior to Figure 1-8: a) FTIR spectra of the base forms of polyaniline: leucoemeraldine (LB), emeraldine (EB), pernigraniline (PB), adapted from [58]; b) FT-IR spectra of emeraldine base (EB), emeraldine salt (ES) and emeraldine salt after exposure to KOH and reprotonation (adapted from [59]. 9 reprotonation. It must be pointed out that in the pernigraniline salt state, the compound becomes unstable due to the imine groups’ proneness to hydrolysis in an acidic medium [60]. However, there have been successful attempts to stabilize pernigraniline by means of N-alkylation to increase basicity [65] or by employing hydrophobic ionic liquids [66],[67]. Charge carriers in polyaniline The doping phenomenon is closely connected to the formation and nature of charge carriers as well as their transfer along the polymer chains. In literature, the process is most commonly described under the assumption that the doping is initiated by the protonation of imine nitrogen atoms [68]–[71]. At first glance, this may seem controversial because the sp3 hybridization of amine nitrogen atoms suggests a higher basicity with respect to the structurally related, sp2 hybridized imine counterparts, making the former more prone to protonation than the latter. However, it must be considered that in a dynamic system, the exchange of protons between different basic sites can takes place. The exclusive protonation of amine sites cannot be followed by any redistribution of charge along the polymer chain, whereas the protonation of imine nitrogen atoms leads to electronic rearrangements that result in an energetically more favourable structure. As shown in Figure 1-10, the initial protonation causes the formation of (di)cationic quinone diiminium fragments, named bipolarons. They are then transformed into their resonant radical cationic polaron form which is energetically more favourable due to the extension of the aromatic system [72]. The electrical conductivity of the material is given by the decoupling of the created neighbouring charge carriers and their delocalization along the polymer chain so as to minimize Coulomb repulsion. The presence of the polaron as a state with unpaired electrons has been proven by means of magnetic susceptibility measurements [73],[74] and electron paramagnetic resonance (EPR) [75]. The coexistence of polarons and bipolarons was further proven by X-ray photoelectron spectroscopy Figure 1-9: Mutual transformations of the forms of PANI by means of redox reactions and chemical doping. 10 (XPS) in combination with elemental analysis and optical spectroscopy [76],[77]. Inoue et al. [76] estimated a polaron to bipolaron ratio of about 5:1 in a perchlorate doped PANI sample. Furthermore, the XPS technique was employed to gather evidence for the proposed protonation mechanism, taking advantage of the fact that the position and shape of N(1s) line are sensitive to the electronic environment of the nitrogen atoms which changes upon oxidation and doping. The fitted line components in the N(1s) core-level spectrum for the quinoid imine, benzenoid amine and positively charged nitrogen atoms have been identified by Kang et al. [78],[79], with band energies at 389.1 eV, 399.3 eV and > 400.5 eV, respectively. To study the doping process, the authors exposed the emeraldine base polymer to non-volatile acids (H2SO4 or HClO4) and evaluated the signal intensities from the N(1s) and S(2p) or Cl(2p) core-level spectra, which allowed to estimate the doping level at different acid concentrations [69]. H2SO4 and HClO4 were chosen as doping agents because volatile dopants such as HCl are not suited for measurements in ultra-high vacuum and the Cl-/N ratio seldom exceeds 0.5 even at high acid concentrations [78]. According to their data, upon protonation, the imine signal disappears in favour of the emerging charged nitrogen signals. The intensity of the amine signal remains largely unchanged at doping levels up to 50 % (in 1 N acid) but diminishes in favour of the signals of charged species at higher doping levels (in concentrated acid), as demonstrated in Figure 1-11 a−d). This suggests that amine nitrogen atoms participate in the acidbase reaction only after all the imine nitrogen atoms are protonated. The authors also report that the conductivity of the excessively doped polymer (anion/N ratio > 0.5) diminishes with respect to a doping level of 50 %, which is due to the distortion of the π-conjugation and the polaron lattice by protonated amine units (−NH2+−). The same is reported by Yue et al. [80] for a sulfonated, self-doping polyaniline sample: the conductivity is lower when more than 50 % of quinoid and benzenoid rings are sulfonated. In another work, Snauwaert et al. [70] have correlated the percentage of imine nitrogen atoms with the oxidation of PANI starting from its leucoemeraldine base form, recording XPS spectra Figure 1-10: Scheme of the process of doping of emeraldine base, anions are omitted (adapted from [72]). a) Emeraldine base polymer, b) formation of the bipolaron lattice upon protonation, c) electronic rearrangement to polaron state , d) resulting polaron lattice as a consequence of charge delocalization. 11 after the application of different potentials (Figure 1-11 f). Their results coincide with the abovementioned electronic transformation of the polymer, although it must be pointed out that according to the imine percentage (about 75 %), the product of the second oxidation step at +0.8 V appears to be nigraniline rather than pernigraniline. The data also shows that the oxidation occurs at more or less discrete values rather than gradually over the potential range. In classical semiconductor terms, the radical cationic polarons can be considered as electron holes and thus, PANI is denoted as a p-type organic semiconductor. Contrary to common inorganic semiconductors, PANI, like other electrically conductive polymers, requires high levels of doping, reaching highest conductivity at 50 % protonation which corresponds to a pH of 2 or lower [81]. The reason is the depinning of charge carriers which is best understood considering their environment. Although delocalization is possible throughout the highly conjugated electronic system, at low doping levels the charge defects are widely separated from one another. In order to maintain charge neutrality, each charge carrier migrates as a pair together with its associated counterion, having the relatively slow diffusion of the latter as the limiting factor. At high doping levels, each polaron finds itself in the vicinity of multiple anions, not being pinned to any particular one, and can migrate faster along the polymer chain in the mean field of the surrounding counterions [82]. In the bulk of the polymer, the limiting factor for the conductivity of PANI is the electron transfer from one chain to another, which suggests a dependence of charge mobility from the morphology of the material. In their work, Focke et al., [83] propose a mechanism according to which interchain charge transfer can be facilitated by protonation, as shown in Figure 1-12. Figure 1-11: N(1s) and Cl(2p) core-level XPS spectra of HClO4 doped emeraldine films with a ClO4−/N ratio of a, b) about 0.50 and c, d) about 0.78. e) N(1s) core level spectrum of an emeraldine base film (a−e adapted from [69]). f) XPSdetermined imine content in polyaniline as a function of the applied potential at pH = 0 (black dots) in correlation with the CV oxidation scan (solid li ne), adapted from [70]. 12 1.2.1.2 Poly(ethylene oxide) as the solid polyelectrolyte In order to enable electrochemical reactions between the gate electrode and the polyaniline channel that would cause resistive switching, the two must be connected through a material that is a dielectric itself but allows for charge transport from one electrode to another. The realization of memristive devices in microelectronic appliances requires a preferably solid electrolyte for better manageability. To fulfil these requirements, high molecular poly(ethylene oxide) (PEO) (with Mw = 8 MDa) was chosen as it is a solid organic material that swells in water to form a viscous gel which can be easily cast manually. The swelling occurs due to the solvation of the PEO chains with water molecules. Once cast, a pristine PEO gel gradually loses some of its water content due to evaporation. With decreasing water content, the interactions of PEO chains with each other prevail more and more, leading to aggregation and crystallization which can be prevented by adding dopants (i.e., prepare the gel in an aqueous solution of dopant salts). A lithium salt is used as the main dopant in SPEs for OMDs, which is inspired by the advances in the field of lithium-ion batteries (LIB) [84]–[86] that take advantage of lithium’s small ionic radius and its ability to intercalate in carbon or metal oxide matrices. In the operating mechanism that was proposed for the OMD in earlier works, a key role in the switching of the device is attributed to lithium ions as they were thought to be essential for the charge transport between polyaniline and the gate electrode within the active zone of the device. The lithium ions are coordinated by PEOs etheric oxygen atoms through ion-dipole interactions and the charge transport is thought to be due to the hopping of Li+ between different coordination sites. Analogous to LIB technology, one of the reasons why lithium salts have been the dopant of choice is lithium’s small ionic radius which implies a higher ionic mobility. Evidently, to enable a sufficient charge carrier mobility, the PEO gel must be maintained in an amorphous state. Thus, another property of salts such as LiClO4 or LiBF4 is taken advantage of, namely their hygroscopicity that maintains a certain necessary amount of water in the system, preventing the aggregation of PEO chains. Due to the acid-base reactivity of polyaniline it is also necessary to create an acidic environment in the PEO gel that would otherwise deprotonate the Figure 1-12: Mechanism of interchain charge transfer assisted by protonation (adapted from [83] ). Chain II is protonated, enabling a single-electron transfer from chain I to chain II through resonance, after which chain I is deprotonated. 13 polyaniline salt to its nonconductive base form. This requires a second, acidic dopant such as HCl. The aspect of doping in PEO is one of the major topics of the present work, discussed in detail in section 4.1. 1.2.2 OMD switching mechanism Understanding the processes that take place in the active zone of the device is key to improving it in terms of performance and stability and making it more appropriate for complex applications. For this reason, the OMD’s switching mechanism has been studied extensively in the past, providing FTIR spectroscopic [87] or X-ray fluorescence (XRF) [88] data to support the constructed model. Before proceeding to outline the details of the device operation, it must be pointed out that in scope of the present work, the study of the role of SPE dopant ions that is presented in section 4.1 led to a number of findings which reveal several inconsistencies in the formerly assumed model. However, in this section, it appears necessary to present the details of the formerly proposed switching mechanism including the respective supporting data as to allow for a clearer discussion of the new findings reported in section 4.1. As was stated before, the resistive switching occurs as a consequence of electrochemical reactions of the conductive polymer polyaniline which is transformed from its electrically insulating leucoemeraldine state (PANI-LE) to its conductive emeraldine salt state (PANI-ES) upon oxidation (or vice versa upon reduction). The process can be monitored electrically by plotting the output current against the bias applied to the drain electrode with respect to the grounded source and gate electrodes. For a more precise characterization, two currents are recorded during the I-V measurements – the total current (ISD or ITOT) flowing through the device and separately, the gate electrode current (or ionic current, IG). The gate current represents the amount of charge that passes through the gate electrode during the electrochemical reaction. Calculating the difference between the two registered currents allows to evaluate the evolution of the so-called electronic current (IEL) that flows between the drain and source electrodes and can be expressed as IEL = ISD − IG. As presented in Figure 1-13, the I-V curve (panel a) shows a hysteresis loop as a characteristic feature of a memristive device, which is due to the difference in applied potentials at which the electrochemical oxidation and reduction take place. Typically, oxidation starts at around +0.3 V during the anodic scan and reduction occurs at around +0.1 V during the cathodic scan. These potentials correspond to the peaks of the ionic current IG (panel b) which resembles a voltammogram of polyaniline in the given potential range. The peaks mark the growing or decreasing trend of the IEL curve. The position and width of these peaks is determined by the retention time tR per voltage bias value, i.e., the scan rate. Faster scanning causes a shift to higher potential values for the oxidation peak(s) and lower values for the reduction peak(s), which is due to the fact that the system is not given enough time to equilibrate. On the contrary, the peaks converge at higher retention times, as the respective reactions are finalized sooner at lower applied voltage biases. The shape and width of the hysteresis loop both change accordingly. 14 Another characteristic feature of the OMD’s I-V curve is the rectification at lower potential values. This is given by the fact that once in its insulating PANI-LE form, the change of the voltage dependent channel current is very low in comparison to the transition to PANI-ES, during which the conductivity increases by up to four orders of magnitude [89]. Ideally, only one couple of redox peaks is expected as the result of the fully reversible transition between these two forms of polyaniline. For comparison, a cyclic voltammogram with the transitions of polyaniline at the respective potentials is shown in Figure 1-14. It can be seen that when the voltage range is extended beyond +0.8 V, a second anodic peak occurs as a result of the oxidation of emeraldine to pernigraniline. As mentioned above, this second oxidation step should be avoided because, with respect to PANI-ES, the fully oxidized pernigraniline form contains twice as many imine groups that are prone to degradation in acidic media [59], with no remaining amine groups to buffer their protonation that, on its turn, facilitates hydrolysis. A smaller oxidation peak may occur at around +0.6 V and is attributed to crosslinking reactions between PANI chains, forming phenazine structures as demonstrated in Figure 1-15. According to Geniès et al. [90], a reaction can occur between growing PANI chains and aniline nitrenium ions that are formed as an intermediate during the polymerization [91]. The authors also assume a crosslinking reaction between different PANI chains that carry nitrenium ion moieties. Since the formation of nitrenium ions within polyaniline chains appears to be unlikely, due to the likelihood of electron rearrangement in a conjugated system, the proposed mechanism has been slightly modified and oxidized emeraldine chains are assumed to react with one another. The phenazine structures disrupt the polaron lattice, reducing the material’s maximum conductivity over time as the process is not fully reversible, even though the presence of a slight B’ peak suggests some degree of reversibility [92]. However, this process takes place more commonly during the electrochemical synthesis of PANI from aniline by means of cyclic voltammetry and is less likely to be observed during OMD characterization. Figure 1-13: Typical I-V characteristic of an Organic Memristive Device, with curves for a) the electronic current IEL featuring a counter-clockwise hysteresis and a rectification at negative voltages and b) the ionic current IG. 15 Figure 1-14: Cyclic voltammogram of a HClO4 doped PANI film recorded at a 200 mV s−1 sweep rate, in a 1 M aqueous HClO4 . A/A’: peak corresponding to the transition between PANI-LE and PANI-ES, B/B’: peak corresponding to crosslinking reactions betw een PANI chains and their partial reversion, C/C’: peak corresponding to the transition between PANI-ES and PANIPS (adapted from [92]). Figure 1-15: Mechanism of crosslinking a) between a growing PANI chain 1 and the nitrenium ion of aniline 2 during the synthesis by means of CV and b) between two PANI chains during oxidation (adapted from [90]). 22 2 Materials and methods 2.1 Chemicals aluminium chloride AlCl3 Acros Organics aluminium bromide AlBr3 Alfa Aesar ammonium persulfate (NH4)2S2O8 Acros Organics ammonium thiocyanate NH4SCN Carlo Erba calcium chloride CaCl2 Fluka cerium chloride CeCl3 Carlo Erba caesium chloride CsCl VWR Chemicals ferric chloride FeCl3 Acros Organics ferrous sulfate FeSO4 VWR Chemicals lithium chloride LiCl Fluka lithium perchlorate LiClO4 Sigma Aldrich lithium tetrafluoroborate LiBF4 Merck lithium hexafluorophosphate LiPF6 Aldrich lithium thiocyanate LiSCN Carlo Erba magnesium chloride MgCl2 Fluka magnesium sulfate MgSO4 VWR Chemicals potassium chloride KCl Fluka potassium thiocyanate KSCN Sigma Aldrich rubidium chloride RbCl Sigma Aldrich sodium chloride NaCl Sigma Aldrich sodium bilsulfate NaHSO4 VWR Chemicals sodium dodecylsulfate NaSO4C12H25 Sigma Aldrich sodium sulfate Na2SO4 Carlo Erba silver (wire, d = 50 µm and 125 µm) GoodFellow silver (foil, d = 125 µm) Alfa Aesar acetic acid (100 %) CH3COOH VWR Chemicals hydrobromic acid (48 %) HBr Carlo Erba hydrochloric acid (1 N) HCl Carlo Erba orthophosphoric acid (85 %) H3PO4 VWR Chemicals perchloric acid (60 %) HClO4 Fisher Scientific sulfuric acid (98 %) H2SO4 Carlo Erba 23 acetone Carlo Erba acetonitrile VWR Chemicals aniline Merck chitosan (low molecular weight) Sigma Aldrich citric acid VWR Chemicals 1,1,1,3,3,3-hexafluoroisopropanol Alfa Aesar isopropanol Carlo Erba NEA 121 Norland Products Inc. N-methyl-2-pyrrolidone VWR polyaniline (emeraldine base, Mw = 105 Da) Sigma Aldrich poly(ethylene oxide) (Mw = 8∙106 Da, 200−500 ppm BHT) Sigma Aldrich poly(ethylene oxide) (Mw = 106 Da) Alfa Aesar p-toluenesulfonic acid Alfa Aesar toluene Carlo Erba xylene Carlo Erba The solvent purity was analytical grade or ACS reagent grade. All chemicals were used without further purification. LiSCN was dried at 80 °C for 15 h prior to weighing. 2.2 Instruments Agilent B2902A electrical characterization (SMU) Bruker Syrius MALDI-TOF mass spectrometry Diener Electronic Femto low-pressure O2/Argon plasma surface treatment Elma Elmasonic P ultrasound treatment Hettich Roto Silenta III centrifugation Merck Millipore Reference A+ Milli-Q water preparation Keithley 236 electrical characterization (SMU) Keithley 6514 electrical characterization (electrometer) KSV-5000 trough Langmuir-Schaefer thin film deposition of PANI Leica DMS300 optical microscopy National Instruments PXIe-1073 electrical characterization (SMU) Nikon H550L optical microscopy Novascan PSD Pro Series UV ozone surface treatment Optomec Aerosol Jet 200 Aerosol Jet Printing Perkin Elmer Spectrum Two FTIR spectroscopy Veeco Dimension 3100 Nanoman Atomic Force Microscopy Zeiss Auriga Compact Scanning Electron Microscopy 24 2.3 Experimental 2.3.1 LS-deposition of pristine PANI Solutions of polyaniline (1.0 mg/ml in NMP) were prepared by adding PANI-EB to the solvent portion wise under vigorous stirring. Eventual precipitate was allowed to settle for least 30 min and small portions were taken for dilution to 0.1−0.2 mg/ml in NMP/toluene (9:1 v/v). Standard samples were made from rectangular 15 x 7 mm (unless specified otherwise) quartz substrates covered by chromium (deposited by CVD). Electrodes on each side of the substrate were made by covering 2 mm wide stripes with polyimide tape and etching away the remaining chromium with a proprietary, ceric ammonium nitrate-based etching agent (Sigma Aldrich). The surface was covered by strips of polyimide tape, leaving a 2−3 mm free channel. The substrates were washed and thoroughly rinsed with isopropanol and water prior to use. The deposition of PANI thin films was carried out on a KSV-5000 trough using a modified Langmuir-Schaefer technique. Milli-Q water was used as the subphase, and the PANI solution in NMP/toluene was injected onto its surface and compressed by Teflon barriers at a constant rate of 0.5 cm/min until a target surface pressure of 10 mN/m. The toluene in the solution prevented PANI from being dragged into the subphase by NMP. The formed PANI layer was separated by an acrylic glass grid and 30−60 LS monolayers were deposited on the substrate, alternating the direction of deposition after every 5th layer. The deposited layers were doped for 30−40 s in 1 N HCl twice with an interval of ca. 30 min. Lastly, the polyimide tape strips were removed, leaving a PANI-ES channel stretching between the chromium electrodes. 2.3.2 Synthesis of the chitosan/PANI composite The chitosan/PANI composite material (CPA) was synthesized by oxidative polymerization of aniline in the presence of chitosan. First, 0.3 g chitosan were dissolved in 50 ml 4 % acetic acid under stirring for 30 min. The viscous solution was poured in a 250 ml round bottom flask. Separately, 0.9313 g (10 mmol) aniline were dissolved in 50 ml 1 N hydrochloric acid, the solution was cooled in an ice bath and added dropwise to the cooled flask under stirring. A few drops of a 1M FeSO4 solution were added to the mixture. Lastly, a solution of 2.7382 g (12 mmol) ammonium persulfate (APS) in 40 ml 1 N hydrochloric acid was cooled and added dropwise to the reaction mixture in the flask over ca. 1 h under stirring, maintaining the temperature below 5 °C. The mixture was stirred for 12 h and its colour changed from pale yellow over blueish green to dark green. The excess APS was quenched by adding 5 ml of 1 M FeSO4 solution. The resulting suspension was centrifuged at 4000 RPM for 5 min. The light green supernatant was discarded, and the dark green solid product was resuspended in 1 N HCl and centrifuged. This was repeated three times. The pellet was then resuspended once more in acetone, centrifuged, and dried at 50 °C for 24 h. Lastly, the dark green, brittle solid product was grinded in an agate mortar. The yield was 1.160 g (94 % relative to the mass sum of aniline and chitosan educts), although it is likely that some traces of solvents remained in the material. 25 Product analysis FTIR spectra were kindly taken by Dr. Silvia D’Auria on a Perkin Elmer Spectrum Two spectrometer in the Department of Chemistry, University of Parma, preparing the sample as a KBr pellet. MALDI-TOF measurements were kindly taken by Dr. Mirko Buttrini on a Bruker Syrius mass spectrometer in the Department of Medicine and Surgery, University of Parma. The spectra were acquired in the positive linear mode in the range of 2–20 kDa with a 60 Hz laser frequency and an ion source voltage of 20 kV, using α-Cyano-4-hydroxycinnamic acid (HCCA) as the matrix material. 2.3.3 Preparation of PANI and CPA inks Inks of pristine PANI were made by pre-dissolving PANI-EB in NMP under stirring. Xylene and/or IPA were slowly added to the stirred solution in different proportions. The resulting solution was treated with ultrasound (P = 120 W, f = 37 kHz, as in all following applications) for 30 min and stored in a refrigerator. In some cases, the solution was filtered through a 0.2 µm PTFE syringe filter. The inks were treated with ultrasound for 20 min prior to use. CPA-inks were prepared by adding CPA to water/IPA (8:2 v/v) and treating with the sample with ultrasound for 10 s, resulting in a fine suspension (a filtering test with a 0.45 µm PTFE syringe filter yielded a clear, colourless filtrate). The inks were stored in a refrigerator and treated with ultrasound for 30 s prior to use. The concentrations of PANI and CPA as well es solvent ratios are given in section 6.2. 2.3.4 Aerosol Jet Printing experiments 15 x 7 mm quartz slides with chromium electrodes or silicon slides (1 µm thermal SiO2 layer) featuring T-shaped gold electrodes (by photolithography, d = 200, 10 or 5 µm) were used as substrates. Prior to use, the samples were treated with ultrasound in isopropanol for 1 h, rinsed with acetone and water (Milli-Q). The surface was treated with plasma (O2/Argon, P = 3 W, t = 9 min) and UV/ozone (t = 3 min, T = 50 °C) immediately before printing. The deposition of PANI and CPA inks was carried out with an Optomec Aerosol Jet 200 system with various printing parameters, such as: - nozzle diameter: 200 µm - deposition plate temperature: 90 °C - speed: 0.5−2 mm/s - sheath gas: 40 sccm - carrier gas: 30 sccm - atomizer current: 0.5 mA The printed layers were doped with 1 N hydrochloric acid for ca. 30 s. 26 2.3.5 Preparation of PEO gels PEO gels were in 3 ml screw-neck glass vials with Teflon gasket spacers. Aqueous solutions of various dopants with concentrations between 0.1 and 1.0 mol/l were added portion wise and filled up to 3 ml with water (Milli-Q, ρ = 18.2 MΩ℅cm, was used in all further steps). Different amounts of PEO (Mw = 8℅106 Da, 200−500 ppm BHT as inhibitor, 90−120 mg depending on final concentration) were added portion wise to the vial under gentle agitation, followed by the next portion of dopant solution or water. The suspension was vortexed for 30−60 s and stored for 1−2 days to swell until homogeneity, being vigorously mixed with a stirrer a few times by hand. The resulting gels were stored in a refrigerator unless specified otherwise. 2.3.6 OMD fabrication All substrates were treated with ultrasound in isopropanol for 15-60 min and rinsed with acetone and water. In the case of quartz samples, the active zone of the OMD (in contact with the (poly)electrolyte was delimited by covering the PANI layer with strips of polyimide tape, leaving a 1 mm wide uncovered stripe in the middle. In the case of T-shaped Si/SiO2 samples, the channel was active across the whole length. Two steps of 2−3 layers of narrow polyimide tape strips were placed to both sides of the active zone, parallel to the PANI channel. Then, 1−3 layers of PEO gel were cast manually on top of the active zone with a plastic stirrer or micropipette tip. A silver wire (d = 50 µm), with a piece of tin was soldered to one of its ends as a cable connection point, was stretched and fixated over the polyimide tape steps within the active zone, perpendicular to the PANI channel. The wire was covered by another 1−2 layers of cast PEO gel. The device was ready to be used once the polyelectrolyte solidified. In some cases, coating with NEA followed, as described in section 4.3. 2.3.7 Electrical characterization The acquisition of electrical data was carried out using an SMU to apply voltage biases and register output currents (both the total current ISD and the gate current IG were registered simultaneously). Cyclic I-V characteristics were recorded by scanning the device in the potential range [−0.4 V; +0.8 V] or [−0.4 V; + 0.6 V], starting with an anodic scan from 0.0 V. The scan rate was determined by the scanning step (0.1 V or 0.02 V) and the respective dwell time, as specified in the experiments. Kinetic measurements were carried out at a constant bias (e. g. +0.5 V for oxidation and −0.2 V for reduction) for 5−10 min. All measurements were preceded by an at least 2 min long conditioning period at −0.2 V to prepare the sample, by complete reduction of PANI to the LE state. Additional steps were added when performing characterizations of OMDs with a series of different liquid electrolytes as described in section 4.1. When changing the electrolyte, the sample and the trough were carefully rinsed with 0.1 N HCl and then rinsed twice more with the next electrolyte. Each experiment was carried out at least in triplicate. 27 3 Organic memristive devices interfacing biological systems One of the general long-term goals of our research is the development of systems featuring Organic Memristive Devices for information processing and integration with bioelectronic systems and sensing. Such applications imply the use of large arrays of elements, mimicking biological neural networks and possibly interfacing and/or being usable as parts of a natural nervous system. Since the aim is to mimic some functionalities of biological systems, as a prerequisite for the applicability of OMDs and their compatibility with biological elements, a lot of previous work has been dedicated to the study of relevant analogous properties. In particular, the similarities between an OMDs and biological synapses have been demonstrated. According to Hebb’s principle [2], one of the fundamental properties of synaptic transmission between neuronal cells is the so-called spike-timing-dependent plasticity (STDP), meaning the that the strength of connections between neurons is reinforced by the prior activity of said connections. This is often simplified as “What fires together, wires together”, meaning that the connection between two cells becomes stronger the more often these cells are simultaneously activated. This concept has been generally proven to be applicable to memristive devices, since they display a change in conductance as a response to a sequence of received electrical stimuli [97]–[99]. Recently, Battistoni et al. demonstrated this property in Organic Memristive Devices [100], showing long-term potentiation and depression caused by series of electrical stimuli with various biologically relevant frequencies. The operability of the device in a pulse mode has been demonstrated earlier by Smerieri et al., including the ability to generate an oscillating response that is assumed to originate from capacitive effects [93],[101]. In the context of biocompatibility, such a behaviour can be seen as analogous to the spike train-like propagation of action potentials in living cells. Lastly, in a work by the groups of Parma and Kazan [102], the OMD has been successfully employed to directly connect two living cortical neurons of a rat. The artificial synapse enabled the communication between the two stimulated cells where the signal propagation fully followed the Hebbian principle, mimicking the natural ones in fine details. This is another great step towards biological and biomedical applications since it displays the ability of OMDs to directly emulate the functionality of biological synapse. Furthermore, exploratory research has been carried out for the implementation of OMDs in the hardware realization of Artificial Neural Networks (ANNs) in the form of single and double layer perceptrons [103],[104], as schematically shown in Figure 3-1. Such systems allow to solve analogue tasks such as input signal classification, paving the way the perspective to carry out more complex tasks at the hardware level with multilayer structures. Furthermore, it could be demonstrated that OMD-based systems are capable of emulating simple biological behaviour patterns, as it occurs in the feeding of the great pond snail Lymnaea stagnalis [105]. Essentially, it resembles the well-known Pavlov’s dog experiment which has been later emulated on the basis of inorganic memristive devices by the group of Ziegler et al. [106]. The equivalent electronic circuit mimicking the homoand 28 heterosynaptic learning process were realized with only one and two OMDs, respectively. The results are an adequate representation of the increased response to a conditioned stimulus (in this case, the presence of amyl acetate) after the conditioning procedure, when the conditional stimulus is applied together with the unconditioned one (in this case, the presence of food particles), resulting in an increased output signal (corresponding to the movement of feeding muscles) (see Figure 3-2). 3.1 Electrical noise and OMD operation When going towards large, complex arrays of interconnected devices, it is necessary to consider the occurrence of some degree of crosstalk and interferences including electrical noise. This is especially true when attempting to integrate OMDs with biological systems consisting of a myriad of interconnected elements, where so called noise artifacts are inevitable. The term “electrical noise” means the fluctuation of a signal within a certain distribution, characterized by a spectrum of frequencies and an amplitude, within which range this fluctuation occurs. In the case of a random distribution covering all possible frequencies, the resulting unpredictable fluctuations are called “white noise” in analogy to white light consisting of the whole spectrum of visible wavelengths. Similarly, it is called “coloured noise” when its frequency distribution by a single value over a period of time, or “low pass filtered noise” for a certain interval of frequency values (see Figure 3-3). Figure 3-1: Schematic representation of an elementary (left) and a double layer perceptron (right) [104]. The inputs Xi corre spond to different stimuli that are weighed by weight functions Wi represented by OMDs, the sum of which results in an output signal Y. Figure 3-2: Scheme representing a part of the nervous system* of the great pond snail Lymnea staglalis , responsible for its feeding behaviour (a) with an equivalent electronic circuit featuring OMDs (b). c) Circuit output signal as a response to the conditioned stimul us (CS) before and after conditioning. *CGCs: cerebral giant cells; SN: sensory neurons; CBI: cerebrobuccal interneurons ; CPG: central pattern generator interneurons. The arrows point at synaptic interconnections, emulated by OMDs in panel (b) (adapted from reference [105]). 29 Although it is seen mostly as a nuisance and an interference, there are works in the field of neurosciences discovering positive, constructive effects of this phenomenon [107]. In a work by Hunter et al. [108] the effect of the parameters of electrical noise on the reliability of the firing pattern of single neurons is studied. The authors distinguish between the contribution of the frequency range and the amplitude of noise (expressed in the form of the so-called coefficient of variability, CV). To this end, the response of slowly adapting motoneurons (aplysia) to bursts of large amplitude inputs alternating with low amplitude inputs is analysed. The experiments show a frequency-independent increase in spike time reliability in the presence of fluctuations of the input current with sufficiently high noise amplitude (CV = 1). The dependence on the frequency of the input shows when comparing otherwise equivalent experiments in which a broadband input (Figure 3-4 a) is compared to one lacking the frequencies around the neuron’s firing rate fDC (Figure 3-4 b). It can be seen that the firing reliability is maintained when the neuron receives a broadband input, but diminishes soon after the modulation of the noise amplitude to lower values (CV = 0.15) when the frequencies around fDC are not represented. The explanation behind such a response is the stabilization of the firing pattern through interference of the firing rate with harmonics and subharmonics of fDC from the frequency spectrum of the current input, which could be relevant when operating OMDs in the oscillator mode. Similar results were obtained in a work by Ermentrout et al. [107], where populations of interconnected living neuronal cells have been studied by comparing their firing patterns generated by the injection of current pulses with and without superimposed noise signals. The study shows that in analogy to single neurons, despite its chaotic nature, noise can create order in a system, causing a certain degree of synchronization of the response of an array of neurons, as illustrated in Figure 3-4 c). Similar to the previous case, the authors report that the reason behind this phenomenon is a delay or advancement of the next action potential, which is caused by positive or negative interference determined by the phase at which the noise stimulus arrives. Figure 3-3: Graphical representation of low pass filtered and white noise with amplitude values drawn from a Gaussian distribution (from reference [107]). 30 In the framework of both mimicking natural functional behaviour and of interfacing ANN with the biological systems it is quite interesting to study the operation of OMD under the exposure to noise. More specifically, we investigated the exposure to noise as a source of potential spikes that could drive OMD switching in analogy to the experiments demonstrating the STDP of the device [100],[105]. To this end, we studied the kinetic response of the device at voltages close to the switching onset potential. As it is known, the electrochemical transition of the PANI channel into its conductive state only occurs above this threshold value (typically at around +0.3 V bias), with faster kinetics at higher applied potentials (within the stability range of PANI-ES). Our expectation was that by applying electrical noise with a sufficient amplitude, a premature switching of the device could be achieved at sub-threshold potentials due to a potentiation by the noise’s positive potential spikes, as illustrated in Figure 3-5 a, b). Figure 3-5 c) shows the kinetic response of an OMD, fabricated using the classical procedure [109], at a constant +0.3 V bias, superimposed by a coloured electrical noise signal of different frequencies, measured in duplicate. A continuous, fast decay of the output current as well as switching kinetics is observed both between and within each set of measurements, indicating some short-term material degradation. Our assumption was that this is due to the solid polyelectrolyte as the most labile component of the OMD, being prone to excessive drying as well as acid dedoping due to HCl volatility. On this basis we carried out an extensive study that is presented and discussed in detail in chapter 4. Within the framework of this experiment, we attempted to circumvent the decay of the device’s characteristics by providing a stable humidity environment to the active zone of the device by means of capping, which is described in more detail in section 4.2. As it can be seen in Figure 3-5 Figure 3-4: Spike time reliability as a function of the magnitude of current fluctuations with a) a broadband current input and b) a current input lacking frequencies around fDC (adapted from [108]). c) Synchronization effect of an electrical noise stimulus on the firing pattern of an array of uncoupled model neurons. The neurons’ regular and independent firing in a diagonal pattern is transformed into a clustered pattern with more synchronous firing across the population (adapted from [107]). 31 d), this approach indeed gives a relevant stabilization of the system that delivers a much more stable and reproducible output. We hence proceeded by carrying out kinetics measurements at different applied constant voltages close to the onset potential, comparing the response to a bias with and without the application of a 20 Hz noise signal with an amplitude range of 0.1 V, acquired in succession. Two sets of results are presented in Figure 3-6. In the first series of measurements, no difference is visible between the mean output current in the case of the stimulus with superimposed noise and the response at the respective bias without noise application. This suggests that the postulated effect of switching enhancement by noise did not apply, which could be caused by a negation of the potentiation by positive voltage spikes through the inverse effect of negative voltage spikes, before the switching can initiate. On the other hand, a small difference can clearly be seen in the second set of measurements. Considering the constant behaviour in the first series, we assume that the slight decay of characteristics without the application of noise in the second series might, again, be caused by slow material degradation. Figure 3-5: a) Graphical representation of a premature initiation of OMD switching at subthreshold potential values, caused by a potentiation by the positive potential spikes of a superimposed noise signal (b). c) Kinetic response of an OMD at a constan t bias of +0.3 V and an applied coloured noise of different frequencies and an amplitude range of 0.1 V, exposed to ambient conditions. d) Kinetic response of the device at a constant +0.3 V bias on a capped device, providing humid conditions around the active zone. 38 as the related reaction kinetics are the most relevant aspects to be considered in the context of the device functioning. Hence, it becomes clearer which components are essential for a proper OMD operation and which ones, in turn, can be varied in order to improve the system in terms of stability and performance. Although a significant progress was made by the study of Cifarelli et al., there is still the need for a better understanding of the device operation that is crucial for further enhancing the performance of the system. In particular, one challenge is to overcome the present OMD limitations in terms of reproducibility and endurance that are not yet fully adequate for developing a robust technology. Some problems arise from the material degradation of the device’s components, concerning PEO, PANI and the silver gate electrode. In all likelihood, it is the deterioration of the former that is mostly responsible for the rather quick decay of the OMD’s performance over time. In view of previous findings, in the following section, the revised operation mechanism is further investigated. In particular, the goal is to provide further proof for the assumptions made above for a better understanding of the effects of doping additives in the solid polyelectrolyte, and to suggest viable approaches to improve OMD performance and reliability. 4.1.1 Li+ free SPE containing other hygroscopic salts It must be pointed out that, despite not being directly involved in OMD switching, the lithium salt plays a very important role as an SPE additive due to its hygroscopicity, as it provides the necessary water content within the polyelectrolyte gel to lower the fraction of crystallized PEO. This is of paramount importance for the device’s performance since, as it is well known, a PEO-based polyelectrolyte demonstrates its best ionic conductivity in a fully amorphous state when sufficiently hydrated [124]. Thus, the concentration of the dopant remains a significant parameter in the fabrication of OMD. From now on, when dopant concentrations are mentioned, they are always related to the preparation of the gel, i.e., the step in which the gel is prepared by adding PEO to an aqueous solution of the respective dopant salt. As the SPE is deposited, it solidifies by losing the majority of its water content upon evaporation, so that the final proportion of the dopant in the SPE increases. A determination of the final water content and PEO or dopant concentration could be made by precise weighing of the gel upon deposition and after its solidification. To better qualify the role of hygroscopic additives, in the present work we first examined how the device operates when substituting LiClO4 or LiCl in PEO-based polyelectrolyte with other hygroscopic salts. Our experimental finding are the subject of a literature work [125]. To this end, we prepared various PEO gels containing various salts, namely MgCl2, CaCl2, RbCl, CsCl, CeCl3, Mg(NO3)2, MgSO4 and Na2SO4. We also added an acid to the PEO gel to maintain the PANI sufficiently doped upon contact, thus preventing its transition to the nonconductive emeraldine base state. Hydrochloric acid was used mainly to ensure a sufficiently low pH level while in the case of the gel containing Mg(NO3)2 the HCl also provided the chloride ions which are necessary for both half-reactions to occur. An exception was the gel containing MgSO4 where H2SO4 was used instead of HCl. Some of the gels, namely the ones with RbCl, CsCl, Mg(NO3)2, Na2SO4 and MgSO4 salts, 39 proved to be insufficiently hydrated in the solid state, leading to intense PEO and/or salt crystallization. The worst case is the one based on MgSO4. In these SPEs, switching was not achievable because of the limitations on ionic transport. The change from the amorphous state is due to the hygroscopicity of the mentioned salts that is not enough to maintain the needed level of hydration in the gel that, hence, undergoes a drying process when exposed to the surrounding atmosphere in ambient condition. In the context of switching kinetics, this effect was shown by Burganova et al. [126] when varying the LiClO4 concentration in a standardly fabricated OMD. In this work, too low dopant concentrations (0.025 M LiClO4) led to stronger PEO crystallization, because the amount of water that could be retained in the gel was not enough to maintain its amorphous state in the whole bulk of the polyelectrolyte. Conversely, the formerly common concentration of 0.1 M LilO4 also appeared to be too high, as in this case, the time required for all of the ionic current to pass through the polyelectrolyte was the highest. In this case, the obstruction of current probably was due to the formation of LiClO4 salt clusters that, in turn, may have served as a source of nucleation for PEO. An intermediate concentration of 0.05 M LiClO4 in the gel preparation showed the fastest kinetics in this work, making the first step towards the optimization of the device. The behaviour of the gel containing MgSO4 seems odd at first glance, because this salt is known to be very hygroscopic, so it should guarantee a sufficient water content. In this case, the crystallization is due to another type of process, the so-called salt-induced precipitation (“salting out”) of PEO in the presence of sulfate. The underlying concept of kosmo-and chaotropicity of dopant ions is presented in section 4.2 of the present thesis work as it will be more relevant for the development of PEO gels with improved durability. The occurred hindrance with insufficiently hydrated gels was preliminarily resolved by creating an atmosphere with increased humidity around the sample. This was achieved by placing a cap containing a wet ball of cotton on top over the active zone. As shown in Figure 4-5, the gels partially cleared and liquefied within few minutes, allowing to carry out a standard electrical characterization of the memristive device. Examples of current-voltage curves of OMDs prepared with the mentioned SPE are shown in Figure 4-6 a). All devices that were fabricated with polyelectrolytes containing chloride salts displayed typical memristive behaviour. It is noteworthy that the OMD containing MgCl2 shows particularly satisfactory characteristics in terms of the hysteresis shape and, more importantly, reproducibility. This is a good indication of the quality and stability of the PEO gel due to a favourable combination of hygroscopicity and chaotropicity of the salt ions. For the device containing Mg(NO3)2, no memristive behaviour was observed even by placing the cap with the wet cotton above the active zone, used to maintain the PEO in its amorphous state. This may be well understood by considering the oxidizing character of the nitrate ions, which prevents PANI from being reduced (or maintained) to the leucoemeraldine form when attempting to switch off the device. The same effect, although weaker, likely occurs in the devices with PEO containing MgSO4 and Na2SO4. When applying the cap to liquefy the SPE and carrying out standard current-voltage measurements, these devices display characteristics that are somewhat similar to 40 those of a typical OMD. Figure 4-6 b) shows an I-V curve of a device fabricated with PEO containing MgSO4, recorded with a dwell time of 20 s, which exhibits some switching, characterized by the rectification of the curve at negative applied biases and a narrow hysteresis. This indicates that the device shows faster overall kinetics as compared to those in panel a) at similar measuring conditions. When the dwell time is decreased to 5 s per voltage step, a significant broadening of the hysteresis is observed, yet the device is still fast enough to switch. On the contrary, increasing the dwell time to 60 s causes an even stronger narrowing of the hysteresis to a point where it is barely present. This is because during the backward sweep, the kinetics of PANI reduction is faster than the dwell time. It is noteworthy that the electronic current in the “off” state at negative voltage biases remains in the Figure 4-5: a) Photograph of a commonly fabricated OMD, b) covered with a plastic cap with a wet cotton ball under its sealing; Topview microphotographs of the OMD c) before and d) after the exposure to the humid atmosphere through capping (SPE made with 3 0 mg/ml PEO concentration and 0.1 mol/l RbCl). Figure 4-6: a) Hysteresis curves of a series of OMDs with SPE made from PEO solutions (as colorcoded and labelled in the figure). The samples with RbCl and CsCl were mea sured under humid atmosphere by applying a cap with a wet ball of cotton on top of the active zone , dwell time for each acquisition step of 0.1 V = 30 s. b) I-V characterization of an OMD with SPE made from a PEO solution containing 0.05 M MgSO4, measure d under humid atmosphere with different dwell times per voltage step as labelled. 41 order of µA (compared to nA for a typical OMD) in all three cases. Comparable results were obtained for the devices made with PEO containing Na2SO4. They indicate that the switching-off took place only partially and an equilibrium state was established in which some of the material remains in its oxidized, conductive form. In section 4.1.3, the role of the sulfate ion as an oxidizing agent will be discussed in further detail. 4.1.2 The role of dopant cations Based on the above experimental results, we observed that lithium ions, in specific, are not necessary for OMD operation, since devices fabricated with SPE containing other hygroscopic chloride salts also showed switching. However, this does not exclude the possibility that in these cases, the initially suggested role of Li+ [93],[94] is performed by another cation (Mg2+, Ca2+, Cs+ etc.), as it has been implied in the fundamental work dedicated to exploring the device via XRF [88]. As discussed by Cifarelli et al. [114], the presence of Li+ ions in the polyelectrolyte is not necessary for the functioning of the OMD, whereas the role of the silver gate electrode and the presence of chloride ions is crucial. We aim at demonstrating with additional evidence the notion that dopant cations do not play a decisive role in the electrochemical reaction defining the OMD’s working principle shown in Equation 7 and Equation 8. Our approach is to show the independence of the OMD’s switching kinetics from the type of cation present in the electrolyte. From the XRF experiments of reference [88] it is known that the migration velocity of the bulkier Rb+ in PEO is lower with respect to Li+ ions. Hence, if the cations participate in the OMD’s switching mechanism, we expect to observe a difference in the characteristics of the corresponding devices. A direct comparison of devices with various compositions of the PEO-based polyelectrolyte (as discussed in section 4.1.1) is difficult because of the different hygroscopicity of the dopant salts defining the SPE’s final phase composition. Even with equal ion concentrations and providing a humid atmosphere using the cap, the conditions of the polyelectrolytes could still be significantly different from each other. For this reason, we decided to make the comparison using equally concentrated liquid electrolytes where the salt hygroscopicity has no effect at all, leaving whichever differences may occur to be attributable solely to the dopant ions with different ionic radii and mobilities. This goal required a modified experimental setup similar to the one proposed in reference [88], which is shown Figure 4-7. Figure 4-7: a) Photograph and b) schematic representation of the setup for kinetics characterizations of devices with liquid electrolytes. The well depth and width is 3 mm. 42 A series of solutions of monovalent metal chlorides (LiCl, NaCl, KCl, RbCl, CsCl) in aqueous HCl were chosen to both simplify and unify the system, the latter meaning that the only difference in SPEs would be the added cation. Relatively low concentrations for both HCl and the dopant (0.01 N and 0.001 M, respectively) were chosen because higher concentrations could possibly even out gradients and mask eventual kinetics differences caused by different dopant ions. Because of the need for an acidic environment, we could not reduce further the HCl concentration, since it would lead to the reduction of PANI-ES to PANI-EB and, hence, to a loss of conductivity. To provide even better comparability, all measurements were performed on the same PANI film, thoroughly rinsing it when changing the electrolyte and carrying out each measurement repeatedly in order to monitor the presence of a possible aging of the sample. We carried out three series of experiments using the same PANI film, repeating kinetic measurements when varying polyelectrolyte solutions one by one in each unique series. The obtained kinetics of all the series show the same trend. The typical kinetic responses of the OMD for different metal chloride electrolytes are shown in Figure 4-8 a). They appear to be quite similar with differences in the normalized curves that are within the margin of error, as confirmed by the fit parameters in Table 2. We could not observe any trend in correlation with the ionic radii and the established ionic mobilities [127]. However, when considering the curves without normalization, one particularity must be pointed out. The saturation current results to be slightly higher for the electrolytes containing caesium and rubidium chloride in most cases. It is especially peculiar that no trend with respect to ionic radii is observed throughout the rest of the series. This observation could not be adequately explained in the framework of the showed study. As a perspective, it would be interesting to assess whether this effect could be associated with the strength of the interactions of these cations with chloride ions. A stronger interaction with chloride ions compensating the positive charges on oxidized PANI chains could cause a stronger depinning of polarons from their counterions, increasing charge mobility and conductivity. Figure 4-8: a) Normalized and b) not normalized kinetic response of OMD wit h liquid electrolytes containing different alkali metal chloride salts at an applied voltage bias of +0.5 V. Reduction periods at −0.2 V are omitted. 43 Table 2: Double exponential fit parameters ((τ represent rate constants) for curves resulting from kinetics measurements as in Figure 4-8 a). Fit equation:         At least nine single measurements have been carried out for each electrolyte with a total number of 52 measurements used for the statistics. In the last row, a mean value of the calculated average τ parameters can be seen with the corresponding standard error. 4.1.3 SPE containing Li+ with an addition of [12]crown-4 The results shown up to now indicate that there is no significant effect of the cation radius on the kinetic response of the device. However, the observed inconsistencies in the saturation currents of devices fabricated with SPE containing different cations could give rise to some doubts. We hence carried out another experiment, similar in design to the previous one, with the idea to observe the kinetic response of a device when adding a substance that interferes significantly with the cationic mobility. In detail, we compared the response of devices with two equal electrolytes containing LiCl as the solute, one of which contained [12]crown-4 (12C4) of equimolar concentration. Crown ethers are representatives of the class of coronands that are known to form stable complexes with alkali metal ions (among others) depending on the size of their cavity. The 12C4 is one such coronand, having a cavity radius of 0.72−0.81 Å [128], matching the ionic radius of Li+ (0.74−0.86 Å), hence suitable to bind lithium ions, as shown in Figure 4-9. There are numerous reports in literature showing that such a complexation affects the mobility of ionic species [129]–[132]. Following the former assumption that lithium ions play a role in the operation mechanism of the OMD, a significant difference should be expected in the kinetic response of the devices when comparing the two above mentioned electrolytes. We then carried out the same measurements shown in Figure 4-9 on the same substrate, using the approach discussed in the previous section. We therefore chose the setup with the liquid electrolytes for the same reasons discussed in section 4.1.2 (Figure 4-7). The normalized curves correspond rather well with one another, keeping the same shape and trends, suggesting that the addition of [12]crown-4 to the electrolyte causes no significant change in the kinetic response. The corresponding double exponential fit parameters are reported in Table 3. Some insignificant differences are observed in the I/Imax saturation values between the series characterizing the electrolyte with and without the crown ether, where the latter are slightly lower than their counterparts. This is likely due to changes in the material itself because all the measurements were carried out in succession and a consistent, small decay can also be observed within both separate series. dopant avg. τ 1 σ(τ 1 ) avg. τ 2  LiCl 7.37 0.78 108.02 7.15 NaCl 7.64 1.03 124.92 6.48 KCl 7.53 0.76 123.18 4.82 RbCl 8.03 1.02 113.65 6.52 CsCl 6.76 0.81 106.03 5.47 mean 7.38 0.19 115.16 3.85 44 Table 3: Double exponential fit parameters (τ represent rate constants) for curves resulting from kinetics measurements as in Figure 4-9. Fit equation:         4.1.4 The role of dopant anions In the above-described experiments, the choice of a liquid electrolyte may lead to think that that eventual differences in the device characteristics could be masked or even neutralized by the fast (cationic) diffusion processes occurring in the liquid state, even at relatively low concentrations of dopants. In order to provide a further insight on the role of ionic species, we also investigated whether changing the anion would affect the memristive characteristics of the OMD. For a better comparison of the results, we decided to confront anions with the same charge but of different size. For our system, the choice of monoprotic acids to be compared with the standardly used HCl is somewhat limited. Common organic acids such as acetic acid are too weak to provide a pH that would dope PANI at appropriate concentrations, while the hydrogen halides HF and HI are toxic, corrosive or could display some redox chemistry of their own, which could interfere with the measurement. Other possible candidates as dopants for PANI are strong organic (sulfonic) acids. Although there are reports [132],[133] of polyanilines doped with CSA, TSA or DBSA, these are typically synthesized compounds, made by polymerization of aniline in the respective acid solution, yielding the self-doped electrolyte curve τ 1 τ 2 LiCl 1 15.64 164.00 2 12.24 117.51 3 13.03 107.76 LiCl + 12C4 1 11.45 106.08 2 12.37 101.91 3 12.76 98.61 Figure 4-9 : Kinetic response of OMD with liquid electrolytes containing LiCl with (red) and without (blue) the equimolar addition of [12]crown-4 at an applied voltage bias of +0.5 V. Reduction periods at −0.2 V are omitted. The molecular structure of the lithium12C4 complex is shown in the inset. Sandwich complexes of one lithium ion with two 12C4 molecules are also common. 45 polymer form. Our devices are fabricated with pristine PANI with higher mean molecular weight (Mw = 105 Da) in its emeraldine base form, the doping of which was not successful with any of these three sulfonic acids in our experiments. We assume that this is due to the high density of the deposited PANI film and the low mobility of the respective anions. For these reasons, the range of electrolytes that we used was limited to HCl, HBr, HClO4 and H3PO4. In a first series of experiments, a comparison of the device’s kinetics in liquid electrolytes featuring different anions (i.e., different diluted acids) was carried out following the measuring routine described in Chapter 2. To provide equal conditions in each experiment, we prepared electrolytes at the same pH and concentration of 0.1 mol/l, assuming the full dissociation for the strong acids HCl, HBr, and HClO4. In the case of the relatively weak H3PO4 (pKa1 = 2.124 [134]), the concentration of choice was 1.43 mol/l in order to provide equal H3O+ and anion (H2PO4-) concentrations of 0.1 mol/l. While the second and third dissociation steps introduce HPO42and PO43to the system, their concentrations are negligibly low and hence are not considered. As it is shown in Figure 4-10 a), the device switches on and off in each case with a clear difference in the kinetics depending on the electrolyte. This becomes even more evident upon fittings the acquired curves by a double exponential function, giving the rate constants τ1 and τ2 listed in Table 4. We observed that in the case of HCl and HBr, the maximum current was decaying during the series. We found that this is probably due to the volatility of these compounds because the same was not observed for the nonvolatile HClO4 and H3PO4. It appears that the response is quite similar in HCl and HBr, which is in good correlation with the quite similar relative ionic mobilities of chloride and bromide, reported in Table 4 [135]. The small difference observed in the onset time can be considered as negligible. It occurs a few seconds earlier in the case of HBr, likely due to the different reaction kinetics of chloride or bromide on either one of the electrodes (PANI, Ag or both). An instant switching onset is observed for HClO4, which we assume to be due to the oxidizing character of perchloric acid, which promotes the transition of PANI to its conductive state even at low applied voltages. This is further confirmed by the relatively high currents during the reduction period at −0.2 V, suggesting that the transition from the conductive emeraldine salt state to the nonconductive leucoemeraldine state is incomplete and some conductivity is maintained, analogous to solid state devices produced with sulfate salts, mentioned in section 4.1.1. The switching kinetics, however, appears to be slower than in HCl or HBr (see slopes in the inset of Figure 4-10 a), which corresponds to the lower relative ionic mobility of perchlorate with respect to chloride or bromide [135]. A clearer indication of the correlation of kinetics with the ionic mobility is observed in H3PO4 which does not feature any additional redox activity of its own. The onset of the switching occurs later than that observed for HCl and HBr, again, possibly because of slower reaction kinetics on Ag and/or PANI. Once the switching is initiated, the current increases significantly slower in comparison to the other acids. This is in good correlation with the significantly lower ionic mobility of dihydrogen phosphate (more than a factor 2). 46 Table 4: Double exponential fit parameters (τ1 and τ2 represent the rate constants of the fittings) for curves resulting from kinetics measurements in Figure 4-10 a). The ionic mobility values uX/uK+ (relative to that of K+) are taken from reference [135]. To further support our findings, in a second series of experiments, we compared the kinetics of the devices in a monoand a diprotic acid (HCl and H2SO4). Although the experiments with MgSO4 discussed in section 4.1.1 show that H2SO4, like HClO4, also has an oxidizing character, PANI doped with H2SO4 shows very similar conductivity to HCl doping, suggesting that the second oxidation step towards the nonconductive pernigraniline state does not take place without any applied bias voltage. Keeping in mind the divalent character of the sulfate ion, the respective potassium and magnesium salts (KCl and MgCl2 for HCl, K2SO4 and MgSO4 for H2SO4) were added to the electrolyte for a more meaningful comparison. To keep the pH equal in each experiment, we chose an acid concentration of 0.01 N, assuming complete dissociation of H2SO4. Therefore, the amount of salt additive was chosen to obtain that the total concentration of the chloride and sulfate ions, respectively, were also equal in each case (0.02 M). As a consequence, due to the different valencies of the cations, the respective cation concentrations also differed (e.g., when measuring with HCl, 0.005 M MgCl2 or 0.01 M KCl were added to achieve an equal total chloride concentration of 0.02 mol/L). This step allowed us to further assess the role of the concentration, size and electric charge of the cations. As shown in Figure 4-10 b), no significant difference could be observed when comparing electrolytes containing the same acid but different cations, which further proves the conclusions of section 4.1.2. At the same time, electrolyte τ 1 τ 2 u X /u/ K + 0.1 M HCl 1.89 49.86 1.04 0.1 M HBr 1.91 53.24 1.06 0.1 M HClO 4 3.35 80.66 0.92 1.43 M H 3 PO 4 7.59 79.20 0.45 Figure 4-10: a Kinetic response of OMD in HCl, HBr, HClO4 and H3PO4 at +0.5 V (oxidation period, 5 min) and − 0.2 V (reduction period, 5 min). b) Kinetic response of OMD with HCl (blue, cyan) and H2SO4 (red, magenta) based electrolytes at +0.5 V (oxidation period, 10 min) and −0.2 V (reduction period, 5 min). For each electrolyte, from a series of three consecutive measurements, only the first is shown in the figure. 47 the difference between the performance of the OMD with electrolytes containing different anions is very pronounced – a significantly faster response is observed in H2SO4. This could be in a first instance attributed to the bivalent nature of the sulfate ion, suggesting that a larger charge per unit of time is carried from one electrode to another compared to the chloride ion. However, this is only one part of the picture since the sulfate ion is also significantly less mobile in aqueous solutions [135]. The final balancing is unclear and hence it is difficult to come to a conclusive understanding. As shown in Figure 4-10 b), there is a clear decay of the current after the maximum achieved after the switching-on. This suggests that a portion of the PANI-ES is further oxidized to the nonconductive pernigraniline state at the applied voltage bias (+0.5 V). Besides, when applying a negative drain voltage, the measured negative current is comparatively high, indicating an effect similar to that observed for the device featuring HClO4 (see Figure 4-10 a). The oxidizing character of diluted sulfuric acid must be considered, since it presumably hinders the reduction of PANI-ES to PANI-LE when switching off the device, giving rise to the large current at negative drain voltage. On this basis, it is debatable whether the remarkable acceleration of the kinetics is due to the nature of the anion alone or assisted by this extraneous electrochemistry. Further experiments with sulfuric acid go beyond the scope of this work since we established that sulfate salts are not suited to be used in SPE formulas based on PEO. To summarize, we provided a deeper insight and further detailed data that are consolidating and validating the mechanism proposed by Cifarelli et al. [114]. The ordinary functioning of the device was successfully shown in a series of experiments with PEO gels containing a series of different dopant salts. Our findings [125] strongly indicate that LiClO4, commonly used in OMD devices, has the sole function of providing favourable conditions for PEO-based polyelectrolyte to maintain its amorphous state, and hence it can be replaced by another sufficiently hygroscopic salt. Therefore, it does not play a specific role in the memristive mechanism. Additional evidence is also gained from kinetics studies proving that the nature of the cation (i.e., its size and valence) have no significant effects on the memristive response of the OMD. A further, possibly more convincing confirmation of this conclusion is given by the fact that no significant difference could be registered when comparing OMD with electrolytes containing Li+ and its 12C4 complex. Based on our findings, one could conclude that in the OMD switching mechanism, the relevant charge transfer is provided by the chloride ions, confirming what was proposed by Cifarelli et al. Additional evidence to support this mechanism is given by the experiments discussed here where the choice of the anion has a pronounced effect on the OMD’s kinetics. The effect is evident when chloride ions are replaced by sulfate ions with a different ionic charge and/or mobility in the electrolyte, indicating the relevance of the anion for the OMD’s operation. Overall, this work contributes to a more detailed understanding of the chemical processes underlying the switching of the device that is absolutely necessary to both improve the system’s performance and overcome its weak points in terms of performance and endurance. 54 Overall, the attempts to replace aluminium chloride as the acidic component of the SPE were unsuccessful as this dopant appears to have a superior combination of properties such as acidity and hygroscopicity. One last acidity-enhancing additive, namely p-toluenesulfonic acid (TSA), has also been employed, and the related device performance will be discussed below, in section 4.2.3.3. 4.2.2 The concept of kosmotropicity, chaotropicity and the Hofmeister series As it has been established above, the hygroscopicity of the dopant is a crucial parameter to provide a necessary content of water to the SPE system. However, experiments carried out with gels containing very hygroscopic sulfate salts, such as Na2SO4 and MgSO4, resulted in SPEs that were nevertheless very prone to crystallization upon drying. This important piece of information shows that there must be another parameter to be considered when developing a new polyelectrolyte mixture. Conceptually, the issue with sulfate ions lies in the same area as hygroscopicity, being the interactions of sulfate ions with water molecules. As it has been mentioned above, the kosmoor chaotropicity of the dopant ions has been established as the second crucial concept for the selection of appropriate additives in a PEO gel composition. The concepts of kosmotropicity (from “kosmos” – order) and chaotropicity (from “chaos” – disorder) was conceived in the field of protein purification [142]–[144]. According to these concepts, the addition of certain salts to an aqueous solution of macromolecules, such as proteins, can influence the stability of their native state. A distinction is made between the so-called “salting out” caused by kosmotropic agents, meaning that the added solute causes the precipitation of the protein, and “salting in” caused by chaotropic agents and implying a stabilization in the dissolved state. Depending on their kosmotropic or chaotropic activity, ionic solutes as well as organic amphiphilic compounds, hydrotropes, surfactants and even organic solvents can be categorized and ordered in a so-called lyotropic series. The lyotropic series for cations and anions is commonly referred to as the Hofmeister series, according to which different ions are ordered according to their capacity to precipitate proteins. Overall, anions appear to have a stronger effect than cations. In earlier studies, the kosmotropic and chaotropic effects were said to stem from the ordering or disordering of the structure of water and increasing or decreasing its surface tension, respectively [145]. More recent studies indicate that ions do not affect the bulk water properties [146]–[149], indicating that direct interactions between the ions and macromolecules must be considered to understand lyotropic phenomena [150]. The individual effects of the ions depend on the macromolecule. For instance, in proteins, many different amino acid moieties are attached to the peptide backbone and may be hydrophobic, polar, or charged. The sum of the interactions of said side chains dictates the native folding and the tertiary/quaternary structure of the macromolecule. Chaotropic ions interact with the backbone or side chains of the protein in such a way that the natural framework of water-mediated interactions is disrupted. The hydrophobic effect is weakened and the tertiary structure unfolds, promoting the solubilization and causing a denaturation of the protein. Conversely, kosmotropes, such as the commonly used 55 ammonium sulfate, strengthen the hydrophobic effect, causing the precipitation of the protein, ideally preserving its native structure. In the case of the SPE containing MgSO4, correspondingly, the precipitation of PEO is caused by sulfate ions as an early member of the Hofmeister series. These ions interact more readily and frequently with water molecules than the PEO chains. As water becomes less prone to interact with the polar −O− units of PEO molecules, the hydrophobic effect is strengthened, decreasing the chain mobility and hydration, and hence causing stronger PEO aggregation [151],[152]. Conversely, the former standard gels containing LiClO4 displayed a very favourable morphology as the chaotropic ClO4− ion decreases the hydrophobic effect and prevents PEO chains from aggregating. The effect of kosmotropic and chaotropic ions on the morphology of PEO is shown in Figure 4-15. The chloride ion is usually listed as a neutral anion in terms of lyotropicity, meaning that the tendency of uncapped AlCl3-based gels to crystallize, as described in section 4.2.1, is likely due to the Al3+ ion, despite the high hygroscopicity of the salt. In reference [153], Al3+ is featured as a kosmotrope, given its small size, high ionic charge and low polarizability that suggest strong interactions with water. However, this study focuses on the interaction of ions with water and the presented lyotropic series for cations severely contradicts the trend given in other works [154]. For comparison, two different sets of anionic and cationic Hofmeister series are shown in Figure 4-16. The different results obtained in various works indicate that the prevalent effects of ions cannot always be generalized and depend on the system at hand. However, there appears to be a more solid consensus concerning the effect of anions. Sharma et al. [155] presented a study of the effects of inorganic ions and other compounds (such as surfactants) on the cloud point of a solution containing a PEO-PPO-PEO-block-copolymer. Their results are consistent with the anionic Hofmeister series. This confirms that the concept of kosmoand chaotropicity can be applied to the interactions of additives in PEO-based SPEs. However, caution must be taken when evaluating the applicability of some of the results presented in that reference. For instance, the authors report an increase of the cloud point when sodium dodecyl sulfate (SDS) is added, whereas in our case, the addition of SDS to the PEO gel caused rapid Figure 4-15 : Simplified schematic representation of the interactions of chaotropic and kosmotropic ions with water and PEO chains. Chaotropes tend to interact more with PEO chains and less with water molecules, PEO chains are maintained hydrated and amorphous. Kosmotropes tend to interact more frequently with water, making it less available for interactions with PEO, leading to a str engthening of the hydrophobic effect and aggregation of PEO chains. 56 precipitation after its deposition. The two systems must be differentiated on the basis of the different molecular weight of the (co)polymers and its water content. Because water is significantly less abundant in our system, the hygroscopicity of the solutes starts to play an important role, as it was established above. The lack of hygroscopicity may cause that the solute claims a significant portion of water for its own hydration without contributing to its overall content, thus cancelling possible positive lyotropic effects. It is also worth noting that specific preferential interactions such as the formation of ion-specific coordination complexes may cause deviations from the classical Hofmeister series [142],[152]. This circumstance makes comparisons of the kosmoand chaotropic character of ions difficult, especially in the case of cations. For instance, the Mg2+ ion is presented as a kosmotrope in some studies [151], while conversely, it is said to have rather chaotropic properties in other works [144]. The following section shows the usefulness of the Hofmeister series as a concept that complements the abovementioned importance of the hygroscopicity of the dopant salt. It paves the way to the perspective of further exploring favourable SPE compositions containing solutes that will eliminate the negative effects of salt or polymer aggregation and precipitation caused by drying or salting out processes. These developments are fundamental for fabricating devices suitable for advanced applications in neuromorphic technologies, as reproducibility and endurance remain some of the most relevant challenges in current state OMDs. 4.2.3 Optimization of intrinsically acidic SPE The implementation of aluminium chloride as an additive in the first generation of novel PEO gels is an important step in the development of high performance SPEs. While fulfilling the hygroscopicity requirement, the dopant also endows the gels with intrinsic acidity, allowing to avoid the addition of a volatile acid to the preparation. SPEs containing AlCl3 enable faster switching kinetics and show increased durability, leading to a more stable operation, as the polyaniline remains doped at all times. However, problems emerge as the hygroscopicity of the salt competes with the Figure 4-16: Two sets of anionic and chaotropic series from references [154] (some anions from the original publicatio n are omitted) and [153]. Note that the cationic series strongly differs between the two references. 57 kosmotropic character of the aluminium ion, which can cause the PEO to aggregate when the surrounding atmosphere is insufficiently humid. The development of stable and performing SPEs requires the fulfilment of some fundamental criteria for the selection of dopant candidates:  Solubility: the additive must be water-soluble to be employed in water-based PEO gels, especially considering the comparatively low level of hydration after deposition;  appropriate anion: the anion must participate in both half-reactions in the active zone (i.e. it both serves as a counterion for PANI-ES and reacts with the Ag gate electrode);  intrinsic acidity: avoiding the addition of volatile acid compounds ensures a stable pH;  hygroscopicity: upon solidification of a deposited gel, the dopant must maintain enough water in the system to keep PEO in its amorphous state under ambient conditions;  lyotropicity: the ions constituting the dopant must have a net chaotropic character, in order to decrease the hydrophobic effect of PEO chains and to prevent aggregation. 4.2.3.1 AlCl3-based SPE with secondary doping Among all possible dopant candidates, aluminium chloride has been selected as the main dopant because it meets all the above requirements besides the lyotropicity, which may be compensated by the introduction of a secondary additive. For this, we used the Hofmeister series as an aid to choose suitable salts showing lyotropicity. While it is not necessary that the secondary solute participates in PANI switching or provides acidity, it must be taken into account that it also must remain hydrated after the drying process. As it claims a part of the water content for itself, a combination of sufficient hygroscopicity and chaotropicity is desirable also for the secondary salt. For example, in PEO gels with a mixture of AlCl3 and MgCl2, an intense aggregation still occurred, as it can be seen in Figure 4-17. In this particular series of gels, it appears that a higher fraction of MgCl2 improves the quality of the gel. However, even though the magnesium salt is both hygroscopic and has a chaotropic effect [144],[152],[154], either one or both properties appear to be insufficiently pronounced to fully compensate the drawback of aluminium. It has been reported that the affinity of Mg2+ to proteins may be influenced by the pH because the ion affinity depends on the charge of the Figure 4-17: Micrographs of solidified PEO gels (20 mg/ml) made from aqueous solutions of a) 0.09 M AlCl3 , 0.01 M MgCl2, b) 0.75 M AlCl3, 0.25 M MgCl2, c) 0.05 M AlCl3, 0.05 M MgCl2 . The quality of the gel increases with a decreasing Al:Mg ratio. The AlCl3 concentration was not lowered further to maintain an acceptable pH. 58 protein [156]. A lowered pH created by aluminium chloride might cause a similar effect, altering the interactions between Mg2+ and PEO. In the following experiments, the gels’ morphology was studied under the microscope and used as a measure for the assessment of their quality. The general correlation between morphology and performance is quite reliable, since excessively drying gels with aggregates always perform worse in terms of switching kinetics and achievable peak currents. Considering the good experience made with the lithium perchlorate-based gels, widely used so far, we decided to use mixtures of AlCl3 and LiClO4 in the second generation of high performance SPEs. In fact, lithium perchlorate is an excellent dopant by itself if we disregard the necessity to add the volatile HCl to provide both acidity and chloride ions in earlier preparations. The salt has a favourable combination of both hygroscopicity and lyotropicity since the perchlorate ion is a late member of the Hofmeister series (see Figure 4-16). We have observed that the gels prepared with mixtures of salts were notably more fluid than their single-doped counterparts, which is probably due to the extra added water content. Increased fluidity poses the risk of poorly enveloping the gate electrode, thus inhibiting the proper switching of the device. For this reason, the PEO concentration was increased from 20 to 30 mg/ml with the aim of increasing the gels’ viscosities and facilitating their deposition. Three gels have been prepared; The first gel was made with a PEO concentration of 30 mg/ml in a 0.05 M AlCl3 and 0.05 M LiClO4 solution; we named it as P30-Alli-0111, (P30 stands for the concentration of PEO; Alli stands for Aluminium and Lithium; the first two digits, 01, denote the overall salt concentration (0.1M) and the last two digits (11) indicate their ratio. The second and third gels are made of 0.075 M AlCl3 and 0.025 M LiClO4 (P30-Alli-0131) and 0.025 M AlCl3 and 0.075 LiClO4 (P30-Alli-0113). Figure 4-18 demonstrates a comparison of the morphology of these three gels. It is worth noting that the quality of the gel increased with increasing LiClO4 content which is in accordance with its chaotropicity. P30-Alli-0113 and P30-Alli-0111 remained amorphous and transparent after Figure 4-18: a) Gel P30-Alli-0111 30 min after deposition; b),c) progressing drying of P30-Alli-0111. d) P30-Alli0113 30 min after deposition. e) P30-Alli-0131 30 min after deposition; f) progressing drying of P30-Alli-0131. 59 deposition while P30-Alli-0131 crystallized after a short period of time. Devices made with the first two gels both showed promising results in terms of endurance with respect to AlCl3-only SPEs. Figure 4-19 and Figure 4-20 show the I-V characteristics of two series of three different devices made with gels P30-Alli-0113 and P30-Alli-0111, respectively, demonstrating a good level of cycle-to-cycle reproducibility over 300−500 cycles. One device was cycled over a total of 1500 times (two series of 500 and 1000 consecutive cycles), which corresponds to over 80 hours of continuous operation. It is noteworthy that despite an eventual decline of peak current, any 20 consecutive cycles were almost identical to one another. In general, the shape of the I-V curve remained unchanged over the whole series of measurements. A shorter series of 20 measuring cycles was carried out on some devices 2−3 days after the endurance series to show that their functionality is maintained even when the device is left in ambient conditions in an idle state. The eventual fluctuations of output characteristics are caused by slight changes of the measuring conditions throughout the series (temperature, humidity) since the experiments were intentionally carried out in a non-controlled environment. Such an endurance level is precluded to standard devices with SPEs containing LiClO4 and HCl. In our experience, such devices usually lasted for only about 50 equivalent cycles before no switching could Figure 4-19: Endurance series of I-V characterizations of different OMDs fabricated with gel P30-Alli0111 as the SPE, measured with a dwell time of 2 s/0.02 V. a,b) 500 consecutive cycles, followed by a short series of 20 cycles acquired after 2 and 3 da ys, respectively. c) Series of 500 consecutive cycles followed by another 1000 cycles acquired 2 days after the first series. The inset demonstrates cycles 501-1500. d) Development of the peak output current Iel of the series in panel c). The discontinuity at 500 cycles is caused by the break between the two series of measurements, during which partial drying of the sample has occurred. 60 be observed, due to the PANI dedoping caused by HCl evaporation. The endurance of our modified SPE is unprecedented for solid state OMDs, surpassing the record set by Lapkin et al. [157] who have measured a PANIand PEO-based microdevice over 11000 short on/off cycles (10.5 s each, amounting to ca. 32 h in total). Arguably, the comparison is not entirely correct as we confront kinetic on/off cycling with hysteresis measurements. However, the reproducibility of the shape of the hysteresis and the related stability of the output current over time leads to think that our device may work properly on very long timescales. The increase of the output current in the beginning of the series (best seen in Figure 4-20 c) and Figure 4-19 c)) is due to a conditioning process during which the SPE morphology adjusts to the environmental conditions and the device parameters stabilize. Eventually, each series showed a decreasing trend of the output current which is caused by the slow drying of the AlCl3/LiClO4 gel. We attribute this to the lowered ratio of dopants and PEO, since the concentration of PEO was increased from 20 to 30 mg/ml. It appears that the resulting decreased overall salt:PEO ratio of about 3.3∙10-3 mmol/mg is insufficient for the hydration of the Figure 4-20: Endurance series of I-V characterizations of different OMDs fabricated with gel P30-Alli0113 as the SPE, measured with a dwell time of 10 s/0.1 V. a) 200 cycles in the bias interval [−0.6 V; +0.8 V] followed by 200 cycles in the bias interval [ −0.4 V; +0.6 V]. The strong dec rease of peak current after the first 200 cycles is due to the lower applied voltage at the end of the anodic scan. The switching onset remains unaffected by the changed voltage range. b,c) 300 and 500 consecutive cycles, respectively. d) Development of the peak output current Iel of the series in panel c). 61 SPE under changing surrounding conditions. For this reason, the concentrations of the added salts have been adjusted accordingly to a summed dopant concentration of 0.15 mol/l, corresponding to the initial salt:PEO ratio of 5∙10-3 mmol/mg. Figure 4-21 shows two of such gels with different salt ratios. The gel P30-Alli-01521 remains transparent with minimal crystallization on the stripe profile, while P30-Alli-01512 tends to crystallize in the whole bulk of the layer. This suggests that the benefit of added moisture in the gel with a 2:1 AlCl3:LiClO4 ratio is higher than that of the excess of chaotropic agent in the gel with a reversed ratio of 1:2. With a further increased LiClO4 concentration, the gel solidifies very slowly and forms inhomogeneous aggregates. Given the improved stability over time, we fabricated a device with the P30-Alli-01521 gel as the solid polyelectrolyte and an endurance measurement series was carried out to determine the long-term stability/reproducibility of the device response. We established that the humidity of the surrounding atmosphere has a strong influence on the quality of the SPE, which is why it is desirable to keep the measuring conditions stable. In addition, the conductivity of PANI is also dependent on humidity, meaning that a gel with a higher water content should enhance the performance. This property has been used to fabricated PANI-based humidity sensors as water is said to facilitate the protonation of PANI’s imine nitrogen atoms through hydrogen bonding [158],[159]. To show the potential of our modified PEO gels under stable conditions, we carried out another endurance experiment in which a device fabricated with the gel P30-Alli-01521 was subjected to an even longer series of measurements. In order to provide stable and favourable humidity levels throughout the series, the device was placed in a home-made closable chamber together with a petri dish filled with water. The experimental setup is shown in Figure 4-22. After installing the device and closing the chamber, the system was let to equilibrate for an hour. This was done in order to avoid large fluctuations of the electrical response, as it was the case during the endurance series in Figure 4-19 c) and Figure 4-20 c). The results of the continuous, 3000 cycles long endurance series corresponding to almost 7 full days (about 167 h) of continuous operation are presented in Figure 4-23. As it can be seen in panel a), some equilibration still took place over the first 500 cycles, during which the onset voltage shifted towards lower values. Figure 4-21: Gels a) P30-Alli-01512 , b) P30-Alli-01521 and c) P30-Alli-0211 six hours after deposition. 62 Figure 4-22: Experimental setup for the 3000 cycles endurance series. Holes were drilled in the perimeter of the chamber to insert cables. A water-filled petri dish was used as a source of humidity to provide more stable measuring conditions. Figure 4-23: The results of the endurance test of a device fabricated with the gel P30-Alli01521 as the SPE which was cycled 3000 times with a dwell time of 2 s per step of 0.02 V in the voltage range [ −0.4 V; +0.6 V]. The measurements took place in a partially air-conditioned laboratory (turned on during weekdays and off during the night). Shown are the IV characteristics of the device during cycles a) 1-3000, b) 501-2500 and c) 1201-1400. d) I-t profiles for the maximum and minimum currents IEL,max and IEL,min throughout the series. 63 The middle 2000 cycles (501-2500) display a relatively stable response in terms of the shape of the hysteresis and onset voltage. Although a slow decay of peak output current over time is evident, the differences between any 200 successive cycles within this span are negligible, displaying a remarkable reproducibility of I-V characteristics. The development of the maximum and minimum registered output currents during the anodic and cathodic scans in panel d) reveals an interesting detail about the performance of the device. Five plateau-like spikes (one in the first and four in the second half of the series) can be seen in the profile of the maximum output current IEL,max. These spikes are caused by temperature differences between day and night since the air conditioning in the laboratory was not continuous throughout the day. It is known that the conductivity of polyaniline as an organic semiconductor is temperature-dependent, increasing at higher temperatures (valid for values around room temperature) [160]. The collected data reflect this property, as the conductivity rises every other day except for days 2 and 3 when the air conditioning remained off and the temperature remained low. On the other hand, IEL,min also displays spikes towards less negative values (lower absolute values) at the same times as IEL,max spikes. Higher (absolute) currents during the cathodic scan are an indication for an incomplete reduction of the polymer, causing some residual conductivity. Since the reaction kinetics depend on the diffusion of ions through the SPE and PANI layer, it can be argued that the viscosity of PEO, which also depends on temperature, is another factor for such behaviour. This seems particularly plausible because the device is measured close to the scan rate limit with a dwell time of 2 s/0.02 V. A high scan rate means that the reduction and oxidation of the PANI layer will start lagging behind the scan at suboptimal conditions. This is further confirmed by the fact that IEL,min is gradually restored to lower absolute values once the temperature is increased. Conversely, disregarding the temperature spikes, IEL,max displays a steady decay throughout the whole series, indicating a slow degradation process, the cause of which is not fully understood. On the one hand, a gradual drying of the SPE could be a reason of a progressing worsening of the device’s performance. However, it is unlikely in this case, as this endurance test was carried out in a closed chamber of relatively low volume, with a controlled humidity. This indicates that some form of material degradation is the more likely cause. The decomposition of PANI appears to be more probable due to the hydrolysis and cleavage of the polymer chains in its emeraldine salt state, as it has been discussed above. With a lower average chain length, the charge transfer between different chains becomes more predominant as the limiting factor for conductivity. The decomposition of PANI seems especially plausible when considering the increased water content of the PEO gel given by the humid environment. Other undesired reactions cannot be excluded, such as crosslinking [90],[91], and covalent binding of chlorine to benzenoid or quinoid rings in PANI [161], both of which disturb the regular electronic structure of the polymer. According to reference [161], the latter reaction can take place at sufficiently high acid concentrations, even though the chloride ion is a poor nucleophile. Another source of degradation could be the silver wire as it is also subjected to continuous redox cycles during which the porosity of the AgCl layer may gradually change. Though, given the fact that 70 hexafluorophosphate have proven themselves as yet another class of dopants unsuitable for OMDs. The application of LiBF4 and LiPF6 in LIB technology is possible because, in this case, nonaqueous solvents that do not react with the solutes at such meaningful rates, such as ethers or organic carbonates, are employed [169]. 4.2.3.3 SPE with bromide as the primary or secondary dopant anion For the fourth generation of our SPEs with increased stability, we decided to switch our attention back towards alternatives for aluminium chloride as the primary dopant. With the experience gained from our previous research, adhering to the criteria that qualify an additive as an appropriate dopant, our idea was to replace AlCl3 by a salt with similar properties. We were searching for a Lewisacidic, hygroscopic salt with higher chaotropicity, whose anion would participate in the redox reactions involving both the PANI channel and the silver electrode. To minimize the number of systematic changes brought by the replacement of the dopant, we used a homologue of aluminium chloride. At the same time, the salt should have no redox chemistry of its own in the potential range between −0.4 V and +0.6 V, i.e., where the OMD is operated. For this reason, we refrained from using aluminium iodide because of the low standard electrode potential of the couple 3 I−/ I3− (E° = +0.53 V [170]). On the other hand, a similar side reaction is not expected for bromide because it remains stable under our experimental conditions. According to the electrochemical series of standard electrode potentials, the oxidation of Br− to Br2 occurs at potentials above +1.0 V (+1.087 V in aqueous solutions [171]. Aluminium bromide is comparably hygroscopic, with the [Al(H2O)6]3+ ion providing acidity while having a larger, more polarizable and hence more chaotropic anion than chloride. OMD operation mediated by bromide ions has already been shown in section 4.1.4 of this work, where measurements in aqueous HBr have been carried out successfully. The experiments have also shown an overall similar behaviour in comparison with an equimolar HCl electrolyte, given that chloride and bromide have comparable ionic mobilities [135]. Nevertheless, those results (obtained using liquid electrolytes) must be handled with caution, since the obstruction of the bulkier bromide (ionic radii: r(Br-) = 1.95 Å, r(Cl-) = 1.81 Å [172]) could be stronger in a solid polyelectrolyte. Another source of concern is the bromide’s higher chemical activity, since it is a much better nucleophile than chloride, possibly causing substitution reactions or the cleavage of PEO chains, which will be assessed below. Figure 4-30: Microphotographs of gel P30-AlliBF0211 a) 15 min after solidification, b) 30 min after solidification; c) gel P30-AlliPF-01521 30 min after solidification. 71 We started by preparing a 30 mg/ml PEO gel containing 0.1 M AlBr3 as the sole dopant (P30-AB-01). Surprisingly, although chloride and bromide are neighbouring each other in the lyotropic series, the deposited gel was considerably more stable than its equimolar AlCl3 analogue from section 4.2.1, remaining transparent after solidification. The gels displayed satisfactory longterm stability, showing some moderate drying on the edges of the deposited layer after 11 days of storage in ambient conditions. Moreover, a noteworthy property of the AlBr3-based gel was a superior resistance to heat, as it remained stable upon heating the sample to 60 °C on a hot plate, as it is shown in Figure 4-31 c). However, OMDs assembled with P30-AB-01 as the SPE displayed no switching. We assume that the acidity of the gel was insufficient to maintain polyaniline in its conductive PANIES form, because AlBr3 solutions had higher pH levels than the equimolar solutions of AlCl3. In an attempt to lower the gel’s pH, we increased the dopant concentration, preparing gels with 0.2 M (P30AB-02), 0.5 M (P30-AB-05) and 1.0 M AlBr3 (P30-AB-10). Figure 4-31 d-f) show that the resulting SPE became more homogeneous and transparent, while also becoming more and more soft, to the point where P30-AB-05 and P30-AB-10 remained in the form of a viscous liquid. This also meant that it became more difficult to ensure that the gate electrode was in good contact with the SPE when assembling devices, as it would flow down from it rather than enveloping it, as shown in Figure 4-32. Thus, such gels were not used further. The issue with the assumed insufficient acidity of the gel could not be solved by using mixtures of AlBr3 and AlCl3. Low concentrations of AlCl3 had no considerable effect on the acidity of the gel while higher concentrations caused excessive hygroscopicity, so that the gels would not solidify sufficiently. On the contrary, low concentrations of AlBr3 did not provide enough chaotropicity, leading to poor stability and crystallization over time. For this reason, it was decided to use another additive as a source of acidity. As we have established above, the additive must be Figure 4-31: Microphotographs of AlBr3-doped SPE with 30 mg/ml PEO. Gel P30-AB-01 (0.1 M AlBr3) a) 30 min after deposition, b) 11 days after deposition, c) after 15 min at 60 °C. d) P30-AB-02 (0.2 M AlBr3), e) P30-AB-05 (0.5 M AlBr3), f) P30-AB-10 (1.0 M AlBr3) 1 h after deposition. 72 non-volatile and have no salting-out effect on PEO. When discussing the concept of lyotropicity in section 4.2.2, hydrotropes have been mentioned among other substances (besides chaotropic salts) that are capable of salting-in of macromolecules. One such hydrotrope is the organic, non-volatile ptoluenesulfonic acid (TSA or p-TsOH, see Figure 4-33), whose ability to salt-in stems from its amphiphilic structure, allowing it to interact with both water and unipolar moieties. TSA is known to be somewhat soluble in water and forms sols in alcohols and ethers as well as saline solutions. Its pKa is estimated by different techniques to be between −1.4 and −6.2 [173], which is in any case sufficiently low to maintain PANI in its doped state in the active zone. The AlBr3and TSA-based PEO gels required some optimization in terms of the content of both salts. The micrographs of some examples are shown in Figure 4-34 a-d). However, the best SPEs were obtained using concentration ratios similar to the ones familiar from our previous research, i.e., the gels P30-ABT-01521 (0.1 M AlBr3, 0.05 M TSA) and P30-ABT-0211 (0.1 M AlBr3 and 0.1 M TSA). While we can empirically determine the dopant:PEO ratio to be optimal around 5-7∙10-3 mmol/mg, a thorough examination of optimal levels of hygroscopicity and lyotropicity require the study of various parameters of a wide array of salts, which goes beyond the scope of this work of thesis. The two mentioned gels have shown outstanding long-term and thermal stability and withstood the continuous exposure to temperatures as high as 110 °C for over 16 hours, as it is shown in Figure 4-34 e, f). Such a property is of particular interest for the standpoint of automated manufacturing, since one of the steps (the printing of polyaniline) requires the substrate to be heated during the film deposition, as will be shown in Chapter 4. The operation of OMDs assembled with AlBr3and TSA-based SPEs was unsuccessful. We assumed that the reason was one of the concerns expressed when we first employed aluminium bromide, namely that it is bulkier than chloride which might hinder it from migrating through PEO to enable the switching reaction. For this reason, given the exceptional thermal stability of polyelectrolytes containing TSA, we decided to return to AlCl3 as the primary dopant with an addition Figure 4-32: Schematic representation of the d eposition of a PEO gel in the active zone of an OMD a), followed by an ordinary drying process with the gate electrode staying emerged in the SPE (b). c) Supposed drying process of gels wit h excessive dopant hygroscopicity and/or chaotropicity (such as P30-AB-05, P30-AB10), leading to insufficiently contact of the gate electrode with the electrolyte. Figure 4-33 : Chemical structure of p -toluenesulfonic acid (TSA). 73 of the sulfonic acid. As expected, due to the lower overall chaotropicity of the dopants, such gels were generally less stable than their bromide-based analogues, which can be seen in the structure of the solidified P30-ACT-01521 gel (0.1 M AlCl3, 0.05 M TSA) in Figure 4-35. Since the P30-ACT0211 gel remained stable, we can assume that the TSA itself exerts a salting-in effect, even without the presence of another chaotropic salt. Moreover, the transparent appearance of the gel P30-ACT022101 reveals that this effect must be very pronounced, since a relatively low concentration of TSA (0.02 mol/l) is enough to counteract the effect of the 0.2 M AlCl3. As it was shown earlier in Figure 4-28, such high concentrations of AlCl3 in gels without a secondary dopant caused the crystallization of the SPE after deposition. However, its thermal stability was not as high as that of AlBr3-/ TSAbased ones. Gel P30-ACT-0211 on the other hand showed remarkable stability even after 15 h at Figure 4-34: Microphotographs of AlBr3and TSA-doped SPE with 30 mg/ml PEO. a) Gels P30-ABT01251 30 min after deposition; b) P30-ABT-01521 30 min after deposition; c) P30-ABT-0211 30 min after deposition; d) P30-ABT011101 30 min after deposition; e) P30-ABT-01521 after 16 h at 110 °C; f) P30-ABT-0211 after 16 h at 110 °C. Figure 4-35: Microphotographs of various AlCl3and TSA-based PEO gels. a) P30-ACT01521 30 min after deposition; b) after 30 min at 90 °C. c) P30-ACT-022101 30 min after deposition; d) after 2 h at 70 °C. e) P30-ACT0211 30 min after deposition; f) after 2 h at 70 °C; g) after 15 h at 110 °C; h) after 30 min at 10 °C. 74 110 °C on a hot plate. The sample also remained stable after 30 min at 10 °C. It is noteworthy that during the long heating at 110 °C, the substrate became covered by a pale deposit to each side of the cast SPE layer. The same could be seen in Figure 4-34 e, f), although less pronounced. This is likely due to the sublimation of aluminium chloride in the form of Al2Cl6 (although, according to literature, this normally occurs in harsher conditions [174]. Despite the excellent stability of the SPE, devices assembled with P30-ACT-0211 were not functional, as it was the case with analogous bromide-based SPEs. Given the fact that bromide-based liquid electrolytes allowed switching in section 4.1.4, it is hence more probable that the switching is prevented by TSA in bromideand chloride-based gels. In a study by Stöffler and Luft [175], various routes of chemical oxidation of TSA with hydrogen peroxide are shown. Considering the electronwithdrawing character of the −SO3− group, a similar electrochemical process (possibly assisted by AlCl3 or AlBr3) seems to be plausible in our case, preventing the oxidation of PANI-LE to PANI-ES. Although the PEO gels discussed in this section could not be used to fabricate OMDs, the impressive stability of gels containing bromide and/or TSA is encouraging for further research in this direction. For instance, TSA could be replaced by other, more stable sulfonic acids such as camphorsulfonic acid, which would provide the necessary level of humidity, chaotropicity and, possibly, thermal stability. 4.2.3.4 AlCl3-based SPE with a cationic chaotrope as a secondary dopant Our previous attempts to replace LiClO4 as the secondary dopant led to the development of new SPEs based on different chaotropic anions. Despite the remarkable stability of some of these SPE, they could not be employed in OMDs because the switching did not occur for various reasons. Thiocyanate and TSA have been shown to display some redox chemistry of their own which could have prevented polyaniline from switching; bromide, although shown to be functional in liquid electrolytes, probably did not provide the necessary level of acidity in form of AlBr3 to maintain PANI protonated; tetrafluoroborate and hexafluorophosphate most likely decomposed due to hydrolysis, releasing kosmotropic fluoride ions that led to precipitation. Perchlorate itself has shown the disadvantage of being a strong oxidant, which likely leads to the degradation of PEO over time. As shown by a few examples of gels such as P30-ACT-022101, a relatively low concentration of the (anionic) chaotropic agent can be enough to stabilize a gel with a high concentration of AlCl3. This leads us to the conclusion that although cations are generally known to have a weaker salting in-/out effect on macromolecules than anions, a strong cationic chaotrope might be able to provide the necessary conditions as well. Hence, in the fifth generation of our modified PEO-based SPEs, we decided to focus on introducing the chloride salt of a chaotropic cation to the system in order to stabilize PEO’s structure. According to literature (mostly dedicated to protein research), guanidinium chloride (GndCl) is a strong chaotrope [176]–[179], capable of salting-in of proteins, decreasing the hydrophobic effect and causing the unfolding of their tertiary structure, hence denaturalizing them. While this effect is undesired in protein purification and isolation, it is the goal 75 in our case as it keeps PEO from precipitating and crystallizing. The cause of guanidinium’s chaotropicity is its interactions with the macromolecules and water. The cation is highly polarizable, since its charge is evenly distributed over the whole molecule, as shown in Figure 4-36. Unlike some of the dopants we used before (TSA, AlBr3), guanidinium chloride has a high pKa of 13.71 [180], which is why the acidity in the SPE had to be provided entirely by the primary dopant AlCl3. Furthermore, based on previous results, adjustments of the dopant concentrations had to be made to make up for guanidine hydrochloride’s lack of hygroscopicity. We found out empirically that concentrations of AlCl3 above 0.2 mol/l or GndCl above 0.1 mol/l caused crystallization due to an oversaturation of the mixture. We prepared a series of 30 mg/ml PEO gels that are presented in Table 5. Once cast, most of these gels dried out and crystallized after a few hours. Some of the more stable SPEs are shown in Figure 4-37. The gel P30-ACG-0231 emerged as the one with the most favourable composition as it remained transparent 2 days after deposition, while all the other ones (including those in Figure 4-37 a-d) eventually began to dry out during the same period. Table 5: Compositions of 30 mg/ml PEO gels based on AlCl3 and GndCl, and their visual appearances 2 hours and 2 days after casting on quartz substrates in ambient conditions. gel symbol [AlCl 3 ] (mol/l) [GndCl] (mol/l) appearance after 2 h after 2 d P30-ACG-0131 0.025 0.075 dry edges − P30-ACG-0111 0.05 0.05 dry edges − P30-ACG-01512 0.05 0.1 very dry edges − P30-ACG-01521 0.1 0.05 dry ends; slightly dry edges − P30-ACG-0213 0.05 0.15 overall turbid; dry edges − P30-ACG-0211 0.1 0.1 transparent; slightly dry ends dry, crystallized P30-ACG-0231 0.15 0.05 transparent transp.; one end slightly dry P30-ACG-02532 0.15 0.1 transparent * dry, crystallized P30-ACG-02541 0.2 0.05 transparent dry patches P30-ACG-0321 0.2 0.1 transparent small, dry patches *casting of this gel was particularly difficult due to low viscosity/density Figure 4-36: Chemical structure of the guanidinium cation. 76 The thermal stability of the gel P30-ACG-0231 (0.15 M AlCl3, 0.05 M GndCl) was tested by placing the cast sample on a heating plate at 60 °C for 5 min. Much unlike the gels containing aluminium bromide and/or TSA, this gel crystallized within tens of seconds, as shown in Figure 4-37 f). However, the sample regenerated almost completely after 10-15 min in a humid chamber (with a water-filled petri dish, as used in the endurance series in section 4.2.3.1). Looking ahead, such a behaviour is satisfactory when planning the implementation of such gels in automated OMD manufacturing. The gel P30-ACG-0231 was then chosen to assess the operability of OMDs featuring AlCl3and GndCl-doped PEO. It was done using the gel as the electrolyte in its semiliquid form in the setup shown in Figure 4-7. The devices operated ordinarily in the bias range of [−0.4 V; +0.6 V] with a regular hysteresis shape and kinetics similar to earlier examples. This means that the pH of the SPE was sufficient and no undesired side reactions inhibiting the switching took place. A PEOconcentration of 30 mg/ml has proven to be too low in the case of AlCl3and GndCl-based gels because their viscosity and density was generally rather low, making the deposition by manual casting more difficult. Thus, we increased the concentration of PEO to 35 or 40 mg/ml while maintaining the overall salt:PEO ratio (6.67∙10-3 mmol/mg) and proportion of dopants. The advantage of denser gels besides the better applicability is that they solidify faster than the more liquid 30 mg/ml analogues. The gel P35-023331 (0.175 M AlCl3, 0.058 M GndCl) was used to assemble an OMD that was subjected to an endurance test over 1000 cycles. The device was placed in a protective chamber, but without humidifying the atmosphere inside with a water-filled petri dish like it was the case with the device featuring gel P30-Alli-01521 in section 4.2.3.1. The series started off with relatively slow kinetics and a low output current which normalized within the first 10 cycles through conditioning of the system (for instance, refreshing Ag/AgCl surface). A slow but steady decay of peak current throughout the series of 1000 continuous I-V-cycles can be observed, becoming almost linear after ca. 27 h. It is noteworthy that no plateau-like peaks were observed like in the 3000-cycle series in Figure 4-23. The current fluctuations in that I-t profile Figure 4-37: Microphotographs of various AlCl3and GndCl-based PEO gels 2 h after deposition: a) P30-ACG0211; b) P30-ACG-02532; c) P30-ACG-02541; d) P30-ACG-0321; e) P30-ACG-0231. f) P30-ACG-0 231 5 min after 60 °C on the heating plate; g) P30-ACG-0231 regenerated for 15 min in humid conditions. 77 were clearly dependent on the time of day and the air conditioning in the laboratory. We argued that the cause for such behaviour was either the temperature or humidity and emphasized the former, reasoning that the water remaining in the petri dish within the chamber should have provided an equal level of humidity throughout the series. However, the more recent results for the device shown in Figure 4-38 indicate that the humidity has a much larger effect on the performance of the OMD than the temperature. This entire endurance 1000-cycles series was recorded without any air conditioning, however, some temperature fluctuations between day and night still occurred. The lack of an additional source of humidity in the chamber combined with the linear decay of the current suggests that the temperature fluctuations alone had no visible effect. On the contrary, in the presence of a source of humidity, an increase of temperature has a visible effect in that the relative humidity increases as well, causing a stronger hydration of PEO and facilitating diffusion processes. Regardless of this detail, the OMD featuring the SPE P35-ACG-023331 displays quite good endurance, considering that the measurements took place in non-optimized, ambient conditions. The gradual decay of output characteristics is most likely caused by partial drying and nucleation of aggregates from which the crystallization could propagate, as demonstrated in Figure 4-38 d). The shape of the Figure 4-38: Endurance series of I-V characterizations of an OMDs fabricated with gel P35-ACG0231 as the SPE, measured with a dwell time of 2 s/0.02 V. a) 1000 consecutive cycles with the bias interval [ −0.4 V; +0.6 V] followed b) Cycles 400−600. c) Development of the peak output current IEL,max throughout the series. d) Microphotograph of the active zone of the device with partial nucleation/ crystallization by the end of the endurance series. 78 hysteresis showed good reproducibility throughout the whole series, recognizable by the switching onset at the exact same bias in any 200 consecutive cycles (except the first few). Summarizing the results of section 4.2 it can be stated that the introduction of intrinsically acidic PEO dopants was very successful at eliminating one of the major sources of instability. The best achieved results are compiled in Table 6. The following attempts to improve the SPE composition for better performance and endurance led to a deeper understanding of the effects that affect the stability of the gel, as well as factors that can cause material degradation. The most notable results were obtained with gels containing aluminium bromide and/or p-toluenesulfonic acid, displaying outstanding thermal stabilities. Although these SPEs did not allow OMD switching in one way or another, their interesting properties motivate to work with other, similar substances to create optimized, functioning gels with superior endurance. Finally, the introduction of guanidinium chloride to AlCl3-doped SPEs allowed to fabricate devices with satisfactory stability in ambient conditions, enabling more sophisticated experiments in the future. Table 6: Summary of the most representative results of our research of novel, multiply doped, PEO-based SPE. gel symbol gel composition (30 mg PEO/ml) stability applicability in OMD prim. dopant sec. dopant short-term (2 h) long-term (2 d+) Alli-0113 0.025 M AlCl3 0.075 M LiClO4 good poor yes Alli-01521 0.1 M AlCl3 0.05 M LiClO4 very good good yes AlAS-0111 0.05 M AlCl3 0.05 M NH4SCN bad bad n. a. AlKS-0113 0.025 M AlCl3 0.075 M KSCN poor bad n. a. AlliS-0211 0.1 M AlCl3 0.1 M LiSCN very good very good no AlliBF-0211 0.1 M AlCl3 0.1 M LiBF4 very bad bad n. a. AlliPF-01521 0.1 M AlCl3 0.05 M LiPF6 bad bad n. a. ABT-0211 0.1 M AlBr3 0.1 M p-TsOH excellent excellent** no ACT-0211 0.1 M AlCl3 0.1 M p-TsOH excellent very good** no ACG-0231 0.15 M AlCl3 0.05 M GndCl* very good good yes * guanidinium chloride, ** very good thermal stability at 90 °C 4.3 Increased device stability through coating with NEA So far, we have shown that some of the gels such as AlCl3-/ LiClO4-based ones display a remarkable reproducibility of device characteristics over long periods of time when provided with a favourable level of humidity surrounding the sample. In this respect, the degree of swelling is a property of PEO-based SPEs that poses a certain cause of instability, as it makes the system dependent on the available content of water. As it was shown in some of our experiments, the relative humidity of the atmosphere surrounding the sample can significantly affect the SPE’s structure and the performance of the related OMD, specifically the output current and the switching kinetics. This 79 source of fluctuations in the device response is intrinsic to PEO-based SPEs and is inevitable, independent from the dopants. Recently, microdevices featuring liquid electrolytes (aqueous HCl solutions) have been presented by Battistoni et al. [181]. While such an approach bears advantages like increased switching kinetics and overall reproducibility, it is rather more suited for model devices for theoretical, proof-of-concept experiments because of the impracticality of liquid electrolytes in electronic devices and systems. Besides, the issues of dopant evaporation (in this case HCl), which was solved in the present work, would persist in such devices, albeit probably less pronounced than in HCl-doped, PEO-based SPE. One way of circumventing the issue of the environmental influence could be the sealing of the devices after the assembly, maintaining the SPE in an optimal state and leaving it unaffected by ambient humidity. Sealing would also hinder the diffusion of air oxygen into the system, preventing oxidative degradation. To apply the concept of device sealing to our OMDs, we have chosen the commercially available NEA 121 (Norland Electronic Adhesive). According to the manufacturer (Norland Products Inc.), the product is a urethane-related resin-based formulation and consists of a mixture of benzophenone and a number of mercapto esters (the exact composition is not disclosed). At room temperature, it appears as a viscous, colourless liquid with a distinct smell. It contains a catalyst that allows it to be cured by irradiation with UV light or at elevated temperatures (e.g., 10 min at 125 °C in a convection oven or 3 h at 80 °C). Curing through heating has the advantage of a more homogeneous process, i.e., the polymerization progresses evenly throughout the whole thickness of the deposited layer. Curing by UV light depends on the thickness of the layer and acts stronger on the layers that are closer to the light source. Nevertheless, due to the sensitivity of most of our SPEs to high temperatures, we resorted to UV-irradiation. The polymerization of monomers occurs in the wavelength range from 320 to 380 nm with peak sensitivity around 365 nm, resulting in a transparent, solid polymer coating. The curing is said to be a very exothermic process. The manufacturer claims that NEA 121 displays very good adhesion on glass, metals, printed circuit boards and many plastics, making it an appropriate candidate for the type of manufacturing techniques envisioned for our devices. Some of the properties of the coating material are presented in Table 7. Table 7: Properties of NEA 121 (source: Norland Products Inc. data sheet). property viscosity at 25 °C 300 cps elongation at failure 30 % modulus of elasticity 160000 psi tensile strength 3500 psi dielectric constant (1 MHz) 4.04 86 5 Kinetic and dynamic aspects in OMD operation 5.1 The role of the device’s geometry The kinetic behaviour of OMDs has been a subject of theoretical research in the past. The electrochemical model proposed by Smerieri et al. [93] and further developed by Demin et al. [94] (see section 0) adequately describes the OMD’s switching process. Although their arguments related to the role of lithium ions is flawed (as demonstrated in section 4.1), the main concepts of their model remain the basis for the theoretical background on the OMD’s switching mechanism. The present chapter deals with relevant kinetic and dynamic aspects of OMD operation. While some related aspects have been discussed in previous works, the available literature does not address some relevant features. Hence, it appears beneficial to provide a summary and explain said effects and their cause. One of the fundamental aspects of OMD’s operation, already discussed in this chapter is the discrepancy between the rates of oxidation and reduction. The reason for such behaviour is essentially the geometry of the OMD. We have shown that in our three-terminal device, the on/off state is determined by the conductivity of the PANI channel. The conductivity of PANI can be controlled through redox reactions that are triggered by the applied electrical potential. This takes place in the active zone (AZ) of the device, which is de facto an electrolytic cell. A central concept that needs to be understood is the distribution of the voltage along the PANI channel, as it is shown in Figure 1-16. At any given source-drain voltage bias, the potential profile in PANI decays towards the grounded source electrode. As a consequence, during the anodic scan the oxidation is triggered in the section that is closest to the drain once the threshold value Vox is reached there, progressing towards the source as the bias increases. On the contrary, during the cathodic scan, the reduction in the part of the AZ closest to the source will be triggered at a drain voltage bias that is higher than the threshold value (Vbias > Vred). Furthermore, during the anodic scan, the PANI-channel as a whole becomes conductive only once the majority of PANI-LE in the active zone transitions into PANI-ES. Conversely, during the cathodic scan, the conductivity of the channel starts decreasing as soon as a fraction of the active zone transitions into the insulating PANI-LE. In other words, when comparing the applied bias with the theoretical redox potentials, during a standard I-V characterization we observe a somewhat delayed switching-on and a premature switching-off of the device. This effect may be better understood from the so-called kinetic characterizations, when the switching occurs at a constant bias. Typical applied voltage values are +0.5 V or +0.6 V for switching-on and −0.2 V for switching-off. As it was mentioned above, the transition of the OMD from the insulating to the conductive state occurs gradually as the oxidation progresses from the drain towards the source electrode. The transition back to the insulating state occurs faster because the conductivity of the whole channel decreases immediately after the reduction of PANI initiates. Furthermore, unlike with positive bias values, when applying a negative voltage, the whole channel is at a potential below Vred, so that PANI is reduced 87 simultaneously in the entire active zone. This is also the reason for the seemingly paradoxical appearance of the IEL-t curve, displaying positive current values at a negative applied voltage, as shown in Figure 5-1 b). Such behaviour could be observed in the kinetic measurements from section 4.1.4. As the switching-off initiates, the S-D-current drops rapidly by a few orders of magnitude, while the gate current remains high (in absolute values) as the reaction progresses. Because the electronic current is calculated as the difference between the two, the resulting value is positive, as it was the case in Figure 4-10. Interestingly, the shape of the curve for HClO4 resembled the one in Figure 5-1 a), which is due to the higher off-state currents caused by an inhibited reduction process. The role of the position and width of the active zone Another factor that emphasizes the importance of the geometry of the device is the influence of the position and width of the active zone, which again is connected to the distribution of the voltage across the PANI channel. This mostly affects the switching-off process, when the entire active zone is in the conductive state. According to Figure 5-2 a), the potential difference within the AZ decreases with its width, resulting in a more uniform response. Similarly, positioning the active zone closer to the drain causes PANI's potential within it to match more closely the applied bias (Figure 5-2 b). This leads to a more predictable and responsive switching behaviour, since the applied voltage required for switching better matches the theoretical redox potentials. The switching-on is less affected by such geometrical changes because, in any case, all the voltage drop mostly occurs in the active zone. Overall, it becomes clear that, in principle, the position and width of the AZ can be used as a tool to control the I-V characteristics of the OMD. However, in practice, a precise control of the position of the active zone becomes difficult in application-oriented, miniaturized devices as it is more convenient to resort to a geometry where the whole PANI channel is covered with the (poly)electrolyte and is hence active. In this case, another effect becomes relevant when considering the switching behaviour at constant biases. As it was mentioned in section 0, the switching rate also depends on the distance between the gate electrode and the conductive channel. This effect becomes more significant with an increasing ratio of the Figure 5-1: Schematic representation of an OMD’s kinetic response at + 0.5 V (oxidation, switching-on) and −0.2 V (reduction, switching-off). a) Expected c urve shape with a negative electronic current at a negative bias. b) Typically observed curve shape with a positive electronic current at a negative bias. 88 channel thickness and its length. In a bottom-contact/top-gate configuration, the upper layers of PANI react sooner than the lower ones, as shown in Figure 5-3. This may be an additional source of delay for both the switching-on and -off because the conductive state of the lowest PANI layers are the most relevant, since they are in direct contact with the source and drain electrodes. In this respect, a bottom-contact/bottom-gate configuration could be beneficial in terms of switching rates (see Figure 5-3 b). Such a configuration can only be realized with a solid polyelectrolyte that could better support a layer of PANI deposited on top it. Together with the fact that this effect only applies for short channels, this makes the idea of a bottom-gate OMD configuration better suited to be implemented by high-resolution printing techniques, which will be discussed in chapter 4. 5.2 The relationship between electrode potentials and the applied bias It has been established that the actual electrode potential of PANI does not correspond to the applied drain voltage bias. However, assuming a central position of a narrow active zone in the onstate and an applied bias of +0.4 V, the true electrode potential of PANI in that region would not be Figure 5-2: a) The effect of the width of the active zone, affecting the potential range between the parts close to the drain and source. b) The effect of the position of the active zone with respect to the source and drain electrodes, affecting the deviation of the actual potential values within the AZ from the voltage bias applied at the drain (schematic). Figure 5-3: Schematic representation of t he oxidation and reduction process for an organic memristive microdevice a) in the standard configuration and b) in the bottom-gate configuration. The arrows represent the propagation of the transition of PANI. 89 Equation 14 Equation 15 Equation 16 +0.2 V, as suggested solely by the potential distribution profile. As we mentioned above, from an electrochemical point of view, the PANI and the silver electrode that are in contact with the SPE in the active zone form an electrolytic cell. Hence, the applied bias can be seen as the potential difference between the anode and cathode that is imposed upon the system: In our case, when a positive bias is applied to the drain, the PANI electrode represents the anode, and its potential can be calculated as the sum of the applied bias (corrected by the voltage distribution profile) and the electrode potential of the Ag/AgCl cathode. At this point, the fact that the OMD’s gate electrode is not a true reference electrode becomes relevant, as its electrode potential is not constant throughout the measurement. It rather depends on the activity of chloride ions whose concentration is also variable due to the formation of ion gradients caused by the applied voltage. The electrode potential can be calculated following the Nernst equation (Equation 15). However, it would be necessary to determine the concentration and activity coefficients of the chloride ions at different bias values.   for Ag/AgCl:  R … universal gas constant (R = 8.314 J mol -1 K -1 ) F … Faraday constant (F = 9.6485℅10 4 As/mol) E°… standard electrode potential (E° = 0.222 V for Ag/AgCl [182]) a i … ion activity (a i = f i ℅c i ) f i … activity coefficient According to this relationship, the OMD switching onset at around +0.3 V (theoretical Vox of PANI) in some cases is rather coincidental, and means that under the given conditions, this bias equals the difference between the electrode potentials of PANI and Ag/AgCl. This also explains how the switching may sometimes initiate at low voltage biases such as ca. + 0.14 V in one of our endurance series Figure 4-23. Since the chloride ion partakes in both half-reactions, both electrode potentials depend on chloride concentration. This means that the dopant concentration in PEO gels is not only important from the point of view of SPE stability but may also be used as a tool to control the switching onset potential. It is possible that the improved kinetics observed with AlCl3-doped SPEs were caused by the increased chloride concentration. It should be noted that since chloride ions partake in both halfreactions, the electrode potential of both Ag/AgCl and PANI depends on their concentration. However, due to the applied electrical field, the chloride concentration is higher at the positive pole and lower at the negative one. According to the Nernst-equation, during the switching-on with a 90 positive applied bias, the concentration gradient causes a correction of the electrode potential of Ag/AgCl to slightly higher values, and that of PANI to slightly lower ones. Consequently, a lower applied bias should suffice to trigger the reaction according to Equation 14. For the switching-off reaction, the situation is reversed because of the inversed polarization of the electrodes and orientation of the concentration gradient. 5.3 On-line observation of OMD switching Although the importance of the effects connected to the potential distribution across the conductive channel has been established in early works on PANI-based OMD, there is still little evidence for the proposed switching mechanism. In section 1.2.4, we presented some early spectroscopic data aimed to reaffirm the alleged participation of lithium ions in the switching process, corrected by our own data in section 4.1. However, those results are merely indirect indications of the involvement of one or the other ionic species. A more direct approach has been taken by Battistoni et al. [183], featuring a spectrophotometric observation of the transition of PANI from the insulating to the conductive state, correlated with simultaneously recorded electrical characteristics. As a development of this approach, we have carried out a study to optically visualize in real time the switching between the transient states of PANI in the active zone. The results of this study were presented at the MEMRISYS 2019 conference in Dresden (2019) [184]. The experimental equipment resembled the one used in section 4.1.2 for measurements in liquid electrolytes (see Figure 4-7), featuring a Teflon trough on top of which the OMD was mounted “face-down”. The transparent quartz substrate allows to observe under a microscope the 52 LS layers thick PANI channel from below, without the interference of the gate electrode. A silver wire or a foil (125 µm in diameter or thickness) were used as the gate electrode, attached to the bottom of the Teflon well, filled with a semiliquid PEO gel with 0.05 M LiClO4 and 0.1 N HCl or aqueous 0.1 N HCl as the electrolyte. The transient state of PANI was determined by the colour of the channel in the active zone on which the microscope’s digital camera was focused. The electrical characteristics were acquired simultaneously in the kinetic mode at constant applied biases. The predictable colour transitions of PANI between yellow and green, corresponding to PANI-LE and PANI-ES, are shown in Figure 5-4 b) and e), with the propagation of the PANI-ES front from the drain to the source electrode presented in panel f). Meanwhile, panels a), c) and d) display an unexpected second transition from green to blue and back. Initially, this second transition was attributed to the temporary oxidation of PANI-ES to PANI-PS, with an explanation based on the voltage distribution across the channel, as schematically shown in Figure 5-5. The initial potential profiles (lighter lines) indicate that at high enough voltage biases, there are regions of the AZ close to the drain that are at a potential exceeding Vox(ES-PS), triggering the second oxidation step towards the purple-blue pernigraniline form. As the voltage is redistributed (darker lines), eventually, the potential in these regions drops below Vox(ES-PS) again, so that the green PANI-ES is reformed. An indication for such behaviour was interpreted in the 91 Figure 5-4: a-e) Series of snapshots from digital on-line r ecords of the transition of PANI in the active zone of an OMD during kinetic measurements at different measuring conditions. No change was observed after 20 s in panel e). f) Representation of the propagation of the PANI-ES front from the drain to the source electrode. The colour saturation in series a) and f) was digitally edited for clarity. Figure 5-6: Recorded electric characterization of the OMD corresponding to the measurement in Figure 5-4 c). Figure 5-5: Schematic illustration of the proposed potential distribution across the PANI channel during kinetic measurements at high voltage biases (the position of the two oxidation steps is chosen arbitrarily). The arrow represents the propagation of the oxidation of PANI-LE. The red lines are a simplified representation of the potential profile at different times, with the respective wouldbe oxidized region of the active zone indicated by the vertical lines. 92 recorded I-t characteristics (Figure 5-6), displaying a negative peak in the ionic current at a positive applied bias that was attributed to a reduction of PANI-PS to PANI-ES. This is accompanied by a temporary increase of resistivity, corresponding to the insulating state of PANI-PS. The fact that no such transition is observed in Figure 5-4 b) is explained by the lower applied bias. In panel e), the measurement takes place in a liquid electrolyte in which the second transition remains undetected due to faster kinetics, favoured by faster diffusion processes. The reasoning behind these observations was later reconsidered as some of the above arguments are flawed. The first and foremost source of doubt is the colour of the polymer after the alleged second oxidation step. Although colours might appear altered due to the superimposition of different transient states of PANI in the same film, the blue colour of the polymer corresponds much rather to emeraldine base than pernigraniline which should show a shade of purple. Secondly, although the voltage (re)distribution profile is illustrated schematically, the insulating character of PANI-PS is not represented. This fact cannot be ignored because it would significantly change the picture, since the potential drop mostly occurs within the insulating regions. Lastly, it becomes clear that this explanation does not take into account the geometry of the device with a central position of the active zone. This would mean that PANI’s geometry-corrected potential values within the AZ would be significantly lower than the theoretical oxidation potential for the transition from PANI-ES to PANI-PS, with the possible exception of Figure 5-4 d) and e) where the applied bias is +1.6 V. We hence assume that the occurrence of the blue colour at high biases and the increase of resistance in Figure 5-6 correspond to the appearance of emeraldine base as a consequence of PANI dedoping. The negative peak of the ionic current is attributed to the polarization of the electrode rather than to a chemical reaction. It is reasonable to assume that such dedoping was caused by the strong channel electric field expelling the protons from the positively charged PANI electrode. This also explains the absence of a green-blue transition in the cases with lower applied voltages, where the electric field is weaker. The returning green colour, meaning the re-doping of PANI, is due to the gradient of chloride ions that is inverse to the one of protons. The accumulating negative charge is gradually reduced through the reprotonation of progressively formed imine nitrogen sites, providing the possibility for the formation for ionic couples with chloride. This assumption is consistent with the fact that a slight blue colour appears in panel a) of Figure 5-4 while none is observed in panel b), although the applied bias is equal in both cases. The reason is that in the first case, the strength of the electric field is amplified by the surface area and parallel orientation of the silver foil as opposed to the wire. Lastly, no transition to PANI-EB is observed in panel e) because of the increased ionic mobility of protons (following the Grotthuss mechanism) and chloride ions in an aqueous electrolyte. This effect is also likely the reason for the shape of the hysteresis curve in Figure 4-19 b). The slope of the curve slightly decreases after +0.4 V, indicating a lower conductivity of the channel that may have been caused by deprotonation driven by the electric field. The influence of the counterion, pH and the ingress/egress of H+ or anions in and out of the polymer on the switching process and on the conductivity of PANI have been discussed in the works by Focke et al. [83] and Kalaji et al. [185]. 93 The reasoning behind the behaviour of the OMD’s active zone is confirmed by similar results by Xia et al. [186] who have studied a NiO/PANI composite material that responded in a similar manner to the application of different voltage biases. In a combination of our work and that of Battistoni et al. [183] Lapkin et al. [187] presented a study where the absorbance of the PANI layer in the active zone of the OMD has been registered by the CCD matrix of an optical microscope. These results are in accordance with the ones presented in our work, although no intermediate transition was observed there, since the measurements were carried out at lower biases. In conclusion, the concepts introduced in Chapter 4 add to the understanding of the factors affecting the stability and endurance of OMDs. Specifically, the demonstrated dependence of the performance of the devices on the pH and the concentrations of chloride ions gives rise to speculation as to whether more extensive modification of the materials could further increase the stability. Modified polyanilines such as poly(N-ethyl aniline) [158],[159] could partially solve the pH dependency because the trisubstituted nitrogen atoms do not require protonation for the formation of polarons upon oxidation. Besides, the susceptibility to chain cleavage through hydrolysis would be lowered because the nitrogen atoms would be stabilized by the +I-effect of another substituent. Consequently, the development of more stable SPEs would also be facilitated, since the focus could be completely shifted towards the stability of the PEO gel, with no regard to the addition of acidic dopants. On the other hand, sulfonated [80] or otherwise self-doping polyanilines with a carefully controlled ratio of dopant groups and nitrogen atoms would eliminate the channel’s dependency from anion gradients within the electrolyte. Thus, a sulfonated, N-alkylated polyaniline (see Figure 5-7) seems to be an interesting candidate for further research. Figure 5-7: Chemical structure of a sulfonated, Nalkylated polyaniline as a potential candidate as a conductive polymer for future OMDs with increased endurance. 94 6 Towards automated OMD manufacturing – printed neuromorphic devices In the previous chapter, we explored ways to prolong the lifetime of the OMD by improving the solid polyelectrolyte as one of its most labile components and by sealing the devices with NEA to protect them from environmental effects. These approaches allowed us to significantly increase the endurance and cycle-to-cycle reproducibility of individual OMDs with respect to earlier standards. So far, the device assembly has mostly been done manually, inherently bringing a certain degree of variability with each manufacturing step. For instance, although the Langmuir−Schaefer technique enables a precise control over the thickness of the deposited polyaniline layer, the quality of each layer may vary and depends on the experience and proficiency of the operator. Another source of variability during manual fabrication is the positioning and width of the active zone, as well as the thickness of the SPE layer that affect the kinetics, as we have shown earlier. These and other factors lead to a relatively poor device-to-device reproducibility of electrical characteristics, even after improving each individual device’s endurance. The statistics in our case gives the peak output current ranging between 5 and 30 µA, with most recurring values of 10−20 µA. While it is mostly required a stable kinetics, among other features the reproducibility of the on/off ratio value also plays an important role insofar as its control represents a key for implementing more complicated, sophisticated systems such as ANNs and sensing interfaces. Hence, from a technological standpoint, there has long been a need for steps towards the automation of the manufacturing process. In fact, besides minimizing variability among different devices, other advantages of automated manufacturing are the timesaving (i.e. the so-called rapid prototyping) and ease of downscaling routes, the latter being a prerequisite for higher performance, faster kinetics and mass production. While OMD downscaling has been attempted earlier by Lapkin et al. [157] and Battistoni et al. [181] (both worked with channel lengths of 200 µm instead of a few millimetres), those works still featured a manual deposition of PANI with the LS technique. The present chapter is hence devoted to developing methods for automated deposition of the OMD’s individual components by means of manufacturing protocols based on 3D printing techniques. 6.1 Printing techniques (state of the art) A rough overview of the various printing techniques is shown in Figure 6-1. Because of their versatility and efficiency, we focused on non-contact, direct writing (DW) technologies, such as Inkjet (IJP) and Aerosol Jet Printing (AJP), that are considered as promising methods in organic microelectronics manufacturing. Direct writing techniques stand out as purely additive methods alternative to standard techniques such as photolithography widely employed for the manufacturing of standard (e.g. metal/metal oxide junctions) and organic-based electronic devices. In particular, they compare favourably to other printing techniques, e.g. screen printing, nanoimprinting or gravure printing, due to their gentleness and versatility in terms of compatible substrates, including flat, 95 flexible and even fragile ones, such as optical fibre glass [188]. Despite the relative simplicity combined with the industrially scalable character, these techniques are highly efficient in that the material waste is minimized and the fabrication cost is reduced, while still allowing to generate complex patterns. Major drawbacks come from the difficulty to achieve fast mass-production of devices and systems. The basic principles of the Inkjet and Aerosol Jet Printing methods, as well as their advantages and limitations, are presented in the following sections. 6.1.1 Inkjet Printing Inkjet Printing (IJP) is a contactless, direct-writing method based on the ejection of microdroplets containing the functional material to be deposited. The ink (a mixture of solvents and the functional material) is fed into the deposition head as a solution or colloidal suspension and it is directly patterned onto the substrate, heated to assist solvent evaporation. The ink is released through a nozzle in the form of microdroplets, controlled by a piezoelectric, thermal or electrohydrodynamic actuator [189], as shown in Figure 6-2. The droplet jet is produced due to pressure pulses, generated through deformation of the fluid cavity, through thermal expansion of the ink and the sudden formation followed by a collapse of vapour bubbles, or by generating an electric field between the nozzle and the substrate, respectively. The ejection of droplets can occur in either of the two following modes. The continuous ink jet (CIJ) mode involves the application of a DC field between the nozzle and the substrate plate, resulting in an uninterrupted release of charged droplets in a stream that breaks down into a column of individual droplets, due to Rayleigh instability [190],[191]. The ink is recycled by directing the jet towards a gutter or deflected towards the substrate by applying an electric field between two metallic plates. The droplet-on-demand (DOD) mode is characterized by a pulse-like release of droplets when requested by the pattern to be printed, thus saving ink material when the nozzle is not actuated. Ejection frequencies of 1−20 kHz can be achieved with piezoelectric transducers [190]. Electrohydrodynamic actuators enable the nozzle operation in both the CIJ and DOD mode, depending on the type of applied current (DC or AC, respectively) [189],[192]. Figure 6-1 : Overview of common printing techniques employed for organic materials (adapted from [189]). 102 drying of the droplets caused by the sheath gas flow. The optimal NMP/xylene/IPA ratio was found to be 1:1:2 v/v, since lower percentages of NMP led to PANI precipitation. This new ink was printed on quartz and polyimide tape substrates with the same CGF and ShGF settings, whereas the higher PANI concentration allowed a faster deposition at a plate speed of 2 mm/s. The new ink formulation in combination with the UV/ozone treatment of the substrate (which is not suited for fibroin substrates) allowed to obtain well-defined, 160 µm wide lines without any notable overspray or spotting, as shown in Figure 6-4 c, d). Other preparations, such as ink filtering with a 0.2 µm PTFE syringe filter, were excluded because it was detrimental for the printing process, causing spotting. The reason is that due to the low portion of NMP, some of the PANI precipitated, forming a colloidal suspension that was removed by filtration, lowering too much the material load of the ink. Hence, filtering was omitted in the following tests and ultrasound treatment was found to be the only process allowing a better PANI dispersion in the chosen solvent mixture. With this approach, six stacking layers of PANI were printed onto a 15x7 mm quartz substrate equipped with Cr electrodes defining the OMD channel. Large-area films on quartz have proven problematic as we observed a poor layer adhesion as well as gaps between the printed lines, as shown in Figure 6-5 a). Hence, high resistance values, in the order of MΩ, were found after the film doping with HCl. For comparison, LS-deposited channels (30 layers) of Figure 6-4: Upper panels: micrographs of aerosol jet prints of a pristine 0.1 mg/ml PANI ink in NMP/IPA (1:1 v/v) on a) quartz and b) fibroin, with line widths of ca. 200 µm. Printing on quartz resulted i n inhomogeneous drying and the formation of PANI islands while the features were better defined on fibroin. Lower panels: micrographs of aerosol jet prints with a 1.0 mg/ml PANI ink in NMP/xylene/IPA (1:1:2 v/v) on c) quartz and d) polyimide tape, with line widths of ca. 160 µm. The sharp lines coming off the printed line in panel d) are scratches on the tape surface. 103 equivalent dimensions displayed resistance values between 5 and 50 kΩ. The same ink was printed onto Si/SiO2 substrates (1−2 µm thermal oxide layer) with gold electrodes prefabricated by photolithography (dSD = 200, 10 or 5 µm; w = 6 mm). The line definition was largely improved because of the better wettability of the substrate, as shown in Figure 6-5 b). After doping in HCl, the measured channel resistances were in the ranges of 10−50 kΩ for 200 µm channels, 450−1500 Ω for 10 µm channels and 230−1000 Ω for 5 µm channels. Hence, resistance values scaling was consistent with the scaling of channel length, indicating the films’ quasi-ohmic behaviour. These values are comparable to standard OMDs, albeit lower values are to be expected considering the geometry of the channel. Despite the improved coverage, narrow gaps between the printed lines were still recognizable upon optical microscopy. Furthermore, the SEM images and micrograph in Figure 6-7 a, b, e) revealed that the deposited layer was again quite inhomogeneous, with cracks and pores up to 1 µm (with the deposition of a few larger ones up to 10 µm) in diameter. Using the above PANI films, the first ever made OMD with a printed PANI layer was fabricated on test patterns with dSD = 200 µm and using P30-Alli-0113 as the SPE. The scheme and I-V characteristics of the device are shown in Figure 6-6, displaying an ordinary memristive characteristics, featuring a hysteresis and rectification of the I-V curve and proving the concept of printed organic memristive devices. Figure 6-5: Micrographs of PANI printed with the 1.0 mg/ml ink in NMP/xylene/IPA (1:1:2 v/v) onto a) a standard quartz substrate (6 printed layers) and b) a silicon substrate (1−2 µm oxide layer) with prefabricated gold electrodes (dSD = 200 µm, 10 printed layers). Figure 6-6: a) Schematic representation and b) I-V characteristics of an OMD fabricated with P30-Alli-0113 as the SPE, featur ing an AJ printed PANI layer on Si/SiO2 with Au electrodes (dSD = 200 µm). Scanning was done with a dwell time of 2 s per step of 0.02 V. 104 Figure 6-7: SEM images of aerosol j et printed pristine PANI on Si substrates with the 1.0 mg/ml ink in NMP/xylene/IPA (1:1:2 v/v): a, b) 10 printed layers without plasma pre-treatment of the substrate at 1K and 30K magnifica tion, respectively; c, d) 4 printed layers with plasma pretreatment of the substrate at 10K and 30 K magnification, respectively. Noteworthy is the improved coverage and reduced pore amount and average diameter in c, d). Panel e) shows a micrograph of the sample from panels a, b). Note the profile difference between the bulk and edges of the lines as well as the cracks within the printed layer. For comparison, panel f) s hows a SEM image of an LS-deposited PANI channel (24 LS-layers) from reference [109]. The LS layer has a granular morphology while the printed structures are rather fibrous. 105 Nevertheless, such defective films imply the need of a new printing pattern, consisting of an enhanced overlapping between adjacent printed lines, as demonstrated in Figure 6-8. This pattern has been adopted in all the following deposition runs, aiming to reduce the gaps between adjacent lines and the porosity of the layer. Additionally, to further improve the wettability of the substrate, it was subjected to a O2-plasma pre-treatment, aimed at both removing eventual residual organic contaminants and making it more hydrophobic by eliminating surface-bound water molecules. Using this approach, the resistance of the printed layers was further lowered to about 340−6500 Ω and 30−500 Ω for substrates with dSD = 200 and 10 µm, respectively. As shown by the SEM images in Figure 6-7 c, d), the morphology of the deposited layer is significantly improved, giving a much better conducting channels made of fewer printed layers. 6.2.2 Aerosol jet printing of a chitosan:PANI composite material Although our approach produces PANI films with reduced defects and lower resistance values, it seemed to be too much dependent on the processing route. So, it appeared to be more convenient to change the strategy and develop a new PANI-based ink that would allow a better coverage of the substrate. The dewetting issues connected to the ink composition are difficult to overcome even by means of surface treatment, with the main problem being the presence of NMP. As a nonvolatile solvent, NMP is generally not well-suited for printing applications. Besides, it poses severe health risks being classified as toxic and teratogenic. For this reason, we proceeded by searching to modify the printed material itself since pristine PANI is insoluble in other solvents. We took inspiration from the work by Ratuchne et al. [236], describing a chitosan:PANI composite material that is reported to be more conductive than polyaniline (due to a higher surface area). According to that work, this material is also more thermally and electrochemically stable, making it a promising material for electrochromic devices and capacitors. Most importantly for the present thesis work, the material is reported to be water-soluble, making it an interesting candidate as a PANI surrogate for printing applications. The water-solubility is imparted to the blend by chitosan – a biocompatible polymer that has already been used in OMDs as an SPE [237]. Figure 6-8: Schematic representation of the new adopted printing pattern, a) top view, b) side view. In one case each next layer is printed on top of the previous one and in the other, there is an offset of half a step with each next layer. 106 According to the procedure from literature [236], the chitosan:PANI blend (CPA, see Figure 6-9) was synthesized from aniline in hydrochloric medium in the presence of pre-dissolved, low molecular chitosan, following an oxidative polymerization mechanism. The full synthesis instruction is given in section 2.3. A dark green product was obtained, indicating the emeraldine salt form of PANI. The IR spectrum in Figure 6-10 features the characteristic absorption bands: δop(C−H) and δip(C−H) in 1,4-disubstituted rings at 828 cm-1 and 1106−1164 cm-1, respectively; bipolaron ν(=NH+−) at 1140 cm-1 (as a shoulder band); polaron ν(C−N•+) at 1238 cm-1; ν(C−N) of secondary aromatic amines at 1307 cm-1; ν(C=N) in the vicinity of quinoid rings at 1379 cm-1; benzenoid ν(C=C) at 1495 cm-1; quinoid ν(C=C) at 1586 cm-1 [17]. The absence of a sharp band around 3610−3645 cm-1 indicates that no free −OH groups are present, suggesting that they participate in intermolecular hydrogen bonds, characterized by a broad band around 3200−3400 cm-1. The same applies to a broad band around 3000−3200 cm-1 which can be attributed to H-bound NH groups. The band at 3385 cm-1 corresponds to free NH groups of PANI that do not participate in H-bonding. The small band at 1009 cm-1 corresponds to ν(C-O) of the secondary cyclic alcohol moieties in chitosan. Overall, the chitosanrelated signals are weak with respect to those of PANI, indicating that the blend most likely consists mainly of polyaniline. Figure 6-9: Chemical structure of the chitosan:PANI composite (on the left) and a photograph of the obtained product (on the right). Figure 6-10: FTIR spectrum of the synthesized CPA composite (KBr pellet). 107 Additional structural information could be gained from MALDI-TOF measurements, although no precise information could be given about the molecular weight distribution of the formed PANI chains. The analysis has proven difficult because the composite is insoluble in the most common solvents like water, NMP, isopropanol, ethanol, DMSO, acetonitrile (ACN), chloroform or THF. Dispersions on the other hand adhered too weakly on the substrate/matrix, often making an effective ionization difficult. As exemplified in Figure 6-11 a), there was a group of three signals around 3400 u that reoccurred in the majority of all the acquired spectra. It appears as though this particular formation is especially abundant in our material and/or more stable under the ionization conditions. Although information could be gained about the structure of this species, it is reasonable to assume that the signals arise from chain-like or cyclic polyaniline structures. An indication of this are the differences in the m/z of the individual peaks, being ca. 71.3 u between the second and the first, and ca. 115.8 between the third and the first one. These atomic weights correspond to a doping of imine nitrogen atoms by two HCl units (ca. 72.9 u) or two CH3COOH units (ca. 120.1 u, remaining solvent traces from the synthesis). The periodically occurring, weaker signals in Figure 6-11 a) also have a difference of ca. 72 between them, further indicating the presence of PANI chains with various levels of HCl doping. Similar results can be seen in Figure 6-11 b), exhibiting the same periodicity of signals in the inset and indicating chain lengths of up to 8 kDa. However, considering the solubility issues, it is likely that the majority of the product has a significantly higher molecular weight, since only a fraction of it could be dissolved in HFIP. In the future, a more detailed analysis of the product (e. g. deposited as a solution or directly printed by AJP) could deliver more insight about its structure. Figure 6-11: MALDI-TOF mass spectra of the synthesized CPA composite deposited on the substrate a) as a dispersion in H2O/ACN (1:1 v/v), b) as a solution in H2O/HFIP (1:1 v/v). The baselines are subtracted, the margin of error is 1000 ppm. 108 Despite the solubility issues of CPA, it could easily be dispersed within 10 s by means of ultrasound treatment in water/IPA (8:2 v/v), forming a dark green suspension that remains stable over long periods of time, with minimal reprecipitation after weeks. The fact that the material is fully dispersed and not dissolved is proven by filtration with a 0.45 µm PTFE syringe filter, resulting in a clear, colourless filtrate. We attribute such behaviour to the relatively small fraction of chitosan in the blend, as suggested by IR, as well as its presumably high molecular weight. In the original paper [236], the molecular weight of the utilized chitosan was not specified, but the reported particle size and plate-like morphology of the material obtained by Ratuchne et al. (see SEM image in Figure 6-12 d) suggest that it might have been larger than in our case (low molecular weight, 20−50 kDa). For comparison, Figure 6-12 a, b) shows SEM images of CPA printed with the water/IPA-based, 2.0 mg/ml ink (8:2 v/v) on Si/SiO2 substrates. The morphology results to be fibrous and rather similar to the deposited pristine PANI samples from the previous section. The coverage of the substrate is largely improved, even though the surface has not been pre-treated with plasma, suggesting a much better wetting of the substrate and evaporation rate of the ink. The resulting printed lines were very well defined, with widths down to 120 µm. The SEM images show that the printed layer is quite porous and rough, with randomly occurring sites where some aggregation of the material occurs. The thickness of the layer was estimated from ATM measurements of single-layer films on the edge of Figure 6-12: a, b) SEM images of single layer printed CPA on a Si/SiO2 substrate at 1K and 20K magnification, respectively. c) Profile of the printed layer measured by means of AFM. d) SEM image of the blend from reference [236], for comparison. 109 the printed layer (see Figure 6-12 c). Typically, values around 1 µm per layer were obtained. Despite the excellent definition of features, the adhesion of the layer was quite poor as it could be mechanically removed with ease. Plasma treatment of the substrate could possibly solve this issue in future experiments. Resistance as low as 14 Ω could be achieved after doping in HCl. Considering a channel width of 4 mm, length dSD of 10 µm and estimated thickness of 2 µm (1 µm per layer), this corresponds to a conductivity of ca. 1.8 S/cm. For comparison, pristine PANI films deposited with the LS technique (30 layers) on standard 15x7 mm quartz substrates show conductivity levels around 50 S/cm. In its doped, emeraldine salt state, the layer displays nearly ideal ohmic behaviour, as shown in the inset of Figure 6-13 a). A memristive device has been fabricated by printing two CPA layers as the conductive channel on a Si/SiO2 substrate equipped with gold electrodes (dSD = 10 µm), using the gel P30-Alli01521 as the SPE and a silver wire as the counter electrode. The I-V characterization are shown in Figure 6-13 a). The device displays a very high performance with a well-defined, reproducible hysteresis of the I-V curve. Further tests have shown that the OMD remains fully functional at scan rates as high as 100 mV/s, with minor widening of the hysteresis, although the oxidation reaction starts to notably lag behind the applied bias at scan rates above 40 mV/s (Figure 6-13 b). Figure 6-13: a) 40 consecutive I-V cycles measured with a dwell time of 2s per step of 0.02 V (10 mV/s scan rate) on an OMD fabricated with 2 printed CPA layers as the conductive channel and P30-Alli-01521 as the SP E (on Si substrate with Au electrodes and dSD = 10 µm). In the inset, the almost ideal ohmic behaviour of the layer is demonstrated. b) IV characterization of the device at different scan rates. c) Photograph of the device (only the middle left layer is involved). 110 6.3 Towards AJP-compatible polyelectrolyte materials Although the quite good and novel results obtained from devices fabricated with the CPAbased ink from the previous section, there is a switching kinetics that is not matching the expectation based on the reduced channel dimensions. Such a behaviour can be explained considering the thickness of the printed CPA layer. With a thickness of 1 µm, on a scale of a channel length of only 10 µm, the layer attains bulk properties in that its conductivity strongly depends on the charge transfer between the single deposited layers, or in the case of printed CPA, rather between the single fibre microstructures. This is especially relevant for the bottom-contact/top-gate geometry used so far, where the switching of the channel initiates closer to the gate electrode, but only comes into effect once the conductive connection is established with the source and drain terminals. As was discussed in section 5.1, this issue could be circumvented by adopting a different kind of device architecture with a bottom-contact/bottom-gate arrangement, where the bulk properties of the channel play a minor role. The versatility of AJP provides the necessary tools to approach the sought-after goal of fabricating fully printed devices, including all the components of an OMD. With the ability to print conductive metal nanoparticle inks, AJP allows us to fabricate a silver gate [218] and gold source and drain electrodes [238], with the final step of interconnecting them with the conductive CPA ink developed above. The missing link is the deposition of the polyelectrolyte material to create the junction between the gate and the conductive channel. By consequence, it would be required to print CPA onto the latter, however it is not expected to be an issue since AJP is suitable for the deposition on a variety of substrates, including flexible, uneven, and soft ones. The following section is dedicated to making steps towards developing printable polyelectrolyte materials. 6.3.1 Low-viscosity PEO-based hydrogels Our development of printed polyelectrolytes relies on the knowledge acquired when improving the PEO-based SPE composition in chapter 4. Since the properties of the gel remain unchanged when maintaining the same dopant:PEO ratios, it seems possible to prepare a less concentrated gel with a low enough viscosity for being printed. However, gels fabricated with highmolecular PEO (Mw = 8 MDa) have proven too viscous to be processed at reasonable concentrations, even down to 5 mg/ml. We hence proceeded to prepare PEO gels with lower average chain length (Mw = 1 MDa), yielding more fluid solutions of sufficiently low viscosity at 30 mg/ml. The durability of the gels has been assessed optically as in the previous cases. Out of all the prepared gels, the three most promising candidates were P1M30-Alli-01521 (0.1 M AlCl3, 0.05 M LiClO4), P1M30-Alli-01532 (0.09 M AlCl3, 0.06 M LiClO4) and P1M30-ACG-0231 (0.15 M AlCl3, 0.05 M GndCl). The P1M30ACG-0231 gel has proven the most durable in comparison to the LiClO4 doped counterparts, as shown in Figure 6-14 a, b), whereas both types of doping were on par in analogous gels with 8 MDa PEO. We assume that in the case of lower molecular, 1 MDa PEO, the overall hygroscopicity of the dopant 111 ensemble becomes less critical because the chains become easier to hydrate, while the benefit of the strong chaotropic character of the guanidinium cation becomes more significant and comes into effect more. Despite the low viscosity of the gel, the atomization of the SPE ink could not be achieved and hence it could not be used in the AJP. This is likely due to the still very high mean molecular weight of PEO chains, which prevents them from effectively entering the aerosol created by the ultrasonic atomizer. On the other hand, further lowering of the molecular weight of PEO also was unsuccessful as the prepared gels of comparable concentrations (30−150 mg/ml) did not solidify even overnight, potentially making it unsuitable for the deposition at room temperature. Assisting the water evaporation through mild heating at 40 °C resulted in rapid and irreversible drying out of the samples, as shown in Figure 6-14 c). It appears that the system becomes less effective at containing humidity with decreasing PEO Mw, while the interactions between the shorter PEO chains are too strong to invert the formation of aggregates by increasing the humidity. While the application of P1M gels was unsuccessful by means of AJP, it is possible that printing could succeed with the otherwise less advanced IJP technique, thanks to its more straightforward mechanism of the formation of the ink jet that requires no provisional atomization of the material. Issues might arise from the need to heat the deposition plate for a better deposition and definition of printed features, regarding the temperature and humidity sensitivity of the SPE system. This does not concern the deposition process of the PEO gel itself, since in this case, the heating of the plate in particular and the precision of printing in general is not required, since it is applied as a bulk and only serves to provide the medium for the electrochemical reactions between the channel and the gate. Much rather, it is the last manufacturing step, being the deposition of CPA in a bottomcontact/bottom-gate geometry, which is most likely to cause drying out and crystallization of PEO. However, it has been shown in section 4.2.3.4 that gels based on guanidine chloride can be quickly regenerated in a humid atmosphere after drying, making P1M30-ACG-0231 it a promising candidate for future attempts of printing with IJP. Figure 6-14: Photographs of lower molecular PEO gels with Mw = 1 MDa, 6 h after deposition. a) Gel P1M30-Alli01521 (30 mg/ml PEO, 0.1 M AlCl3, 0.05 M LiClO4) with incipient drying in the centre of the cast drop. b) Gel P1M30-ACG0231 (30 mg/ml PEO, 0.15 M AlCl3, 0.05 M GndCl). The aggregates on the perimeter did not grow considerably ove r time. c) An even lower molecular PEO gel with Mw = 12.5 kDa after drying at 40 °C. The crystallized PEO could not be regenerated in a humid atmosphere or by directly adding water to the system. 118 study of abovementioned noise-related and other effects, but also make advances in the biotechnological, biomedical, computer science, neuroscience and robotics fields. In conclusion, with our research work we believe we achieved important, fundamental steps in this direction by developing materials and method of fabrications that allow to produce much better performing, more stable, reliable and reproducible OMDs with materials that make the interfacing with the bio-world viable. 119 Bibliography [1] Y. Li, Z. Wang, R. Midya, Q. Xia, J. J. Yang. “Review of memristor devices in neuromorphic computing: materials science and device challenges.” J. Phys. D: Appl. Phys. 51, 503002 (2019). [2] D. O. Hebb. “The Organization of Behavior: A Neuropsychological Theory.” (Wiley, 1949). [3] N. Rochester, J. H. Holland, L. H. Haibt, W. L. Duda. “Tests on a cell assembly theory of the action of the brain, using a large digital computer.” IRE Trans. Inform. Theor. 2, 80–93 (1956). [4] F. Rosenblatt. “The perceptron: A probabilistic model for information storage and organization in the brain.” Physiolog. Rev. 65, 386–408 (1958). [5] B. Widrow, W. H. Pierce, J. B. Angell. “Birth, life, and death in microelectronic systems.” IRE Trans. Mil. Electron. MIL–5, 191–201 (1961). [6] L. O. Chua. “Memristor − The missing circuit element.” IEEE Trans. Circuit Theor. 18, 507–519 (1971). [7] D. B. Strukov, G. S. Snider, D. R. Stewart, R. S. Williams. “The missing memristor found.” Nature Lett. 453, 80–84 (2008). [8] A. Asamitsu, Y. Tomioka, H. Kuwahara, Y. Tokura. “Current switching of resistive states in magnetoresistive manganites.” Nature 388, 50–52 (1997). [9] H. Pagnia, N. Sotnik. “Bistable switching in electroformed metal–insulator–metal devices.” Phys. Status Solidi A 108, 11–65 (1988). [10] Y. B. Nian, J. Strozier, N. J. Wu, X. Chen, A. Ignatiev. “Evidence for an oxygen diffusion model for the electric pulse induced resistance change effect in transition-metal oxides.” Phys. Rev. Lett. 98, 146403 (2007) (2007). [11] S. Q. Liu, N. J. Wu, A. Ignatiev. “Electric-pulse-induced reversible resistance change effect in magnetoresistive films.” Appl. Phys. Lett. 76, 2749–2751 (2000). [12] A. Beck, J. G. Bednorz, C. Gerber, C. Rossel, D. Widmer. “Reproducible switching effect in thin oxide films for memory applications.” Appl. Phys. Lett. 77, 139–141 (2000). [13] V. Erokhin, T. Berzina, M. P. Fontana. “Hybrid electronic device based on polyanilinepolyethyleneoxide junction.” J. Appl. Phys. 97, 064501 (2005). [14] J. Yao, Z. Sun, L. Zhong, D. Natelson, J. M. Tour. “Resistive switches and memories from silicon oxide.” Nano Lett. 10, 4105–4110 (2010). [15] A. Mehonic, S. Cueff, M. Wojdak, S. Hudziak, O. Jambois, C. Labbé, B. Garrido, R. Rizk, A. J. Kenyon. “Resistive switching in silicon suboxide films.” J. Appl. Phys. 111, 074507 (2012). [16] J. W. Han, M. Meyyappan. “Copper oxide resistive switching memory for e-textile.” AIP Adv. 1, 032162 (2011). 120 [17] L. Alekseeva, T. Nabatame, T. Chikyow, A. Petrov. “Resistive switching characteristics in memristors with Al2O3/TiO2 and TiO2/Al2O3 bilayers.” Jap. J. Appl. Phys. 55, 08PB02 (2016). [18] Y. E. Syu, T. C. Chang, J. H. Lou, T. M. Tsai, K. C. Chang, M. J. Tsai, Y. L. Wang, M. Liu, S. M. Sze. “Atomic-level quantized reaction of HfOx memristor.” Appl. Phys. Lett. 102, 172903 (2013). [19] J. Sun, E. Lind, I. Maximov, H. Q. Xu. “Memristive and memcapacitive characteristics of a Au/Ti-HfO 2-InP/InGaAs diode.” IEEE Electron. Device Lett. 32, 131–133 (2011). [20] W. Gee Kim et al. “Dependence of the switching characteristics of Resistance Random Access Memory on the type of transition metal oxide; TiO2, ZrO2, and HfO2.” J. Electrochem. Soc. 158, H417–H422 (2011). [21] A. Prakash, D. Jana, S. Maikap. “TaOx-based resistive switching memories: Prospective and challenges.” Nanoscale Res. Lett. 8, 418 (2013). [22] X. Cao, X. Li, X. Gao, W. Yu, X. Liu, Y. Zhang, L. Chen, X. Cheng. “Forming-free colossal resistive switching effect in rare-earth-oxide Gd2 O3 films for memristor applications.” Journal of Applied Physics 106, (2009). [23] A. Younis, D. Chu, S. Li. “Oxygen level: The dominant of resistive switching characteristics in cerium oxide thin films.” J. Phys. D Appl. Phys. 45, 355101 (2012). [24] A. Kumar, Y. Rawal, M. S. Baghini. “Fabrication and characterization of the ZnO-based memristor.” International Conference on Emerging Electronics 2012, Proceedings 4–6 (2012) doi:10.1109/ICEmElec.2012.6636244. [25] F. M. Simanjuntak, D. Panda, K. H. Wei, T. Y. Tseng. “Status and prospects of ZnO-based resistive switching memory devices.” Nanoscale Res. Lett. 11, 368 (2016). [26] G. Milano, L. Boarino, C. Ricciardi. “Junction properties of single ZnO nanowires with asymmetrical Pt and Cu contacts.” Nanotechnology 30, 244001 (2019). [27] V. Jousseaume et al. “Back-end-of-line integration approaches for resistive memories.” IEEE International Interconnect Technology Conference Proceedings 41–43 (2009) doi:10.1109/IITC.2009.5090335. [28] Q. Wang et al. “Reliable Ge2Sb2Te5 based phase-change electronic synapses using carbon doping and programmed pulses.” J. Materiomics in press, (2021). [29] S. I. Oh, I. H. Im, C. Yoo, S. Y. Ryu, Y. Kim, S. Choi, T. Eom, C. S. Hwang, B. J. Choi. “Effect of electrode material on the crystallization of GeTe grown by atomic layer deposition for phase change random access memory.” Micromachines (Basel) 10, 281 (2019). [30] D. Ielmini, A. L. Lacaita. “Phase change materials in non-volatile storage.” Mater. Today 14, 600–607 (2011). [31] G. Betti Beneventi et al. “Carbon-doped GeTe: A promising material for Phase-Change Memories.” Solid-State Electron. 65–66, 197–204 (2011). 121 [32] B. S. Lee, R. M. Shelby, S. Raoux, C. T. Retter, G. W. Burr, S. N. Bogle, K. Darmawikarta, S. G. Bishop, J. R. Abelson. “Nanoscale nuclei in phase change materials: Origin of different crystallization mechanisms of Ge2Sb2Te5 and AgInSbTe.” J. Appl. Phys. 115, 063506 (2014). [33] G. Liu, Y. Chen, C. Wang, W. Zhang, R.-W. Li, L. Wang. “Polymer memristor for information storage and neuromorphic applications.” Mater. Horiz. 1, 489–506 (2014). [34] L. Yuan, S. Liu, W. Chen, F. Fan, G. Liu. “Organic memory and memristors: from mechanisms, materials to devices.” Adv. Electron. Mater. 7, 2100432 (2021). [35] T. Zhang et al. “Negative differential resistance, memory, and reconfigurable logic functions based on monolayer devices derived from Gold nanoparticles functionalized with electropolymerizable TEDOT units.” J. Phys. Chem. C 121, 10131–10139 (2017). [36] S. Goswami et al. “Robust resistive memory devices using solution-processable metalcoordinated azo aromatics.” Nature Mater. 16, 1216–1224 (2017). [37] G. Zhou, Z. Ren, L. Wang, B. Sun, S. Duan, Q. Song. “Artificial and wearable albumen protein memristor arrays with integrated memory logic gate functionality.” Mater. Horiz. 6, 1877–1882 (2019). [38] R. Kumar, R. G. Pillai, N. Pekas, Y. Wu, R. L. McCreery. “Spatially resolved Raman spectroelectrochemistry of solid-state polythiophene/viologen memory devices.” J. Am. Chem. Soc. 134, 14869–14876 (2012). [39] J. Q. Yang et al. “Fluorenone/carbazole based bipolar small molecules for non-volatile memory devices.” Org. Electron. 78, 105584 (2020). [40] A. A. Minnekhanov, A. v. Emelyanov, D. A. Lapkin, K. E. Nikiruy, B. S. Shvetsov, A. A. Nesmelov, V. v. Rylkov, V. A. Demin, V. v. Erokhin. “Parylene based memristive devices with multilevel resistive switching for neuromorphic applications.” Sci. Rep. 9, 10800 (2019). [41] A. N. Matsukatova et al. “Resistive switching kinetics and second-order effects in parylenebased memristors.” Appl. Phys. Lett. 117, (2020). [42] A. N. Matsukatova, A. A. Minnekhanov, V. v. Rylkov, V. A. Demin, A. v. Emelyanov. “Resistive switching kinetics of parylene-based memristive devices with Cu active electrodes.” J. Phys. Conf. Series 1758, 012025 (2021). [43] N. Raeis Hosseini, J. S. Lee. “Resistive switching memory based on bioinspired natural solid polymer electrolytes.” ACS Nano 9, 419–426 (2015). [44] M. K. Hota, M. K. Bera, B. Kundu, S. C. Kundu, C. K. Maiti. “A natural silk fibroin proteinbased transparent bio-memristor.” Adv. Funct. Mater. 22, 4493–4499 (2012). [45] M. K. Ram, M. Adami, M. Sartore, M. Salerno, S. Paddeu, C. Nicolini. “Comparative studies on Langmuir-Schaefer films of polyanilines.” Synth. Met. 100, 249–259 (1999). 122 [46] R. C. Sanfelice, D. T. Balogh, F. Lederle, J. Adams, S. Beuermann. “Studies of Langmuir and Langmuir-Schaefer Films of Poly(3-Hexylthiophene) and Poly(Vinylidene Fluoride).” J. Phys. Chem. B 124, 7037–7045 (2020). [47] M. Navarro, J. Benito, L. Paseta, I. Gascón, J. Coronas, C. Téllez. “Thin-Film Nanocomposite Membrane with the Minimum Amount of MOF by the Langmuir-Schaefer Technique for Nanofiltration.” ACS Appl. Mater. Interfaces 10, 1278–1287 (2018). [48] R. Y. N. Gengler, A. Veligura, A. Enotiadis, E. K. Diamanti, D. Gournis, C. Józsa, B. J. van Wees, P. Rudolf. “Large-yield preparation of high-electronic-quality graphene by a Langmuir-Schaefer approach.” Small 6, 35–39 (2010). [49] Z. U. Khan, A. Kausar, H. Ullah, A. Badshah, W. U. Khan. “A review of graphene oxide, graphene buckypaper, and polymer/graphene composites: properties and fabrication techniques.” J. Plast. Film Sheet. 32, 336–379 (2016). [50] A. v. Kazak, M. A. Marchenkova, K. S. Khorkov, D. A. Kochuev, A. v. Rogachev, I. v. Kholodkov, N. v. Usol’tseva, M. S. Savelyev, A. Y. Tolbin. “Ultrathin Langmuir–Schaefer films of slipped-cofacial J-type phthalocyanine dimer: Supramolecular organization, UV/Vis/NIR study and nonlinear absorbance of femtosecond laser radiation.” Appl. Surf. Sci. 545, 148993 (2021). [51] J. Stanley, Y. Dai, L. Boucheron, B. Lin, M. Meron, O. Shpyrko. “Novel comparison of microscopy and diffraction techniques on the structure of iron oxide nanoparticle monolayers transferred by Langmuir-Schaefer method.” Rev. Sci. Instrum. 86, 063704 (2015). [52] S. Erokhina, V. Sorokin, V. Erokhin. “Polyaniline-based organic memristive device fabricated by layer-by-layer deposition technique.” Electron. Mater. Lett. 11, 801–805 (2015). [53] A. G. Green, A. E. Woodhead. “Aniline-black and allied compounds. Part I.” J. Chem. Soc. 97, 2388–2403 (1910). [54] A. G. Green, A. E. Woodhead. “Aniline-black and allied compounds. Part II.” J. Chem. Soc. 101, 1117–1123 (1912). [55] E. T. Kang, K. G. Neoh, K. L. Tan. “Polyaniline: a polymer with many interesting intrinsic redox states.” Prog. Polym. Sci 23, 277–324 (1998). [56] J. G. Masters, Y. Sun, A. G. Macdiarmid, A. J. Epstein. “Polyaniline: allowed oxidation states.” Synth. Met. 41, 715–718 (1991). [57] J. G. Masters, J. M. Ginder, A. G. MacDiarmid, A. J. Epstein. “Thermochromism in the insulating forms of polyaniline: Role of ring-torsional conformation.” J. Chem. Phys. 96, 4768–4778 (1992). [58] T. Fukuda, H. Takezoe, K. Ishikawa, A. Fukuda, H. S. Woo, S. K. Jeong, J. Oh, J. S. Suh. “IR and Raman studies in three polyanilines with different oxidation levels.” Synth. Met. 69, 175–176 (1995). 123 [59] L. Brožová, P. Holler, J. Kovářová, J. Stejskal, M. Trchová. “The stability of polyaniline in strongly alkaline or acidic aqueous media.” Polym. Degrad. Stab. 93, 592–600 (2008). [60] G. E. Asturias, A. G. Macdiarmid, R. P. Mccall, A. J. Epstein. “The oxidation state of ‘emeraldine’ base.” Synth. Met. 29, E157–E162 (1989). [61] K. Stejskal, R. G. Gilbert. “Polyaniline. preparation of a conducting polymer (IUPAC Technical Report).” Pure Appl. Chem 74, 857–867 (2002). [62] F. Lux. “Properties of electronically conductive polyaniline: a comparison between wellknown literature data and some recent experimental findings.” Polymer (Guildf) 35, 2915– 2936 (1994). [63] J. Stejskal, P. Kratochvil, A. D. Jenkins. “The formation of polyaniline and the nature of its structures.” Polymer (Guildf) 37, 367–369 (1996). [64] J. W. Jeon, Y. Ma, J. F. Mike, L. Shao, P. B. Balbuena, J. L. Lutkenhaus. “Oxidatively stable polyaniline:polyacid electrodes for electrochemical energy storage.” Phys. Chem. Chem. Phys. 15, 9654–9662 (2013). [65] G. D’Aprano, M. Leclerc, G. Zotti. “Stabilization and characterization of pernigraniline salt: the ‘acid-doped’ form of fully oxidized polyanilines.” Macromol. 25, 2145–2150 (1992). [66] W. Lu, A. G. Fadeev, B. Qi, B. R. Mattes. “Fabricating conducting polymer electrochromic devices using ionic liquids.” J. Electrochem. Soc. 151, H33–H39 (2004). [67] F. F. C. Bazito, L. T. Silveira, R. M. Torresi, S. I. Córdoba de Torresi. “On the stabilization of conducting pernigraniline salt by the synthesis and oxidation of polyaniline in hydrophobic ionic liquids.” Phys. Chem. Chem. Phys. 10, 1457–1462 (2008). [68] A. Ray, A. F. Richter, A. G. Macdiarmid, A. J. Epstein. “Polyaniline: protonation/deprotonation of amine and imine sites.” Synth. Met. 29, E151–E156 (1989). [69] E. T. Kang, K. G. Neoh, K. L. Tan, B. T. G. Tan. “Protonation of the amine nitrogens in emeraldine-evidence from X-ray photoelectron spectroscopy.” Synth. Met. 46, 227–233 (1992). [70] P. Snauwaert, R. Lazzaroni, J. Riga, J. J. Verbist, D. Gonbeau. “A photoelectron spectroscopic study of the electrochemical processes in polyaniline.” J. Chem. Phys. 92, 2187–2193 (1990). [71] K. G. Neoh, E. T. Kang, K. L. Tan. “Protonation of leucoemeraldine in the solid state and in solution.” J. Polym. Sci. B Polym. Phys. 31, 395–401 (1993). [72] S. Stafström, J. L. Brédas, A. J. Epstein, H. S. Woo, D. B. Tanner, W. S. Huang, A. G. MacDiarmid. “Polaron lattice in highly conducting polyaniline: theoretical and optical studies.” Phys. Rev. Lett. 59, 1464–1467 (1987). [73] A. J. Epstein, J. M. Ginder, F. Zuo, R. W. Bigelow, H.-S. Woo, D. B. Tanner, A. F. Richter, W.-S. Huang, A. G. Macdiarmid. “Insulator-to-metal transition in polyaniline.” Synth. Met. 18, 303–309 (1987). 124 [74] A. J. Epstein, J. M. Ginder, F. Zuo, H.-S. Woo, D. B. Tanner, A. F. Richter, M. Angelopoulos, W.-S. Huang, A. G. Macdiarmid. “Insulator-to-metal transition in polyaniline: effect of protonation in emeraldine.” Synth. Met. 21, 63–70 (1987). [75] S. Kumar Gupta, V. Luthra, R. Singh. “Electrical transport and EPR investigations: a comparative study for d.c. conduction mechanism in monovalent and multivalent ions doped polyaniline.” Bull. Mater. Sci 35, 787–794 (2012). [76] M. B. Inoue, K. W. Nebesny, Q. Fernandoa, M. Lnoueb. “X-Ray photoelectron spectroscopy of new soluble polyaniline perchlorates: evidence for the coexistence of polarons and bipolarons.” J. Mater. Chem. 1, 213–216 (1991). [77] T. Hagiwara, T. Demura, K. Iwata. “Synthesis and properties of electrically conducting polymers from aromatic amines.” Synth. Met. 18, 317–322 (1987). [78] E. T. Kang, K. G. Neoh, S. H. Khor, K. L. Tan, B. T. G. Tan. “Structural determination of polyaniline by X-Ray photoelectron spectroscopy.” J. Chem. Soc. Chem. Commun. 695–697 (1989). [79] K. L. Tan, B. T. G. Tan, E. T. Kang, K. G. Neoh. “X-ray photoelectron spectroscopy studies of the chemical structure of polyaniline.” Phys. Rev. B 39, 8070–8073 (1989). [80] J. Yue, A. J. Epstein. “Synthesis of Self-Doped Conducting Polyaniline.” J. Am. Chem. Soc. 112, 2800–2801 (1990). [81] A. G. MacDiarmid, J. C. Chiang, A. F. Richter, A. J. Epstein. “Polyaniline: a new concept in conducting polymers.” Synth. Met. 18, 285–290 (1987). [82] W. W. Focke. “Conduction mechanisms in polyaniline.” (1987). [83] W. W. Focke, G. E. Wnek, Y. Wei. “Influence of oxidation state, pH, and counterion on the conductivity of polyaniline.” J. Phys.Chem. 91, 5813–5818 (1987). [84] G. B. Appetecchi, F. Alessandrini, M. Carewska, T. Caruso, P. P. Prosini, S. Scaccia, S. Passerini. “Investigation on lithium-polymer electrolyte batteries.” J. Power Sources 97, 790–794 (2001). [85] Z. Xue, D. He, X. Xie. “Poly(ethylene oxide)-based electrolytes for lithium-ion batteries.” J. Mater. Chem. A 3, 19218–19253 (2015). [86] S. Zhao, Q. Wu, W. Ma, L. Yang. “Polyethylene oxide-based composites as solid-state polymer electrolytes for lithium metal batteries: a mini review.” Front. Chem. 8, 640 (2020). [87] T. Berzina, V. Erokhin, M. P. Fontana. “Spectroscopic investigation of an electrochemically controlled conducting polymer-solid electrolyte junction.” J. Appl. Phys. 101, 024501 (2007). [88] T. Berzina, S. Erokhina, P. Camorani, O. Konovalov, V. Erokhin, M. P. Fontana. “Electrochemical control of the conductivity in an organic memristor: A time-resolved Xray fluorescence study of ionic drift as a function of the applied voltage.” ACS Appl. Mater. Interfaces 1, 2115–2118 (2009). 125 [89] V. Erokhin, T. Berzina, A. Smerieri, P. Camorani, S. Erokhina, M. P. Fontana. “Bio-inspired adaptive networks based on organic memristors.” Nano Commun. Netw. 1, 108–117 (2010). [90] E. M. Geniès, M. Lapkowski, J. F. Penneau. “Cyclic voltammetry of polyaniline: interpretation of the middle peak.” J. Electroanal. Chem 249, 97–107 (1988). [91] E. M. Geniès, M. Lapkowski. “Spectroelectrochemical evidence for an intermediate in the electropolymerization of aniline.” J. Electroanal. Chem 236, 189–197 (1987). [92] S. Pruneanu, E. Veress, I. Marian, L. Oniciu. “Characterization of polyaniline by cyclic voltammetry and UV-Vis absorption spectroscopy.” J. Mater. Sci. 34, 2733–2739 (1999). [93] A. Smerieri, V. Erokhin, M. P. Fontana. “Origin of current oscillations in a polymeric electrochemically controlled element.” J. Appl. Phys. 103, 094517 (2008). [94] V. A. Demin, V. v. Erokhin, P. K. Kashkarov, M. v. Kovalchuk. “Electrochemical model of the polyaniline based organic memristive device.” J. Appl. Phys. 116, 064507 (2014). [95] N. S. Sariciftci, H. Kuzmany, H. Neugebauer, A. Neckel. “Structural and electronic transitions in polyaniline: A Fourier transform infrared spectroscopic study.” J. Chem. Phys. 92, 4530–4539 (1990). [96] S. I. Zheludeva, M. v. Kovalchuk, S. Lagomarsino, N. N. Novikova, I. N. Bashelkhanov, V. v. Erokhin, L. A. Feigin. “Observation of evanescent and standing X-ray waves in region of total external reflection from molecular Langmuir-Blodgett films.” JEPT Lett. 52, 170–175 (1990). [97] S. H. Jo, T. Chang, I. Ebong, B. B. Bhadviya, P. Mazumder, W. Lu. “Nanoscale memristor device as synapse in neuromorphic systems.” Nano Lett. 10, 1297–1301 (2010). [98] L. Deng, G. Li, N. Deng, D. Wang, Z. Zhang, W. He, H. Li, J. Pei, L. Shi. “Complex learning in bio-plausible memristive networks.” Sci. Rep. 5, 1–10 (2015). [99] Z. Wang et al. “Memristors with diffusive dynamics as synaptic emulators for neuromorphic computing.” Nat. Mater. 16, 101–108 (2017). [100] S. Battistoni, V. Erokhin, S. Iannotta. “Frequency driven organic memristive devices for neuromorphic short term and long term plasticity.” Org. Electron. 65, 434–438 (2019). [101] A. Smerieri, T. Berzina, V. Erokhin, M. P. Fontana. “Polymeric electrochemical element for adaptive networks: Pulse mode.” J. Appl. Phys. 104, 1–7 (2008). [102] E. Juzekaeva, A. Nasretdinov, S. Battistoni, T. Berzina, S. Iannotta, R. Khazipov, V. Erokhin, M. Mukhtarov. “Coupling cortical neurons through electronic memristive synapse.” Adv. Mater. Technol. 4, 1800350 (2019). [103] A. v. Emelyanov et al. “First steps towards the realization of a double layer perceptron based on organic memristive devices.” AIP Advances 6, 111301 (2016). [104] S. Battistoni, V. Erokhin, S. Iannotta. “Organic memristive devices for perceptron applications.” J. Phys. D Appl. Phys. 51, 284002 (2018). 126 [105] V. Erokhin, T. Berzina, P. Camorani, A. Smerieri, D. Vavoulis, J. Feng, M. P. Fontana. “Material memristive device circuits with synaptic plasticity: learning and memory.” Bionanoscience 1, 24–30 (2011). [106] M. Ziegler, R. Soni, T. Patelczyk, M. Ignatov, T. Bartsch, P. Meuffels, H. Kohlstedt. “An electronic version of Pavlov’s Dog.” Adv. Funct. Mater. 22, 2744–2749 (2012). [107] G. B. Ermentrout, R. F. Galán, N. N. Urban. “Reliability, synchrony and noise.” Trends Neurosci. 31, 428–434 (2008). [108] J. D. Hunter, J. G. Milton, P. J. Thomas, J. D. Cowan. “Resonance effect for neural spike time reliability.” J. Neurophysiol. 80, 1427–1438 (1998). [109] K. Gorshkov, T. Berzina. “On the hysteresis loop of organic memristive device.” BioNanoSci. 1, 198–201 (2011). [110] S. Battistoni, R. Sajapin, V. Erokhin, A. Verna, M. Cocuzza, S. L. Marasso, S. Iannotta. “Effects of noise sourcing on organic memristive devices.” Chaos, Solitons Fractals 141, 110319 (2020). [111] D. A. Bernards, G. G. Malliaras. “Steady-state and transient behavior of organic electrochemical transistors.” Adv. Funct. Mater. 17, 3538–3544 (2007). [112] X. Strakosas, M. Bongo, R. M. Owens. “The organic electrochemical transistor for biological applications.” J. Appl. Polym. Sci. 132, 41735 (2015). [113] J. Rivnay, S. Inal, A. Salleo, R. M. Owens, M. Berggren, G. G. Malliaras. “Organic electrochemical transistors.” Nat. Rev. Mater. 3, 17086 (2018). [114] A. Cifarelli, T. Berzina, A. Parisini, S. Iannotta. “Memristive response and electrochemical processes in polyaniline based organic devices.” Org. Electron. 83, 105757 (2020). [115] D. Orata, D. A. Buttry. “Determination of ion populations and solvent content as functions of redox state and pH in polyaniline.” J. Am. Chem. Soc. 109, 3574–3581 (1987). [116] N. Agmon. “The Grotthuss mechanism.” Chemical Physics Letters 244, 456–462 (1995). [117] E. Ventosa, W. Schuhmann. “Scanning electrochemical microscopy of Li-ion batteries.” Phys. Chem. Chem. Phys. 17, 28441–28450 (2015). [118] H. Ha, J. Payer. “The effect of silver chloride formation on the kinetics of silver dissolution in chloride solution.” Electrochim. Acta 56, 2781–2791 (2011). [119] V. I. Birss, C. K. Smith. “The anodic behavior the formation and reduction thin silver chloride films.” Electrochim. Acta 32, 259–268 (1987). [120] S. Jaya, T. P. Rao, G. P. Rao. “Monoand multilayer formation studies of silver iodide on silver electrode from iodide-containing solutions.” J. Appl. Electrochem. 17, 997–1001 (1987). [121] B. M. Jović, V. D. Jović, D. M. Dražić. “Kinetics of chloride ion adsorption and the mechanism of AgCl layer formation on the (111), (100) and (110) faces of silver.” J. Electroanal. Chem 399, 197–206 (1995). 127 [122] P. Bro, N. Marincic. “The high rate oxidation of silver electrodes in chloride solutions.” J. Electrochem. Soc. 116, 1338 (1969). [123] C. Cachet, I. Epelboin, M. Keddam, R. Wiart. “The Ag/Ag+ system: an impedance model for nucleation and growth.” J. Electroanal. Chem 100, 745–757 (1979). [124] S. B. Aziz, T. J. Woo, M. F. Z. Kadir, H. M. Ahmed. “A conceptual review on polymer electrolytes and ion transport models.” J. Sci. Adv. Mater. Dev. 3, 1–17 (2018). [125] R. Sajapin, T. Berzina, R. Burganova, S. Iannotta. “New insight in the operation mechanism of organic memristive devices: The role of PEO-based polyelectrolyte solute ions.” Org. Electron. 94, 106173 (2021). [126] R. Burganova, A. Parisini, S. Vantaggio, R. Sajapin, T. Berzina. “The role of the polyelectrolyte composition in kinetic behaviour of organic memristive device.” Microelectron. Eng. 239–240, 111527 (2021). [127] S. H. Lee, J. C. Rasaiah. “Molecular dynamics simulation of ion mobility. 2. Alkali metal and halide ions using the SPC/E model for water at 25 °C.” J. Phys. Chem. 100, 1420–1425 (1996). [128] R. E. A. Dillon, D. F. Shriver. “Ion transport in cryptand and crown ether lithium salt complexes.” Chem. Mater. 11, 3296–3301 (1999). [129] M. L. Kaplan, E. A. Rietman, R. J. Cava, L. K. Holt, E. A. Chandross. “Crown ether enhancement of ionic conductivity in a polymer-salt system.” Solid State Ion. 25, 37–40 (1987). [130] M. Jóźwiak, L. Madej. “Complex formation of crown ethers and cations in water-organic solvent mixtures: The thermodynamic functions of complex formation of benzo-15-crown5 with Na+ in water + ethanol at 298.15 K.” J. Chem. Eng. Data 55, 1965–1970 (2010). [131] L. C. Manege, T. Takayanagi, M. Oshima, S. Motomizu. “Ion association reactions in aqueous solutions between non-UV-absorbing crown ether-alkali metal complexes with picrate ion by capillary zone electrophoresis.” Analyst 125, 1928–1932 (2000). [132] M. Rashid, A. A. Rahim, M. J. Noordin. “Electropolymerization of p-toluene sulfonic acid doped polyaniline on copper and its application as a corrosion inhibitor.” Anti-Corros. Method. Mater. 58, 131–139 (2011). [133] M. Babazadeh. “Aqueous dispersions of DBSA-doped polyaniline: one-pot preparation, characterization, and properties study.” J. Appl. Polym. Sci. 113, 3980–3984 (2009). [134] L. F. Nims. “The second dissociation constant of phosphoric acid from 20 to 50°.” J. Am. Chem. Soc. 55, 1946–1951 (1933). [135] P. H. Barry, J. W. Lynch. “Liquid junction potentials and small cell effects in patch-clamp analysis.” J. Membrane Biol. 121, 101–117 (1991). [136] U. W. Grummt, A. Pron, M. Zagorska, S. Lefrant. “Polyaniline based optical pH sensor.” Anal. Chim. Acta 357, 253–259 (1997). 134 [231] G. Tarabella, D. Vurro, S. Lai, P. D’Angelo, L. Ascari, S. Iannotta. “Aerosol jet printing of PEDOT:PSS for large area flexible electronics.” Flex. Print. Electron. 5, 014005 (2020). [232] J. Fan, A. A. Forero Pico, M. Gupta. “A functionalization study of aerosol jet printed organic electrochemical transistors (OECTs) for glucose detection.” Materials Advances 2, 7445– 7455 (2021). [233] D. Majak, J. Fan, S. Kang, M. Gupta. “Delta-9-tetrahydrocannabinol (Δ9-THC) sensing using an aerosol jet printed organic electrochemical transistor (OECT).” J. Mater. Chem. B 9, 2107–2117 (2021). [234] K. Parate et al. “Aerosol-jet-printed graphene immunosensor for label-free cytokine monitoring in serum.” ACS Appl. Mater. Interfaces 12, 8592–8603 (2020). [235] C. Fisher, B. J. Warmack, Y. Yu, L. N. Skolrood, K. Li, P. C. Joshi, T. Saito, T. Aytug. “Allaerosol-jet-printed highly sensitive and selective polyaniline-based ammonia sensors: a route toward low-cost, low-power gas detection.” J. Mater. Sci. 56, 12596–12606 (2021). [236] F. Ratuchne, M. Danczuck, E. G. de Castro. “Enhanced stability and conductivity of (polyaniline/chitosan) composites.” Orbital 10, 239–246 (2018). [237] A. Cifarelli, A. Parisini, T. Berzina, S. Iannotta. “Organic memristive element with chitosan as solid polyelectrolyte.” Microelectron. Eng. 193, 65–70 (2018). [238] A. Lesch, D. Momotenko, F. Cortés-Salazar, I. Wirth, U. M. Tefashe, F. Meiners, B. Vaske, H. H. Girault, G. Wittstock. “Fabrication of soft gold microelectrode arrays as probes for scanning electrochemical microscopy.” J. Electroanal. Chem 666, 52–61 (2012). [239] A. Cifarelli, T. Berzina, A. Parisini, V. Erokhin, S. Iannotta. “Polysaccarides-based gels and solid-state electronic devices with memresistive properties: Synergy between polyaniline electrochemistry and biology.” AIP Advances 6, 111302 (2016). [240] A. Cifarelli, A. Parisini, S. Iannotta, T. Berzina. “Organic memristive devices based on pectin as a solid polyelectrolyte.” Microelectron. Eng. 185–186, 55–60 (2018). [241] L. D. Koh et al. “Structures, mechanical properties and applications of silk fibroin materials.” Prog. Polym. Sci. 46, 86–110 (2015). [242] D. N. Rockwood, R. C. Preda, T. Yücel, X. Wang, M. L. Lovett, D. L. Kaplan. “Materials fabrication from Bombyx mori silk fibroin.” Nature Protoc. 6, 1612–1631 (2011). [243] D. H. Kim et al. “Dissolvable films of silk fibroin for ultrathin conformal bio-integrated electronics.” Nat. Materials 9, 511–517 (2010). [244] C. Vepari, D. L. Kaplan. “Silk as a biomaterial.” Prog. Polym. Sci. 32, 991–1007 (2007). [245] X. Wang, E. Wenk, A. Matsumoto, L. Meinel, C. Li, D. L. Kaplan. “Silk microspheres for encapsulation and controlled release.” J. Control. Release 117, 360–370 (2007). [246] M. A. Tomeh, R. Hadianamrei, X. Zhao. “Silk fibroin as a functional biomaterial for drug and gene delivery.” Pharmaceutics 11, 494 (2019). 135 [247] H. J. Kim, H. S. Kim, A. Matsumoto, I. J. Chin, H. J. Jin, D. L. Kaplan. “Processing windows for forming silk fibroin biomaterials into a 3D porous matrix.” Aust. J. Chem. 58, 716–720 (2005). [248] Y. Wang, H. J. Kim, G. Vunjak-Novakovic, D. L. Kaplan. “Stem cell-based tissue engineering with silk biomaterials.” Biomaterials 27, 6064–6082 (2006). [249] U.-J. Kim, J. Park, C. Li, H.-J. Jin, R. Valluzzi, D. L. Kaplan. “Structure and properties of silk hydrogels.” Biomacromol. 5, 786–792 (2004). [250] C. H. Wang, C. Y. Hsieh, J. C. Hwang. “Flexible organic thin-film transistors with silk fibroin as the gate dielectric.” Adv. Mater. 23, 1630–1634 (2011). [251] Y. Xiao, K. Kalaitzidou, D. Yao, W. H. Yeo, T. A. L. Harris. “Challenges and advances in Aerosol Jet Printing of regenerated silk fibroin solutions.” Adv. Mater. interfaces 7, 1902005 (2020). [252] J. W. Seo, H. Kim, K. H. Kim, S. Q. Choi, H. J. Lee. “Calcium-modified silk as a biocompatible and strong adhesive for epidermal electronics.” Adv. Funct. Mater. 28, 1800802 (2018).