Neuro-inspired information processing: from novel materials concepts for neuromorphic computing to local processing of biological signals

Resume : Neuromorphic systems achieve high energy efficiency by computing with spikes, in a brain-inspired way. However, finding spike-based learning algorithms that can be implemented within the local constraints of neuromorphic systems, while achieving high accuracy, remains a formidable challenge. Equilibrium Propagation is a hardware-friendly counterpart of backpropagation which only involves spatially local computations and applies to recurrent neural networks with static inputs. So far, hardware-oriented studies of Equilibrium Propagation focused on rate-based networks. In this work, we develop a spiking neural network algorithm called EqSpike, compatible with neuromorphic systems, which learns by Equilibrium Propagation. Through simulations, we obtain a test recognition accuracy of 97.6% on MNIST, similar to rate-based Equilibrium Propagation, and comparing favourably to alternative learning techniques for spiking neural networks. We show that EqSpike implemented in silicon neuromorphic technology could reduce the energy consumption of inference by up to three orders of magnitude and training by up to two orders of magnitude compared to GPUs. Finally, we also show that during learning, EqSpike weight updates exhibit a form of Spike Timing Dependent Plasticity, highlighting a possible connection with biology.

Resume : Novel computational paradigms in combination with proper hardware solutions are required to overcome the limitations of our state-of-the-art computer technology, in particular regarding energy consumption. Due to the inherent complex and non-linear nature, spintronics offers the possibility of energy-efficient, non-volatile hardware solutions for various unconventional computing schemes. [1-3] In this talk, I will address the potential of topologically stabilized magnetic whirls ? so-called skyrmions - for reservoir computing, a computational scheme that allows to drastically simplify spatial-temporal recognition tasks. We have shown that random skyrmion fabrics provide a suitable physical implementation of the reservoir [4,5] and allow to classify patterns via their complex resistance responses either by tracing the signal over time or by a single spatially resolved measurement. [6] In any type of hardware-based computation, finally, some sort of readout of the system is needed. While often a significant effort is made in enhancing the resolution of an experimental technique to obtain further insight into the sample and its physical properties, advantageous data analysis has the potential to provide a deeper insight into given data set. This is particularly relevant when the signal is close to the resolution limit, i.e., where the noise becomes at least of the same order as the signal. [7] [1] J. Grollier, D. Querlioz, K.Y. Camsari, KES, S. Fukami, M.D. Stiles, Nat. Elect. 3, 360 (2020) [2] E. Vedmedenko, R. Kawakami, D. Sheka, ..., KES, et al., J. of Phys. D, 53, 453001 (2020) [3] G. Finocchio, M. Di Ventra, K.Y. Camsari, KES, et al., JMMM 521, 167506 (2021) [4] D. Prychynenko, M. Sitte, et al, KES, Phys. Rev. Appl. 9, 014034 (2018) [5] G. Bourianoff, D. Pinna, M. Sitte and KES, AIP Adv. 8, 055602 (2018) [6] D. Pinna, G. Bourianoff and KES, Phys. Rev. Appl. 14, 054020 (2020) [7] I. Horenko, D. Rodrigues, T. O?Kane and KES, arXiv:1907.04601

Resume : Spin-Torque Nano-Oscillators (STNOs) can act both like microwave emitters (oscillators) and microwave rectifiers (diodes). These oscillators are very promising as they are compatible with CMOS technology and we showed previously that it was possible to leverage their non-linear behavior to emulate neurons and be used for Reservoir Computing. In this work, we show that STNOs acting as diodes can be used as trainable synaptic connections, and that we can assemble a Deep Neural Network (DNN) based on microwave detection and emission. In Artificial Neural Networks, synapses are used to do Multiply-And-Accumulate (MAC) operations, which corresponds to weighted sums. In order to demonstrate a Multiply-And-Accumulate operation (MAC) with STNOs, we use microwave emitters of different frequency and send their microwave signals into a chain of spin-diodes wired in series. This MAC operation leverages the frequency-multiplexing of the input and the frequency-selectivity of the spin-diodes: each diode rectifies its matching input signal. First, we prove that the voltage of each diode is proportional to the power emitted by one of the oscillators (Multiply), and that the total rectified voltage is the sum of the voltage of each diode (Accumulate). We show that each synaptic weight can be chosen by tuning the resonance frequency of each diode. These experiments prove the connectivity between a layer of artificial neurons and a layer of artificial synapses using similar devices at the nanoscale. Then, we simulate a layer of these synapses emulated by spin-diodes in order to solve the dataset “digits” (8x8 pixels handwritten digits) in order to prove that we can train a network of these spin-diodes. Even considering the non-linear behaviors of the spin-diodes, our network achieves recognition with an accuracy as good as software neural networks. Finally, we simulate these spin-diodes to solve a larger dataset in order to discuss the scalability of this MAC operation to very large chains of spin-diodes. This novel concept of DNN relying on microwave signals allows us to route the information between neuron and synapses with a simplified spatial architecture, thus achieving a high density of connections. It provides a fast, compact and low power solution to process microwave encoded information with Artificial Intelligence methods. Such implementation could reduce the size and the dependency of embedded Radio-Frequency systems. This work was supported by the European Research Council ERC under Grant bioSPINspired 682955, the French ANR project SPIN-IA (ANR-18-ASTR-0015) and the French Ministry of Defense (DGA).

Resume : Neuromorphic circuits are designed to process information via spike-based computation and plasticity, similarly to the human brain. To develop neuromorphic systems with high energy efficiency and high synaptic connectivity, novel devices must be developed that can closely mimic the cognitive processes of the brain. One example is the volatile resistive switching memory (RRAM) based on voltage-induced Ag diffusion, which can mimic the short-term memory and short-term plasticity in the biological neural network. In this talk I will summarize the device physics and applications of Ag-based RRAM, covering the device structure, the atomistic modeling and the neuromorphic circuits that can leverage volatility for efficient reservoir computing and motion detection.

Resume : The importance of bio-inspired computing arrays for the development of programmable non-Von-Neumann computer architectures that avoid the separation of computing and memory structures has been demonstrated in recent investigations. Since the introduction by Chua and Yang in 1988, Cellular Neural Networks (CNN) have become a paradigm for complexity. These bio-inspired networks, which are characterized by local couplings of nonlinear dynamic systems of comparatively low complexity, are regarded as the basic structure in a CNN Universal Machine (CNN-UM) architecture, which is a complete dynamic array stored program computer, very often realized on a single chip with optical sensors in CMOS technology. CNN-UM high-speed computing systems are programmed by CNN templates as instructions e.g. in image processing applications or in multidimensional medical signal processing. Since CMOS implementations of programmable cellular sensor-processor array architectures cannot exploit the full potential of this paradigm and do not exceed a resolution of 176X144 cells, M-CNN structures have been proposed that can be implemented with an increased resolution as compared to the pure CMOS counterparts. M-CNN are proposed in this contribution as a bio-inspired paradigm for universal mem-computing in future sensor-processor systems. A detailed introduction to the theory of memristors and M-CNN is given in the presentation. Especially, newly developed methods [3] for programming such structures are presented and discussed in detail. [1] R. Tetzlaff, A. Ascoli, I. Messaris, and L.O. Chua, ?Theoretical Foundations of Memristor Cellular Nonlinear Networks: Memcomputing with Bistable-like Memristors,? IEEE Trans. on Circuits and Systems?I: Regular Papers, 2019, 10.1109/TCSI.2019.2940909 [2] A. Ascoli, R. Tetzlaff, I. Messaris, and L.O. Chua, ?Theoretical foundations of Memristor Cellular Nonlinear Networks: Stability Analysis with Dynamic Memristors,? IEEE Trans. Circuits and Systems?I: Regular Papers, 2020, Volume:67, issue: 4 [3] A. Ascoli, R. Tetzlaff, S.M ?Steve? Kang, and L.O. Chua, ?Theoretical foundations of Memristor Cellular Nonlinear Networks: A DRM2-based Method to Design Memcomputers with Dynamic Memristors, IEEE Trans. Circuits and Systems?I: Regular Papers, 2020, Volume:67, issue: 8

Resume : Neuromorphic engineering is poised to revolutionise information technologies by developing electronic devices that emulate neuro-biological architectures at the hardware level. An example of neuromorphic device is the “artificial synapse”, which can be represented by non-volatile variable resistance memory elements or memristors. Another example is smart imaging, where light sensors are designed to mimic the spatio-temporal nature of human vision, not only by turning light into electrical signals but also by capturing and sending the useful-only information to the processing unit in an extremely efficient manner. The commonly employed device structure of optoelectronic and memristive devices is a vertically aligned configuration, where one or more layers of the active material(s) are “sandwiched” between the metal electrodes. To achieve fast response speed in light-sensing ability or resistive switching of the memristor, one has to perform extreme downscale of the active layer thickness. Alternatively, coplanar nanogap electrode architectures may be employed, which offer many advantages, such as lower power consumption, faster operating speed, greater sensitivity and higher-level of integration. Herein, we show high speed light-sensing and memristive devices based on coplanar nanogap (<15 nm) separated electrodes, fabricated with a high throughput, inexpensive nanopatterning technique, compatible with flexible substrates, named adhesion lithography. We demonstrate memristors based on metal oxide, organic and hybrid perovskite materials, discuss the mechanism dominating their operation and showcase their ability to be integrated in arrays in order to perform learning/training operations and other synaptic functionalities.

Resume : Abstract: As CMOS scaling is approaching atomic boundaries, the end of Moore's era [1] seems inevitable. It is for this reason that engineers worldwide are investigating new strategies [2] to keep the integrated circuit (IC) performance growth rate, predicted by Moore, in the years to come, without the need to shrink the transistor size any further. The search for novel devices, capable to combine within a single physical nanoscale volume multiple functionalities, is one of the approaches of greater interest. In this respect, the memory resistor, memristor for short [3], offers extraordinary opportunities for the electronics of the future. Given the inherently-nonlinear mechanisms at the origin of the dynamical behaviors of memristors [4]-[5], the applicability of linear system-theoretic methodologies for the investigation and design of circuits based upon them is rather limited [6]. In this presentation we show how the generalization of a powerful nonlinear system-theoretic technique, known as Dynamic Route Map (DRM) [3] and instrumental to analyze first-order mathematical descriptions, such as those governing the dynamics of memristors with scalar state, to ordinary differential equations with two degrees of freedom [7] allows to investigate [8] bio-inspired memcomputing memristor circuits. Furthermore, shaping the generalized DRM of a second-order memristive circuit, through the application of a comprehensive stability analysis to its mathematical description, allows to guide the circuit solutions from prescribed initial conditions toward desired attractors, allowing the implementation of a predefined memory or computing task [9]. This study reveals the significant role, which nonlinear system theory is expected to play in the years to come, for the development of a systematic approach to memristor circuit design. Bibliography: [1] G.E. Moore, "Cramming more components onto integrated circuits," Electronics, vol. 38, no. 8, pp. 114-117, 1965 [2] R.S. Williams, "What's next? [The end of Moore's law]," IEEE Computing in Science & Engineering, vol. 19, no. 2, pp. 7-13, 2017, DOI: 10.1109/MCSE.2017.31 [3] L.O. Chua, "Five Non-Volatile Memristor Enigmas Solved," Applied Physics A, vol. 124, no. 8, 563(43pp.), 2018 [4] A. Ascoli, S. Slesazeck, H. Mähne, R. Tetzlaff, and T. Mikolajick, "Nonlinear dynamics of a locally-active memristor," IEEE Trans. Circuits and Systems--I (TCAS--I): Regular Papers, vol. 62, no. 4, pp. 1165-1174, 2015 [5] A. Ascoli, R. Tetzlaff, L.O. Chua, J.P. Strachan, and R.S. Williams, "History Erase Effect in a Non-Volatile Memristor," IEEE Trans. on Circuits and Systems-I (TCAS-I): Regular Papers, vol. 63, no. 3, pp. 389-400, 2016 [6] A. Ascoli, R. Tetzlaff, and M. Biey, "Memristor and Memristor Circuit Modelling based on Methods of Nonlinear System Theory," Springer Lecture Notes on Nonlinear Dynamics in Computational Neuroscience, F. Corinto, and A. Torcini eds., pp. 99-132, 2018, DOI: https://doi.org/10.1007/978-3-319-71048-8 7 [7] R. Tetzlaff, A. Ascoli, I. Messaris, and L.O. Chua, "Theoretical Foundations of Memristor Cellular Nonlinear Networks: Memcomputing with Bistable-like Memristors," IEEE Trans. on Circuits and Systems-I: Regular Papers (TCAS-I), 2019, DOI: 10.1109/TCSI.2019.2940909 [8] A. Ascoli, I.Messaris, R. Tetzlaff, and L.O. Chua, "Theoretical Foundations of Memristor Cellular Nonlinear Networks: Stability Analysis with Dynamic Memristors," IEEE Trans. on Circuits and Systems-I: Regular Papers (TCAS-I), 2019, DOI: 10.1109/TCSI.2019.2957813 [9] A. Ascoli, R. Tetzlaff, S. Kang, and L.O. Chua, "Theoretical Foundations of Memristor Cellular Nonlinear Networks: a DRM2-based method to design memcomputers with dynamic memristors," IEEE Trans. on Circuits and Systems-I: Regular Papers (TCAS-I), 2020

Resume : Biological sensory nervous systems can sense and process external stimuli. In this regard, artificial sensory nervous systems which mimic detecting and information processing functions of biological counterparts are able to process detected stimulation and thus track stimulation history such as intensity, duration, and frequency of the stimuli. Thus, the artificial systems can replicate complicated functions of biological counterparts and are applicable for neuromorphic computing, soft robots, and neural prostheses. Here, various types of artificial nervous systems are presented using organic electronics to emulate biological sensory nervous systems. Firstly, pressure sensors (artificial mechanoreceptors), organic ring oscillators (artificial nerve fibers), and synaptic transistors were integrated to emulate the functions and operating principles of biological counterparts [1]. Secondly, by connecting an organic photodetector (an artificial photoreceptor), a stretchable artificial synapse, and a polymer actuator (an artificial muscle), a stretchable photo-sensitive artificial sensorimotor nervous system was developed [2]. Furthermore, a retina-inspired photo-sensitive synaptic transistor was developed by using ultraviolet-responsive 2-dimensional carbon nitride nanodot layers as a floating gate to selectively detect and process ultraviolet light exposure information [3]. Lastly, for an artificial auditory system, organic artificial synapse was integrated with a triboelectric nanogenerator [4]. References [1] Y. Kim, A. Chotors, W. Xu, Y. Liu, J.Y. Oh, D. Son, J. Kang, A. M. Foudeh, C. Zhu, Y. Lee, S. Niu, J. Liu, R. Pfattner, Z. Bao, and T.-W. Lee, A bioinspired flexible organic artificial afferent nerve, Science, 360, 998?1003 (2018). [2] Y. Lee, J. Y. Oh, W. Xu, O. Kim, T. R. Kim, J. Kang, Y. Kim, D. Son, J. B.-H. Tok, M. J. Park, Z. Bao, and T.-W. Lee, Stretchable organic optoelectronic sensorimotor synapse, Sci. Adv., 4, eaat7387, (2018). [3] H.-L. Park, H. Kim, D. Lim, H. Zhou, Y.-H. Kim, Y. Lee, S. Park, and T.-W. Lee, Retina-Inspired carbon nitride-based photonic synapses for selective detection of ultraviolet light, Adv. Mater., 1906899 (2020). [4] D.-G. Seo, Y. Lee, G.-T. Go, M. Pei, S. Jung, Y. H, Jeong, W. Lee, H.-L. Park, S.-W. Kim, H. Yang, C. Yang, and T.-W. Lee, Versatile neuromorphic electronics by modulating synaptic decay of organic synaptic transistors; from artificial neural networks to neuro-prosthetics, Nano Energy, 65, 104035, (2019).

Resume : Sensory inputs are critical for making intelligent decisions by humans. The five senses, including our sense our touch via skin, provide high-fidelity information about the state of our environment and convey them rapidly to the respective cortexes in the human brain via our nervous system. In artificial systems, such as robots or prosthetics, the ability to asynchronously process and transmit sensory information can help to provide similarly fast, real-time feedback in order to make control decisions. Moreover, asynchronously networked sensors can provide greater robustness to damage if each sensor node operates independently from one another. In my talk, I will discuss our recent work in developing neuromorphic electronic skins inspired by the somatosensory system(1). Such neuromorphic skins can enable greater scalability without sacrificing transmission speed. In addition, they can take advantage of machine learning algorithms to rapidly determine surface features for faster robotic controls. Recently demonstrated computer hardware such as neuromorphic chips can also be used to process the neuromorphic signals and lower the power requirements for processing large amounts of tactile information simultaneously. References: 1. Lee, W. W. et al. A neuro-inspired artificial peripheral nervous system for scalable electronic skins. Sci. Robot. 4, eaax2198 (2019). 2. Yang, W., Hon, M., Yao, H. & Tee, B. C. K. An Atlas for Large-Area Electronic Skins: From Materials to Systems Design. (2020). doi:DOI: 10.1017/9781108782395 3. Yao, H. et al. Near-hysteresis-free soft tactile electronic skins for wearables and reliable machine learning. Proc. Natl. Acad. Sci. U. S. A. (2020). doi:10.1073/pnas.2010989117

Resume : Spiking neural networks (SNNs) are a type of artificial neural network that encodes information in a frequency domain, in the form of neural spike trains. These networks typically use Integrate-and-Fire (I&F) spiking neurons, hence they mimic biological neural networks more closely than the artificial neural networks based on neurons with graded response. Hardware SNN implementations are almost exclusive implemented using CMOS technology. While such circuits are very fast, they are also hard and rigid. In contrast, biological neural networks are slow and soft. This creates many levels of difficulty for implantable brain-machine interfaces. Organic electronics are new types of electronics based on organic and polymeric materials. Compared with silicon electronics, they are much slower, but softer and often biocompatible. They can be used to implement artificial neurons with compatible speeds and mechanical form factor, making for much better implementation of implantable brain-machine interfaces (BMI). Here, we update on our efforts to fabricate physically flexible and biocompatible spiking artificial neurons. We fabricated Axon-Hillock (AH) I&F neurons using flexible and biocompatible organic electronics. The circuit employs both n- and p-type organic transistors, based on N,N?-bis(n-octyl)-x:y,dicyanoperylene3,4:9,10-bis(dicarboximide) (PDI8-CN2) and dinaphtho[2,3b:2?,3?-f]thieno[3,2-b]thiophene (DNTT) as n- and p-type organic semiconductors, along with resistors and capacitors. We demonstrate AH circuit producing series of spikes with frequency proportional to magnitude of input current, akin to the functioning of biological neurons. We show the frequency-current transfer function demonstrating neural dynamics with upper and lower spiking boundaries. The static and dynamic power dissipations are also evaluated. Finally we demonstrate a basic current summing capability of the neuron. We believe that this is the first demonstration of a functioning of a flexible and biocompatible organic electronic integrate-and-fire spiking neuron, and shows a path towards long-term implantable SNNs for future BMI.

Resume : Organic electrochemical transistors (OECTs) have recently attracted tremendous attention from the scientific community due to their unique set of features. Especially, the lack of an insulating layer that separates gate from channel, like in conventional field effect transistors, and their feature to support electrolyte gating open new pathways in the field of bioelectronics. In particular, they render the device ideal for measurements in the aqueous environments that dominate the biological world. Additionally, they provide extra possibilities in terms of the device architecture as the gating electrode can be patterned on the same plane as the conducting channel. In a set up like that, what is of extreme interest is the effect of the gate position with respect to the channel. In this work, the spatiotemporal response of a poly(3,4-ethylenedioxythiophene) polystyrene sulfonate (PEDOT:PSS) OECT is examined and its device physics is studied. This work opens new routes for exploiting the physics of multi-gate OECT devices in bioelectronics, sensing and neuromorphic electronics.

Resume : 2D networks of molecularly functionalized nanoparticles (NPs) (hereafter called NMN : nanoparticle molecule network) have emerged as an interesting approach in molecular electronics to understand fundamental electron transport mechanisms, as well as to develop potential applications in electronics, sensing and computing circuits.(1) We study the NMN approach with functional molecules self-assembled in high density 2D networks with topographic structures that are intrinsically similar to the structure of a "reservoir computing". We use specific molecules which can change their electronic properties upon a given excitation (optical) to study multifunctional systems at the nanoscale (< 100 nm). These approaches, without direct analogs in semiconductor nanoelectronics, would open new perspectives to molecular electronics in unconventional computing. Here, we report and discuss several key features of these NMNs (2-4) to assess their possible use for reservoir computing: highly non-linear electron transport, variability, complex/rich dynamics such as harmonic and interharmonic generations, intermodulation distortion, co-tunneling, noise and plasmonic response. We discuss these properties based on NMN made of several suitable molecules (photo-switches (2-4), ion and proton-switches (5,6), polyoxometalates (7,8),...). (1) Liao, J. et al., Chem. Soc. Rev. 2015, 44, 999-1014. (2) Viero, Y. et al., J. Phys. Chem. C 2015, 119, 21173-21183. (3) Viero, Y. et al., Adv. Funct. Mater. 2018, 28, 1801506. (4) Stievenard, D. et al., Nanoscale 2018, 10, 23122-23130. (5) Tran, T. K. et al., Adv. Mater. 2013, 25, 427-431. (6) Audi, H. et al., Nanoscale 2020, 22, 133001. (7) Laurans, M et al., Nanoscale 2018, 10, 17156-17165. (8) Dalla Francesca, K. et al., Nanoscale 2019, 11, 1863-1878.

Resume : The employment of artificial neural networks (ANN), inspired by the biological nervous system, enables signal processes to elaborate solutions for specific issues. The high performance of such ANNs is achieved through the dynamic change of the synaptic weights by applying learning algorithms for self-optimization. Despite of the simple operations for each single elements in an ANN, a network with a huge amount of simulated elements consumes lots of computing capacity using the von Neumann architecture. To overcome this issue, neuromorphic devices facilitate the designing of hardware ANNs, which emulates the synaptic functions analogously. Here we demonstrate the viability of such device using photonic/plasmonic devices in combination with the photochromic molecules. In one example we will show photonic waveguides in combination with the photochromic diarylethene (DAE) molecules, in which positioning and irradiating DAEs on single waveguides we modulate the intensity of the guided light to emulate the plasticity of the synaptic weights and basic AND- or OR logic gate operations by using specific training set. In a second example we demonstrate the capability to emulate synaptic functionalities by exploiting surface plasmon polaritons modulation. Such plasmonic devices exhibits the fundamental functions of a synapse, such as potentiation, depression, and long-term plasticity. This work was supported by the Deutsche Forschungsgemeinschaft through the CRC 951.

Resume : Photochromic molecules provide an intriguing and relatively untapped alternative to traditional materials utilized in organic memory devices. We have recently reported on a new prototype of a nonvolatile organic memory (OMEM) that integrates a layer of dithienylethene photochromes (DTE) into a solution-processed, multilayer device [1-3]. The DTE molecules undergo a change in both their UV-visible absorption and energy level position due to a photo- and/or electrically-induced ring-opening/-closing reaction. Exploiting the difference in HOMO and LUMO energies of both isomers and the subsequent change in hole-injection barrier we use this DTE layer to control the hole injection and transport within our OMEM layer stack. Optimized devices have displayed ON/OFF ratios in both current and electroluminescence approaching 106. We investigate both, optical and electrical programming of 12x12 passive matrix OMEM devices and show that precise control of the ratio of both isomers in the active layer enables access to a multitude of intermediate states, demonstrating the potential of these devices for future multi-level memory applications. We report a dynamic range of ca. 200.000, thus, storage of up to 16 bits per pixel seem feasible. These devices are candidates for neuromorphic networks and computing. [1] P. Zacharias, M. C. Gather, A. Köhnen, N. Rehmann, K. Meerholz, Angew. Chem. Int Ed. 48, 4038 (2009). [2] R.C. Shallcross, P.O. Körner, E. Maibach, A. Köhnen, K. Meerholz, Adv. Mat. 34, 4807 (2013). [3] Körner, P. O., Shallcross, R. C., Maibach, E., Köhnen, A. & Meerholz, K. Org. Electronics 15, 3688, (2014).

Resume : As an alternative to conventional crossbar arrays of memristive devices realized with a top-down approach, bio-inspired nanoarchitectures based on self-organized nanowires (NWs) have attracted great attention for implementing unconventional types of computing [1]. In this work, we report on the synaptic behavior of memristive Ag NW networks realized with a bottom-up approach. The emergent behavior of the network is shown to be related to two different physical phenomena: i) rupture/rewiring of single NWs (wiring plasticity) and ii) resistive switching behavior of single NW cross-point junctions (weight plasticity). These effects result in a structural plasticity of the system that reconfigures when subjected to external electrical stimulation. The network measured in two terminal configuration under voltage pulse stimulation exhibited typical features of short-term plasticity such as Paired-Pulse facilitation (PPF). More importantly, the functional connectivity of the system is shown to be responsible for an intrinsic heterosynaptic behavior where direct stimulation of a synaptic pathway is responsible also for modulation of the synaptic weight of non-stimulated synapses, as modelled and experimentally demonstrated in multiterminal configuration [2]. References [1] Milano, Gianluca, et al. "Recent Developments and Perspectives for Memristive Devices Based on Metal Oxide Nanowires." Advanced Electronic Materials (2019): 1800909. [2] Milano, Gianluca, et al. "Brain‐Inspired Structural Plasticity through Reweighting and Rewiring in Multi‐Terminal Self‐Organizing Memristive Nanowire Networks." Advanced Intelligent Systems (2020): 2000096.

Resume : One of the central endeavours in the fields of bioelectronics and neuromorphic computing is the development of logic elements that can transduce and process ionic and electronic signals. Motivated by this challenge, we report fully monolithic, nanoscale logic elements based upon n- and p-type nanowires as electronic channels, and proton-gated by electron-beam patterned Nafion [1]. In particular, we demonstrate inverter circuits with state-of-the-art performance for ion-electron systems with DC gain exceeding 5, and frequency response up to 2 kHz for sine wave signals and hundreds of Hz for square wave signals – far surpassing the 1-200 Hz range requirement for neural sensing applications. A key innovation facilitating the logic integration is a new electron-beam process for patterning Nafion with linewidths down to 125 nm. This delivers feature sizes compatible with low voltage, fast switching elements thereby expanding the scope for Nafion as a versatile patternable high-proton-conductivity element for bioelectronics, neuromorphic computing or other applications where nanoengineered protonic membranes and electrodes are required. [1] J.G. Gluschke et al., Materials Horizons doi: 10.1039/d0mh01070g (2020)

Resume : The realization of bioinspired circuits that mimic the signal processing capabilities of the brain demands the reproduction of both short- and long-term aspects of synaptic plasticity on a single device level. The emergence of organic electronics has made available a host of new materials and devices with properties that are well-suited to address this challenge. We show how mixed conductivity in organic semiconductors can be leveraged to control neuromorphic functions similar to the basic short-term and long-term synaptic plasticity. Namely, we demonstrate that the timescales for functions such as depression, adaptation and dynamic filtering can be controlled by tuning the ratio of electronic/ionic mobility. We further show that 3D printing can be leveraged to develop new architectures for neuromorphic computing.

Resume : Various manufacturing strategies have been employed to realize organic electronics, where photolithographic techniques, printing and coating are some of the prime ones. As organic electronics is applied to, and also configured to mimic several of the key signal processing features of neuro-systems, news strategies that goes beyond traditional 2D-manufacturing protocols are desired. Here, novel materials that enable in vivo-manufacturing protocols in 3D are reported along with electrode and device functionalities to define a seamless interface between neuronal systems and electronics. The goal of this effort is to define an amalgamation of pharmacological and electronic functionalities in attempt to derive new neuro-technology devices for medical applications and for engineering.

Resume : Artificial intelligence applications have demonstrated their enormous potential for complex processing over the last decade. However, they are mainly based on digital operating principles while being part of an analogue world. Moreover, they still lack the efficiency and computing capacity of biological systems. Neuromorphic electronics emulate the analogue information processing of biological nervous systems (1). Organic artificial synapses exhibit volatile as well as tunable/non-volatile conductance states that replicate the behavior of their biological counterparts (2-6). In this work we present a path planning robot that uses a locally trained organic neuromorphic circuit to navigate through a maze. The neuromorphic circuit responds and adapts to environmental stimuli directly, as it is integrated with the robot sensors. The fusion of sensor signals with low-power organic neuromorphic electronics paves the way toward stand-alone, brain-inspired computing circuitry in autonomous and intelligent robotics (7). References 1. C. Mead, Neuromorphic electronic systems. ‎Proc. IEEE 78, 1629-1636 (1990). 2. P. Gkoupidenis et al. Synaptic plasticity functions in an organic electrochemical transistor. Appl. Phys. Lett. 107 (26), 263302 (2015). 3. P. Gkoupidenis et al. Neuropmorphic functions in PEDOT:PSS organic electrochemical transistors, Adv. Mater. 27 (44), 7176 (2015). 4. Y. van de Burgt et al. A non-volatile organic electrochemical device as a low-voltage artificial synapse for neuromorphic computing. Nat. Mater. 16, 414 (2017). 5. E. J. Fuller et al. Parallel programming of an ionic floating-gate memory array for scalable neuromorphic computing, Science 364 (6440), 570 (2019). 6. A. Melianas et al. Temperature-resilient solid-state organic artificial synapses for neuromorphic computing, Sci. Adv. 6, (27) eabb2958 (2020). 7. C. Wan et al. Artificial Sensory Memory. Adv. Mater. 1902434, 1–22 (2019).

Resume : Artificial intelligence (AI) has been heralded as the flagbearer of the fourth industrial revolution. But it comes with a cost and that is computing power. It is projected that by 2040, we will need more computing energy than the total energy we can produce now. So, we need devices that can offer higher computing/ storage density with low energy consumption like neuronal computation. We are addressing these challenges using a molecular-electronic route. Historically, organic electronic devices have stimulated scientific excitements in OLEDs but are yet to make any other significant technological impact. The reasons behind their limited success are their poor robustness, stability, endurance and most importantly, the lack of mechanistic understanding that restricts the emergence of approaches to solve these problems. We have overcome each of these difficulties in our memristors based on transition metal complexes of azo-aromatic ligands that exhibit high reproducibility (~350 devices), fast switching (?30 ns), excellent endurance (~1012 cycles), stability (>106 s) and scalability (down to ~60nm2)1-5. Using in-situ Raman spectroscopy we can track the electronic changes in molecules in-operando at every point of our voltage sweep providing a clear picture of our molecular mechanism that enables us to do different molecular and device engineering to achieve targeted functionalities. Using devices of this genre we are addressing the existing computing challenges via three routes, (i) By designing devices with ultra-low power: We can design memristors with switching voltage as low as 70mV, with energy ~36aJ/ 60nm2. The current and voltage levels of these devices meet the requirements specified in ITRS road map. (ii) By designing memristors and memcapacitors with multiple discrete plateaus3: We have developed memristors with 3- 4 distinct conducting plateaus which also shows mem-capacitance. Their concomitant occurrence is enabled by symmetry breaking of our film-molecules driven by voltage, a new paradigm in condensed matter physics. (iii) Brain inspired computing: Using devices that exhibit concomitant memristive and memcapacitive functions we can simulate biological actions such as neuronal action potential and even cardiac myocyte pulsing. References 1. Goswami S, Matula AJ, Rath SP, Hedström S, Saha S, Annamalai M, et al. Robust resistive memory devices using solution-processable metal-coordinated azo aromatics. Nature Materials 2017, 16(12): 1216. 2. Valov I, Kozicki M. Non-volatile memories: Organic memristors come of age. Nature Materials 2017, 16(12): 1170. 3. Sreetosh Goswami, Santi P. Rath, Damien Thompson, Svante Hedström, Meenakshi Annamalai, Rajib Pramanick, B. Robert Ilic, Soumya Sarkar, Christian A. Nijhuis, Jens Martin, Sreebrata Goswami, and T. Venkatesan, A Ternary Resistive Memory Device Based on Charge Disproportionate Molecular Redox, Nature Nanotechnology, (2020) 4. Sreetosh Goswami, Sreebrata Goswami and T. Venkatesan, An Organic approach to low energy memory and Brain Inspired Electronics, Applied Physics Reviews- Invited Review 7, 021303 (2020) 5. Sreetosh Goswami, Debalina Deb, Agnès Tempez, Marc Chaigneau, Manohar Lal, R. Stanley Williams, Sreebrata Goswami and T. Venkatesan, Nanometer Scale Uniform Molecular Switching in Organic Memristors, Advanced Materials, 32 (42), 2070318

Resume : Neuromorphic systems must have at least 5 unavoidable features that are present in living beings. First, neuromorphic systems must perform memorizing and processing functions, using same elements. Second, it must allow data acquisition from sensors with preliminary processing and recording. Third, it must allow non-equilibrium processes, such as oscillator behavior at fixed values of input stimuli. Fourth, it must mimic some features of nervous systems. Fifth, it must permit possibility of coupling with living beings. In this presentation these features are considered with a special attention to how memristive devices can be implemented for reaching the goal. Comparison of characteristic properties of organic and inorganic memristive devices are discussed.

Resume : Sensory information processing in robot skins currently rely on a centralized approach where signal transduction (on the body) is separated from centralized computation and decision-making, requiring the transfer of large amounts of data from periphery to central processors, at the cost of wiring, latency, fault tolerance and robustness. We envision a decentralized approach where intelligence is embedded in the sensing nodes, using a unique neuromorphic methodology to extract relevant information in robotic skins. Here we specifically address pain perception and the association of nociception with tactile perception to trigger the escape reflex in a sensorized robotic arm. The proposed system comprises self-healable materials and memtransistors as enabling technologies for the implementation of neuromorphic nociceptors, spiking local associative learning and communication. Configuring memtransistors as gated-threshold and -memristive switches, the demonstrated system features in-memory edge computing with minimal hardware circuitry and wiring, and enhanced fault tolerance and robustness.

Resume : Associative learning, a critical learning principle to improve an individual’s adaptability, has been emulated by few organic electrochemical devices. However, complicated bias schemes, high write voltages, as well as process irreversibility hinder the further development of associative learning circuits. Here, we present a non-volatile organic electrochemical transistor (OECT) based on a poly(3,4-ethylenedioxythiophene):tosylate (PEDOT:Tos)/ Polytetrahydrofuran (PTHF) composite. This device can continuously and reversibly change its conductance state at a write bias less than 0.8 V and the state retention time can be longer than 200 min without decoupling the write and read operation. By incorporating a pressure sensor and a photoresistor into the gate terminal of volatile and non-volatile OECTs, a neuromorphic circuit is demonstrated with the ability to associate two physical inputs (light and pressure) instead of normally demonstrated electrical inputs, which may have implications for biomimetic devices like electronics-skin and neuroprosthetics. To unravel the non-volatility of this material, UV-Vis-NIR spectroscopy, X-ray photoelectron spectroscopy (XPS) and grazing-incidence wide-angle X-ray scattering (GIWAXS) are used to characterize the oxidation level variation, compositional change and the structural modulation of the PEDOT:Tos/PTHF films in various conductance states. Non-crystalline PTHF is believed to be capable of trapping cations which is the origin of the memory behavior of OECT devices based on such composites. The implementation of the associative learning circuit as well as the understanding of the non-volatile material represent critical advances for organic electrochemical devices in neuromorphic applications.

Resume : Neuroprostheses are neuroengineering devices that have an interface with the nervous system and supplement or substitute functionality in disabled people. In the collective imagination, neuroprotheses are primarily used to restore sensory (e.g. acoustic prostheses) or motor capabilities (e.g. artificial limbs), but in the recent years new devices to be applied directly at the brain level are taking place. To realize energy-efficient and real-time processing devices, closed-loop neuromorphic systems are envisaged as the core of next-generation neuroprosthetics for brain repair. In this talk, I will present the first exploitation of a real-time hardware neuromorphic prosthesis to restore bidirectional interactions between two neuronal populations. We developed an in vitro experimental model of neuronal sub-populations to mimic the mutual interaction between neuronal assemblies and performed a focal lesion to functionally disconnect them. Then, we employed our neuroprosthesis for two potential clinical applications: bidirectional bridging to artificially reconnect two disconnected populations and hybrid bidirectional bridging to replace the activity of one population with a real-time neuromorphic Spiking Neural Network. Further examples of closed-loop neuroprostheses for promoting neuroplasticity in vivo are introduced and discussed. Closed-loop neuroprosthetics based on neuromorphic computation will form the base of novel bioelectrical therapeutics for healthcare.

Resume : The interface between biological cells and non-biological materials has profound influences on cellular activities, chronic tissue responses, and ultimately the success of medical implants and bioelectronic devices. The optimal coupling between cells, i.e. neurons, and materials is mainly based on surface interaction, electrical communication and sensing. In the last years, many efforts have been devoted to the engineering of materials to recapitulate both the environment ( i.e. dimensionality, curvature, dinamicity) and the functionalities (i.e. long and short term plasticity) of the neuronal tissue to ensure a better integration of the bioelectronic platform and cells. On the one hand, here we explore how the transition from planar to pseudo-3D nanopatterned inorganic and organic materials have introduced a new strategy of integrating bioelectronic platforms with biological cells under static and dynamic conditions. Although a spontaneous penetration does not occur, adhesion processes are such that a very intimate contact can be achieved. On the other hand, we investigate how organic semiconductors can be exploited for recapitulating electrical neuronal functions such as long term and short term potentiation. In this way, both the topology and the material functionalities can be exploited for achieving in vitro biohybrid platforms for neuronal network interfacing.

Resume : Responsive modulation of neural networks is increasingly being used to treat patients with auditory-neurological disorders and neuropsychiatric diseases. Yet, current technology burdens neurostimulation tools with bulky, non-biocompatible electrical components that require rigid encapsulation for long-term implantation in body. Recently, we created a novel transistor architecture (internal ion-gated organic electrochemical transistors; IGT) that can be an efficient building block for integrated bioelectronics. These transistors include all the key features required for safe, efficient, and prolonged use of transistors in biological environments: i) they are made out of biocompatible and stable materials; ii) they are soft and conformable; iii) they show high speed and amplification to detect potentially low-amplitude ionic signals of the body; iv) they can perform certain computations. Here, I am presenting the vision of our newly founded lab towards designing, based on that emerging technology, and developing novel fully implantable, contained and responsive neural interface devices that will allow long-term acquisition and closed-loop manipulation of neural circuits with high spatiotemporal resolution over extended period of time to reveal neural dynamics in the auditory-neurological pathway.

Resume : There is no shortage of examples showing how neuromorphic systems can be applied to process biological signals or to interface biological tissue [1, 2, 3]. Usually, in such contexts, the neuromorphic system is utilized for anomaly detection. The automation of the long-term monitoring of biological signals holds promise for lightening the burden placed on clinicians. At the same time, the adoption of such devices allows processing to be performed in situ, without the need to transfer the data to an external processor. In turn, on-site signal analysis renders closed-loop intervention feasible, to correct the source of the anomalies. So far, the common approach has been to implement the network of spiking neurons either on multi-core neuromorphic platforms [1, 2] or on programmable units (FPGA) [3]. However, if the aim is to develop wearable or even chronically implantable devices, it is imperative to move in the direction of embedded solutions, tailored and optimized for the specific application. To this end, we propose a neuromorphic device implemented in CMOS technology for the detection of epileptic seizures from Local Field Potential (LFP) signals. The LFP data have been acquired by a Multi-Electrode Array (MEA) in a slice of mouse hippocampus-EC (entorhinal cortex). The system includes a spike encoder converting the recorded signals into trains of spikes, and a small network of 2x1 neurons with online biological-plausible learning. The encoder yields two spike-trains: UP spikes that account for the positive slope signal and DOWN spikes that correspond to the negative slope signal. The synapses among the input and output layers are plastic and follow the STDP rule. Early results show a significant increase in the spiking rate of the postsynaptic neuron during seizure-like activity. [1] Bauer, Felix Christian, Dylan Richard Muir, and Giacomo Indiveri. "Real-time ultra-low power ECG anomaly detection using an event-driven neuromorphic processor." IEEE Transactions on Biomedical Circuits and Systems 13.6 (2019): 1575-1582. [2] Donati, Elisa, et al. "Discrimination of EMG signals using a neuromorphic implementation of a spiking neural network." IEEE transactions on biomedical circuits and systems 13.5 (2019): 795-803. [3] Joucla, Sébastien, et al. "Generation of locomotor-like activity in the isolated rat spinal cord using intraspinal electrical microstimulation driven by a digital neuromorphic CPG." Frontiers in neuroscience 10 (2016): 67.

Resume : Monitoring of bioelectric signals in peripheral sympathetic nerves of small animal models is crucial to gain understanding of how the autonomic nervous system controls specific body functions in health and disease. Advances in the understanding of autonomic nerves function depend on the availability of low-invasive electrodes for the chronic recording of nerve activity in conscious animals. Here we report a highly stretchable low-impedance electrode realized by microcracked gold films as metallic conductors covered with electrodeposited, stretchable conducting polymer composite to facilitate ion-to-electron exchange. The conducting polymer composite based on poly(3,4-ethylenedioxythiophene) polystyrene sulfonate (PEDOT:PSS) obtains its adhesive, low-impedance properties by controlling thickness, plasticizer content and deposition conditions. Atomic Force Microscopy measurements under strain show that the optimized conducting polymer coating is compliant with the micro-crack mechanics of the underlying Au-layer, necessary to absorb the tensile deformation when the electrodes are stretched. In conclusion, we demonstrate the applicability of the proposed stretchable electrodes by performing in vivo high quality recordings of renal sympathetic nerve activity in conscious rats.

Resume : Most of today?s strategies to interface biology with electronic hardware are based on layered architectures where the front-end of sensing is optimized separately from the back-end for processing/computing signals. Alternatively, biological systems are capitalizing on distributed architecture where both sensing and computing are mix together and co-optimized. In this talk, we will present our strategy to implement bio-sensing of electroactive cells in a neuromorphic perspective. We will present how organic electrochemical transistors can be used to record electrical signals from neural cells. We will show various strategies capitalizing on the versatility of organic materials synthesis and organic device fabrication to tune and adapt the functionalities of such bio-sensors. We will then present how these strategies can be efficiently used to realize computing functions directly at the interface with biology. Notably, we will illustrate how a network of ionic sensors can implement the reservoir computing concept, a powerful neuromorphic computing approach of particular interest for dynamical signal processing.

Resume : Organic electrochemical transistors (OECTs) are in a stage of rapid development as novel applications that make use of these versatile devices continue to emerge. OECTs are characterized by the coupling of both ionic and electronic inputs to modulate transistor channel conductance. This attribute renders OECTs ideal for interfacing electronics with biological systems, which make use of ionic and biochemical currents and gradients for signaling. Here we will present a new concept of OECT that evolves with use. The transistor channel can be formed, modulated, and obliterated under operation. The strength of the transistor response to a given stimulus can be modulated within a range that spans several orders of magnitude, introducing behaviors analogous to neuroplasticity into electronic systems. This evolvable transistor can be incorporated into a simple circuit that mimics classical conditioning. Also, we will present several biomimetic logic circuits that link sensory inputs to the growth of the transistor channel, thus to achieve higher order processes like self-regulation and coincidence detection. OECTs that physically and electronically evolve under operation will bring about a new paradigm of machine learning based on evolvable organic electronics.

Resume : Memristors are the most suitable components for energy-efficient neurohardware applications as they can combine information processing with memory storage in a single component, as neurons do. A large variety of memristive materials are available, but practically all research is focused on the use of extended inorganic materials. Despite the success of this type of materials, they are inherently limited by a low chemical variability and functionalization, by difficulties in nanostructuration, by poor reproducibility because of large parameter dependences and by limited cycling endurance. Molecular materials are potential memristors which excel in some of these aspects. First, they allow using the versatility of molecular chemistry to design the materials. Second, they permit to reduce the size and the energetic requirements. And third, they improve the processability by means of low-cost solution fabrication processes (as surface deposition for example). However, so far no clear design roadmap for molecule-based memristive materials exists and very few examples have been reported. My work in this area intends to develop a theoretical methodology for the design of molecular memristive systems. From the material analysis, my team aims to obtain reliable criteria to procure new proposals for Neuromorphic applications. The driving idea is to build a multiscale approach where each theoretical step is built upon the results obtained in the previous step, along with successive increase of the size of the problem. From a material science perspective, understanding the mechanisms that govern memristive behavior in Molecular Materials is mandatory to design and build robust and energy efficient molecular memristors.

Resume : Electronic synapses constitute an auspicious building block for neuromorphic computing systems & bioinspired electronic applications. They are of low-cost & power consumption, while significantly reducing device’s structure complexity. Their minimal nature, enabling the fabrication through solution-processed based techniques, such as inkjet printing. Inkjet-printing is an environmentally friendly method with minimum material waste. Also, inkjet-printing technology facilitates fast prototyping and enables large area & flexible electronics fabrication. Herein, fully inkjet-printed electronic synapses (Ag/a-TiO2/Ag, 80-200 nm thicknesses, active area ~ 80 μm x100 μm) based on a custom TiO2 nanoparticle suspension solution/ink are developed. The devices are characterized primarily through voltage pulses, due to agreement with the stimulation in biological synapses, and matching the operation process of realistic devices. The synthesized ink characterized in term of its fluidic properties. The results of the printed elements present biomimetic synaptic dynamics, and the developed ink shows optimal jetting & printing characteristics, with shelf life over 5 months.

Resume : Recently, research on wearable intelligent electronics that detects, remembers, and learns external stimuli in real-time is receiving great interest. And a tactile synaptic device studies have emerged because they have the potential for human-interactive neuromorphic applications capable of large-scale parallel processing. However, the artificial tactile synaptic devices reported so far have complex physical connections between the sensor unit and the memory unit, which are inevitably not suitable for the wearable and patchable device due to the complex and costly manufacturing steps. Here, we demonstrate an artificial organic tactile synaptic device based on the integrated single device that enables the sensing, storing, and learning of a variety of tactile information. The synaptic device is able to be programmed with various tactile input pressures, by using a ferroelectric field-effect transistor structure with a pressure-sensitive ball-shaped top gate electrode. The synaptic device reliably and stably operates with high tactile reception sensitivity of 88 KPa-1 under bending conditions. And it was confirmed that synaptic plasticity was stably implemented for 10,000 inputs by various electrical/tactile stimuli, which allows for precise and robust tactile perception learning. Furthermore, we demonstrated that an integrated 4 x 4 tactile synaptic array allows for 2-dimensional tactile learning and proof-of-concept recognition simulations for diverse handwriting patterns with an outstanding error tolerance. As a result, our study proposes the novel platform for a single, integrated tactile neuromorphic system, which can simultaneously sense and learn a variety of external information.

Resume : Synaptic transistors have recently been proposed as promising devices for brain inspired computing, allowing efficient synapse simulation by realizing learning and signal transmission functions simultaneously. Several dielectric families, such as electrolytes, proton conductors, and ferroelectric materials have been employed as insulating layers in synaptic transistors. Traditionally, these dielectrics are obtained via vacuum deposition methods. Printing technology offers low-cost fabrication of electronic devices and circuits by eliminating the need for lithographic patterning of the deposited layers. Up to now, printed synaptic transistors with electrolyte dielectrics have been reported with operation frequencies limited by 1kHz due to the ionic nature of electrolytes. We report synaptic transistors employing ink-jet printed aluminum oxide dielectrics for the first time. Paired pulse facilitation, excitatory and inhibitory post-synaptic signals at the output of the devices are successfully demonstrated up to at least 50kHz improving the synaptic operation frequency of printed transistors by nearly two orders of magnitude. Origin of the synaptic behavior is shown to be the positive charge trapping in the insulator, which is caused by hydroxide rich and porous nature of the printed AlOx layers. Finally, an acoustic system employing synaptic transistors responding to sound signals at frequencies ranging from few Hz to several kHz is demonstrated.

Resume : Around thirty percent of epileptic patient do not respond to medication. In drug-resistant partial epilepsies, resective surgery is the treatment of choice to suppress seizures, provided that the epileptogenic zone (EZ) is clearly identified and safely removed. In this context, the capacity to rely on objective biomarkers of the EZ is fundamental to define the optimal surgical approach in the specific context of each patient. One of the most important biomarkers are electrophysiological ones which represent local field potential variations generated by a network of neurons. Such variations can be captured by small-scale electrodes implanted in the brain tissue. However, the smaller the electrodes are the higher their impedance will be. One of the strategies to overcome this problem, is to improve the impedance of the electrodes with conductive and biocompatible polymers such PEDOT:PSS. In this work, microwire ultra-flexible electrodes were electrodeposited with PEDOT:PSS and characterized in terms of impedance, coating structure and stability. The equivalent circuit of electrode-electrolyte interface were modeled as well to have better undertesting of coating properties. This process enables a controlled tuning of the electrode surface impedance which is essential in the brain – electrode interfacing. Finally, these electrodes are tested in vivo to record High Frequency Oscillations (HFOs), which are transient brief interictal signals with a frequency band of 250-600Hz, from the epileptic rat hippocampus. PEDOT:PSS coated electrodes are also compared with classical ones to demonstrate their higher performance in HFOs detection.

Resume : Memristive devices have many potential applications, including low-power computer memories and neuromorphic computing. A mechanistic understanding of the fundamental processes underlying the switching behaviour in memristive devices is necessary to guide device design and optimisation, however there is a big gap between our macroscopic understanding of these phenomena, as obtained experimentally, and their microscopic interpretation. In this work, we aim at filling this gap with a focus on electrochemical metallization memories with a Ag/ZnO electrolyte/Pt architecture. More specifically, we study from first principles the desorption process of the Ag metal ions from the anode and their subsequent adsorption onto the semiconducting electrolyte. We developed a model for the interface between the Ag anode and ZnO electrolyte and studied the energetics of Ag spill over to the electrolyte as a function of the adsorption site, the orientation of the metal surface and external factors, such as an applied bias. We also evaluated the energy pathway for migration of the metal atoms on the electrolyte surface. The presented results are the basis for a new theoretical model of the mechanisms underpinning memristive properties.

Resume : New ways of processing information are being developed as the reduction of the scale of the technology reaches the prediction of Moore’s law. In the last years, advances in materials engineering based on adaptive heterostructures has found a new paradigm to optimize drawbacks in signal processing. This permit us to enabling new operating modes where an electronic structure can be designed with different properties and incorporation of internal states that are capable of reconfiguration. The heterostructure reaction to an external stimulus (like a change in frequency) will self-adjust the parameters of those properties to carry out certain electronic operations. In this work, a study of Mn thin films (100 nm) deposited by sputtering is presented. The films are oxidized at different temperatures (250, 350 and 450 °C) and Raman spectroscopy as well as X-Ray Diffraction (XRD) is done to examine the phases of oxidation obtained. An experimental bandgap is calculated by the Kubelka-Munk relation for each sample in order to understand the transport phenomena of the MnxOy layer. After the initial deposition of Mn oxides, a second layer of ZnO:Zn is deposited by Co-Sputtering on top. An electrical characterization of the final structure is performed, and the transport phenomena at the interface are explained. The investigation of MnO/ZnO:Zn Heterostructure will allow us to describe how the internal states of the films can be of use for adaptive electronics.

Resume : Resistive random access memory (ReRAM) is a promising device for a new generation of flash memory. These devices can also be utilized in neuromorphic systems. Formation and rupture of conductive filaments in a dielectric layer of metal-insulator-metal (MIM)-structures is the most common and well-known mechanism for resistive switching (RS). However, filamentary RS has some major problems, such as the requirement of a forming process and device to device instability, which can be crucial for cross-bar integration. Recent studies have shown that there are possible ways to eliminate a forming process. One of them is the development of devices exhibiting stable filamentary RS without any forming step (forming-free devices). Another way is to design devices with non-filamentary (or homogenous) RS mechanism. In this work, we show that the oxygen-deficient switching layer (ex. TaOx or WOx) in combination with appropriate electrodes could be used for the realization of both filamentary and non-filamentary resistive switching. However, addition of a rectifying layer with high work function (ex. HfO2) to a switching layer is required for the realization of a second approach. It is worth noting that non-filamentary RS devices can also demonstrate a “self-rectifying” behavior, which is a useful way to fight sneak current issues in 3D structures.

Resume : Metal halide perovskite materials are attracting significant attention for use as the active layer in memristive switch architectures due to their solution processability, high electronic mobilities and low switching energies. The ease of ion and defect migration through these materials that gives rise to hysteresis in perovskite solar cells is thought to underly the change of resistive state seen in other two terminal sandwich device architectures. However, the details of the underlying switching mechanism in these perovskite memristors is not well established. The migration of silver ions to or from electrodes to form or dissolve conductive silver halide filaments through the methyl ammonium lead iodide (MAPI) films has been proposed as the mechanism that switches a device between a high resistive state and a low resistive state [1]. In this study we examine the factors that determine whether memristive switching is observed in a device. We find that devices prepared from MAPI can be classified with two general types of behaviour, either that of a memristor or a ‘memdiode’. Previously identified, memristive behaviour shows a transition between two clearly different resistive states on application of a sufficient forward or reverse bias [2]. We observed this when devices were prepared using reactive electrode materials such as Ag. We find that this behaviour occurs once a minimum threshold voltage has been applied to the device enabling it to transition into a resistive switching regime with different ohmic states. In this mode, the devices show a high cycling endurance (after 50 cycles approximately 7% change of the resistance in high resistivity state and low resistivity state) and state retention with on/off ratios of approximately (10^4). A study of the effect of electrode contact area (from 0.002 mm^2 to 3.14 mm^2) and the active layer film thickness shows currents in the ‘on’ state appear approximately independent of electrode size but inversely proportional to film thickness, consistent with the formation/dissolution of single bridging filaments to explain switching. When devices with reactive electrode metals were operated below a threshold voltage of approximately 2 V memdiode like behaviour is observed, similar to devices with more inert electrode materials such as aluminium operated over a wider voltage range. Memdiodes display reversible (or switchable) rectifying behaviour which is more volatile than the memristive behaviour with much greater overall resistance. We hypothesise that the mechanism for the change in memdiode resistance in these devices can be explained by the diode junctions formed at the interfaces. The differential resistance of these junctions varies as the electrostatic barrier of the interfaces change in response to the accumulation of mobile ionic defects in the perovskite layer [3]. In addition to memristor behaviour we suggest that the memdiode mode of operation may have many interesting potential applications in neuromorphic computing. 1) Yoo, Eunji, Miaoqiang Lyu, Jung-Ho Yun, Chijung Kang, Youngjin Choi, and Lianzhou Wang. "Bifunctional resistive switching behavior in an organolead halide perovskite based Ag/CH 3 NH 3 PbI 3− x Cl x/FTO structure." Journal of Materials Chemistry C 4, no. 33 (2016): 7824-7830. 2) Zhao, Xiaoning, Haiyang Xu, Zhongqiang Wang, Ya Lin, and Yichun Liu. "Memristors with organic‐inorganic halide perovskites." InfoMat 1, no. 2 (2019): 183-210. 3) Xiao, Zhengguo, Yongbo Yuan, Yuchuan Shao, Qi Wang, Qingfeng Dong, Cheng Bi, Pankaj Sharma, Alexei Gruverman, and Jinsong Huang. "Giant switchable photovoltaic effect in organometal trihalide perovskite devices." Nature materials 14, no. 2 (2015): 193-198.

Resume : Shallow feed-forward networks are incapable of addressing complex tasks such as natural language processing that require learning of temporal signals. To address these requirements, we need deep neuromorphic architectures with recurrent connections such as deep recurrent neural networks. However, the training of such networks demand very high precision of weights, excellent conductance linearity and low write-noise- not satisfied by current memristive implementations. Inspired from optogenetics, here we report a neuromorphic computing platform comprised of photo-excitable neuristors capable of in-memory computations across 980 addressable states with a high signal-to-noise ratio of 77. The large linear dynamic range, low write noise and selective excitability allows high fidelity opto-electronic transfer of weights with a two-shot write scheme, while electrical in-memory inference provides energy efficiency. This method enables implementing a memristive deep recurrent neural network with twelve trainable layers with more than a million parameters to recognize spoken commands with >90% accuracy.

Resume : Memristive devices are attracting a great deal of attention for memory, logic, neural networks, and sensing applications due to their simple structure, high density integration, low-power consumption, and fast operation. In particular, multi-terminal structures controlled by active gates, able to process and manipulate information in parallel, would certainly provide novel concepts for neuromorphic systems. Hence, transistor-based synaptic devices may be designed, where the synaptic weight in the postsynaptic membrane is encoded in a source-drain channel and modified by presynaptic terminals (gates). In this work, we show the potential of reversible field-induced metal-insulator transition (MIT) in strongly correlated cuprates for the design of robust and adjustable multi-terminal memristive transistor-like devices. We have studied different structures patterned on YBa2Cu3O7-? films, which are able to display gate modulated non-volatile volume MIT, driven by field-induced oxygen diffusion within the system. The key advantage of these materials is the possibility to homogeneously tune the oxygen diffusion not only in a confined filament or interface, as observed in widely explored binary and complex oxides, but also in the whole material volume. We show several device configurations in which the lateral conduction in a drain-source channel is effectively controlled by active gate-tunable volume resistance changes, thus emulating the synaptic weight. Large design flexibility can be obtained by changing the switching performance of different gates, thus offering the possibility to locally adjust the conductance response as required to implement neuromorphic functionalities.

Resume : Memristor, which simply consists of a switching layer inserted between two electrodes, is one of the most strong candidates to become a device-platform for imitating the principal of biological neural network due to its nonlinear and dynamic electrical characteristics depending on the history of applied electrical programming [1-3]. In this study, we fabricated a nanorod structured SiOx memristor using E-beam evaporator with glancing angle deposition at the wafer-scale and utilized the device as an artificial neuron for probabilistic computing applications. The device can exhibit a low forming voltage (< 2 V), a high ON-OFF ratio (> 105), reliable switching performances, and fast switching time (~40 ns), where the switching event is attributed to the transition between two Si phases (amorphous Si and Si nanocrystal). Notably, the nanorod structured SiOx can lead to the considerable reduction of forming voltage and enhancement of stochastic switching characteristics, when compared with the typical SiOx memristor. Moreover, using voltage pulse trains, the SiOx nanorod memristor with different glancing angles has successfully mimicked fundamental neuronal dynamics called integrate-and-fire processes and stochastic functionalities for the bayesian network in which each node is probabilistic variables. Then, as a proof of concept, we simulated the probabilistic inference for the correlation between three biological genes. Taken all together, the designed SiOx memristor neuron could pave the way for stochastic artificial neurons and its based probabilistic computing technology. References: [1] S. Choi, J.-W. Choi, J. C. Kim, H. Y. Jeong, J. Shin, S. Jang, S. Ham, N.-D. Kim, and G. Wang, Nano Energy, Under review (2021) [2] S. Choi, J. Yang, G. Wang, Adv. Mater., 32, 2004659 (2020) [2] S. Choi, S. Jang, J.-H. Moon, J. C. Kim, H. Y. Jeong, P. Jang, K.-J. Lee, and G. Wang. NPG Asia Mater. 10, 1097–1106 (2018)

Resume : After more than 40 years of continuous evolution, our computing systems are reaching their limits. Indeed, the architecture of Von-Neumann, on which our computers are based, physically dissociates the hearts of calculations from the memory. The sequential processing of information is thus confronted with a bottleneck, more commonly known as "Memory Bottleneck". A solution is to draw inspiration from the natural mathematical paradigms of the human brain, in which the data are massively parallelly processed with high energy efficiency, realizing the hardware implementation of neuromorphic networks. This approach opens the possibility to bring the information storage sites (synapses) closer to the treatment sites (neurons). The discovery of memristor, theorized in 1971 by L. Chua, has led to the development of novel artificial neuromorphic concepts and devices, including ferroelectric-based ones. Ferroelectric Tunnel Junction (FTJ) type memristors based on zirconium-doped hafnium oxide, Hf0.5 Zr0.5 O2 (HZO) have recently displayed synaptic learning capabilities [1]. In addition, HZO processes are already fully compatible with silicon CMOS industry with oxide layers thinner than 10 nm. In the present work, the HZO layer is realized by room temperature magnetron sputtering of a Hf0.5 Zr0.5 O2 ceramic target and subsequently crystallized by rapid thermal annealing [2]. Using different bottom electrode layers (germanium, titanium nitride, platinum) grown on silicon and different substrates (n-doped silicon, n-doped germanium), we studied the effect on the stabilized crystalline phase and microstructure, band structure alignment and electrical properties of thin HZO films. Furthermore, we explored the effect of ultra-thin buffer layers between the electrodes and the HZO layer, trading on the material, the position and the thickness. We exploited X-ray photoemission spectroscopy to analyze the chemistry and the electronic state of the electrodes/HZO interface. X-ray reflectometry and grazing incidence X-ray diffraction were used to probe the thickness and structural characteristics of the HZO layer, whose ferroelectricity is associated to the polar orthorhombic phase. We will discuss our results in the framework of structural, chemical and physical properties of the different electrode/ferroelectric interfaces and their effect on the electrical properties of thin HZO-based junctions. References: [1] L. Chen et al., “Ultra-low power Hf0.5Zr0.5O2 based ferroelectric tunnel junction synapses for hardware neural network applications,” Nanoscale, vol. 10, no. 33, pp. 15826–15833, 2018 [2] J. Bouaziz, et al., “Huge Reduction of the Wake-Up Effect in Ferroelectric HZO Thin Films,” ACS Appl. Electron. Mater., vol. 1, no. 9, pp. 1740–1745, 2019

Resume : The development of novel brain-inspired neuromorphic computing architectures is anticipated to play a key role in addressing the strict requirements of the artificial intelligence (AI) era. In order to obtain a high degree of learning accuracy within an artificial neural network (ANN) that operates with the backpropagation algorithm, a highly symmetric synaptic weight distribution is desired. Along these lines, we present here a novel device engineering approach that enables analog synaptic properties in completely forming free SiO2-conductive bridge memories (CBRAM). This is achieved, by either incorporating a dense layer of Pt nanoparticles (NPs) as a bottom electrode (BE) or by fabricating bilayer structures using a second switching layer of VOx. Interestingly, compared with the reference sample (Sample B) that manifests both threshold and bipolar switching modes, the Pt NCs sample (Sample A) exhibits only the threshold switching pattern whereas the bilayer configuration (Sample C) operates only under the bipolar switching mode as illustrated by direct current (DC) measurements. These characteristics have a direct, while different impact, on the conductance modulation pattern and determine the analog nature of the synaptic weight distribution. While the Pt NPS incorporation yielded quite steep transition slopes in the hysteresis patterns (~1 mV/dev (A)), which is important for selector applications, the bilayer configuration displayed quite good linearity responses during the potentiation/depression processes. The incorporation of the layer of VOx can modulate the CF growth mechanism and reduce the total power consumption by limiting the diffusion of the silver cations. The transition from the high resistance state (HRS) to the low resistance state (LRS) takes place for all Samples at about VSET ~ 180 – 200 mV, while for Sample B & C the reverse transition occurs at VRESET ~ 100 mV. The switching ratio for Pt-NCs sample that operates under the threshold switching mode is about ~106 while a slightly smaller memory window is recorded for the other two samples (~105). Moreover, the hysteresis patterns exhibit no significant degradation after the application of 300 consecutive DC cycles. Valuable insights regarding the origin of these effects and in particular of the symmetric and linear conductance modulation processes are gained through the implementation of a self-consistent numerical model that takes into account both the impact of the electrodes’ thermal conductivity on the switching pattern as well as the different diffusion barriers for silver ion migration. Our approach provides useful guidelines towards the realization of high yield ANNs with biological-like dynamic behavior, by controlling the conducting filament (CF) growth mechanism. Further research could be carried out in order to assess the diffusion limiting properties of other material configurations in order to enhance even more the switching linearity.

Resume : The association of sensory stimuli in the environment forms the basis of logic and learning. Classical conditioning is a key benchmark of associative learning capabilities. While classical conditioning had been well emulated in artificial synaptic devices, the inhibition of conditioning is yet to be explored. Latent inhibition of conditioning is an adaptation for noise reduction by suppressing learning of non-salient cues and only processing novel cues in the environment. The challenge of inducing such inhibitory behaviour in memory devices can be resolved with a hybrid device with two distinct temporal characteristics. Here, we first emulate classical conditioning with an associatively coupled optoelectronic synapse – photomemristor. Optical stimuli, inducing persistent memory, act as biologically salient unconditioned stimulus (US). Electrical stimuli, with volatile kinetics, act as neutral conditioned stimulus (CS) which could subsequently be paired with the US in a conditioning procedure. Interestingly, the forgetting rate of the photomemristor device can be tuned quantitatively with the intensity of the electrical stimulus. Similarly to priming effects observed in memristors, electrical priming stimulus prior to optical training stimulus can tune the decay of the persistent photoconductivity. As a result, we were able to demonstrate an inhibitory behaviour in the optical memory by a pre-exposure of electrical stimuli, for the first time. Such adaptable synapses are crucial building blocks in neuromorphic circuits capable of learning and performing in-memory processing.

Resume : Memristive devices are of great interest owing to a number of their attractive properties, in particular, plasticity (the multilevel character of resistive switching), which allows them to emulate synapses in hardware artificial neural networks (ANNs). The use of local learning rules for such ANNs, for example, bio-like spike-timing-dependent plasticity (STDP), has firmly established itself in recent years. At the same time, there is a growing interest in wearable and biocompatible computing systems that are safe for the human body. We have previously shown that such safe spiking ANNs can be created using memristors based on the FDA-approved parylene (PPX) polymer [1]. Besides the fact that STDP-based learning was shown to be well working in various memristors, in biological systems, the basic STDP can be modified in the presence of neuromodulators (e.g., dopamine) [2]. This effect is believed to be essential for important biological functions such as reinforcement learning. Recently we have shown that the memristive STDP window shape for (CoFeB)x(LiNbO3)1-x memristors with anionic type of switching could be simply modulated by introducing coefficients that control the neuron spike amplitudes [3]. These coefficients could be considered as dopamine concentration indicators, and the dopamine-like modulation would serve as a reward in learning. Obviously, such simple and effective emulation of the presence of dopamine during the learning process of two-terminal memristive devices would significantly advance the development of hardware ANNs, e.g. in direction of the reinforcement learning algorithms realization. It would be especially useful to understand whether such modulation can be used in memristors with cationic type of switching, based on biocompatible materials, such as PPX. The goal of this work was to shed light on this question. We have studied memristors of Metal(Cu, Ag)/PPX/ITO structure. It was found that, in addition to stable memristive characteristics suitable for ANNs, these devices can also change their conductance by means of bio-like STDP rules, including dopamine-like modulated STDP. The amplitude coefficients from -1 (inhibitory mode) to 1 (excitatory mode) of pre- and post-spikes, reflecting the “dopamine” concentration, in various combinations allow observing the STDP window not only of the usual shape, but also of the anti-STDP, bell and anti-bell shapes, as well as intermediate ones. The use of such modulated STDP in a simulated model of reinforcement learning of a simple cartpole-type ANN is also discussed. The obtained results demonstrate that the development of memristors based on PPX provides prospects for hardware realization of bio-inspired operation ANNs with reinforcement learning ability. The study was supported by the Russian Science Foundation (Project No. 20-79-10185). [1] Minnekhanov et al., Sci. Rep. 9, 10800 (2019). [2] Gurney et al., PLoS Biol. 13, e1002034 (2015). [3] Nikiruy et al., AIP Adv. 9, 065116 (2019).

Resume : Brain-inspired neuromorphic computing (NC), which mimics the functionalities of our human brain, is vital in building a low-power artificial intelligence system. Hardware implementation of such as artificial synaptic devices is an essential path towards this goal. Electrolyte gating has been widely used in neuromorphic devices because of massive carrier injections at a very low gate voltage and higher ionic mobility. Much progress has been made in three-terminal devices based on oxides, organic materials, and 2D materials. However, studies based on metallic ferromagnetic systems remain highly unexplored. Additionally, a synaptic device based on magnetic metal has a spin degree of freedom that can be tuned. In this work, we have demonstrated a three-terminal coplanar synaptic transistor based on permalloy (Ni80Fe20) by electrolyte gating. A giant anticlockwise loop opening in the transfer curve shows a non-volatile and reversible change of channel conductance suitable for emulating synaptic functions. Multilevel states are a criterion for achieving analog computing. We have realized multilevel, non-volatile, reversible conducting states by applying gate pulses of different amplitudes. This permalloy based artificial synapse imitates the biological synapse where the gate pulses act as the external stimuli, the ions as the neurotransmitters, and the channel conductance act as the synaptic weight. We have emulated the essential synaptic functions at room temperature. Important forms of short-term plasticity like excitatory postsynaptic conductance (EPSC), inhibitory postsynaptic conductance (IPSC), and paired-pulse facilitation (PPF) have been emulated successfully. By varying the gate pulse amplitude, duration, and number, we have shown a transition from short-term memory (STM) to long-term memory (LTM). The modulation of channel conductance and retention is more prominent at higher amplitude, duration, and number due to the intercalation/extraction of more oxygen ions into the channel. Important forms of long-term plasticity like long-term potentiation (LTP), long-term depression (LTP), and multiple potentiation and depression cycles have been demonstrated. The energy consumed for reading and writing per synaptic event is calculated at different gate amplitude and duration, showing a linear escalation. The lowest writing energy consumed in our device is 0.6 nJ and can be further reduced by reducing the channel dimension and the pulse width. Along with the emulation of synaptic functions, the device showed dynamic filtering behavior and can perform as a high-pass filter when worked in positive gate polarity. These results provide an insight into the potential usage of a ferromagnetic metal-based synaptic transistor for a large-scale, low power exhausting, silicon friendly neuromorphic computing system.

Resume : On the horizon of a new era in computing, the need for better, faster, and higher density data storage and processing is increasing more than ever. Many candidates have been proposed, amongst which resistive random-access memory (RRAM) is predicted to outperform competitors in the long run. Not only is RRAM scalable to the nm length scale, but it is also characterized by low-power operation, fast switching, and has the potential to establish multilevel data storage and to display synaptic behaviour through exploiting gradual switching dynamics. Oxygen-engineered yttrium oxide-based RRAM is of particular interest to both non-volatile memory and neuromorphic applications. Via controlling the oxide stoichiometry, we are able to tailor the switching dynamics from two-level/digital to multilevel/analog characteristics.[1] An enhanced accessibility of stable intermediate resistive states in the set process that also exhibit non-linear conductance quantization is achieved through decreasing the oxygen content of the yttria layer. These states can be used to tune the device resistance in a quasi-analog manner, which has great significance for multi-state and synaptic operations. A physics-based theoretical framework is used to describe charge transport in the tunneling and ballistic regimes and to explain the observed quantum effects. [1] Petzold and Piros et al., Adv. Electron. Mater. 6, 2000339 (2020)

Resume : Memristors are circuit elements that can change their conductivity under applied voltage bias and preserve this state even after external impact is over. This property is the reason why memristors are widely used in hardware implementations of neuromorphic systems (NSs), mainly as synaptic weights. NSs, including spiking neuromorphic systems, are subjects of interest because of their high performance and low power consumption in solving complex cognitive tasks. Memristors can be made of variety of materials and hence many different mechanisms can result in their resistive switching. One of these materials is LiNbO3 with nano particles of metal inside. It has multifilamentary switching mechanism, when switching between resistive states is caused by forming and destruction of conductive filaments in isolating matrix. Introduction of metal nanoparticles into isolator makes this process more pre-defined, hence decrease both cycle-to-cycle and device-to-device variability of resistive switching. This type memristors are shown to have good operating characteristics, such as: up to 10^6 switching cycles, 10^5s retention time, 256 resistive states; spike-timing-dependent plasticity, dopamine-like modulated plasticity [1], noise-assisted learning, etc. were demonstrated using them. Crossbar structures with this type of memristors were also fabricated. However, nanocomposite memristors have major disadvantage – relatively high switching voltage (≈ 6V). It may be a result of relatively big voltage decrease on filaments. This issue can be serious, especially in NSs, distinctive property of which is low power consumption. It can also increase complexity of silicon electronics, cooperating with memristors because of higher currents required. This is why we developed nanocomposite memristors with thin working layer – 30nm of pure LiNbO3 under 60nm of CoFeB LiNbO3 nanocomposite. In this work we demonstrate their switching properties (switching voltage ≈ 2-3V), evaluate endurance, retention time, device to device and cycle to cycle diversities, demonstrate basic and dopamine-modulated spike-timing dependent plasticity (STDP), examined possibility of noise-assisted learning in NSs with these memristors as synaptic weights. For latter experiments we developed self-made analog spiking neuron with square form of spikes and internal support of dopamine modulation of them. The study was supported by the Russian Science Foundation (Project No. 20-79-10185). [1] Kristina E. Nikiruy, Igor A. Surazhevsky, Vyacheslav A. Demin, and Andrey V. Emelyanov, «Spike-Timing-Dependent and Spike-Shape-Independent Plasticities with Dopamine-Like Modulation in Nanocomposite Memristive Synapses», Phys. Status Solidi A 2020, 1900938, DOI: 10.1002/pssa.201900938

Resume : Organic mixed ionic-electronic conductors (OMIECs) have recently emerged as promising materials for neuromorphic devices. OMIECs allow for ionic gating of electronic charge transport and render memory switching in neuromorphic devices highly energy efficient. In addition, OMIECs show high biocompatibility, electrochemical stability and soft mechanical properties that make OMIEC based neuromorphic devices particularly suitable for direct integration with biomedical transducer functionality.1 Despite the interest, charge transport properties of OMIECs are not well understood and experimental characterization is difficult due to the mixed conductor properties and contact resistance in organic electrochemical transistors.2 At the same time understanding and optimizing of carrier mobilities in OMIECs will be crucial to achieve fast neuromorphic operation. To address the issue, we introduce the electrolyte-gated van der Pauw (EgVDP) method for the simple and accurate determination of the transport characteristics of OMIEC thin films, independently from contact effects. Our technique is applied to the most widespread OMIEC blend, PEDOT:PSS. Device fabrication is fully compatible with common OECT process flows and EgVDP method creates transport conditions similar to thin film transistor operation, allowing for an accurate extraction of the mobility and the threshold voltage. By comparing with transistor measurements, we find that gate voltage dependent contact resistance effects lead to systematic errors in OECT based transport characterization. These observations confirm that a contact-independent technique is crucial for the proper characterization of OMIECs, and the EgVDP method reveals to be a simple, elegant but effective technique for this scope. 1. Tuchman, Y. et al. Organic neuromorphic devices: Past, present, and future challenges. MRS Bull. (2020) doi:10.1557/mrs.2020.196. 2. Kaphle, V., Paudel, P. R., Dahal, D., Radha Krishnan, R. K. & Lüssem, B. Finding the equilibrium of organic electrochemical transistors. Nat. Commun. (2020) doi:10.1038/s41467-020-16252-2.

Resume : We present our recent realisation of a magnetic Galton board [1], a nanoscale recreation of Sir Francis Galton’s paradigmatic experiment in which balls fall through an array of pegs to show that randomness can lead to order in the form of a binomial distribution. With the nanoscale experiment presented here, we demonstrate that magnetic domain walls (the point-like magnetic textures forming during magnetization reversal) can be employed as stochastic information carriers in magnetic networks, with a degree of stochasticity that is continuously tuneable in a controlled manner. Using high-resolution immersion Kerr microscopy, we have imaged magnetic domain wall propagation through a nanowire array a very large number of times, observing that complex domain wall dynamics can generate high-quality stochasticity. By varying the direction of an externally-applied field, we were also able to demonstrate the continuous tuneability of the stochastic domain-wall decision process, a critical step to the realisation of functional computing systems. These observations open an interesting route to cheaply deploy stochasticity into the wide range of magnetic domain-wall devices currently under investigation for their promise of fast, energy-efficient and non-volatile information processing. Interestingly, in these non-volatile devices the movement of an information carrier through the circuit naturally performs computations, eliminating completely the separation between memory and logic. Combined with the emergence of neuromorphic stochastic computing concepts, the possibility of turning domain-walls into information carriers with tuneable stochasticity opens a fascinating route for the development of new computing frameworks and device architectures. [1] Sanz-Hernández, Dédalo, et al. "A nano-magnetic Galton board." arXiv preprint arXiv:2010.10389 (2020).

Resume : A recently discovered electronic element, the memristor (short name for memory resistor) is proposed for performing non-linear electrical tasks related to neurohardware applications, given that it can combine information processing with memory storage in a single component, similar to how biological neurons work. A recent proposal in the molecular scientific community claims that organic semiconductors and molecular materials could play a key role as neuromorphic materials for being tunable, cheap and easy to process. In this work, an organic ion-polymer based device is proposed and the analog memristive behavior is described and characterized in detail. Further theoretical discussion on the transport mechanism is included to explain the behavior in these ionic devices under an external electric field. We present a step further in the fabrication, electrical characterization, and learning process of this ion-polymer based device for memristive applications. The active material proposed consists of a mixture of electron and ion transporting polymers (Super yellow and Hybrane), together with an embedded ionic salt (lithium triflate, LiCF3SO3). The particular advantage of using a combination of two conductive polymers with an ionic material has never been presented as a prospective memristive material up to now. Here, we show the study of the analogue memristive behaviour to reveal a quasi-continuous scheme of achievable states. We report multiple controllable pinched hysteresis loops in the I-V curves, while also explaining how both spiking potentiation and depression in terms of neuroplasticity can be introduced in the system.

Resume : The recent development of doped Hafnium oxide-based ferroelectrics has opened the opportunity to integrate ferroelectric devices with CMOS technology. Among them, ferroelectric tunnel junction (FTJ) devices have the potential for extremely low power consumption memories for in-memory computing and neuromorphic applications [1]. FTJ devices rely on the difference in tunneling rate across the ferroelectric layer, for different polarization states. A high direct-tunneling current necessitates a thin ferroelectric layer, typically 1-4 nm. The development of ultra-thin doped Hafnium dioxide-based ferroelectric layers still an active area of research. Therefore, an FTJ design based on composite stack of Metal-Ferroelectric-Dielectric-Metal can enable devices with high tunnel electro-resistance ratio (TER) while utilizing thick ferroelectric layer (~10 nm) [2]. In most works so far, the crystallization temperatures utilized for doped Hafnium dioxide-based ferroelectrics is in the range of 500-600°C. These temperatures can be detrimental for CMOS back-end-of-line integration. In this work, we demonstrate a CMOS backend compatible FTJ stack with a 400°C crystallized Hf0.5Zr0.5O2 (HZO) ferroelectric layer and TiN-Al2O3-HZO-W stack. The devices show reasonable TER with multi-level resistance tenability necessary for neuromorphic applications. We will discuss the cycling behaviour of these multiple resistance states. Additionally, the dielectric-ferroelectric interface in the composite stack plays a very crucial role in polarization stability of the ferroelectric and hence the resistance states [3]. We will discuss a detailed investigation of the FTJ device characteristics with varying programming pulse widths. These measurements shed light on the dielectric-ferroelectric interface and allow multiple resistance state programming. Additionally, we will also discuss device characteristic dependence on the dielectric material, its thickness and identify parameters to improve the multiple resistance state stability, on-state current and retention. [1] M. Benjamin et al., “Direct correlation ferroelectric properties and memories charecteristics in ferroelectric tunnel junctions,’’ IEEE Journal of Electron Devices Society (J-EDS), vol. 7, pp. 1175-1181, 2019. [2] R. Hojoon et al., “Ferroelectric tunnel junctions based on aluminium oxide/zirconium-doped hafnium oxide for neuromorphic computing,” Scientific Reports, vol. 9, pp. 1-8, 2019. [3] M. Si et al., “Ferroelectric Polarization Switching of Hafnium Zirconium Oxide in a Ferroelectric/Dielectric Stack”, ACS Applied Electronic Materials, vol. 1, issue 5, pp. 745–751, 2019