Mechanisms for Enhanced State Retention and Stability in Redox‐Gated Organic Neuromorphic Devices

Abstract

units (GPUs) and >1000 central processing units (CPUs) for some of the highest performance demonstrations. [4] In order to approach the massively parallel and energy efficient operation of the brain (≈25 W), [7] neuromorphic computing architectures have been proposed which utilize physical processes in materials to emulate synaptic behavior. [8][9][10] Among these, resistive memory (memristive) devices [11,12] have gained traction as a highly suitable option, offering projected efficiency gains of up to 10 6 over von Neumann architectures when implementing ANN algorithms. [13,14] Efficiency gains originate from parallel computation and scale with array size (N) [15] : an N × N sized neuromorphic array can simultaneously perform N 2 operations using Ohm's and Kirchoff's laws (multiply and accumulate, respectively), whereas GPUs can perform only N operations simultaneously.Nonvolatile memristive devices such as phase change memory (PCM) and resistive random-access memory (ReRAM) have been proposed for use as nonvolatile artificial synapses due to their ability to tune the resistance state by applied voltage pulses. [16][17][18] By arranging these devices in a crossbar array architecture, vector-matrix multiplication (VMM) can be performed in an analog fashion where the input vector is represented by voltages, the operator matrix is represented by the conductance of each memristive element, and the output vector is represented by currents. These crossbar arrays have recently been implemented in a dot-product engine which uses VMM operations to classify handwritten digits with ≈90% accuracy. [19] However, this demonstration requires a timeconsuming feedback programming scheme which reduces the parallelism of the write operation to the array and ultimately its overall efficiency.Recently, electrochemical nonvolatile redox memory (NVRM) devices [20,21] have emerged as an ideal candidate for neuromorphic arrays as they decouple read and write operations, thereby allowing for low-energy programming and accurately tunable conductance states while potentially avoiding the time-voltage dilemma. [22,23] Furthermore, organic materials are an attractive alternative to conventional resistive memory due to their linearly tunable conductance, [22] biocompatibility, [24] and unique switching mechanisms which significantly differ from their inorganic counterparts. [25][26][27][28][29][30] In particular, there has been interest in mixed ionic-electronic conductors, Recent breakthroughs in artificial neural networks (ANNs) have spurred interest in efficient computational paradigms where the energy and time costs for training and inference are reduced. One promising contender for efficient ANN implementation is crossbar arrays of resistive memory elements that emulate the synaptic strength between neurons within the ANN. Organic nonvolatile redox memory has recently been demonstrated as a promising device for neuromorphic computing, offering a continuous range of linearly programmable resistance states and tunable electronic and elect...

Energetic Control of Redox‐Active Polymers toward Safe Organic Bioelectronic Materials

“…If OMIECs are required to develop OECTs with low turn‐on voltage versus Ag/AgCl (e.g., p(g2T‐TT)), research efforts should be undertaken to redirect the pathway for ORR to H 2 O formation instead of H 2 O 2 . The findings of this work are also relevant for other applications including energy storage devices, nonvolatile memories or sensor devices where undesired oxidation of the active materials and H 2 O 2 formation can affect the retention of the charge, [ 8 ] modify the charge of a memory state, [ 42 ] increase device degradation or possibly interfere with the sensing mechanism of biosensors.…”

Functional Connectivity of Organic Neuromorphic Devices by Global Voltage Oscillations

“…This behavior is attributed to the faradaic response of the global gate electrode. When a voltage is applied at an electrode with a faradaic response (i.e., an ohmic contact between the gate and the electrolyte), the potential drop at the electrolyte is negligible, and the overall device behavior is practically independent of the distance between the gate electrode and the device.…”

Recent Advances in Transistor‐Based Artificial Synapses

“…In the past few years, scientists around the world have tried to mimic the signal transmission process and various types of synaptic plasticity in the brain, although the brain mystery has not yet been fully revealed . The realization of “brain‐like computing” can start from simulating the structure and function of neural networks and artificial synapses, without waiting for neuroscientists and cognitive scientists to fully understand the brain's mechanism, which may even need the exploration process to be longer.…”