Coupled Ionic–Electronic Charge Transport in Organic Neuromorphic Devices

Abstract

Conductive polymer devices with tunable resistance allow low-energy, linear programming for efficient neuromorphic computing. Depolarizing impurities, however, are difficult to exclude and limit device performance through nonideal writes and self-discharge. It is shown that these phenomena can be numerically described by combining two-phase charge transport models with electrochemical self-discharge. The simulations accurately reproduce the experimental data, including cyclic voltammetry and standard neuromorphic functions, such as linear programming of discrete states and short-term potentiation. Impurities affect device write accuracy significantly for long programming times above 1000 ms. The effect is reduced to 0.03% for shorter times. Self-discharge is impacted by device potential as well as impurity concentration. A model-based trade-off between operating parameters nearly triples the number of usable conductance states at ambient conditions. Understanding these device limitations as well as workarounds is a vital step toward the implementation of neuromorphic device networks.

Hydration of a Side-Chain-Free n-Type Semiconducting Ladder Polymer Driven by Electrochemical Doping

“…Conjugated polymers showing mixed ionic–electronic transport, − more generally organic mixed-ionic electronic conductors (OMIECs), have emerged in the past few decades as promising materials for applications in bioelectronics, − energy storage, , logic circuit elements, , and neuromorphic computing. − Organic electrochemical transistors (OECTs) are widely used as a model testbed to study mixed ionic–electronic transport in OMIECs. − Unlike conventional organic field-effect transistors (OFETs), an OECT’s source–drain current ( I D ) is modulated by the gate ( V G ) through voltage-driven redox processes and associated ion uptake from the electrolyte into the entire volume of the semiconductor channel, resulting in bulk doping/dedoping. Hence, the efficiency of this modulation, described by the transconductance ( g m ≡ ∂ I D /∂V G ), is governed by the carrier transport mobility (μ) and the volumetric capacitance ( C* ) in the channel .…”

A High-Strength Neuromuscular System That Implements Reflexes as Controlled by a Multiquadrant Artificial Efferent Nerve

“…Under a negative RP, the read/write noise are higher than that under a positive RP, so we only extracted 31 valid memory states with RP = −0.5 V, but extracted 42 states at RP = 0.5 V (Figure 5f). 35 Electrical spikes were used to change the synaptic weight properties (potentiation weight: ΔW p ; depression weight: ΔW d ) of each artificial neuron. For this purpose, 55 000 figures from the Modified National Institute of Standards and Technology (MNIST) data set were chosen for training with multilevel neural networks, and 10 000 different figures from the same data set were chosen for testing (Figure 5g).…”

“…The noise becomes dominant when weight changes are small. Under a negative RP, the read/write noise are higher than that under a positive RP, so we only extracted 31 valid memory states with RP = −0.5 V, but extracted 42 states at RP = 0.5 V (Figure f) …”

Spiking neural networks compensate for weight drift in organic neuromorphic device networks

“…Electrochemical self-discharge is described through a Butler-Volmer-type equation. The fully coupled system is solved with a custom simulation framework CP-E n PE n [24,25] based on PETSc [26] and is described in detail in [24]. The model is validated against experimental data that include cyclic voltammetry, device charging and self-discharge [24].…”