Anion-Engineered Organic Electrochemical Transistors With Multi-Timescale Synaptic Dynamics for Task-Adaptive Spiking Neural Networks.

Biological synapses process complex temporal information through a diverse range of excitatory postsynaptic current (EPSC) decay scales. However, conventional neuromorphic hardware often employs homogeneous synaptic models, limiting its capacity for multi-scale temporal tasks. Here, we present a material-to-system co-design strategy to emulate biological temporal heterogeneity using anion-engineered organic electrochemical transistors (OECTs). By tailoring the anion species within a solid-state ion-gel electrolyte utilizing BF4 -, TFSI-, and MeSO4 -, we demonstrate precise control over ionic reverse diffusion kinetics, realizing the controllable tuning of synaptic decay constants (from 0.275 s to 1.059 s). These stateful synapses act as intrinsic leaky integrators, embedding short-term memory directly into the device hardware. We evaluate these devices across diverse spiking neural network (SNN) architectures, including feedforward SNNs and liquid state machines (LSMs), using four benchmark datasets. Our results reveal a critical timescale-matching effect: fast-decaying synapses optimize spoken digit recognition (TI-46, 89.4%), while slow-decaying synapses are essential for long-term event integration (DVS Gesture, 88.5%). By transforming synapses from passive transmitters into active physical regulators of network computation, this work provides a versatile hardware platform for constructing task-adaptive neuromorphic systems capable of processing complex spatiotemporal information.