Flexible Tunable-Plasticity Synaptic Transistors for Mimicking Dynamic Cognition and Reservoir Computing

Abstract

Inspired by biological systems, neuromorphic computing can process extensive data and complex tasks more efficiently than traditional architectures. Artificial synaptic devices, serving as fundamental components in neuromorphic computing, needto closely mimic synaptic characteristics and construct neural network computing systems. However, most existing multifunctional synapse devices are structurally complex and lack tunability, making them unsuitable for building smarter computing systems. In this work, a flexible tunable-plasticity synaptic transistor (TST) is realized with memory modulation and neuromorphic computing capabilities by using indium gallium zinc oxide as channel and a hybrid layer of polyimide and Al₂O₃ as dielectric. The TST exhibits a novel transition from short-term plasticity to long-term one by adjusting stimulus amplitude, mirroring dynamic human memory and forgetting behaviors across various scenarios. A neural network system with low non-linearity and a wide range of conductance variations is constructed, and it demonstrates a 94.1% recognition rate on classical datasets. A reservoir computing system for 4-bit coding is also developed, which significantly reduces computational complexity and network size without sacrificing recognition accuracy. The devices and the system work as the foundation of more intelligent and more efficient computing systems.