Low-Voltage Organic Transistor-Based Reservoir Computing for Multitask Applications

Physical reservoirs, particularly those based on transistors, exhibit nonlinear dynamic responses and are well-suited for dynamic temporal tasks. In this work, we present a reservoir computing system utilizing a low-voltage organic field-effect transistor (OFET). At the system level, we establish a device-specific spike current map that can be reused across multiple tasks, enabling consistent hardware pipelining. Diverse inputs are preprocessed to conform to a unified spike protocol, while the intrinsic dynamics of the reservoir project these inputs into separable high-dimensional states for the readout layer. Using the OFET device and mapping, we validate our approach on a range of recognition tasks, from static image classification to dynamic gesture classification with various gesture input modalities. The high accuracy achieved in these tasks demonstrates the potential of our neuromorphic device for gesture classification across diverse input types, and highlights a promising avenue for transistor-based physical reservoirs to process a wide variety of temporal signals using a unified protocol and single-device architecture.