In-Liquido Reservoir Computing with Distributed and Globally Reconfigurable Dynamics for Task-Adaptable Computing

Reservoir computing (RC) is an energy-efficient neuromorphic paradigm that exploits the intrinsic short-term dynamics of reservoirs for temporal information processing. Effectively processing temporal tasks with varying characteristic time scales requires physical reservoirs that offer both sufficient state richness and adaptable dynamical regimes. However, existing physical reservoirs still face significant challenges in fulfilling both requirements at the hardware level, limiting their self-adaptive processing across tasks with varying temporal characteristics. Here, we present an in-liquido physical reservoir based on a gating distance- and shared electrolyte-mediated organic electrochemical transistor (OECT) array, enabling distributed and globally reconfigurable dynamics for adaptive computing. The gating distance establishes a gradient distribution of relaxation times across the array, while modulation of the shared electrolyte globally reconfigures the overall dynamical regime, yielding a relaxation time distribution spanning over 2 orders of magnitude (0.013-2.3 s). Owing to its physically distributed dynamics, the in-liquido reservoir demonstrates powerful processing capability across diverse temporal classification and prediction tasks. Real-time feedback-driven operation further demonstrates physically global reconfiguration, enabling self-adaptive matching of reservoir dynamics to tasks with different temporal characteristics. These results establish the in-liquido reservoir as a versatile and reconfigurable hardware platform for task-adaptive neuromorphic computing.