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

Area of Science:

• Neuromorphic Engineering
• Materials Science
• Computational Neuroscience

Background:

• Reservoir computing (RC) is an energy-efficient approach for temporal information processing using physical systems.
• Existing physical reservoirs struggle to balance state richness and adaptable dynamics for diverse temporal tasks.
• Hardware limitations hinder self-adaptive processing in current neuromorphic systems.

Purpose of the Study:

• To develop a physical reservoir with distributed and reconfigurable dynamics for adaptive computing.
• To overcome the limitations of existing hardware in processing temporal tasks with varying time scales.
• To create a versatile platform for task-adaptive neuromorphic computing.

Main Methods:

• An in-liquido physical reservoir was designed using an organic electrochemical transistor (OECT) array.
• Gating distance and shared electrolyte modulation were employed to control reservoir dynamics.
• The reservoir's processing capabilities were evaluated on temporal classification and prediction tasks.

Main Results:

• The OECT array reservoir achieved a relaxation time distribution spanning over two orders of magnitude (0.013–2.3 s).
• The in-liquido reservoir demonstrated effective processing for diverse temporal tasks due to its distributed dynamics.
• Real-time feedback enabled global reconfiguration, allowing self-adaptive matching of reservoir dynamics to task requirements.

Conclusions:

• The in-liquido reservoir offers a powerful and adaptable hardware platform for neuromorphic computing.
• This approach enables physically distributed and globally reconfigurable dynamics for enhanced adaptive computing.
• The developed OECT-based reservoir advances the field of task-adaptive neuromorphic hardware.