Energy-efficient time series processing in real-time with fluidic iontronic memristor circuits

Area of Science:

• Neuromorphic Engineering
• Materials Science
• Computational Neuroscience

Background:

• Iontronic neuromorphic computing is a rapidly growing field.
• Angstrom-confined iontronic devices offer low power consumption and natural signal alignment.
• Challenges exist in comparing iontronics to conventional substrates and identifying applications.

Purpose of the Study:

• To propose a pathway for iontronic circuits to tackle time series benchmark tasks.
• To enable performance comparisons between iontronic and conventional neuromorphic systems.
• To highlight potential application domains for efficient real-time time series processing.

Main Methods:

• Modeling a Kirchhoff-governed circuit with iontronic memristors.
• Using dynamic internal voltages as output vectors for linear readout.
• Logging energy consumption during simulations.
• Integrating models into the open-source pyontronics package.

Main Results:

• Demonstrated prediction performance comparable to solid-state reservoirs without input encoding or virtual timing.
• Achieved exceptionally low energy consumption, over 5 orders of magnitude lower than conventional methods.
• Validated the pathway for iontronic circuits in time series processing.

Conclusions:

• Iontronic technologies offer a promising route for ultra-low-power real-time neuromorphic computation.
• The proposed framework facilitates performance evaluation and application discovery for iontronics.
• This work bridges the gap between iontronic device capabilities and practical neuromorphic tasks.