Photonic Physical Reservoir Computing with Tunable Relaxation Time Constant

Area of Science:

• Materials Science
• Condensed Matter Physics
• Computer Engineering

Background:

• Edge computing demands efficient processing with reduced computational cost.
• Reservoir computing leverages physical dynamics but is limited by system transient characteristics.
• Current physical reservoirs have timescale limitations for input signal processing.

Purpose of the Study:

• To develop a memristor capable of processing information at arbitrary edges with tunable transient characteristics.
• To overcome the timescale limitations of conventional physical reservoir computing.
• To enhance computational performance in edge computing applications.

Main Methods:

• Fabrication of a memristor using an Sn-doped In2O3 /Nb-doped SrTiO3 junction.
• Utilizing the memristor's response to both electrical and optical stimuli.
• Controlling the transient current response timescale via applied voltage.
• Evaluating computational performance in an image classification task.

Main Results:

• The memristor exhibited controlled transient current response timescales over several orders of magnitude.
• Computational performance as a physical reservoir was demonstrated in image classification.
• Optimized learning accuracy was achieved by tuning device transient characteristics.

Conclusions:

• The developed memristor offers a method to control transient response for enhanced physical reservoir computing.
• The findings provide insights into strontium titanate photoconductive properties.
• This work supports the physical implementation of advanced computing systems.