CMOS-Compatible Memristor for Optoelectronic Neuromorphic Computing.

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Science

Background:

• Optoelectronic memristors are key for next-generation data storage and neuromorphic computing.
• Developing devices with light-tunable properties is crucial for these applications.

Purpose of the Study:

• To demonstrate light-tunable resistive switching (RS) in a CMOS-compatible optoelectronic memristor.
• To evaluate the device's performance for neuromorphic applications, including image recognition.

Main Methods:

• Fabrication of an ITO/HfO2/TiO2/ITO optoelectronic memristor.
• Characterization of resistive switching, data retention, endurance, and optical response.
• Simulation of a Hopfield neural network (HNN) using device characteristics for image learning.

Main Results:

• Achieved stable bipolar analog switching with excellent data retention (>10^4 s) and endurance (10^6 cycles).
• Demonstrated light-tunable resistance with an observed current increase under 405 nm light irradiation.
• Successfully trained a simulated HNN to recognize a 10x10 pixel image with 100% accuracy in 13 iterations.

Conclusions:

• The developed optoelectronic memristor exhibits promising characteristics for light-controllable storage and neuromorphic computing.
• The device's ability to achieve synaptic plasticity through both electrical and optical stimulation is vital for its neuromorphic potential.