Light-Programmable IGZO Optoelectronic Memristor for Multifunctional Neuromorphic Computing.

Area of Science:

• Materials Science
• Neuroscience
• Computer Engineering

Background:

• Next-generation neuromorphic systems demand integrated hardware for sensing, memory, and computation.
• Existing platforms often lack the multifunctionality required for advanced artificial intelligence.

Purpose of the Study:

• To develop a light-programmable optoelectronic memristor for versatile neuromorphic applications.
• To emulate synaptic plasticity, neuronal responses, and associative learning using optical stimuli.
• To implement advanced computing architectures like reservoir computing and Restricted Boltzmann Machines.

Main Methods:

• Fabrication of an ITO/IGZO/W optoelectronic memristor.
• Optical stimulation for device programming and function emulation.
• Implementation of reservoir computing and Restricted Boltzmann Machine architectures.
• Analysis of synaptic plasticity and nociceptive responses.

Main Results:

• The device successfully emulates short-term plasticity (EPSC, PPF) and spike-dependent learning.
• Replication of nociceptive responses and Pavlovian associative learning via optical stimuli.
• Achieved 97.005% MNIST classification accuracy using a reservoir computing architecture.
• Reached 96.35% image reconstruction accuracy in a Restricted Boltzmann Machine with probabilistic neurons.

Conclusions:

• The IGZO-based optoelectronic memristor is a versatile, energy-efficient platform for multifunctional neuromorphic computing.
• The device integrates sensory, deterministic, and probabilistic intelligence in a single, reconfigurable unit.
• Optical modulation offers precise control over device behavior for advanced AI applications.