A facile photonics reconfigurable memristor with dynamically allocated neurons and synapses functions

Area of Science:

• Materials Science
• Neuroscience
• Computer Engineering

Background:

• Reconfigurable memristors are key for next-generation neuromorphic computing, enabling artificial neurons and synapses.
• Current memristor limitations include inconsistent turn-on voltage and operating current post-reconfiguration, hindering hardware development.

Purpose of the Study:

• To develop a memristor device with controllable volatile and non-volatile characteristics for dynamic neuromorphic computing.
• To address the inconsistencies in neuromorphic devices by introducing light as a regulatory means.

Main Methods:

• Introduced light as a regulatory mechanism in memristors via photoelectric coupling.
• Achieved reconfiguration of volatile (10^6 cycles) and non-volatile (10^4 s) characteristics using unified working parameters.
• Demonstrated 100% control over switching voltage without current limitations.

Main Results:

• Successfully demonstrated dynamic reconfiguration of memristor characteristics using light.
• Verified the device's capability in various neuromorphic computing tasks, including Morse code decoding, image recognition, and traffic signal recognition.
• Showcased the potential for on-demand dynamic allocation of artificial neurons and synapses.

Conclusions:

• The developed photoelectric coupling method provides a novel solution for memristor inconsistencies in neuromorphic computing.
• This approach enables dynamic adjustment of neuromorphic devices according to specific application needs.
• Offers a new pathway for the integration and advancement of future neuromorphic computing systems.