Chalcogenide optomemristors for multi-factor neuromorphic computation.

Area of Science:

• Artificial Intelligence
• Materials Science
• Neuroscience

Background:

• Neuromorphic hardware accelerates AI by mimicking biological computation.
• Memristive technologies, particularly chalcogenide-based in-memory computing, enhance neural operations.
• Advanced AI requires devices supporting complex operations like reinforcement learning and dendritic computation.

Purpose of the Study:

• To demonstrate multi-factor in-memory computation using nano-scaled chalcogenide semiconductor films.
• To leverage tunable electronic and optical properties for advanced device operations.
• To emulate complex neural functions like plasticity and inhibition in artificial synapses and dendrites.

Main Methods:

• Fabrication of ultrathin photoactive cavities using Ge-doped Selenide.
• Exploitation of joint electronic and optical properties for computation.
• Demonstration of three-factor neo-Hebbian plasticity and shunting inhibition emulation.

Main Results:

• Successfully emulated synapses with three-factor neo-Hebbian plasticity.
• Demonstrated dendritic computation with shunting inhibition.
• Solved a maze game using on-device reinforcement learning.
• Implemented a single-neuron solution for linearly inseparable XOR problems.

Conclusions:

• Nano-scaled chalcogenide films enable advanced multi-factor in-memory computation.
• The developed devices can perform complex AI functions, including reinforcement learning and solving non-linear problems.
• This work advances neuromorphic hardware capabilities for more efficient and powerful AI.