Reconfigurable Hydroxyl Dissociation for Spectrally Decoupled Weight Programming and Photocurrent Computing

Area of Science:

• Optoelectronics
• Materials Science
• Artificial Intelligence

Background:

• The Artificial Intelligence of Things (AIoT) requires compact, power-efficient sensory systems.
• Processing-in-sensor (PIS) offers a solution for edge vision applications by integrating computation within the sensor.
• Direct photocurrent computing (DPC) utilizes photoresponsivity for in-sensor multiply-accumulate (MAC) operations but faces limitations with electrical weight programming.

Purpose of the Study:

• To introduce an optically programmable DPC device for enhanced AIoT sensory systems.
• To overcome the limitations of electrical weight programming in DPC devices.
• To demonstrate a novel material platform for spectrally decoupled weight programming and photocurrent computing.

Main Methods:

• Development of a vacancy-modulated bismuth oxyselenide (BOS) material platform.
• Utilizing reversible surface hydroxyl dissociation triggered by ultraviolet light to reconfigure oxygen vacancy dynamics.
• Implementation of a BOS array for PIS hardware, enabling coarse classification and pre-processing for complex vision tasks.

Main Results:

• Demonstration of an optically programmable DPC device based on BOS.
• Achieved spectrally decoupled weight programming and photocurrent computing.
• Successful implementation of PIS hardware for low-power classification and pre-processing in a processing-near-sensor (PNS) paradigm.

Conclusions:

• The optically programmable BOS DPC device offers a promising solution for compact and power-efficient AIoT sensory systems.
• A hybrid architecture combining PIS and PNS can intelligently balance computational resources for optimal power and performance.
• This work paves the way for advanced edge AIoT applications requiring sophisticated vision processing.