IEEE Study Introduces Photonic AI Hardware for Enhanced Efficiency

Researchers have developed a new AI hardware platform using photonic integrated circuits (PICs), offering improved scalability and energy efficiency, announced in a press release by the IEEE Photonics Society. This novel platform, detailed in the IEEE Journal of Selected Topics in Quantum Electronics, leverages silicon photonics and III-V compound semiconductors to create AI accelerators capable of handling significant workloads with reduced energy requirements.

The study, led by Dr. Bassem Tossoun from Hewlett Packard Labs, highlights the use of optical neural networks (ONNs) that operate at the speed of light, minimizing energy loss compared to traditional electronic distributed neural networks (DNNs). The fabrication process involves a heterogeneous integration approach, starting with silicon-on-insulator wafers and incorporating III-V semiconductors through die-to-wafer bonding.

This innovative photonic platform enables wafer-scale integration of various devices necessary for building an optical neural network on a single chip. It includes active devices such as on-chip lasers, amplifiers, high-speed photodetectors, and energy-efficient modulators. The platform aims to address computational and energy challenges, paving the way for robust and sustainable AI accelerator hardware in the future.