An optical chip that can train machine learning hardware

A multi-institutional research team has developed an optical chip that can train machine learning hardware. Their research is published today in Optics.

Machine learning applications have risen to $165 billion annually, according to a recent McKinsey report. But before a machine can perform intelligence tasks like recognizing the details of an image, it must be trained. Training modern artificial intelligence (AI) systems like Tesla’s autopilot costs several million dollars in electrical power consumption and requires supercomputer-like infrastructure.

This increasing AI “appetite” leaves an ever-widening gap between computer hardware and demand for AI. Photonic integrated circuits, or simply optical chips, have emerged as a possible solution to deliver higher computing efficiency, as measured by the number of operations performed per second per watt used, or TOPS/W. However, although they have demonstrated improved core operations in machine intelligence used for data classification, photonic chips have yet to improve the actual learning and machine training process.

Machine learning is a two-step procedure. First, data is used to train the system and then other data is used to test the performance of the AI ​​system. In a new paper, a team of researchers from George Washington University, Queens University, University of British Columbia and Princeton University set out to do just that.

After one training step, the team observed an error and reconfigured the hardware for a second training cycle followed by additional training cycles until sufficient AI performance was achieved (eg, the system is able to correctly label objects appearing in a movie). Until now, photonic chips have only demonstrated an ability to classify and infer information from data. Now researchers have made it possible to speed up the training step themselves.

This added AI capability is part of a larger effort around photonic tensor cores and other electronic-photonic application-specific integrated circuits (ASICs) that leverage photonic chip manufacturing for machine learning and AI applications.

“This new hardware will speed up the training of machine learning systems and take advantage of the best of what both photonics and electronic chips have to offer. It’s a major leap forward for AI hardware acceleration. These are the kinds of advances, that we need in the semiconductor industry as underscored by the recently passed CHIPS Act,” states Volker Sorger, Professor of Electrical and Computer Engineering at George Washington University and founder of the startup Optelligence.

“The training of AI systems costs a significant amount of energy and carbon footprint. For example, a single AI transformer takes up about five times more CO.₂ in electricity as a petrol car spends in its life. Our training on photonic chips will help reduce this overhead,” adds Bhavin Shastri, Assistant Professor of Physics Department Queens University.