New chip opens door to AI computing at light speed

Engineers at the University of Pennsylvania have developed an innovative chip that utilizes light waves instead of electricity to perform complex mathematical operations crucial for training artificial intelligence (AI). This breakthrough chip has the potential to significantly boost computing speeds while reducing energy consumption.

The silicon-photonic (SiPh) chip merges the nanoscale material manipulation research of Nader Engheta with the silicon-based platform, enabling mathematical computations using light, the fastest communication medium available. By leveraging light waves interacting with matter, this chip represents a promising advancement beyond the current computing limitations rooted in decades-old chip design principles.

Published in _Nature Photonics_, the collaborative effort between Engheta and Electrical Engineering Associate Professor Firooz Aflatouni detailed the development of this novel chip designed for vector-matrix multiplication, a fundamental mathematical operation crucial for neural network functionality in AI systems.

By strategically thinning the silicon in specific regions on the chip, the team controlled the propagation of light for lightning-fast mathematical computations without additional materials. This unique design is already poised for commercial applications like speeding up AI training and classification, with potential adaptations for graphics processing units (GPUs).

Apart from enhancing processing speed and reducing energy consumption, the new chip offers privacy benefits. Multiple simultaneous computations eliminate the need to store sensitive data in a computer’s memory, making future systems nearly impervious to hacking attempts.

Lead by Engheta and Aflatouni, the research team includes Vahid Nikkhah, Ali Pirmoradi, Farshid Ashtiani, and Brian Edwards from Penn Engineering.

Overall, this cutting-edge chip holds promise for revolutionizing AI computing by harnessing the power of light waves to achieve unprecedented processing speeds and energy efficiency, paving the way for future advancements with enhanced privacy safeguards.