New Optical Chip Computes Faster and More Privately for AI
• New chip uses light waves instead of electricity to perform complex math for AI training, promising faster processing speeds, lower energy use, and better privacy
• Chip brings together research on manipulating materials at nanoscale to perform computations with light (by Penn Engineer Nader Engheta) with silicon photonics platform to mass produce chips (by Penn Engineer Firooz Aflatouni)
• Chip design leverages variations in silicon height to control light propagation and perform vector-matrix multiplication, a key operation for neural networks
• Chip could potentially be used to accelerate graphics processing units (GPUs) for faster AI training and classification
• Chip's ability to perform many computations simultaneously means no need to store sensitive data in memory, making it virtually unhackable