Researchers at the University of Pennsylvania have developed a revolutionary computer chip utilizing light instead of electricity, promising faster data transfer and reduced energy consumption, which is particularly beneficial for AI model training.
Unlike traditional computing systems dating back to the 1960s, this silicon-photonic (SiPh) chip leverages light, known for its speed, and scalable silicon technology for rapid implementation.
The researchers focused on designing a chip capable of performing vector-matrix multiplications crucial for AI neural network development. Controlling light propagation through height variations in the chip ensured efficient data transfer in straight lines. The SiPh chip, compatible with existing infrastructure, offers faster computations with lower energy usage, potentially replacing GPUs in AI model training while addressing data privacy concerns through parallel computation without storing data in working memory.