Columbia Engineering develops efficient 3D AI hardware platform

Researchers at Columbia Engineering have developed an innovative 3D photonic-electronic platform. This new technology improves energy efficiency and data transfer speeds, which are key challenges for artificial intelligence (AI) systems. The study, led by Keren Bergman, was published in the journal Nature Photonics. It combines photonics with advanced electronic systems to enhance data communication. This work aims to help AI technologies become faster and more efficient. The team created a compact chip that houses 80 photonic transmitters and receivers. This chip can transmit data at an impressive rate of 800 gigabits per second while using very little energy. It requires only 120 femtojoules per bit, which is significantly better than current standards. Designed for cost-effectiveness, this chip uses components that can be produced in commercial manufacturing facilities. This makes it easier for industries to adopt the new technology. The integration of photonic and electronic chips helps overcome long-standing limitations in energy usage and scalability. This advancement allows AI systems to quickly share large amounts of data, which could enhance various applications, from AI to telecommunications and high-performance computing. The collaborative research included contributions from experts at Cornell University, the Air Force Research Laboratory, and Dartmouth College. This breakthrough may lead to a new era of energy-efficient and fast computing systems.