Microsoft has made a significant breakthrough in AI infrastructure with the development of an experimental analog optical computer (AOC) designed to run large language models more efficiently. This innovative computer uses light to make computations, potentially reducing energy costs for AI workloads.
The AOC calculates using varying light intensities moving through digital sensors. It consists of projectors with optical lenses, digital sensors, and micro-LEDs, utilizing commercially available components like smartphone camera sensors to keep costs down. This technology has the potential to transform various industries, including finance, healthcare, and AI models.
In finance, the AOC's optimization solver algorithm can solve complex transaction settlement problems in banking, such as those used in delivery-versus-payment securities. In healthcare, it can speed up MRI scans, reducing scan times from 30 minutes to just five minutes. Additionally, the AOC can run one type of large language model reasoning, state tracking, at a lower energy cost than traditional GPUs.
The Microsoft Research team is committed to advancing this technology, planning to build a new generation of AOCs every two years. They've published their findings in Nature and released a digital twin repository on GitHub, encouraging other researchers to experiment with the hardware. This collaborative approach could lead to more efficient AI accelerators and have a significant impact on various industries.
