Microsoft's analog optical computer cracks 2 practical problems,shows AI promise (news.microsoft.com)

Microsoft Research’s analog optical computer (AOC), built from off-the-shelf components like micro-LEDs and smartphone camera sensors, has successfully solved two real-world optimization problems in finance and healthcare, showcasing its potential to revolutionize AI and computational workloads. Unlike traditional digital computers, the AOC uses light to perform calculations, offering up to 100 times faster processing and energy efficiency. Key demonstrations include optimizing complex banking transaction settlements involving thousands of parties and accelerating MRI scan reconstructions that could drastically reduce imaging times from 30 minutes to five. These milestones, detailed in a Nature paper, prove the AOC’s practical viability and hint at transformative impacts across sectors reliant on large-scale optimization. Significantly, Microsoft is releasing both the AOC’s “digital twin”—a software model replicating hardware behavior—and its optimization solver algorithm, enabling the broader research community to explore new applications and advance this novel computing paradigm. The digital twin allows tackling larger-scale problems, enhancing problem-solving beyond the prototype’s physical limits. Moreover, the team demonstrated early AI applications, with machine learning tasks mapped onto the AOC, suggesting its potential to run large language models more energy-efficiently by handling state tracking—a challenging task for today’s GPUs—with far lower energy consumption. While still in prototype form with 256 parameters, the AOC’s scalable design promises immense growth, with future versions housing millions of ‘weights’ for tackling increasingly complex problems. Microsoft’s interdisciplinary team envisions this analog optical approach becoming a foundational technology for high-efficiency AI infrastructure and real-world optimization challenges, setting a new direction for sustainable, high-speed computing.