Project AOC is building an analog optical computer that has the potential to accelerate AI inference and hard optimization workloads by 100x. To achieve this, we rely on a physical system to embody the computations and step away from several fundamentally limiting aspects of digital computing – avoiding the separation of compute from memory, operating on both continuous and binary data and adopting asynchronous operation that allows the computer to operate at the “speed of light”. It is built using commodity optical and electronic technologies that are low cost and scalable, showing the potential of analog optical computing in the post-Moore Law’s era.

A key aspect of AOC is its hardware and abstraction have been codesigned with the target applications, i.e., the families of optimization and ML algorithms, to take advantage of the computer’s strengths while accommodating its shortcoming and non-idealities. We have just completed the second-generation of our computer and it is the world’s-first unconventional computer that can accelerate both machine learning inference and hard optimization, and is built using consumer-grade technologies!

We are partnering with M365 Research and Microsoft Research Health Futures on this research. Project AOC is part of our broader focus on Future AI Infrastructure where we are innovating and incubating new hardware technologies that could be deployed or used in our AI and cloud data centers. Our cross-disciplinary team thus spans the entire stack, with experience in systems and networking, optics (system, sub-system and device-level), machine learning, physics, maths, and hardware. 

Update (2023-06-08): We are pleased to announce an online service that will allow participants to experiment with the AOC algorithm. For now, the service provides a performance GPU-based implementation of the algorithm, which will allow the users to experiment with converting problems to the QUMO abstraction and understand the benefits and shortcomings of this new kind of computing. We plan to provide access to the hardware in the near future — stay tuned.
We currently have capacity for a limited number of users. Please let us know if you are interested to participate by writing to us at [email protected].