Microsoft Research announced a breakthrough in analog optical computing with its analog optical computer (AOC), demonstrating extraordinary promise in solving real-world optimization problems and advancing energy-efficient AI processing.
A dedicated research team in Cambridge developed the AOC using commercially available components micro-LED lights, optical lenses, and smartphone camera sensors to ensure affordability and alignment with existing manufacturing supply chains. The AOC achieves computation through physical light-based systems, offering speed and energy efficiency significantly beyond that of traditional digital computing.
Microsoft released two major achievements: successfully applying the AOC to practical optimization problems in banking and healthcare, and publishing a research paper detailing these results in the journal Nature.
The team also unveiled both an “optimization solver” algorithm and a “digital twin” of the AOC, enabling external researchers to explore this computing paradigm, propose new problem domains, and test novel solutions.
Francesca Parmigiani, Principal Research Manager at Microsoft, explained that the digital twin “mimics how the real AOC behaves; it simulates the same inputs, processes and outputs, but in a digital environment like a software version of the hardware.” She noted that this tool allows researchers to scale up optimization and AI problem simulations beyond the capacity of the current prototype.
Parmigiani added, “To have the kind of success we are dreaming about, we need other researchers to be experimenting and thinking about how this hardware can be used.”
Hitesh Ballani, Director of Research on Future AI Infrastructure at Microsoft Research Cambridge, said, “We have actually delivered on the hard promise that it can make a big difference in two real-world problems in two domains, banking and healthcare.” He highlighted that “we opened up a whole new application domain by showing that exactly the same hardware could serve AI models, too.”
In the healthcare domain, researchers applied the AOC’s digital twin to reconstruct MRI scans with promising accuracy. The experiments suggest that the technology could theoretically reduce scan time from 30 minutes to five.
Michael Hansen, Senior Director of Biomedical Signal Processing at Microsoft Health Futures, cautioned, “To be transparent, it’s not something we can go and use clinically right now. Because it’s just this little small problem that we ran, but it gives you that little spark that says, ‘Oh boy! If this instrument was actually in full scale’ …” He emphasized the role of the digital twin, saying, “The digital twin is where we can work on larger problems than the instrument itself can tackle right now. And in that we can actually get good image quality.” Hansen also shared a forward-looking idea: streaming MRI data to an AOC in Azure with results streamed back to clinics.
Also Read: Montage Unveils CXL 3.1 Memory Controller for Data Centers
In the financial services domain, the AOC tackled a delivery-versus-payment (DvP) securities settlement optimization problem. This involved up to 1,800 hypothetical parties and 28,000 transactions representative of clearinghouse operations. The team demonstrated that future, scaled-up AOCs could address such problems on a much larger scale, with new generation AOCs planned roughly every two years.
Ballani remarked, “It is an absolute giant problem with massive real-world finance impact. It’s already a problem where banks need to collaborate, and better algorithms help everyone.”
Shrirang Khedekar, Senior Software Engineer in Advanced Technologies at Barclays and co-author of the Nature paper, affirmed, “We believe there is a significant potential to explore. We have other optimization problems as well in the financial industry, and we believe that AOC technology could potentially play a role in solving these.”
From the project’s inception, the AOC team aimed to adapt the system for AI workloads. A chance conversation during a lunch with AI researcher Jannes Gladrow reshaped the project’s trajectory. Ballani recalled how detailed discussions led to collaboration between Gladrow and AOC team member Jiaqi Chu to map machine-learning algorithms onto the AOC.
Gladrow warned about the prototype’s limited scale: “I think what’s important to understand is the machine is small. It can only run a small number of weights at the moment because it’s a prototype.” Yet he noted the AOC’s inherent advantage: “The most important aspect the AOC delivers is that we estimate around a hundred times improvement in energy efficiency,” And so that alone is unheard of in hardware.
Highlighting the AOC’s unique capabilities, Gladrow explained how its analog design supports energy-efficient reasoning tasks, such as state tracking, that challenge conventional GPUs powering large language models.
Ballani emphasized that while the team has reached a significant milestone, it is only the beginning of a journey toward a commercially viable AOC. He noted that the research has helped demonstrate the tangible potential of the technology, saying, “We’ve been able to convince ourselves and hopefully a broader segment of the world that, well, actually, you know what? There are real applications for the AOC.” He concluded with Microsoft’s vision: “Our goal, our long-term vision is this being a significant part of the future of computing, with Microsoft and the industry continuing this compute-based transformation of society in a sustainable fashion.”