Microsoft Research has developed an analog optical computer (AOC) that uses light to solve complex problems.

According to the team responsible for the project, which is based at the company’s Cambridge, UK, lab, the AOC can be used to solve optimization problems in industries such as finance, logistics, and healthcare, and also has the potential to run AI workloads with a fraction of the energy needed and at much greater speed than the GPUs running today’s large language models (LLMs).

“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,” said Hitesh Ballani, research director for the future of AI infrastructure at the Microsoft Research lab in Cambridge.

The AOC developed by Microsoft Research is the result of a four-year project undertaken by the team to build a device that was faster and more efficient at solving certain problems, but could also operate at room temperature, like a traditional desktop computer. 

As a result, the system was built using commercially available parts, including projectors with optical lenses, digital sensors from smartphone cameras, and micro-LEDs, which the team says are the width of a human hair. 

Unlike a typical binary digital computer, an AOC avoids some of the limiting aspects of digital computing by using physical systems to embody the computations it performs. 

As detailed in a blog post outlining the project, as light passes through the sensor at different intensities, the AOC can add and multiply numbers. It’s these calculations that form the basis for the AOC solving optimization problems.

Further details of the project have been outlined in a paper published in the scientific journal Nature. Microsoft has also publicly released its “optimization solver” algorithm alongside a digital twin developed by the team that will allow researchers from other organizations to further investigate the AOC and propose new problems for the system to solve.

The team said that by solving representative smaller versions of certain problems on the actual hardware, and large ones on the digital twin, it showed that problems could be done at a much larger scale with future generations of the AOC, which the Microsoft Research team envisions creating every two years. 

“The most important aspect the AOC delivers is that we estimate around a hundred times improvement in energy efficiency,” Jannes Gladrow, a principal researcher whose specialty is AI and machine learning, said. “And so that alone is unheard of in hardware.”

Francesca Parmigiani, a Microsoft principal research manager who leads the team that developed the AOC, said that the system is “not a general-purpose computer, but what we believe is that we can find a wide range of applications and real-world problems where the computer can be extremely successful.”