Microsoft team builds simulated optical computer using MicroLEDs

On September 3, Microsoft Research in Cambridge, UK announced that its team successfully used Micro LED light sources, optical lenses and smartphone camera sensors to build a newAnalog Optical Computer (AOC). The research results were published in the latest issue of Nature and are expected to bring breakthrough applications to financial settlements, medical imaging and artificial intelligence.

Image source: Microsoft

It is reported that, unlike digital computers that rely on binary operations, AOC performs calculations through optical physical processes, which can circumvent some of the bottlenecks of traditional digital computing. Researchers imagine that AOC can achieve a hundredfold improvement in speed and energy efficiency on specific problems, and has the advantages of normal temperature operation and low energy consumption.

It is reported that AOC has shown potential in two major areas. In the financial field, Microsoft has cooperated with Barclays Bank to simulate and solve the optimization problem of clearing securities transaction settlement. In the test of 28,000 transactions involving 1,800 participants, AOC demonstrated efficient and accurate solving capabilities.

In addition, in the medical field, the Microsoft research team uses AOC's "digital twin" model to reconstruct MRI scans, which can theoretically shorten the scanning time from 30 minutes to 5 minutes, significantly improving the utilization of medical resources.

The research team emphasized that AOC is not a general-purpose computer, but a dedicated architecture for optimization problems and specific AI tasks, using the intensity changes of photons as they pass through the sensor to achieve addition and multiplication.

Image source: Microsoft

The model parameters currently owned by AOC have been expanded from 64 to 256. In the future, by adding more Micro LEDs, it is expected to reach millions or even billions. AOC can complete more complex calculations and the device size will shrink as the device miniaturizes.

In terms of AI applications, Microsoft researchers have proven that AOC can run simple machine learning tasks and has shown great advantages in high-energy-consuming reasoning tasks such as "state tracking". Research estimates that the energy efficiency of AOC can be improved by about 100 times compared with existing GPUs.

In the future, the research team plans to iterate a new generation of AOC every two years and gradually expand its scale and capabilities. Microsoft Research Group believes that this technology will play a core role in financial transaction optimization, medical imaging acceleration, and AI computing.

It is worth noting that Microsoft has recently continued to apply Micro LED technology to implement its latest research projects. Previously, the Microsoft research team also developed the optical interconnection technology MOSAIC based on Micro LED, using hundreds of low-speed parallel Micro LED channel realizes a wide and slow (WaS) communication architecture mode with extremely low power consumption, longer transmission distance, higher reliability, scalability and compatibility.