Based on a scalable analog optical computer (AOC) architecture designed to use consumer-grade technology, Microsoft has been working with Barclays on a new approach to tackling artificial intelligence (AI) and optimization issues.
The approach explained in the paper published in NatureOvercome the von Neumann bottleneck that occurs in classical computing architectures. Here, performance is limited by the speed at which data can move between memory and the central processing unit. The authors of the paper discuss an approach that eliminates the need to merge digitally and analogue conversion and calculations with memory.
In doing so, they said that AOCs can achieve significant efficiency gains in certain situations. They predicted that AOC would provide performance of about 500 terra operations per watt with 8 bit accuracy.
Microsoft researchers combined 3D optical and analog electronics to build it using a projector with an optical lens, digital sensor and a micro-optical light emitting diode. When the light passes through the sensor at different intensities, the AOC can add and multiply the numbers. This is the basis for solving optimization problems.
in Nature In the paper, the researchers explain how AOC hardware accelerates all computational operations and performs five key operations needed to solve optimization problems. We discuss matrix vector multiplication, nonlinearity, annealing, addition, and subtraction.
This paper also explains how Microsoft works with Barclays to solve the delivery-to-payment securities problem. It aims to find the most efficient way to resolve financial obligations between multiple parties, while minimizing costs or risks within time and available balance constraints.
Shrirang Khedekar, a senior software engineer in Barclays' Advanced Technologies division, worked with a research team at Microsoft UK to create the datasets and parameters used in the study. Co-authored by Khedekar Nature The paper said Barclays is interested in continuing to solve optimization problems as the capabilities of future generations of AOC grow. “We believe there are important possibilities to explore,” he said. “We also have other optimization problems in the financial industry, and we believe AOC technology could play a role in solving these.”
Microsoft said it would release the “optimized solver” algorithm and the digital twin that it developed to allow researchers from other organizations to explore their approach to analog optical computing.
“It's not a general purpose computer, but I believe that computers can find a wide range of applications and real-world problems that can be very successful,” said Francesca Parmigiani, principal research manager at Microsoft, who leads the development team at AOC.
She said that the digital twin mimics how a real AOC works by simulating the same inputs, processes and outputs of the AOC in a digital environment.
“To achieve the success we dream of, we need other researchers experimenting and thinking about how this hardware can be used,” Palmisiani said.
