Academics

MIT Nanophotonic Processor Can Perform Deep Learning Computations [VIDEO]

By

Deep learning has become an important part of almost any industry. Computer systems based on artificial neural networks function like the human brain and can perform complex computations much faster than regular computers. These require more speed and power which is impossible if they rely on electricity alone. MIT researchers have found a much better solution and that is replacing electricity with light.

A team of MIT researchers has developed a type of photonic computer which uses light in performing complex neural computations. According to the researchers, this optic-based neural network system is much faster and more efficient than traditional computers that use electricity.

Neural network tasks calculations are complex and intensive because they require repeated multiplications of matrices. Such tasks are too much for the traditional computer architecture which uses either CPU or GPU chips.

Marin Soljačić, an MIT professor and one of the researchers, used the ordinary eyeglass lens to demonstrate the power and speed of this light-based computer. He said that even a single lens performs complex calculations to the light that passes through it.

The underlying principle behind the photonic chips is similar. In principle, it can perform matrix calculations with zero energy once they are tuned. The team manipulates the direction of multiple light beams so that they can interact with each other. This interaction produces interference patterns that show the result of the intended operation.

Yichen Shen, the author of the paper, said that the photonic chips are much faster but using less than one-thousandth as much energy per operation compared to electronic chips.

The team said that the nanophotonic processor has a lot of practical uses, such as faster signal processing in data transmission. It can also be utilized in security systems, data centers, and even autonomous cars and drones. However, the system still needs to be scaled up to make it fully functioning.

The study was published in the journal Nature Photonics.

© 2024 University Herald, All rights reserved. Do not reproduce without permission.
Join the Discussion
Real Time Analytics