Light Used to Power the First Photonic Neural Network in the World

Image Courtesy: Science Illustrated

A team of researchers at Princeton University found a way to make computer processors work like the human brain. The team successfully integrated a neuromorphic chip to a photonic silicon substrate, creating a prototype of possibly the world’s first and only photonic neural network. Its lightning-fast computing ability uses a ‘neural compiler’ to translate data, beating a traditional CPU in solving a differential equation by a mile – 1,960 times faster to be exact.

Modern day computing requires faster processing speeds while consuming less power. Using light to perform superfast computing is one of the options engineers continue to look into. However, creating processors relying solely on photonic chips prove to be highly expensive.

The team’s breakthrough neuromorphic chip used light and laser to copy the way neurons in a human brain operates. The contraption showcases a semi-conductive silicon board embedded with 49 circular nodes. Each node represents a neural pathway where a certain wavelength of light can pass through. As light travels around the nodes, it modulates the laser’s output, affecting the way data is perceived by a receiver.

The device used photons which have more computing capacity compared to electrons used in today’s CPUs. They can also be used to process information faster, making researchers excited and hopeful in harnessing their abilities. The team was able to combine these features with a neural network resulting in a thousand-fold jump in data processing power.

A neural network is a collection of interconnected neural units mirroring the biological brain’s processing method. Algorithms using this computational approach are self-learning, making them the ideal solution for applications wherein feature detection and interpolation are given priority over direct correlation.

Machines utilizing a neural network helps them perform advanced computing tasks like face, object, and speech recognition, natural language translation, and real-time image and video processing.

With the advancements in AI technology, the need for super-fast computing speeds is inevitable. Many industries can benefit from this development including medicine, agriculture, communication, and military defense.

Aside from cost, development of neuromorphic chips like the Princeton team’s prototype face several challenges preventing engineers from fully realizing the tech’s potential. For one, there’s the issue of losing energy when transforming electronic signals into photons and vice versa. This slows down transmission of data between receiving points.

Although the prototype isn’t yet on par with our smartphone’s computing capabilities, the team believes the results of their study can be utilized by the industry to bring optical computing devices to the market. It may still take years or even decades before we see the technology used widely in our daily computing tasks.