World’s First Light-based Neural Network Arrives For Ultrafast Computing

Share on twitter
Tweet
Share on whatsapp
WhatsApp
Share on facebook
Share
neural-networks-photonic

Short Bytes: Researchers at Princeton University have created the world’s first light-based neural network, also called photonic neural network. They have created an optical computing machine that solves a differential equation 1,960 times faster than a regular processor. This research is expected to push optical computing into the mainstream.

Artificial intelligence is the leading force in today’s technology world. Big companies like Google, Microsoft, IBM, and Amazon are investing heavily in the AI research and development. In Gartner’s top 10 disruptive technology trends of 2017, AI and advanced machine learning is at the top.

With time, the researchers have worked hard to mimic human brains and design efficient circuitry and neural networks. Still, the human brain remains more advanced. Trying to make these circuits operate more like neurons, Alexander Tait and colleagues at Princeton University in New Jersey have designed world’s first photonic neural networks.

The researchers have made the world’s first integrated silicon photonic neuromorphic chip that makes calculations at ultrafast speeds. As the name suggests, neuromorphic chips are the electronic analog circuits that mimic neuro-biological architectures present in the nervous system.

PHOTONIC NEURAL NETWORK SOLVES DIFFERENTIAL EQUATIONs 1,960 TIMES FASTER

Why photonic neural networks? Because photons have more bandwidth as compared to electrons and they can process data at a faster pace. However, due to their higher costs, the research on optical data processing systems didn’t pick speed in the past.

The Princeton researchers have been able to create an optical device which has “nodes that take the form of tiny circular waveguides carved into a silicon substrate in which light can circulate.” Each node is summed by total power detection before it’s fed into the laser. The laser output is again fed back into the nodes, creating a non-linear feedback character.

This results in the formation of a machine that can compute a differential equation 1,960 times faster than a usual processor. The researchers have demonstrated this using 49 photonic nodes.

While these photonic neural networks can’t replace the current CPUs in near future, this development has opened a new door to a new industry that can push the optical computing into the mainstream.

Source: MIT Tech Review
Research Paper: Arxiv.org

Also Read: Learn Neural Networks In A Fun And Open Source Way

Adarsh Verma

Adarsh Verma

Fossbytes co-founder and an aspiring entrepreneur who keeps a close eye on open source, tech giants, and security. Get in touch with him by sending an email — [email protected]

New on Fossbytes

Scroll to Top