Computers powered by light and brain networks

Computers powered by light and brain networks

Researcher Bhavin Shastri believes neuromorphic photonics could solve complex computer optimization challenges and advance machine-learning applications.

By Ishita Aggarwal, Research Promotion and Communications Assistant

February 2, 2021


In a new study published in Nature Photonics, lead author Bhavin Shastri (Physics, Engineering Physics and Astronomy) worked alongside an international team to examine how neuromorphic photonics, a field that uses light (instead of electronics) and neural “brain-like” networks, could improve the speed and efficiency of computing systems.    

The von Neumann bottleneck

Modern-day computers are designed using von Neumann architecture in which the fast processor is physically separated from the slower data and program memories. This arrangement allows for easy reprogramming but limits computer speeds and wastes power and energy as information must be continuously transferred between memory and processor, a concept known as the “von Neumann bottleneck.”

To explore how to address the von Neumann bottleneck, Dr. Shastri and colleagues looked to the field of neuromorphic photonics computing, which leverages the strengths of optical physics and neural networks, inspired by the human brain. Here, signal transmission and processing are powered by light energy (photons) instead of electronics (electrons), allowing for higher bandwidths and lower energy losses.

[Photo of photonic lights]

Neural networks

Unlike computers that process digital information sequentially, neural networks in our brains are highly interconnected, distributed, parallel, and well-suited to processing analogue information. The neural “brain-like” networks that inspire photonics allow for: 1) massive interconnection (photons of different wavelengths or “colours” do not interact); 2) parallelization (all wavelengths can co-exist in a single waveguide); and 3) analogue information can be encoded in the phase or intensity of light.

Shastri and collaborators believe that neuromorphic photonic processors have the potential to solve complex computing optimization problems and further develop artificial intelligence algorithms and machine-learning applications.

“Neuromorphic photonic computing has the potential to revolutionize the speed, energy efficiency, and throughput of modern computing,” says Dr. Shastri. “It’s a very exciting development and it is our hope that this study will help demystify an emerging area that could enable fundamental breakthroughs in fields of computer engineering and physics.”

The research was published in Nature Photonics on Jan. 29, 2021.

Arts and Science