New photonic processor for artificial intelligence

The systems that provide physical support for the deep neural networks used in today’s most demanding artificial intelligence applications have become so large and complex that they are beginning to run into the limits of traditional electronic computing hardware.

Photonic hardware, which can perform the calculations required by machine learning (a form of artificial intelligence) with light, rather than electrons, offers a faster, more energy-efficient alternative.

However, there are some types of neural network calculations that a photonic device cannot perform, as they require the use of off-chip electronics or other techniques that hinder speed and efficiency.

Drawing on a decade of research, a team led by Saumil Bandyopadhyay of the Massachusetts Institute of Technology (MIT) in the United States has developed a new photonic processor that overcomes these obstacles.

Bandyopadhyay and his colleagues have fabricated and demonstrated a photonic processor that can perform all the important calculations of a deep neural network optically on-chip.

In tests, the optical device was able to perform key calculations for a machine learning classification task in less than half a nanosecond, with an accuracy of over 92%, a performance comparable to that of traditional hardware.

Artistic recreation of photonic processor. (Image: Amazings/NCYT)

The chip, made up of interconnected modules that form an optical neural network, is manufactured using processes already used in industry, which would facilitate large-scale production of processors of this type and their integration into existing electronics. (Fountain: NCYT by Amazings)