Page 13 - Spring 2024

FURTHER ADVANCES IN AI

New Chip to provide AI Computing at Light Speed Potential application of the technology is
envisaged most notably in graphics processing
Engineers at the University of Pennsylvania have units, for which there has been a surge in demand
developed a chip for AI computing that uses light following the growth in AI development.
waves rather than electricity for computation,
marking a ‘monumental step forward’ in With the capability for numerous computations
computing technology. to occur simultaneously there is also no necessity
to store sensitive data in a computer’s working
The chip uses silicon-photonic (SiPh) technology
memory, such that computers that are powered
to merge the pioneering work of Benjamin
by this technology can become ‘virtually
Franklyn Medal Laureate and H. Nedwill Ramsey
impervious’ to hacking.
Professor Nader Engheta in nanoscale material
manipulation for mathematical computations Reference: West, P., ‘New Chip opens Door to AI Computing at
using light, with the cost-effective and widely Light Speed’, Electronic Specifier, 23rd. February
available silicon used in mass computer chip
production.

Light’s interaction with matter is identified as a key
pathway to transcending the limitations of chips,
the operational principles of which date back to
the 1960s despite periodic refinements.

Instead of using a silicon wafer of uniform
height the silicon is made thinner, around 150
nanometres, in selected places. The resulting
height variations, which are achieved without
incorporating additional materials, manipulate the
propagation of light through the chip, allowing the
chip to scatter light in precise patterns, so making
for the performance of calculations at the speed
of light.
The aim is to create a platform that can execute
vector matrix multiplication, which is an essential
mathematical operation in neural networks and
hence the AI tools that are powered by such
networks. This operation is therefore crucial for
both the development and functionality of AI
applications.

13
Subscribe to Electron here https://membermojo.co.uk/ie