AI Revives the Drive for Optical Computing

Written by

in

The energy bottleneck in AI inference is prompting multiple startups to explore the use of photonic computing to accelerate matrix operations, solve equations, or process network data. These companies have set targets such as achieving 10 to 80 times greater energy efficiency than GPUs, but data encoding, analog-to-digital conversion, nonlinear computation, and system scaling remain significant obstacles. Industry valuations have risen from US $0.4 billion in 2022 to US $2.22 billion, and future data centers may assign tasks best suited to CPUs, GPUs, and photonic processors to each respectively.

Open hardware box and close-up of intricate optical chip circuitry.
Optical neural network: LEDs through lenses and pixelated attenuator to photodetectors
Microscope probe positioned over a square microchip on a green circuit board
Diagram of light pulses in a waveguide passing phase shifters and photodetectors
Optics lab setup with red laser beams reflecting off lenses and mirrors

Source:IEEE Spectrum · Read original

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *