Getting an all-optical AI to deal with non-linear mathematics
The issue is that this cascading needs huge parallel calculations that, when done on basic computer systems, take lots of energy and time. Bandyopadhyay's group feels this issue can be fixed by carrying out the comparable operations utilizing photons instead of electrons. In photonic chips, info can be encoded in optical homes like polarization, stage, magnitude, frequency, and wavevector. While this would be exceptionally quick and energy-efficient, constructing such chips isn't simple.
Siphoning light
"Conveniently, photonics ended up being especially proficient at direct matrix operations," Bandyopadhyay claims. A group at MIT led by Dirk Englund, a teacher who is a co-author of Bandyopadhyay's research study, showed a photonic chip doing matrix reproduction completely with light in 20...