New AI chip powered by light could save tons of energy
A team at the University of Florida has built a new kind of AI chip—the photonic joint transform correlator (pJTC)—that uses light for its core processing tasks, dramatically reducing reliance on electricity for computation.
This means AI tasks like recognizing images could use up to 100 times less energy than with regular chips.
How the chip works
The chip turns data into laser light, which passes through tiny lenses etched on its surface.
These lenses handle the heavy math before converting everything back to digital signals, making the process super efficient.
The pJTC matched traditional chips in accuracy
In tests, the pJTC matched traditional chips in accuracy but used way less power—pulling off 305 trillion operations per second per watt.
Since it's compact and easy for manufacturers to adopt, this tech could shake up how AI hardware is built, from big data centers to everyday devices.