Light-controlled NPUs promise drastic energy savings in AI data centers

Light-controlled NPUs promise drastic energy savings in AI data centers

The new chip from Q.ANT is said to be 50 times faster than silicon-based chips.

The German company Q.ANT has begun production of a new computer chip that doesn’t run on electricity, but on light. This chip is designed to make AI applications and powerful data centers more energy-efficient. Light seems to be popular: Google also has a test project with internet signals via light beams.

Up to 30 times more efficient than classic chips

According to Q.ANT, the chip would consume up to 30 times less energy than silicon chips and has up to 50 percent faster computing power. The company has invested $15.1 million to convert an existing semiconductor factory into a new facility.

Because production takes place in a repurposed factory instead of a specialized production line, the company believes it can bring the technology to market much faster. The chip is also compatible with existing AI servers. Q.ANT hopes to introduce these chips as standard in AI data centers by 2030.

read also

Etap, Schneider Electric and Nvidia Develop Digital Twin for Energy Management in AI Data Centers

Photonic chips work based on light particles (photons), which can travel much faster than electrons in computer chips. They don’t emit heat and could thus greatly reduce the energy consumption of intensive workloads such as AI.

Q.ANT aims to produce 1,000 wafers per year thanks to a pilot production line. “With this pilot line, we accelerate time-to-market and lay the foundation for photonic processors to become standard coprocessors in high-performance computing,” said Michael Förtsch, CEO of Q.ANT, in a press release.