China Unveils Revolutionary Optical Processor: 2560 TOPS at 50 GHz

Scientists from the Shanghai Institute of Optics and Fine Mechanics at the Chinese Academy of Sciences have announced the creation of the world's first optical processor for ultra-high-parallel computing. The chip's theoretical performance reaches 2560 teraoperations per second (TOPS) at an optical clock speed of 50 GHz — one of the highest figures in the industry.

Unique Architecture: 100 Parallel Light Channels

Unlike traditional optical processors, which operate at a single wavelength, the Chinese design uses more than a hundred parallel optical channels. The laser beam is divided by compact ring resonators — soliton microcombs — into multiple spectral “teeth,” each of which transmits its own stream of bits. All channels pass through an integrated circuit, the size of which is comparable to a fingernail.

According to project manager Se Pen, it is this technology that has made it possible to maintain the miniature size of the chip while dramatically increasing its bandwidth and clock speed.

One of the key features of the optical approach is the absence of resistive heating, which is typical for electronic circuits. This allows parallel channels to be placed closer together, reducing energy losses and the risk of local overheating. The developers claim that they have achieved a bandwidth of over 40 nm and low energy losses with fully reconfigurable signal routing.

New Horizons for Artificial Intelligence

The chip's multi-lane structure makes it particularly promising for accelerating neural networks. AI models that perform huge arrays of similar operations can be easily scaled to the chip's architecture, allowing them to work faster and more energy-efficiently than when using traditional graphics processors.

In addition, low signal latency makes the processor ideal for real-time tasks, from high-frequency trading to swarm control of drones and intelligent peripheral devices.

Energy-Efficient Implementation of AI Computing in Highly Parallelized Environments

In addition to artificial intelligence, the new optical processor can be used in resource-intensive tasks such as medical imaging, scientific modeling, and signal processing. Researchers emphasize that the technology paves the way for new classes of computing that were previously impossible due to the thermal and energy limitations of traditional electronic circuits.