中國科學家研發出執行人工智慧任務速度比 Nvidia A100 快 3000 倍的晶片：研究

A team of Chinese researchers have developed a chip that outperforms current high-performance AI chips in certain tasks, such as image recognition and autonomous driving, according to a recent study. While the new chip isn't ready to replace those found in computers or smartphones, it could soon be integrated into wearable devices, electric vehicles, or smart factories, potentially enhancing China's position in the widespread application of artificial intelligence.

The new chip, dubbed the All-Analogue Chip Combining Electronics and Light (ACCEL), operates using light, specifically photons, to compute and transmit information, resulting in faster processing speeds. The concept of a light-based chip isn't novel, but existing chips rely on electric current for calculations due to the difficulty in controlling photons. In a lab test, the ACCEL chip achieved a computing speed 3,000 times faster than Nvidia's A100, a widely used commercial AI chip. Additionally, the ACCEL chip consumes significantly less energy.

The ACCEL chip was manufactured by China's Semiconductor Manufacturing International Corporation using an inexpensive, two-decade-old transistor fabrication process. The chip's performance could be further enhanced by refining the building process or adopting costlier fabrication processes. Unlike semiconductor chips, photonic chips like ACCEL utilize the inherent physical properties of light, replacing transistors with ultramicroscopes and electrical signals with light signals. However, the chip's analogue computing architecture restricts its application to specific problems and it cannot run various programs or compress files like general computing chips. The chip excels in tasks involving high-resolution image recognition, lowlight computation, and traffic identification.