China has successfully produced a chip ACCEL that is 3,000 times faster than Nvidia’s A100 and consumes 4 million times less power.

According to SCMP, Chinese scientists have successfully developed a chip that is both faster and significantly more energy-efficient compared to current AI chips when performing certain tasks such as image recognition and autonomous driving. While the new product cannot immediately replace the chips currently used in devices like computers or smartphones, it is expected to be applied in wearable devices, electric cars, or smart factories in the near future.

The new chip, called the Electronic and Photonic Integrated Circuit (ACCEL), is based on light and utilizes photon particles for computation and information transmission. It achieves a computing speed of 4.6 PFLOPS, which is over 3,000 times faster than one of the most widely used commercial AI chips, Nvidia’s A100. According to SCMP, the China chip also consumes less energy, approximately 4 million times less.

“The chip’s performance can be further optimized through improvements or the application of costly manufacturing processes,” said an expert representative.

Unlike semiconductor chips, quantum chips leverage the physical properties of light and offer higher efficiency. “Deploying quantum computing systems used to be a challenge due to complex structural design, susceptibility to noise, and system errors. The team has creatively introduced a framework that combines quantum and electronic computing,” Tsinghua’s website states, noting that the use of light significantly increases energy efficiency. Furthermore, “the energy required to operate existing chips for one hour could power the ACCEL chip for over 500 years,” according to SCMP.

However, the new chip is limited in its ability to handle specific tasks. It also cannot run multiple different programs or compress files like conventional computing chips.

According to Tsinghua’s website, the tasks that the new chip can perform include high-resolution image recognition, computation under low-light conditions, and traffic flow identification. It also has certain advantages when performing tasks related to AI vision because passive light from the environment enables direct computation during the sensing process.

According to SCMP, the project is funded by the National Key Research and Development Program of China Ministry of Science and Technology and the National Natural Science Foundation of China. MakeSens, a chip design company based in Beijing and co-founded by one of the researchers involved in the project, is also involved in chip development. The company recently launched a low-power chip utilizing similar computing techniques in May.

“Developing a new computing architecture for the AI era is a remarkable achievement. However, a more significant challenge lies in real-world applications, addressing the significant needs of the nation and the community. That is our responsibility,” said Dai Qionghai, one of the co-leaders of the research group.

Currently, the research team has not yet answered questions regarding the commercial prospects of this chip.

Source: SCMP

Read more: AI generation – a new battleground in phone chip design.