AI

China unveils AI chip that is 3,000 times faster than Nvidia's A100

According to a recent study, Chinese scientists have developed a chip that is substantially quicker and more energy efficient than current high-performance AI processors while doing activities such as image recognition and autonomous driving.

Although the new chip will not be used in devices such as computers or smartphones right away, it may soon be used in wearable devices, electric cars, or smart factories, helping to boost China's competitiveness in the mass application of artificial intelligence, according to researchers in a paper published in the journal Nature.

The country is attempting to catch up with the United States in the AI race after Washington imposed a series of restrictions on China's access to technology, including sophisticated semiconductors.

The new chip, dubbed the All-Analogue Chip Combining Electronics and Light (ACCEL), is light-based and employs photons, a sort of basic particle, to compute and send data at a quicker rate.

The concept of a light-based chip is not new, but existing devices rely on electric current for calculation since photons are more difficult to manage.

In a laboratory test, the new device achieved a computation speed of 4.6 PFLOPS (peta-floating point operations per second), which is 3,000 times faster than Nvidia's A100, one of the most extensively used commercial AI chips. Researchers discovered that the Chinese microprocessor uses 4 million times less energy.

The A100 is subject to US sanctions against China, and it, like other powerful AI chips, is manufactured using advanced lithography machines that China does not have access to.

Instead, China's Semiconductor Manufacturing International Corporation produced the new device utilising a cheap 20-year-old transistor production technology.

"The performance [of the chip] could be further optimized through improvements in the building process or by adopting more expensive fabrication processes under 100 nanometers," the research team from Tsinghua University's automation and electronic engineering departments wrote in a paper published last week.

Photonic chips, as opposed to semiconductor chips, make use of the inherent physical qualities of light by replacing transistors with ultramicroscopes and electrical impulses with light signals.

"Deployment of photonic computing systems was previously difficult due to complicated structural design and vulnerability to noise and system errors." The researchers "innovatively introduced a computing framework that fuses photonic and analogue electronic computing," according to a Tuesday post on Tsinghua's website.

Tsinghua also stated that employing light signals boosted energy efficiency significantly, and that "the energy required to operate existing chips for an hour could power ACCEL for over 500 years."

Its low power consumption may also aid in overcoming the issue of heat dissipation, which is now a substantial impediment to further miniaturizing integrated circuits.

It also has certain benefits for AI vision tasks since passive light from the environment contains information, allowing it to calculate immediately during the sensing process.

The National Key Research and Development Programme of the Chinese Science Ministry and the National Natural Science Foundation of China financed the study.

MakeSens, a Beijing-based semiconductor design business co-founded by one of the project's researchers, was also engaged in the chip's creation. In May, the business introduced a low-power device based on analogue computing.

According to Dai Qionghai, one of the research team's co-leaders at Tsinghua, "developing a new computing architecture for the AI era is a pinnacle achievement." The most essential issue, however, is to translate this new infrastructure into practical applications that address critical national and societal requirements, which is our obligation."

Questions concerning the chip's commercial possibilities have yet to be answered by the research team.