techno-science.net, Sept. 17, 2024 –
A new chip could soon revolutionize the world of artificial intelligence. In China, researchers have developed a new type of processor, using carbon nanotubes instead of silicon.
This innovation, recently published in Nature Electronics, aims to solve a major issue: the energy consumption of artificial intelligence models. Current systems, which are highly computation-intensive, consume vast amounts of energy. But this new chip could potentially change that.
At the heart of this advancement are carbon nanotubes, microscopic structures made of carbon atoms arranged in hexagons. They offer better electrical conductivity than silicon, thereby reducing energy consumption.
The new chip, a tensor processing unit (TPU), contains 3,000 carbon nanotube transistors. These transistors are organized into a systolic array, a configuration that enables the parallel processing of multiple data, increasing computational speed.
Tests have shown that this TPU consumes only 295 microwatts of energy, while performing a trillion operations per watt. This represents a significant step forward compared to current chips, paving the way for more efficient and energy-saving AI systems.
The researchers tested their chip by training a neural network on image recognition tasks. The result? An 88% accuracy, with minimal energy usage. These impressive performances could indeed signal the beginning of a new era for artificial intelligence.
The Chinese team plans to continue their research to further improve the performance of this chip and envisions integrating it with silicon-based processors, thereby creating even more powerful hybrid systems.