Peking University launches carbon nanotube-based tensor processing unit for energy-efficient AI (Image source: Stan Hutter, Unsplash)
Peking University researchers have developed the world’s first tensor processing unit (TPU) using carbon nanotube transistors, a remarkable achievement in the quest for energy-efficient AI hardware. This innovative TPU features high efficiency and low power consumption, demonstrating the potential of carbon nanotube technology to revolutionize the future of AI computing.
Researchers at Peking University have achieved an impressive feat: they have developed the world’s first tensor processing unit (TPU) using carbon nanotube transistors. This is a major step forward in the quest for energy-efficient AI hardware.
We have successfully developed the world’s first tensor processor chip (TPU) based on carbon nanotubes. We were inspired by the rapid development of artificial intelligence applications as well as Google’s TPU.
The team refined their manufacturing process to make the semiconductors 99.9999% pure with super-clean surfaces, resulting in transistors with high current densities and reliable performance. They ran system-level simulations and found that an 8-bit TPU made with these nanotube transistors could operate at 850 MHz and deliver an energy efficiency of 1 tera-operation per second per watt, while consuming less power.
When they assembled a five-layer convolutional neural network using this TPU, it was able to achieve 88% accuracy in recognizing MNIST images while requiring only 295μW of power. This is much better than what we have seen with current convolutional acceleration hardware.
Of course, the practical use of this current 180nm-class TPU might be a bit limited, but the researchers believe their work is a big step toward creating the next generation of energy-efficient AI hardware that relies on carbon nanotube technology.
Translator: Ninh Ngoc Duy – Editorial Assistant – 426843 articles published on Notebookcheck since 2008
Please share our article, every link counts!
