World's first carbon-nanotube-based tensor processor chip, used in high-performance AI computing

https://newsen.pku.edu.cn/news_events/news/research/14371.html

"replaces silicon-based semiconductors... systolic array architecture leveraging carbon nanotube transistor and tensor operation properties... 10X increased speed, reduced power consumption, extends Moore's Law... tensor processing unit based on 3K carbon nanotube field-effect transistors performing convolution operations and matrix multiplication... systolic array supporting parallel 2-bit integer multiply-accumulate operations... 5-layer convolutional neural network 88% accurate with only 295 μW power consumption"