High-performance photonic AI chip design

A team of scientists at Tsinghua University, China has designed an artificial intelligence (AI) chip based on photonics instead of traditional electronic semiconductor components, paving the way for large-scale photonic computing and effective AI applications in the real world (photo).

Accordingly, a large-scale photonic chiplet (chips made up of many different blocks) along with a distributed optical computing architecture called Taichi. Taichi has achieved high accuracy on complex classification tasks, such as the 100-category ImageNet dataset and the 1,623-category Omniglot dataset.

The chip is also capable of high-precision tasks such as composing music and creating paintings, according to the study. Furthermore, Taichi achieved an energy efficiency of 160 tera operations per second per watt, a significant improvement in performance over current photonic integrated circuits and two orders of magnitude higher than traditional AI chips in terms of energy efficiency.

“We predict that Taichi will accelerate the development of more powerful optical solutions, such as critical support for the platform model and the new era of generative AI,” said Fang Lu, an expert at Tsinghua University.

LAM DIEN

Source