Tsinghua team unveils AI chip “Taiji,” inspired by I Ching.

Breakthrough in AI Chip Development by Tsinghua University Research Team

Today (April 13), as a key element in training AI models and carrying out inference tasks, computing power is essential for artificial intelligence.

A recent achievement by a research team at Tsinghua University was published on the latest issue of “Science” on the early morning of April 12. They introduced a groundbreaking distributed breadth-intelligent optical computing architecture and developed the world’s first large-scale interferometric diffraction heterogeneous integrated chip named “Taichi,” achieving a universal intelligent computing power of 160 TOPS/W.

Tsinghua Team Unveils AI Chip Taichi Inspired by I-Ching

Inspired by the classic book “I Ching” during the development of the “Taichi” optical chip architecture, the team members took the inspiration from the phrase “Easy has Taichi, which produces two instruments.” They established a brand-new computational model that unleashed robust performance in optical computing.

Optical computing, as the name suggests, involves transforming the computing carrier from electrical to optical and utilizing light propagation in chips for calculations. With its extremely high parallelism and speed, it is considered one of the most potent competitive solutions for future disruptive computing architectures.

Optical chips possess advantages like high-speed and high parallel computing power, offering hope for supporting advanced AI applications such as large models.

According to the first author of the paper and Ph.D. candidate at the Department of Electronics, Zhiwu Xu, in the “Taichi” architecture, the top-down encoding-split-decoding-reconstruction mechanism simplifies complex intelligence tasks by breaking them down into multi-channel high-parallel sub-tasks. The constructed distributed ‘large receptive field’ shallow optical network addresses sub-tasks individually, overcoming the inherent computing errors of multi-layered deep cascade physics simulators.

The paper reports that the “Taichi” optical chip has an area efficiency of 879T MACS/mm and an energy efficiency of 160 TOPS/N. This chip empowers optical computing for the first time to achieve complex AI tasks like recognizing a thousand classes of objects in natural scenes and generating cross-modal content.

The “Taichi” optical chip is poised to provide computing power support for large model training inferences, general artificial intelligence, and autonomous intelligent unmanned systems.

Tsinghua Team Unveils AI Chip Taichi Inspired by I-Ching

Tsinghua Team Unveils AI Chip Taichi Inspired by I-Ching

Leave a Reply

Your email address will not be published. Required fields are marked *

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.