Tsinghua University's Breakthrough in Intelligent Optical computing

TapTechNews August 8th news, according to the official news of Tsinghua University, the research group of Professor Fang Lu from the Department of Electronic Engineering and the research group of Academician Dai Qionghai from the Department of Automation of Tsinghua University have taken a different approach and initiated the first fully-forward intelligent optical computing training architecture and developed the Taiji-II optical training chip, achieving efficient and accurate training of large-scale neural networks in optical computing systems.

This research result, titled Fully-Forward Training of Optical neural networks, was published online in the Nature journal on the evening of August 7th, Beijing time.

TapTechNews found out that the Department of Electronics of Tsinghua University is the first unit of the paper, Professor Fang Lu and Professor Dai Qionghai are the corresponding authors of the paper, Tsinghua University's doctoral student Xue Zhiwei from the Department of Electronics and postdoctoral Zhou Tiankuang are the co-first authors, and doctoral student Xu Zhihao from the Department of Electronics and Dr. Yu Shaoliang from Zhijiang Laboratory participated in this work. This project is supported by the Ministry of Science and Technology of China, the National Natural Science Foundation of China, the Beijing National Research Center for Information Science and Technology, and the Tsinghua University-Zhijiang Laboratory Joint Research Center.

The reviewer of Nature pointed out in the review comment that The idea presented in this article is very novel, and the training process of such optical neural networks (ONN) is unprecedented. The proposed method is not only effective but also easy to implement. Therefore, it is expected to become a widely adopted tool for training optical neural networks and other optical computing systems.

According to the official introduction of Tsinghua University, in recent years, intelligent optical computing with high computing power and low power consumption has gradually stepped onto the stage of computing power development. The general intelligent optical computing chip Taiji has for the first time promoted optical computing from principle verification to large-scale experimental application and has a system-level energy efficiency of 160 TOPS/W, but the existing optical neural network training relies heavily on the GPU for offline modeling and requires precise alignment of the physical system.

According to the first author, Xue Zhiwei, a doctoral student in the Department of Electronics, under the Taiji-II architecture, the backpropagation in gradient descent is transformed into the forward propagation of the optical system, and the training of the optical neural network can be achieved by using two forward propagations of data-error. The two forward propagations have the inherent alignment characteristic, ensuring the accurate calculation of the physical gradient. Since there is no need for backpropagation, the Taiji-II architecture no longer depends on electrical computing for offline modeling and training, and the accurate and efficient optical training of large-scale neural networks has finally been realized.

The research of the paper shows that Taiji-II can train various different optical systems and has shown excellent performance in various tasks:

In the field of large-scale learning: It has broken through the contradiction between computing accuracy and efficiency, and increased the training speed of optical networks with millions of parameters by an order of magnitude, and the accuracy rate of representative intelligent classification tasks has increased by 40%.

In the complex scene intelligent imaging: It has achieved all-optical processing with an energy efficiency of 5.40×10^6 TOPS/W in the weak light environment (the light intensity per pixel is only sub-photon), and the system-level energy efficiency has increased by six orders of magnitude. Intelligent imaging with a kilohertz frame rate has been achieved in the non-line-of-sight scene, and the efficiency has increased by two orders of magnitude.

In the field of topological photonics: It can automatically search for non-Hermitian singular points without relying on any model priors, providing a new idea for efficient and accurate analysis of complex topological systems.

TapTechNews attached the paper link:

https://www.nature.com/articles/s41586-024-07687-4