In order to solve the problem of large-scale optoelectronic intelligent computing, the research group of Associate Professor Fang Lu of the Department of Electronic Engineering of Tsinghua University and the research group of Academician Dai Qionghai of the Department of Automation abandoned the traditional electronic depth computing paradigm and found another way to pioneer the distributed breadth optical computing architecture, and developed a large-scale interfero-diffraction heterogeneous integrated chip Taichi to achieve 160 TOPS/W general intelligent computing.
In today's era of vigorous development of large-scale model general artificial intelligence, this scientific research achievement explores new inspiration, new architecture, and new path for high-performance computing power with the way of photons. On April 12, the relevant scientific research results were published in the latest issue of the international journal Science.
Turning "Deep" into "Wide": Distributed Breadth Optical Computing Architecture.
As an emerging computing modality, intelligent optical computing has shown far more performance and potential than silicon-based electronic computing in the post-Moore era. However, its computational tasks are limited to simple character classification, basic image processing, etc. The pain point is that the computing advantage of light is trapped in an unsuitable electrical architecture, and the computing scale is limited, which cannot support the intelligent computing of complex large models that require high computing power and high energy efficiency.
Two instruments and one element: interference-diffraction fusion computing chips.
According to reports, the computing energy efficiency of the Taiji light chip exceeds that of the existing intelligent chips by 2-3 orders of magnitude, and will provide computing power support for the intelligent analysis of the speed of light in tens of billions of pixel large scenes, the training and reasoning of tens of billions of parameters and large models, and the milliwatt-level low-power autonomous intelligent unmanned system.
Source: Science and Technology Daily reporter Hua Ling