April 13 News: As one of the three pillars of artificial intelligence, computational power is crucial for training AI models and performing inference tasks.
The latest achievement from the research team at Tsinghua University was unveiled in the April 12th issue of Science. They pioneered a distributed breadth-intelligent optical computing architecture and developed the world’s first large-scale interference diffraction heterogeneous integrated chip named “Taiji,” achieving a universal intelligent computation of 160 TOPS/W.
According to reports, the inspiration for the “Taiji” optical chip architecture came from the classic text I Ching. Taking inspiration from the phrase “The Changes have the Taiji, which engenders the two primary forces,” the team established a new computational model, unleashing powerful performance in optical computing.
Optical computing, as the name suggests, transforms the computing carrier from electricity to light, utilizing light propagation in chips for calculations. With its extremely high parallelism and speed, optical computing is considered one of the most promising candidates for future disruptive computing architectures.
Optical chips possess advantages in high-speed and high-parallel computation, offering hope for supporting advanced artificial intelligence applications such as large-scale models.
According to Xu Zhiwu, the first author of the paper and a doctoral student in the Department of Electronics, in the “Taiji” architecture, the top-down encoding-splitting-decoding-reconstruction mechanism simplifies complex intelligent tasks into multiple channels of highly parallel subtasks. The distributed “large receptive field” shallow optical network deals with subtasks separately, breaking through the inherent calculation errors of multi-layer cascade physical simulators.
The paper reports: “Taiji” optical chip boasts an area efficiency of 879T MACS/mm and an energy efficiency of 160 TOPS/N. It enables natural scene recognition of thousands of objects and complex AI tasks such as cross-modal content generation for the first time in optical computing.
The “Taiji” optical chip is expected to provide computational support for large-model training, inference, general artificial intelligence, and autonomous intelligent unmanned systems.
Source: KuaiKeji (Fast Technology)