China’s Taichi photonic chip ushers in light-speed AI revolution

A team of researchers from Tsinghua University has unveiled a revolutionary photonic chip that could propel artificial intelligence (AI) into a new era of unprecedented speed, efficiency and capability.

China’s Taichi photonic chip ushers in light-speed AI revolution Dashveenjit is an experienced tech and business journalist with a determination to find and produce stories for online and print daily. She is also an experienced parliament reporter with occasional pursuits in the lifestyle and art industries.


A team of researchers from Tsinghua University has unveiled a revolutionary photonic chip that could propel artificial intelligence (AI) into a new era of unprecedented speed, efficiency and capability.

Dubbed ‘Taichi’, the light-based chip leverages integrated photonic circuits to transmit data at blistering speeds while dramatically reducing energy consumption compared to traditional electronics.

It is important to understand that the widespread growth of edge devices and data centres has created significant challenges in terms of bandwidth and efficiency when it comes to image processing, transmission, and reconstruction which plays a significant role in the field of information technology today. So much so that the constant need to convert serial signals between optical and electrical domains, along with the increasing strain on electronic processors, has become a major obstacle for end-to-end machine vision. 

Unprecedented energy efficiency and speed

Taichi builds upon an earlier photonic chip called the optical parallel computational array (OPCA) developed by the same Tsinghua team. While the OPCA demonstrated unprecedented nanosecond processing speeds for image data, Taichi vastly extends this paradigm to general computing workloads. It can process, transmit and reconstruct images at blistering nanosecond speeds – around a million times faster than current methods.

According to the research paper, the OPCA chip represents a fundamental shift in how machine vision systems operate. Instead of first converting optical data into digital electrical signals, as is conventionally done, the OPCA performs all sensing and computing optically on the same integrated chip.

This optical computing approach eliminates the need for energy-intensive optical-to-electronic conversions and bypasses the speed limitations of electronic processors. Now, the OPCA chip achieves blistering processing rates of up to 100 billion pixels per second and a miniscule response time of just 6 nanoseconds.

Overall, as an AI chip, Taichi is over 1,000 times more energy-efficient than the high-performance Nvidia H100 GPU, which is notable given current trade restrictions that prevent the H100 from being available in China​. 

Emphasising how the frequent conversion between optical and electrical signals has been a major bottleneck on improving machine vision capabilities, the researchers at Tsinghua University reckon that the Taichi chip could help unlock a new era of ultrafast edge intelligence for AI applications like autonomous vehicles, industrial inspection, and robotics.

State-of-art AI features and breakthrough architecture

Taichi performs especially well in a broad set of artificial general intelligence (AGI) tasks, such as image recognition or content generation. The chip demonstrated a 91.89% accuracy in correctly identifying images through 1,623 categories in the Omniglot dataset, exhibiting its capacity to tackle more complex tasks.

Moreover, Taichi can deliver exceptional quality content from making an image change its artistic style, or even compose music,  showcasing its versatility and creative potential. Finally, it is the unique architecture of Taichi that distinguishes it from a typical photonic chip. Rather than stacking photonic integrated circuits (PICs), the Tsinghua team developed a new distributed computing architecture that groups PICs in clusters.

This configuration allows for parallel processing, efficiently distributing computing resources across multiple independent clusters to handle subtasks effectively. This architecture addresses the problem of error accumulation across layers, which is often encountered in deep learning structures. It also improves computing capacity and reliability.

While still confined to research labs, the pioneering work on Taichi points towards a future of AI systems powered by light. As demand escalates for intelligent automation at the edge, data centres, and beyond, photonic co-processors could provide a revolutionary solution to the intensifying computational needs. Perhaps, Taichi paves the way for large-scale photonic computing to enable advanced AGI capabilities that would be prohibitively expensive with solely electronic approaches.

Want to learn more about edge computing from industry leaders? Check out Edge Computing Expo taking place in Amsterdam, California and London. 

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

Tags: , ,

View Comments
Leave a comment

Leave a Reply