China Unveils Meteor-1 Optical Chip with 2,560 TOPS Performance

China has introduced its first highly parallel optical computing chip, named Meteor-1, marking a significant advancement in leveraging light-based hardware to manage extensive parallel workloads. This development is part of China's broader strategy to reduce its reliance on US-made technology, particularly in the realm of advanced AI tech. The new chip is designed to accelerate AI and data center tasks, addressing the growing computational demands faced by these sectors.

Developed by teams at the Shanghai Institute of Optics and Fine Mechanics (SIOM) and Nanyang Technological University, the Meteor-1 chip boasts a theoretical peak performance of 2,560 TOPS (tera-operations per second) at a 50 GHz optical clock. This places it in direct competition with Nvidia’s top GPUs, offering a domestically produced solution to meet the increasing computational needs and US chip restrictions.

Historically, optical computing efforts have focused on expanding the dimensions of the internal matrix used for calculations. Larger matrices theoretically allow for more data to be processed in parallel, but they face significant engineering challenges, including complex chip layouts, precise waveguide alignment, and high fabrication costs. Previous efforts by TSMC and academic groups have shown promise in laboratory settings but have struggled to transition into commercial-grade solutions.

Another approach has been to increase the oscillation rate of the light itself. Higher optical frequencies can deliver faster computation but also amplify signal losses, thermal noise, and component tolerances. Until now, no team has successfully combined large matrix scales and ultra-fast optics in a single system without encountering severe trade-offs that degrade real-world performance.

Meteor-1 takes a different approach by multiplying the number of simultaneous tasks rather than enlarging individual components. The chip incorporates over 100 distinct frequency channels within one photonic platform, enabling a hundredfold or greater increase in “optical computility” without expanding the chip’s footprint. This high-order parallelism provides a practical path for next-generation light-powered processors.

With US export restrictions effectively banning Nvidia’s RTX 4090 and RTX 5090 from China, the Meteor-1 chip arrives at a critical moment. Conventional electronic semiconductors are encountering fundamental limits, including heat dissipation, quantum tunneling, and unsustainable power draws. Optical chips, on the other hand, bypass many of these barriers, offering ultra-high speed, broad bandwidth, reduced energy consumption, and minimal latency.

Meteor-1’s architecture is entirely home-designed. Its on-chip light source uses a micro-cavity optical frequency comb that covers more than 80nm of spectrum, spanning upwards of 200 wavelengths. This innovation effectively replaces hundreds of discrete lasers, reducing system complexity, power demands, and costs. The core computing die itself offers a transmission bandwidth above 40nm, facilitating low-latency, massively parallel operations. Coupled with a bespoke driver board featuring over 256 channels for precise signal modulation, the system executed more than 100 simultaneous tasks in benchmark tests, all at a 50 GHz clock, yielding 2,560 TOPS.

Han Xilin, a key researcher on the project, notes that Meteor-1’s cost-performance metrics could soon rival electronic GPUs. Lead researcher Xie Peng, who holds a PhD from the Massachusetts Institute of Technology and has conducted research at Oxford and NTU, attributes the rapid progress to SIOM’s modular team structure under the Chinese Academy of Sciences. “Each critical subsystem had its own dedicated expert, allowing us to integrate full-chain innovation from foundational research through system assembly in a remarkably short span,” Xie Peng said.

Looking ahead, the research group believes their parallel-first design could outpace electronic chips in terms of efficiency, power draw, and latency, meeting AI’s insatiable compute appetite while spawning novel applications in real-time data analysis, autonomous systems, and scientific modeling.