Chinese scientists at the Shanghai Institute of Optics and Fine Mechanics (SIOM) have developed Meteor-1, the first ultra-parallel optical computing chip harnessing over 100 wavelengths of light to execute up to 2560 TOPS at a 50 GHz optical clock rate. This architecture leverages soliton microcomb sources and a Mach-Zehnder interferometer mesh, enabling massive on-chip parallelism without increasing frequency or footprint. Offering an elegant, energy-efficient alternative to traditional GPU-based systems.
Unlike electronic processors that rely on voltage-driven transistors, Meteor-1 computes using photons, dramatically reducing heat and power consumption. It’s theoretical performance rivals top tier NVIDIA GPUs, with vastly lower energy demands, positioning it as a disruptive technology for AI training, real-time signal processing, edge computing (drones, satellites), and high-throughput scientific simulations.
Meteor-1 is currently a lab-grade prototype; full neural network deployment still requires hybrid electronic-photonic integration for non-linear operations. Meanwhile, China has also launched a pilot production line for photonic chips using thin-film lithium niobate wafers – marking the first steps toward scalable manufacturing for AI, 6G, and quantum infrastructure. As export restrictions hinder access to conventional semiconductor tech, optical computing offers a strategic path to hardware autonomy.
Quanta Reads 