AI at the Speed of Light: Tsinghua’s Optical Chip OFE2 Redefines the Foundations of Computing
A research team from Tsinghua University has introduced a technology that may overturn our entire understanding of the computational nature of artificial intelligence. Their development — the Optical Feature Extraction Engine (OFE2) — performs computations not with electricity but with light.
This is not a metaphor. The new optical processor can process data at 12.5 GHz, completing complex machine-learning operations in just 250 picoseconds. In effect, these are computations at the speed of light, where information does not pass through transistors but refracts, interferes, and combines within photonic channels.
At the heart of the system lies a microchip in which traditional electronic circuits are replaced with photonic waveguides and nanostructures that control light flows. This allows signals to be processed in parallel, with almost no thermal losses and with energy consumption orders of magnitude lower than that of modern GPUs and TPUs.
In testing, OFE2 did more than demonstrate speed: it outperformed electronic counterparts in image-recognition accuracy and was even able to make trading decisions in real time — without latency, where milliseconds can cost millions.
This development is not just another chip. It is a new physical language for AI, where light becomes the carrier of intelligence. Optical computing, which scientists have pursued for decades, has finally stepped out of theory and laboratory prototypes into applied reality.
If transistors became the foundation of the digital era of the 20th century, OFE2 may become the symbol of the photonic era of AI, where intelligence runs not on currents but on beams.
More details — at tsinghua.edu.cn and phys.org
