The boffins revealed the initial Taichi chip in April 2024. Unlike traditional chips that rely on electronic components, this compact, modular device is powered by photons or particles of light. These photons activate tiny onboard electrical switches that turn on or off when voltage is applied.
However, the Taichi-II boasts a 40 per cent improvement in classification tasks, which involve sorting and identifying various types of information. Additionally, it delivers a "six orders of magnitude" (i.e., a million-fold) enhancement in energy efficiency under low-light conditions.
The researchers achieved this significant performance boost by training the AI directly on the optical chip rather than relying on digital simulations—a method they termed "fully forward mode." Their findings were published in a study in the journal Nature on 7 August.
Fully forward mode is an AI training method in which data moves in only one direction—forward. This contrasts with traditional training methods, where data is processed in multiple iterative steps. As light passes through the chip, it interacts with tiny components that adjust its direction and modulate its phase and intensity.
These interactions cause immediate changes to the AI model's parameters, enabling it to learn in real-time without repeated processing.
This allows photon chips to operate even faster than before. Light-based chips already have significant advantages over conventional chips, being far less energy-intensive and capable of performing calculations much quicker. This is because photons, unlike electrons, can travel at the speed of light and do not generate heat as they move through the chip, leading to faster and more efficient processing.