December 22, 2025 by Paul Arnold, Phys.org
Collected at: https://techxplore.com/news/2025-12-optical-chip-boost-tier-nvidia.html
Scientists in China have unveiled a new AI chip called LightGen that is 100 times faster and 100 times more energy efficient than NVIDIA chips, the leading supplier of AI chips worldwide. Instead of using electricity to move information, this new optical chip relies on light to perform complex generative tasks.
Traditional general AI models, such as ChatGPT and Stable Diffusion, run on everyday silicon chips and require massive amounts of computing power and electricity, which can generate significant heat. For particularly complex tasks, these chips can struggle with the workload, resulting in slow processing times.
Mimicking the human brain
Photons of light are already known to be much better than electrons for advanced computer chips because they are faster, consume less power, and produce much less heat. Even so, current photonic chips cannot keep up with the massive demands of AI. So, researchers at Shanghai Jiao Tong University and Tsinghua University developed a new kind of fully optical chip with over two million photonic “neurons.”
These neurons are specialized components designed to mimic the human brain by processing information all at once (parallel processing), as opposed to the transistors in standard chips, which act like simple on-off switches that process data serially.
The team created LightGen specifically for large-scale generative AI, such as the creation of high-resolution images, 3D scenes and videos. Typically, older photonic chips have to break them into small “patches” to process them because they are flat. This can result in disjointed and lower-quality images. But LightGen works differently because it has a 3D structure with photonic neurons stacked in layers. This is why it can process an entire complex image at once, rather than piece by piece.
Toward sustainable AI
The research team tested their new chip design on several tasks that usually require massive amounts of energy and time. These included creating high-resolution images of animals and nature, as well as short high-definition videos. It did this 100 times faster, using 100 times less energy than today’s top-tier chips, such as the NVIDIA A100. LightGen also packs 100 times more computing power into the same space. Additionally, it performed some tasks as well as, and even better than, top AI systems such as Stable Diffusion and StyleGAN.
“LightGen provides a new way to bridge the new chip architectures to daily complicated AI without impairment of performance and with speed and efficiency that are orders of magnitude greater, for sustainable AI,” commented the team in their paper published in the journal Science.
While this new chip marks a massive leap forward for generative AI, LightGen is still a prototype. One of the next steps for the researchers is to scale up their design further to handle even larger and more complex AI models.
More information: Yitong Chen et al, All-optical synthesis chip for large-scale intelligent semantic vision generation, Science (2025). DOI: 10.1126/science.adv7434
Journal information: Science
Leave a Reply