Every second, the data behind billions of emails, videos and artificial intelligence queries travels as pulses of light through fibre-optic networks. These signals pass through photonic chips. Photonic chips are small components that direct and combine light so information moves efficiently across networks.
Photonic chips have limitations. They struggle with certain important light-processing tasks. Operations such as converting signals or making them stronger often require extra components. These extra parts take up space, use energy and create heat.
The fast growth of generative artificial intelligence (AI) is increasing the problem. Unlike simple searches, generative artificial intelligence systems require many exchanges between processors. Each exchange needs more signal conversions. This raises energy use in data centres, which already account for about 2 percent of global electricity consumption.
Researchers at Polytechnique Montréal have developed a new approach.
How a new material enables direct light functions
The researchers identified an organic molecule called triphenylamine–dicyanoquinoxaline, known as TPA-QCN. This molecule interacts strongly with light. When deposited as a thin film on silicon through vacuum evaporation, its molecules spontaneously align in an ordered way. This alignment creates a second-order optical nonlinear response. This property allows different light beams to interact with each other as they travel through the material.
As a result, functions such as amplification and modulation of light can occur directly on the chip. The material is compatible with standard manufacturing processes used in the photonics industry and works at low temperature and low cost.
In a demonstration, the researchers built a device that converts infrared light used in telecommunications into visible red light directly on the chip. The advance could lead to better modulators, amplifiers and light sources for quantum technologies. By reducing energy-consuming conversion steps, it may help data centres manage the demands of artificial intelligence more sustainably.
The researchers have described the methods and results of this study in a paper published in Science Advances.