This new AI chip is just the size of a grain of salt

The University of Shanghai for Science and Technology has developed a new AI chip to read data off fibre optic cables using a thousandth of the power requirement of traditional means.

Fibre optic cables can send data at incredible speeds, but it all needs to be decoded at the end by traditional computers, with all the power requirements and speed limitations that come with that. However, researchers at the University of Shanghai for Science and Technology (USST) have developed an AI chip that's smaller than a grain of salt, that can be placed directly on the end of a optic cable and decode the data in realtime. The results have been published in a paper for Nature Photonics.

Traditionally, although fibre cables are the fastest way to currently transmit data, the bottleneck comes at either end, with the requirement for power hungry data processing computers. This new AI chip has the potential to drastically reduce the power required, as well as increase the speed of processing.

The researchers used techniques developed by the University of California, called an all-optical diffractive deep neural network as the basis for their work. According to researcher Haoyi Yu in New Scientist, "the system can be thought of as a passive, well-trained neural network that physically manipulates light to perform computations"

The chip developed by the researchers reconstructed images of Roman numerals the size of a grain of pollen, and has the potential to reduce the resources needed to complete the task, compared to traditional methods, by an order of thousands. Additionally, the task can be completed as fast as light travels through the fibres.

Although currently the system requires custom integration of the chip based upon the specific task it will be asked to do, the principal could lead to more efficient and simplified devices such as endoscopes, and possibly integrated within quantum photonic chips that could form the backbone of a future quantum internet.

tl;dr

• The University of Shanghai for Science and Technology has created a new AI chip that decodes data from fibre optic cables with significantly lower power consumption than traditional methods.
• This chip, smaller than a grain of salt, can be placed directly on fibre optic cables, allowing for real-time data processing and potentially enhancing speed while reducing power requirements.
• By utilizing all-optical diffractive deep neural network techniques, the researchers demonstrated the chip's ability to reconstruct images, achieving processing speeds comparable to the speed of light in the cables.
• Although currently requiring custom integration for specific tasks, this technology could lead to more efficient devices in fields like endoscopy and contribute to the development of quantum photonic chips for a future quantum internet.