This Chinese breakthrough might boost AI speed by 100 times
What's the story
Chinese researchers have made a major breakthrough in artificial intelligence (AI) technology. They have developed an all-optical interconnect system that connects standard electronic chips, increasing distributed inference speeds by over 100 times. This new system also reduces the computational resources needed to just one-ninth of what is typically required. The study was published in the National Science Review journal, with Shu Haowen and Wang Xingjun from Peking University as corresponding authors.
Technology
FPGA chips as building blocks
The researchers used field-programmable gate array (FPGA) chips as the "Lego" building blocks for their new system.
FPGAs are programmable devices that are widely used in applications requiring high parallel-processing capabilities such as missile guidance, autonomous driving, and data centers.
The "joints" connecting these FPGAs are two custom-designed communication hardware components, including a silicon photonic transceiver chip operating at 400 gigabits per second.
Signal conversion
Silicon photonic transceiver chip
The silicon photonic transceiver chip is responsible for converting electrical signals into optical ones and vice versa.
This conversion is key to the new system's ability to boost AI speed while reducing computational needs.
The research team's innovative approach could pave the way for more efficient AI systems by using less energy and resources without compromising performance.
Future implications
Revolutionizing AI systems
The research team's innovation could revolutionize the way we approach AI systems.
By using optical interconnects instead of traditional electronic ones, they have shown that it's possible to achieve higher speeds with less energy.
This could lead to smaller, more efficient data centers and devices that require less power and resources while still delivering high performance.