In today's development of artificial intelligence (AI), data transfer speed has become a significant bottleneck hindering its progress. To break this barrier, a research team led by the University of Michigan (U-M) is developing a brand new chip connection system that uses optical waves instead of traditional cables for data transmission. This innovation is expected to address the "memory wall" problem that limits computing speed and promote further growth of AI models.
The project has received a $2 million grant from the National Science Foundation's Future Semiconductor program, with participation from institutions including the University of Washington, the University of Pennsylvania, Lawrence Berkeley National Laboratory, and four industry partners: Google, HP Enterprise, Microsoft, and Nvidia. Although data processing speeds have increased by 60,000 times over the past 20 years, the data transfer speed between computer memory and processors has only increased by 30 times, creating a disproportionate enhancement that makes data transfer the biggest obstacle to scaling AI models.
The project's chief researcher, Di Liang, a professor of electrical and computer engineering at U-M, stated: "Our technology can keep high-performance computing in sync with the ever-increasing data flow. With optical connections, we expect to achieve data transfer speeds of tens of terabits per second, over 100 times faster than current electrical connections."
Currently, data transfer between multiple memory and processor chips relies on metal connections, which have significant limitations in speed and bandwidth. As AI model sizes continue to grow, the current hard-wired connection model is becoming inadequate. The research team's new design will utilize the transmission properties of light to transfer data between chips through channels called optical waveguides, greatly enhancing data transfer efficiency.
Another highlight of the new technology is its reconfigurability. The researchers plan to use special phase-change materials that change their refractive index when stimulated by lasers or voltage, allowing for flexible adjustment of light paths. As project collaborator Liang Feng, a professor at the University of Pennsylvania, explained: "It's like opening and closing roads; if companies adopt this technology to produce chips, they can rewrite connections for different batches of chips and servers without changing the layout of other components."
Additionally, the research team will develop traffic control software to monitor in real-time which chips need to communicate, allowing for immediate adjustments in connections. This flexible connection method will not only improve data processing efficiency but also allow for dynamic adjustments based on the varying needs of different AI models.
The project will also provide U-M students with opportunities to collaborate with industry, enabling them to gain valuable hands-on experience in the rapidly evolving technology field. Professor Li stated: "Collaboration with the industry helps students better understand modern