The rapid development of artificial intelligence puts forward higher requirements for data transmission speed, and traditional cable connections have become bottlenecks. In order to break through this limit, the research team led by the University of Michigan is committed to developing a new chip connection system based on optical wave transmission, which aims to solve the "memory wall" problem and promote the further development of the AI model. The project has received a large funding from the National Science Foundation, and the support of many universities and science and technology giants has been committed to increasing the speed of data transmission to meet increasingly growing AI computing needs.
In today's development of artificial intelligence (AI), data transmission speed has become an important bottleneck restricting its progress. In order to break this barrier, the research team led by the University of Michigan (UM) is developing a new chip connection system that uses light waves instead of traditional cables for data transmission. This innovation is expected to solve the "memory wall" problem that restricts calculation speed, and promote the further growth of the AI model.
The project has received 2 million US dollars from the National Science Foundation's future semiconductor projects. The participating units include the University of Washington, the University of Pennsylvania, the National Laboratory of Lawrence Berkeley, and four industry partners including Google, HP Enterprise, Microsoft and Nvidia. Although the data processing speed has increased by 60,000 times in the past 20 years, the data transmission speed between computer memory and processors has only increased by 30 times. This increasing proportion has made data transmission the biggest obstacle to the AI model expansion.
Di Liang, chief researcher of the project, Di Yang, a professor of UM Electrical and Computer Engineering, said: "Our technology can keep high -performance computing with the growing data flow. The data transmission speed exceeds the current electrical connection speed of more than 100 times. "
At present, the transmission dependent metal connection between multiple memory and processor chips has serious limitations in terms of speed and bandwidth. With the continuous expansion of the AI model, the current hard connection mode has been difficult to meet the needs. The new design of the research team will use the transmission characteristics of light to transmit data through the channels called the optical wave guide between the chips, which greatly improves the data transmission efficiency.
Another highlight of the new technology is its reconstruction. The researchers plans to use special phase -changing materials. When the material is stimulated by laser or voltage, its refractive index will change, thereby achieving flexible adjustment of the light path. Professor Liang Feng, a project collaborator and the University of Pennsylvania, said: "Just like opening and closing the road, if the company uses this technology to produce chips, they can rewrite it without changing other component layouts. Different batches of chips and server connections. "
In addition, the research team will develop a traffic control software to monitor which chips need to communicate in real time to adjust the connection immediately. This flexible connection method can not only improve data processing efficiency, but also dynamically adjust according to different AI model needs.
The project will also provide UM students with opportunities to cooperate with industry, enabling them to gain valuable practical experience in the rapidly developing technology field. Professor Li said: "Cooperation with the industry allows students to better understand the modernity
This innovative optical chip connection technology is expected to completely solve the bottleneck of data transmission in AI development, provide strong support for the large -scale and performance improvement of AI models in the future, and provide students with valuable practical learning opportunities. Its reconstruction and flexible flow control also makes it show huge potential in terms of application.