In a recent blog post, Intel announced its extensive use of AI in various aspects of its work, including thermal design for products like the Core i9 Ultra processor.
Client products like the Core i9 Ultra processor heavily rely on Turbo Boost functionality during operation. Turbo Boost increases processor frequency, leading to increased heat generation.
To gain a comprehensive understanding of this process, researchers need to precisely analyze the complex workloads of CPU cores, IO, and other components to identify hotspots on the chip during Turbo Boost.
During analysis, researchers must determine the optimal placement of micro heat sensors on the CPU, a task traditionally reliant on experience. The entire testing process could take up to 6 weeks, with only one or two workloads studied at a time.
Intel has adopted internally developed Enhanced Intelligence tools to simplify this issue.
Now, Intel engineers input boundary conditions into the Enhanced Intelligence tool, which can process thousands of variables in minutes and then return ideal placement suggestions.
Additionally, Intel has constructed another complementary tool that can simulate other unmeasured loads using existing test data.
Mark Gallina, Senior Systems Thermal and Mechanical Architect at Intel's Client Computing Group, stated that these two Enhanced Intelligence tools have fundamentally transformed the chip thermal design process, improving design efficiency.
In addition to chip thermal design, Intel's Client Computing Group has also applied AI to several other scenarios, including high-speed IO signal integrity analysis and fault analysis, as well as automatic sampling for energy efficiency analysis.
Semiconductor production is increasingly intertwined with AI, with many companies besides Intel leveraging AI to accelerate chip launch processes:
NVIDIA has introduced an AI-accelerated computational lithography platform called cuLitho, AMD has introduced AI in the chip routing stage as a boost, and Synopsys.ai, a subsidiary of Synopsys, released an AI-driven EDA suite called Synopsys.ai last year.
