Researchers at the University of Technology Sydney, ShanghaiTech University, and Technical University of Munich published a technical paper titled IC-ThermBench: An Open, Progressive Benchmark for Generalizable 2.5D/3D-IC Thermal Learning. The benchmark combines established 3D-IC steady-state, transient, and industrial package tasks with a new 50,000-sample 2.5D chiplet extension designed to evaluate progressively broader represented physics. The IC-ThermBench is the benchmark, and the 2.5D chiplet extension is the new component. The 50,000-sample figure is the size of the extension, and the 2.5D chiplet is the specific configuration that the extension targets.

The progressive aspect is the key: the benchmark evaluates progressively broader represented physics, which is the distinction from a fixed benchmark. The open aspect is the accessibility: the benchmark is open, which is the distinction from a closed benchmark. The thermal learning is the target: the benchmark evaluates the thermal models for the 2.5D and 3D ICs.

This is an open progressive benchmark for thermal learning. The 3D-IC tasks are the established components, and the 2.5D chiplet extension is the new component. The progressive aspect is the key, and the open aspect is the accessibility.

The accuracy of the thermal models, the generalization across nodes, and the practical use in design. The paper presents the benchmark and the extension, but not an accuracy figure or a design application, and the 50,000-sample figure is the size, not a completeness guarantee. The 2.5D chiplet is the specific configuration, and the thermal learning is the target.

Whether the benchmark is adopted in design, whether it generalizes across nodes, and whether the thermal models are accurate. For the semiconductor design community, an open progressive benchmark for thermal learning is a significant tool, and the 50,000-sample 2.5D chiplet extension is the new component that makes it progressive. The progressive aspect is the key: the benchmark evaluates progressively broader represented physics, which is the distinction from a fixed benchmark, and the open aspect is the accessibility that makes it a tool for the design community.

Researchers published IC-ThermBench, an open progressive benchmark for generalizable 2.5D/3D-IC thermal learning, with a new 50,000-sample 2.5D chiplet extension. An open progressive benchmark for thermal learning.