AI & softwareResearch findings
THERMOS: A Thermally-Aware Scheduling Framework for AI Workloads on Heterogeneous PIM Architectures
Original published 2025.08.14 23:35 KSTAdded to Blackdeck 2026.11.22 15:32 KST
What has changed?
A multi-objective reinforcement learning policy that jointly optimizes performance, energy, and thermal constraints in heterogeneous PIM systems.
Key points
- Proposes a thermally-aware scheduling method for AI workloads on heterogeneous multi-chiplet PIM architectures.
- Introduces a single multi-objective reinforcement learning policy that adapts to runtime performance objectives.
- Demonstrates up to 89% faster execution and 57% lower energy consumption compared to baseline methods with minimal overhead.
Connect it to your work
Achieves significant improvements in execution time and energy efficiency while maintaining thermal stability during AI workload execution.
What remains unproven
This research note is based on an abstract; check the original for test conditions and application scope.
Originals used for this note
Links open the original publisher in a new tab.