Abstract
Thermal and power management on a system on a chip (SoC) typically responds to its present state, revising the operating point only after reaching a temperature or power limit. This reactive approach often overlooks the delayed and cumulative heating produced by earlier execution, failing to effectively reconcile performance, power, and completion goals. The disclosed methods forecast temperature ramp rates and trajectories from the present temperature and power of the SoC, form candidate operating points expected to meet a workload completion time, and limits them with caps from thermal and power controllers. A cost function evaluated across a finite prediction horizon then scores each candidate against throttling risk, average power, average frequency, completion time, and remaining headroom, while runtime feedback revises cost weights and horizon length. This approach reduces throttling activations and completion‑time shortfalls.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Heidarinejad, Mohsen; Kenarangi, Farid; and Mittal, Arpit, "Horizon‑Based Operating Point Selection for a System on a Chip Using Predicted Thermal Trajectories and Adaptive Cost Weighting", Technical Disclosure Commons, (September 02, 2026)
https://www.tdcommons.org/dpubs_series/11566