Abstract
A system on a chip (SoC) that runs several workloads at the same time can experience uneven power draw, localized heating, and reduced responsiveness when it relies on a fixed policy that treats the die as uniform. This document describes a closed‑loop framework that observes on‑die temperature and power, estimates power for subsystems without direct sensors, and predicts near‑term spikes and longer‑term demand from queued work. Using these estimates, the framework stretches or compresses task timelines and recommends per‑subsystem power caps, which a control layer applies through operating‑point adjustment, clock and power gating, and task migration. A supervisory layer then compares predicted behavior against measured results and adjusts controller settings and prediction weights. In this way, the framework can improve the balance between responsiveness and energy use while helping the SoC stay within its power and thermal limits.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Heidarinejad, Mohsen; Kenarangi, Farid; Mittal, Arpit; and Khajeh, Amin, "Predictive, Time‑Scaled Scheduling of Concurrent Workloads for Power and Thermal Management in a System on a Chip", Technical Disclosure Commons, (September 02, 2026)
https://www.tdcommons.org/dpubs_series/11552