ML Workload Partitioning Across SoCs for Thermal Balance

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing SOC systems face thermal management challenges due to dynamically varying workloads, leading to performance limitations and increased power consumption, with traditional cooling methods adding bulk and reducing battery life.

Innovation Solution

Implementing dynamic workload partitioning between multiple SOCs using machine learning to monitor and reallocate tasks based on thermal and power characteristics, optimizing thermal balance and reducing the need for active cooling.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Temperature

If traditional cooling methods are used to manage SOC thermal constraints, then thermal performance is maintained, but device bulk increases and battery life reduces

Engineering Contradiction:

ImproveSOC thermal performanceVSAvoiddevice bulk

Core Design Contradiction:

TemperatureVSWeight of moving object

Solution Approach 1:

The patent replaces mechanical cooling systems (fans, heat sinks) with a software-based workload partitioning system that uses machine learning to dynamically allocate tasks between SOCs. This substitution eliminates the need for bulky mechanical cooling components while maintaining thermal performance through intelligent workload management.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system dynamically changes workload parameters by monitoring thermal conditions and adjusting task allocation between SOCs in real-time. When thermal constraints are detected, the machine learning model reallocates workloads to maintain optimal temperature levels without requiring physical cooling infrastructure.

Inventive Principle:

Principle #35Parameter changes

2Temperature

If traditional cooling methods are used to manage SOC thermal constraints, then thermal performance is maintained, but power consumption increases

Engineering Contradiction:

ImproveSOC thermal performanceVSAvoidfan power consumption

Core Design Contradiction:

TemperatureVSUse of energy by moving object

Solution Approach 1:

The patent replaces power-consuming mechanical cooling systems with an intelligent workload management approach. The machine learning model predicts thermal conditions and proactively adjusts workload partitioning, eliminating the need for energy-intensive fan operations while maintaining thermal performance.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system enables the SOCs to self-regulate their thermal conditions through intelligent workload allocation. The machine learning model allows the system to autonomously manage thermal envelopes by dynamically partitioning workloads, eliminating the need for external active cooling mechanisms and their associated power consumption.

Inventive Principle:

Principle #25Self-service

3Device complexity

If static workload partitioning is used, then system simplicity is maintained, but thermal management effectiveness decreases

Engineering Contradiction:

Improveworkload partitioning complexityVSAvoidthermal management effectiveness

Core Design Contradiction:

Device complexityVSTemperature

Solution Approach 1:

The patent transitions from static workload partitioning to dynamic workload partitioning driven by machine learning. The system continuously monitors thermal conditions and dynamically adjusts workload allocation between SOCs, enabling effective thermal management while maintaining relative system simplicity through automated decision-making.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system implements a feedback mechanism where the machine learning model continuously monitors thermal conditions and uses this information to adjust workload partitioning decisions. This closed-loop control enables effective thermal management by responding to real-time thermal feedback without requiring complex manual intervention.

Inventive Principle:

Principle #23Feedback

4Temperature

If dynamic workload partitioning with machine learning is implemented, then thermal management effectiveness improves, but system complexity increases

Engineering Contradiction:

Improvethermal management effectivenessVSAvoidworkload partitioning complexity

Core Design Contradiction:

TemperatureVSDevice complexity

Solution Approach 1:

The machine learning model serves multiple functions: it predicts thermal conditions, determines optimal workload partitioning, and adapts to different usage scenarios. This multi-functionality consolidates what could be multiple separate systems into a single intelligent controller, managing thermal effectiveness while limiting the increase in overall system complexity.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS20260050484A1Dynamic workload partitioning for a system on a chip (SOC)

Publication Date: 2026.02.19 QUALCOMM INC

AI summary

Aspects of the disclosure are directed to dynamic workload partitioning. In accordance with one aspect, the disclosure includes a first system on a chip (SOC) configured to commence a use case execution using a baseline workload partition; and a controller coupled to the first SOC, the controller configured to determine if the baseline workload partition should be reallocated using a machine learning (ML) monitoring of system on a chip (SOC) temperatures.