Embedded Controller Hardware Selection via Machine Learning

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing information handling systems lack the ability to efficiently select between different core classes for executing applications based on performance and power usage, leading to suboptimal system performance and power efficiency.

Innovation Solution

An information handling system that employs a hardware selection machine learning model to determine the most efficient hardware class for an application by calculating a normalized mutual information score for each hardware class, allowing for dynamic migration of workloads to the most suitable core or compute engine based on characteristics such as floating-point and integer operations, instructions per second, and power usage.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If applications are executed on a fixed hardware class without dynamic selection, then system simplicity is maintained, but system performance and power efficiency deteriorate

Engineering Contradiction:

Improvesystem performanceVSAvoidhardware selection complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The system performs self-characterization by automatically analyzing application binary interfaces, instruction sets, and runtime behaviors to determine hardware class requirements without external intervention. The embedded controller autonomously executes machine learning models to match applications with optimal hardware classes, enabling the system to self-optimize performance and power efficiency without adding complex manual configuration processes

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system performs preliminary characterization of applications during binary analysis and loading phases, extracting features such as instruction set architecture, floating-point operation requirements, and memory access patterns before runtime execution. This advance preparation enables the embedded controller to make informed hardware class selection decisions without adding complexity during actual application execution

Inventive Principle:

Principle #10Preliminary action

2Use of energy by moving object

If applications are migrated to optimize hardware class selection, then power efficiency is improved, but system complexity increases

Engineering Contradiction:

Improvepower efficiencyVSAvoidworkload migration complexity

Core Design Contradiction:

Use of energy by moving objectVSDevice complexity

Solution Approach 1:

The embedded controller continuously monitors application runtime behaviors, power consumption metrics, and hardware class performance characteristics, using this feedback to dynamically adjust and refine hardware class selections. The system incorporates feedback loops that track power efficiency gains from migrations and use this information to optimize subsequent migration decisions, reducing the perceived complexity through data-driven automation

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system changes key parameters such as normalized mutual information scores, hardware class affinity values, and power efficiency thresholds to optimize migration decisions. By adjusting these parameters based on application characteristics and system state, the system dynamically determines when and where to migrate workloads, managing complexity through parameter-based control rather than rigid rules

Inventive Principle:

Principle #35Parameter changes

3Productivity

If hardware class selection is based on detailed application characterization, then computation locality is optimized, but measurement and detection difficulty increases

Engineering Contradiction:

Improvecomputation localityVSAvoidapplication characterization difficulty

Core Design Contradiction:

ProductivityVSDifficulty of detecting and measuring

Solution Approach 1:

The system replaces manual or complex analytical methods for characterizing applications with automated machine learning models that process application binaries and runtime data. These models automatically extract relevant features such as instruction set usage patterns, data access behaviors, and computational intensity metrics, substituting complex measurement processes with automated computational analysis that reduces detection difficulty while improving computation locality optimization

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS20250103458A1Computation locality utilization based on an application instruction set

Publication Date: 2025.03.27 DELL PROD LP

AI summary

An information handling system includes a memory and an embedded controller. The memory stores a hardware selection machine learning model. The embedded controller communicates with the memory and executes the hardware selection machine learning model. During the runtime of the information handling system, the embedded controller determines an application to be executed within the information handling system and determines a characterization of the application. Based on the characterization of the application, the embedded controller determines a different normalized mutual information score for each hardware class within the information handling system. The embedded controller determines a different maximum score for each of the hardware classes and migrates the application to a component of a hardware class having a highest score.