Multi-Core Neural Network Scheduling for Latency-Aware Workloads

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Traditional methods for scheduling neural network execution on multi-core devices rely on user-generated schedules, which fail to optimize workload distribution across processing cores, leading to inefficiency and inaccuracy.

Innovation Solution

A technique that identifies workload fragments based on sensor type and desired latency, determines execution times for each fragment, and generates a schedule to ensure timely completion within the specified latency, optimizing the execution across multiple processing cores.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If user-generated schedules are used for neural network execution, then the scheduling process is simple and controllable, but the workload distribution across processing cores is suboptimal leading to inefficiency

Engineering Contradiction:

Improvescheduling simplicityVSAvoidexecution efficiency

Core Design Contradiction:

Ease of operationVSProductivity

Solution Approach 1:

The system automatically generates optimal schedules by analyzing sensor data, workload requirements, and processing core capabilities without requiring manual user input. The scheduler self-adjusts to optimize workload distribution based on real-time conditions, eliminating the need for users to manually create schedules while achieving superior execution efficiency

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system dynamically adjusts scheduling parameters such as time slices, priority levels, and workload allocation based on sensor feedback and system state. By continuously monitoring execution metrics and modifying schedule parameters in real-time, the system optimizes productivity while maintaining ease of use through automated adaptation

Inventive Principle:

Principle #35Parameter changes

2Device complexity

If predetermined execution schedules are used, then the system is easy to implement and control, but it fails to optimize workload distribution and leads to inefficient execution

Engineering Contradiction:

Improvesystem implementation complexityVSAvoidworkload optimization

Core Design Contradiction:

Device complexityVSProductivity

Solution Approach 1:

The system incorporates feedback mechanisms that monitor sensor data, execution performance, and workload status to continuously improve schedule generation. By analyzing real-time feedback loops, the system automatically optimizes workload distribution across processing cores without increasing implementation complexity, as the feedback-driven optimization is integrated into the existing scheduler architecture

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system performs preliminary analysis of sensor data, workload characteristics, and processing capabilities before generating execution schedules. By pre-computing optimal task assignments and preparing execution plans in advance based on predicted system state, the system achieves optimized workload distribution without requiring complex real-time decision-making during execution

Inventive Principle:

Principle #10Preliminary action

3Ease of operation

If manual schedule generation is used, then user control over execution is maintained, but the accuracy and optimization of workload distribution is insufficient

Engineering Contradiction:

Improveuser controlVSAvoidworkload distribution accuracy

Core Design Contradiction:

Ease of operationVSMeasurement precision

Solution Approach 1:

The system replaces manual mechanical schedule creation with automated computational algorithms that analyze sensor data and generate optimal schedules. The mechanical process of manual scheduling is substituted with intelligent algorithms that automatically compute precise workload distributions based on real-time system state, achieving high measurement precision while maintaining ease of operation through automated decision-making

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS20260056801A1Scheduling neural network execution in multi-core environments

Publication Date: 2026.02.26 TEXAS INSTRUMENTS INC

AI summary

Various embodiments of the present disclosure relate to scheduling the execution of one or more neural networks, and in particular, to scheduling the execution of one or more neural networks within the context of a multi-core environment. In one example embodiment, a technique for scheduling neural network execution across multiple processing cores is provided. The technique first includes identifying a plurality of workload fragments of a neural network based on a sensor type and a desired latency associated with each workload fragment. Next, the technique includes determining an execution time for executing each workload fragment. Finally, the technique includes generating a schedule for executing the neural network across multiple processing cores based on the desired latency associated with each workload fragment, and the execution time for executing each workload fragment.