Task Queue Configuration for Neural Network DVFS

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional processor chips face challenges in dynamically monitoring and scheduling task execution states due to difficulties in accessing and modifying task configuration information in real-time, leading to inefficiencies in dynamic voltage and frequency scaling, especially in neural network operations.

Innovation Solution

A data processing device with a task configuration information storage unit and a task queue configuration unit that allows for real-time monitoring and scheduling of tasks, including a DVFS method that dynamically scales voltage and frequency based on neural network topological structure, scale, and accuracy requirements.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If common launching mode is used to execute tasks, then task execution can be completed, but real-time monitoring and scheduling of task execution state becomes difficult

Engineering Contradiction:
Improvetask executionVSAvoidtask execution state monitoring
Core Design Contradiction:
Ease of operationVSDifficulty of detecting and measuring

Solution Approach 1:

A task queue configuration unit is introduced as an intermediary component between the CPU and the task execution system. This unit stores task configuration information including execution states, allowing the CPU to monitor and schedule tasks without direct intervention in the execution flow, thus enabling real-time monitoring while maintaining task execution capability

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The system implements feedback mechanisms where task execution states are continuously updated and stored in the task queue configuration unit. The CPU can query these updated states to make scheduling decisions, creating a closed-loop control system that enables real-time monitoring and dynamic scheduling of neural network processing tasks

Inventive Principle:
Principle #23Feedback

2Loss of energy

If dynamic voltage and frequency scaling is implemented, then power consumption can be reduced, but efficient dynamic monitoring and scheduling becomes challenging

Engineering Contradiction:
Improvepower consumptionVSAvoiddynamic monitoring and scheduling system
Core Design Contradiction:
Loss of energyVSDevice complexity

Solution Approach 1:

Task configuration information including execution states is pre-stored in the task queue configuration unit before dynamic voltage and frequency scaling operations. This preliminary organization of data allows the system to quickly query and analyze task states for scheduling decisions, reducing the complexity of real-time monitoring while enabling effective power management through DVFS

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The task queue configuration unit serves as an intermediary that simplifies the interface between the DVFS controller and the task execution system. By centralizing task configuration information storage and providing standardized query interfaces, it reduces the overall system complexity while enabling efficient dynamic monitoring and scheduling for power optimization

Inventive Principle:
Principle #24Intermediary (Mediator)

3Adaptability or versatility

If task configuration information is stored and accessed in real-time, then dynamic scheduling is enabled, but access and modification efficiency may be reduced

Engineering Contradiction:
Improvedynamic scheduling capabilityVSAvoidinformation access and modification efficiency
Core Design Contradiction:
Adaptability or versatilityVSProductivity

Solution Approach 1:

Task configuration information is extracted and stored in a dedicated task queue configuration unit separate from the main processing flow. This extraction allows efficient access and modification of task states without interfering with the neural network computation pipeline, enabling dynamic scheduling while maintaining high productivity through minimized access overhead

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP3660629B1Data processing apparatus and method
Publication Date: 2023.08.30 SHANGHAI CAMBRICON INFORMATION TECH CO LTD
  • EP3660629B1 patent drawingFigure 1~2
  • EP3660629B1 patent drawingFigure 3~4
  • EP3660629B1 patent drawingFigure 5~6

AI summary

The disclosure provides a data processing device and method. The data processing device may include: a task configuration information storage unit and a task queue configuration unit. The task configuration information storage unit is configured to store configuration information of tasks. The task queue configuration unit is configured to configure a task queue according to the configuration information stored in the task configuration information storage unit. According to the disclosure, a task queue may be configured according to the configuration information.