Systolic Array Clock Frequency Control for Thermal Management

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Machine learning processors, such as systolic array processors, face heat generation issues due to high power consumption from simultaneous operations, limiting their efficiency in performing machine learning tasks.

Innovation Solution

An electronic device with a main processor and a systolic array processor that calculates a switching activity value to dynamically adjust the clock signal frequency based on the switching activity of processing elements, optimizing power usage and managing heat generation by selectively activating and deactivating processing elements across different regions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If a lot of computing units perform a plurality of operations at the same time in a systolic array processor, then machine learning operations are performed quickly, but power consumption increases and heat generation occurs

Engineering Contradiction:
Improvemachine learning operation speedVSAvoidpower consumption
Core Design Contradiction:
ProductivityVSUse of energy by moving object

Solution Approach 1:

The patent implements dynamic frequency adjustment of the clock signal supplied to processing elements based on real-time switching activity monitoring. The operating frequency is varied adaptively to match the actual computational workload, allowing the system to maintain high productivity when needed while reducing power consumption during lower-demand periods

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system changes the operating parameters (clock frequency) of the systolic array processor based on measured switching activity. By adjusting the frequency parameter dynamically, the system optimizes the balance between computational speed and power consumption, resolving the contradiction between high productivity and low energy use

Inventive Principle:
Principle #35Parameter changes

2Speed

If the clock signal frequency is increased to improve processing speed, then machine learning operations are performed faster, but heat generation increases

Engineering Contradiction:
Improveprocessing speedVSAvoidheat generation
Core Design Contradiction:
SpeedVSTemperature

Solution Approach 1:

The patent employs a feedback mechanism where switching activity is continuously monitored and used to adjust the clock frequency. This closed-loop control ensures that the processing speed is optimized without exceeding thermal constraints, as the frequency is dynamically adjusted based on actual operational conditions and power consumption levels

Inventive Principle:
Principle #23Feedback

3Use of energy by moving object

If processing elements are deactivated to reduce power consumption, then heat generation is reduced, but processing capability is limited

Engineering Contradiction:
Improvepower consumptionVSAvoidprocessing capability
Core Design Contradiction:
Use of energy by moving objectVSProductivity

Solution Approach 1:

The patent activates only the necessary subset of processing elements based on the actual computational requirements of the machine learning task. By applying partial action - activating just enough processing units to handle the current workload - the system reduces power consumption and heat generation while maintaining adequate processing capability

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentUS11709795B2Electronic device including main processor and systolic array processor and operating method of electronic device
Publication Date: 2023.07.25 ELECTRONICS & TELECOMM RES INST
  • US11709795B2 patent drawing
  • US11709795B2 patent drawing
  • US11709795B2 patent drawing

AI summary

Disclosed is an electronic device which includes a main processor, and a systolic array processor, and the systolic array processor includes processing elements, a kernel data memory that provides a kernel data set to the processing elements, a data memory that provides an input data set to the processing elements, and a controller that provides commands to the processing elements. The main processor translates source codes associated with the systolic array processor into commands of the systolic array processor, calculates a switching activity value based on the commands, and stores the translated commands and the switching activity value to a machine learning module, which is based on the systolic array processor.