Thermal ROM Simulation Using ML Tile Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data processing systems face challenges in efficiently simulating thermal management due to the high computational load of accurate simulations, which can take weeks to complete, limiting the ability to test various usage scenarios and thermal management techniques effectively.
Innovation Solution
A thermal Reduced Order Model (ROM) trained through machine learning is used to reduce simulation time and computational resources, allowing for efficient simulation of thermal management controls with a spatial resolution of a few microns, enabling faster testing of different usage scenarios and thermal mitigation techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional simulations are used with high spatial resolution (tens of microns), then measurement precision is improved, but productivity deteriorates due to excessive computational time (weeks to months)
Solution Approach 1:
The device is divided into multiple tiles, each representing a portion of the device. Thermal simulations are performed independently for each tile rather than for the entire device at once, allowing parallel processing and reducing overall simulation time while maintaining high spatial resolution across the complete device surface
Solution Approach 2:
Thermal simulations are performed in advance for individual tiles under various power conditions to build a thermal model database. This pre-computed data is then used during actual thermal management operations, eliminating the need to run full-device simulations in real-time and dramatically improving productivity
2Reliability
If conventional simulations are used to test multiple usage scenarios and thermal management techniques, then reliability is improved through comprehensive testing, but loss of time increases due to the need to rebuild simulations for each scenario
Solution Approach 1:
The thermal model is designed to dynamically adapt to different usage scenarios by adjusting power input parameters for individual tiles based on the specific scenario being tested. This allows the same thermal model infrastructure to efficiently evaluate multiple usage scenarios and thermal management techniques without requiring separate simulation builds for each case
Solution Approach 2:
Different usage scenarios are represented by changing power parameters in the thermal model rather than rebuilding the simulation geometry or mesh. By modifying only the power input parameters (e.g., CPU load, GPU usage, display brightness) while reusing the pre-built thermal model, the system can rapidly evaluate numerous scenarios, improving both reliability through comprehensive testing and reducing time loss
Data Source
AI summary
Methods and systems are described that improve simulations which use thermal models to test dynamic thermal mitigation of devices, such as smartphones. These methods and systems can use a thermal Reduced Order Model (ROM) that is trained through machine learning to provide efficient systems that can significantly reduce the time and computational resources required to build a simulation of a device's thermal behavior. The thermal model can be used in different usage scenarios with different power management and thermal management controls to test the device's thermal behavior.