Krylov ROM Dynamic Thermal Management for IC Hotspot Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Dynamic Thermal Management (DTM) technologies face challenges in predicting and managing thermal hotspots in high-power density applications, such as smartphones, due to limitations in computational resources and flexibility in sensing and controlling temperature distributions across IC packages.
Innovation Solution
A Krylov Reduced Order Model (ROM) based system with external control, implemented using Python code, for dynamic thermal management analysis, allowing for co-simulation with temperature-dependent power modeling and flexible time step selection, enabling real-time feedback and improved hotspot detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If 3D transient CFD is applied to generate highly accurate simulation results, then measurement precision is improved, but productivity deteriorates due to unacceptable computing resources and slow simulation speed
Solution Approach 1:
The patent segments the thermal simulation problem into two parts: (1) an offline 3D transient CFD simulation to generate training data and build the ROM, and (2) an online reduced-order model evaluation for rapid temperature prediction. This segmentation allows the computationally intensive accurate simulation to be performed only once during model construction, while the ROM provides fast predictions during runtime, thus resolving the contradiction between accuracy and speed.
Solution Approach 2:
The patent performs preliminary action by pre-computing the 3D transient CFD simulation results offline to create a training dataset. This training data is then used to construct the reduced-order model before actual DTM operations. By performing the accurate but slow simulation in advance, the system achieves both high accuracy in the ROM predictions and fast simulation speed during runtime.
2Productivity
If LTI ROM technology is used to allow speedy simulations, then productivity is improved, but measurement precision deteriorates because it cannot sense where the hot-spot is placed in a system
Solution Approach 1:
The patent applies local quality by using sensor placement information and thermal resistance networks to focus the ROM's predictive capability on specific locations of interest, particularly potential hotspot regions. The system selectively monitors and predicts temperatures at critical locations rather than attempting to predict the entire thermal field, thus achieving both fast simulation speed and accurate hotspot location detection.
3Measurement precision
If distributed temperature sensors are placed around critical locations to track moving hotspots, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent uses copying by creating a virtual thermal model (the reduced-order model) that replicates the thermal behavior of the physical system. This digital twin allows the system to predict temperature distributions and track moving hotspots without requiring physical sensors at every critical location. The ROM copy provides the temperature field information that would otherwise require extensive sensor arrays to measure physically.
Data Source
AI summary
Machine assisted system and method for performing dynamic thermal management (DTM) analysis are described. In one embodiment, the method can include receiving a power profile associated with IP blocks in an integrated circuit (IC) system modeled by a Krylov reduced order model (ROM). The power profile can represent power consumption of each of the blocks based on a predefined operating scenario. The method can additionally include evaluating the temperature of each of the blocks of the IC system for the current time step based on the power profile and the Krylov ROM. The method can further include calculating new power values based on the current temperature field of each of the blocks of the IC system, wherein the power profile can be updated with the new power value for the temperature of each of the blocks of the IC system for the next time step.