Neural Network Thermal Solver for IC Temperature Rise Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for obtaining accurate thermal profiles on integrated circuits (ICs) are time-consuming and inefficient, particularly when using finite element methods, and often result in smeared temperature gradients due to coarse mesh resolution.
Innovation Solution
A neural network-based thermal solver is developed to enhance the resolution of IC thermal profiles by training a temperature rise predictor using thermal simulations and location parameters, allowing for faster and more detailed thermal analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If finite element methods are used to obtain accurate thermal profiles on ICs, then measurement precision is improved, but loss of time increases
Solution Approach 1:
The patent pre-computes and stores thermal response data for various power distribution patterns in a lookup table during an offline phase. During actual thermal analysis, the system quickly matches the current power pattern to pre-computed patterns and retrieves the corresponding thermal profile, avoiding time-consuming finite element simulations while maintaining accuracy
Solution Approach 2:
The patent creates simplified thermal response models that capture the essential thermal behavior of the IC. These models are trained using finite element method data and then used to rapidly predict thermal profiles without running full FEM simulations, effectively copying the thermal response characteristics in a computationally efficient form
2Measurement precision
If mesh density on IC is increased to improve thermal profile resolution, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent divides the IC into multiple discrete power patterns or regions, each with its own pre-computed thermal response characteristics. This segmentation allows the system to handle complex thermal scenarios by combining responses from multiple simpler, pre-analyzed segments without requiring a uniformly fine mesh across the entire IC
Solution Approach 2:
The patent transforms the thermal analysis problem from solving differential equations with fine mesh to a parameter-matching problem where the system selects from pre-computed solutions based on power pattern parameters. This changes the problem from a spatial discretization challenge to a parameter lookup and combination task, reducing model complexity
3Reliability
If system thermal analysis model covers all system components, then reliability is improved, but productivity decreases
Solution Approach 1:
The patent pre-computes thermal responses for various system configurations and power patterns, storing them in lookup tables. When performing system-level thermal analysis, the method quickly retrieves and combines pre-computed results for different components (ICs, packages, PCBs, cooling systems) rather than performing full-system simulations, maintaining comprehensive analysis while dramatically reducing computation time
Solution Approach 2:
The patent merges pre-computed thermal response data from multiple sources (different ICs, packages, PCB configurations, cooling systems) into a unified system-level thermal profile. By combining these pre-analyzed components using superposition principles, the system achieves comprehensive system thermal analysis without the computational burden of analyzing the complete system simultaneously
Data Source
AI summary
Systems and methods use a neural network based predictor, that has been trained to determine a temperature rise across an entire IC. The training of the predictor can include generating a representation of two or more templates identifying different portions of an integrated circuit (IC), each template associated with location parameters to position the template in the IC; performing thermal simulations for each respective template of the IC, each thermal simulation determining an output based on a power pattern of tiles of the respective template, the output indicating a change in temperature of a center tile of the respective template relative to a base temperature of the integrated circuit; and training a neural network. The trained predictor can be used to determine a temperature rise and then can be appended to a system level thermal profile of the IC to generate a detailed thermal profile of the IC.