Machine Learning Model for Circuit Simulation Result Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methodologies for simulating circuit designs under various process, voltage, and temperature (PVT) conditions are computationally intensive and inefficient, requiring a large number of simulations to ensure low probability of failure in semiconductor fabrication, especially for memory ICs, which can take weeks or months and consume significant resources.
Innovation Solution
A machine learning model is generated based on actual simulation results for a minimal set of condition values, allowing for the prediction of simulation results for a circuit design over a broader set of conditions, thereby reducing the need for extensive simulations and saving time and computational resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If extensive simulations are performed under multiple PVT conditions to ensure low probability of failure, then reliability of circuit design is improved, but productivity and time consumption deteriorate
Solution Approach 1:
The patent applies preliminary action by performing simulations on a minimal set of PVT conditions first, generating training data in advance. This preliminary simulation work enables the subsequent machine learning model to predict results for remaining conditions without requiring actual extensive simulations, thus resolving the contradiction between reliability and productivity.
Solution Approach 2:
The patent uses copying by creating a machine learning model that replicates the behavior of actual circuit simulations. The model is trained on simulation results from a minimal set of conditions and then copies/predicts the results for a broader set of PVT conditions, replacing the need for actual extensive simulations while maintaining reliability.
2Measurement precision
If simulations are performed for a broader set of PVT conditions to ensure comprehensive characterization, then measurement precision is improved, but loss of time and computational resources worsens
Solution Approach 1:
The machine learning model serves as a copy of the actual simulation process. It is trained on simulation data from a minimal set of PVT conditions and then generates predicted results for a broader set of conditions, achieving comprehensive characterization without the time cost of actual extensive simulations.
Solution Approach 2:
The patent changes the approach from direct simulation to machine learning prediction. By transforming the problem from computational simulation to statistical prediction, the system achieves comprehensive PVT characterization with significantly reduced time and computational resources while maintaining measurement precision.
3Manufacturing precision
If a large number of simulations are conducted to cover process variations, then manufacturing precision is improved, but productivity deteriorates
Solution Approach 1:
The machine learning model copies the simulation results for process variation analysis. By training on a minimal set of actual simulation data and then predicting results for broader process conditions, the system achieves comprehensive process variation coverage without requiring a large number of actual simulations, thus improving productivity.
Solution Approach 2:
The patent performs preliminary simulations on a minimal set of conditions to capture process variation characteristics. This preliminary action provides sufficient training data for the machine learning model to accurately predict results under broader process variations, eliminating the need for extensive simulations while maintaining manufacturing precision.
Data Source
AI summary
Various embodiments provide for predicting a simulation result for a circuit design using a machine learning model, which can be used as part of a process of an electronic design automation (EDA) system that measures a circuit design (e.g., timing, power, voltage, current, etc.). In particular, various embodiments described herein can enable modeling simulated time measurements of a circuit design, and can enable such modeling with minimal usage of simulation result data.