Monte Carlo Worst Sample Extraction for Yield Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Monte Carlo simulation methods require a large number of samples to verify yield targets, especially when the yield is high, leading to computational inefficiencies and unnecessary testing of non-failing instances, as failures are rare and often occur at extreme unusual value combinations.
Innovation Solution
The method identifies the worst Monte Carlo simulation sample for each design specification by building a performance model, estimating failure probabilities, and dynamically adjusting the order of design specification processing to prioritize samples most likely to fail, using probability-based stop criteria to minimize the number of simulations needed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a large number of Monte Carlo simulation samples are used to verify yield targets, then the reliability of yield verification is improved, but the computational expense increases significantly
Solution Approach 1:
The patent extracts only the most critical aspect of yield verification by identifying and testing only the worst-case sample for each design specification. Instead of simulating all samples to verify yield, the method extracts a single representative worst-case sample that captures the essential failure mode, thereby verifying design robustness with minimal computational expense while maintaining reliability.
Solution Approach 2:
The patent performs preliminary action by using a performance model to predict which sample is most likely to fail before conducting actual simulations. This preliminary prediction allows the method to identify and test only the worst-case sample in advance, avoiding the need for extensive Monte Carlo simulations and significantly reducing computational expense while ensuring reliable verification.
2Reliability
If full Monte Carlo verification is performed to ensure design robustness, then the design reliability is improved, but the simulation time increases
Solution Approach 1:
The patent extracts only the essential information needed for robustness verification by identifying a single worst-case sample for each design specification. This extraction approach maintains design reliability by focusing on the most critical failure scenario while dramatically reducing simulation time compared to full Monte Carlo verification of all samples.
Solution Approach 2:
The patent performs preliminary action by using performance models to predict the worst-case sample before simulation. This preliminary identification allows the method to bypass extensive simulation time by directly testing only the predicted worst-case scenario, thereby ensuring design robustness with minimal simulation time investment.
3Measurement precision
If Monte Carlo simulation is used to estimate failure probability, then the accuracy of yield estimation is improved, but the number of samples required increases
Solution Approach 1:
The patent extracts only the critical failure information by identifying and testing a single worst-case sample for each design specification. This extraction maintains yield estimation accuracy by focusing on the most representative failure scenario while reducing the number of samples required from thousands to just one per specification, thereby improving efficiency without sacrificing measurement precision.
Solution Approach 2:
The patent performs preliminary action by using performance models to predict which sample will exhibit the worst performance before simulation. This preliminary prediction allows the method to achieve accurate yield estimation by testing only the predicted worst-case sample, eliminating the need for large sample sizes and reducing the quantity of samples required while maintaining measurement precision.
Data Source
AI summary
A system, method, and computer program product for reducing the number of Monte Carlo simulation samples required to determine if a design meets design specifications. The worst sample for each specification acts as a design corner to substitute for a full design verification. Embodiments determine the maximum number of samples needed, perform an initial performance modeling using an initial set of samples, and estimate the failure probability of each of the remaining samples based on the performance model. Embodiments then simulate remaining samples with a computer-operated Monte Carlo circuit simulation tool in decreasing design specification model accuracy order, wherein the sample predicted most likely to fail each specification is simulated first. Re-use of simulation results progressively improves models. Probability based stop criteria end the simulation early when the worst samples have been confidently found. A potential ten-fold reduction in overall specification verification time may result.