Adaptive Monte Carlo Sampling for IC Yield Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for estimating failure rates in high-yield integrated circuit (IC) designs require large sample sets for Monte Carlo simulations, leading to increased computational costs and turnaround time, making it inefficient for accurate statistical analysis.
Innovation Solution
A method that performs multiple Monte Carlo simulations using an initial sample set to build a performance model, determines a failure rate and confidence interval, and then selects a second sample set based on an importance distribution to reduce the confidence interval, with iterative updates to the model when necessary.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If large sample sets are used for Monte Carlo simulations, then measurement precision of failure rate is improved, but productivity of design process deteriorates
Solution Approach 1:
The patent changes the sampling parameters by using importance sampling with adaptive sample selection. Instead of uniform random sampling, it dynamically adjusts sample selection based on performance metrics and failure probability estimates, allowing accurate failure rate measurement with fewer samples
Solution Approach 2:
The patent performs preliminary Monte Carlo simulations to build performance models and estimate failure rates before the main analysis. This preliminary action identifies critical regions in the parameter space, enabling more efficient subsequent sampling strategies that focus computational resources on relevant areas
2Measurement precision
If large sample sets are used for Monte Carlo simulations, then measurement precision of failure rate is improved, but loss of time increases
Solution Approach 1:
The patent transforms the sampling approach by changing from fixed uniform sampling to adaptive importance sampling. It dynamically modifies sampling parameters based on observed performance characteristics, concentrating samples in regions that contribute most to failure rate estimation accuracy
Solution Approach 2:
The patent performs a limited number of preliminary simulations to gather sufficient information for model building, then uses this model to guide the main simulation phase. This partial action approach avoids performing excessive full-scale simulations while achieving the same measurement precision
Data Source
AI summary
A method for performing multiple simulations for a circuit using a first plurality of samples is provided. The method includes obtaining a model of the circuit based on a result of the simulations, determining a failure rate and a confidence interval of the failure rate for the circuit with the performance model. The method includes determining an importance distribution based on the failure rate for the first plurality of samples, wherein the importance distribution is indicative of a probability that a sample value for the circuit will fail the simulation, selecting a second plurality of samples based on the importance distribution, performing a second set of simulations using the second plurality of samples to reduce the confidence interval of the failure rate. When the confidence interval is larger than a value, obtaining an updated performance model and performing new Monte Carlo simulations with new samples.