Macro-Model Accelerates Yield Analysis via Importance Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Statistical simulations for yield analysis in memory and logic designs require numerous device instances and simulation runs to observe rare failure events, leading to inefficiencies and prolonged processing times.
Innovation Solution
The method involves constructing a macro-model using machine learning and importance sampling techniques to identify failure regions, allowing for biased sampling and faster prediction of performance indicators, thereby reducing the need for extensive simulation runs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If statistical simulations are performed to observe rare failure events, then measurement precision is improved, but loss of time increases significantly
Solution Approach 1:
The patent pre-identifies failure regions in the parameter space before conducting simulations. By using machine learning to train a model on initial simulation data that defines failure boundaries, subsequent simulations can be strategically directed toward these predefined failure regions, avoiding random exploration and significantly reducing the time needed to observe rare failure events.
Solution Approach 2:
The patent creates a macro-model that replicates the complex failure behavior observed in detailed simulations. This macro-model serves as a simplified copy that can predict failure regions without requiring full-scale statistical simulations, enabling rapid analysis while maintaining accuracy for rare failure event detection.
2Measurement precision
If the number of simulation runs is increased to observe rare failure events, then measurement precision is improved, but productivity deteriorates
Solution Approach 1:
The patent replaces numerous full-scale statistical simulations with a macro-model that copies the essential failure characteristics. This macro-model can rapidly evaluate failure probabilities for different parameter combinations without requiring extensive simulation runs, thereby maintaining measurement precision while dramatically improving analysis productivity.
Solution Approach 2:
The patent transforms the simulation approach by changing from uniform random sampling to targeted sampling based on identified failure regions. By adjusting the sampling parameters to focus on critical regions in the parameter space, the system achieves accurate failure rate measurement with fewer simulation runs, thus improving productivity.
3Reliability
If extensive simulation runs are performed across large variability space, then reliability assessment is improved, but loss of time increases
Solution Approach 1:
The patent performs preliminary analysis to identify and map failure regions in the parameter space before conducting comprehensive reliability assessment. This preliminary step creates a guide that directs subsequent simulations toward critical regions, ensuring accurate reliability assessment without requiring exhaustive simulation of the entire variability space.
Solution Approach 2:
The patent uses a macro-model that captures the reliability characteristics of the system across the variability space. This macro-model serves as a computational copy that can rapidly assess reliability for different parameter combinations without requiring extensive simulations, thereby maintaining accuracy while reducing processing time.
Data Source
AI summary
In one example, a method for evaluating a system includes obtaining a model of the system that defines a boundary between at least one failure region and a non-failure region for a performance indicator with respect to at least one variable of the system. In one embodiment, obtaining the model involves constructing a new model; however, in other embodiments, obtaining the model involves accepting or retrieving a pre-constructed model is input. The method further includes obtaining importance samples for the at least one variable that are biased to the at least one failure region, and calculating indicator values for the performance indicator by applying the importance samples to the model.