Semiconductor Model Parameter Generation for Process Variation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing semiconductor device models fail to accurately capture process-induced variations, leading to inefficiencies in circuit design and performance due to limitations in modeling random variability sources and process-induced device performance.
Innovation Solution
A method for generating semiconductor device model parameters by receiving performance data, extracting and modeling statistics to determine moments and correlations, and applying Generalized Lambda Distribution parameters to produce accurate model parameters that account for non-normal distributions and correlations, enabling the generation of multivariate Gaussian variates for SPICE simulations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If known approaches are used to model semiconductor device behavior, then the modeling process is simple, but the accuracy of modeling process-induced variation is insufficient
Solution Approach 1:
The patent transforms the modeling approach by changing parameters from assuming normal distributions to using Generalized Lambda Distribution (GLD) parameters that can represent non-normal distributions. This allows the model to capture process-induced variations more accurately by adjusting distribution parameters (shape, scale, location) without fundamentally changing the modeling framework
Solution Approach 2:
The patent introduces an intermediary statistical layer between raw device data and final model parameters. By extracting model parameters from multiple statistical instances and modeling their distributions through GLD, the system mediates between complex process variations and simplified circuit-level simulations, achieving both accuracy and efficiency
2Reliability
If traditional model parameters are used, then the model is easy to implement, but it cannot capture non-normal distributions and correlations of process variations
Solution Approach 1:
The patent combines multiple statistical concepts into a composite modeling approach: Generalized Lambda Distribution for marginal distributions, correlation matrices for parameter relationships, and transformation methods for generating realistic device instances. This composite statistical framework captures both non-normal distributions and correlations, significantly improving model reliability
Solution Approach 2:
The patent segments the modeling process into distinct stages: (1) extracting parameters from individual device instances, (2) modeling the statistical distribution of each parameter separately using GLD, (3) establishing correlation structures between parameters, and (4) generating new device instances from the statistical model. This segmentation makes the complex task of modeling process variations manageable and systematic
3Productivity
If process-induced variations are not modeled accurately, then the design process is fast, but circuit design efficiency and performance suffer
Solution Approach 1:
The patent creates statistical copies of real device variations through generated model parameters. By extracting parameters from measured or simulated device instances and reproducing their statistical characteristics (distributions and correlations) in a parametric model, the system efficiently generates virtual device instances that accurately represent process-induced variations without requiring extensive physical measurements or complex simulations
Data Source
AI summary
A method for generating semiconductor device model parameters includes receiving semiconductor device performance data of statistical instances of semiconductor devices, for a plurality of coordinates in a process space with dimensions of process-dependent device parameters Model parameters are extracted to produce individual model instances, each corresponding to the respective statistical instances for the coordinates in the process space. Statistics of the extracted model parameters are modeled by processing the individual model instances to determine, for each coordinate in the process space, moments describing non-normal marginal distributions of the extracted model parameters and correlations between the extracted model parameters. Semiconductor device model parameters are generated for use in simulating a circuit using the determined moments and the determined correlations, for a selected coordinate in the process space.