Integrated Circuit Yield Optimization via Statistical Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for maximizing the yield of integrated circuits, such as the worst-case corner, response surface model, and worst-case distance methods, are inadequate in handling the complex variations and nonlinear performance specifications of analog circuits, often leading to over- or under-constrained designs and inaccurate yield estimation.
Innovation Solution
A method that optimizes integrated circuit design by generating design points that meet predefined specifications, analyzing these points to form clusters, determining a representative design point, running statistical simulations to estimate yield, and iteratively optimizing the design using statistical corners to account for manufacturing process variations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the worst-case corner method is used to optimize analog circuit design, then the design process is simple and fast, but the yield estimation becomes inaccurate and the design may be over- or under-constrained
Solution Approach 1:
The patent uses response surface models as simplified copies of the complex circuit behavior to estimate yield quickly, while also performing detailed Monte Carlo simulations on selected design points to validate and refine the models, combining speed with accuracy
Solution Approach 2:
The patent performs preliminary optimization to identify critical design points before conducting full yield analysis, and uses response surface models to pre-screen design configurations, reducing the need for exhaustive simulations
2Measurement precision
If the response surface model method is used to maximize yield, then the yield estimation accuracy improves, but the computational complexity and time required increase significantly
Solution Approach 1:
The patent divides the design space into regions and uses response surface models for most evaluations, while performing full Monte Carlo simulations only on critical boundary points and selected representative designs, combining both approaches efficiently
Solution Approach 2:
The patent applies response surface modeling selectively to the most critical design parameters and uses it in combination with targeted Monte Carlo simulations rather than applying either method universally to all design points
3Adaptability or versatility
If the number of design parameters and specifications increases, then the design becomes more comprehensive, but the number of worst-case corners increases exponentially making simulation infeasible
Solution Approach 1:
The patent transforms the problem from evaluating discrete worst-case corners to continuous optimization using response surface models, allowing comprehensive parameter coverage without exponential increase in simulation points
Solution Approach 2:
The patent moves from corner-based evaluation in parameter space to surface-based modeling that captures the continuous behavior across all parameter combinations, adding a dimensional perspective that avoids combinatorial explosion
4Reliability
If conventional yield optimization methods are used, then the design meets basic specifications, but the design is not optimized for manufacturing variations and requires more iterations to achieve high yield
Solution Approach 1:
The patent implements iterative optimization where yield simulation results feed back into response surface model updates and design parameter adjustments, continuously improving the design toward maximum yield
Solution Approach 2:
The patent performs preliminary yield analysis using response surface models to identify critical design points before final optimization, and uses statistical corner analysis to pre-assess manufacturing variation impacts
Data Source
AI summary
Method and system for improving yield of an integrated circuit are disclosed. The method includes optimizing a design of the integrated circuit according to a set of predefined design parameters to generating design points that meet a set of predefined design specifications, analyzing the design points to form clusters comprising the design points, determining a representative design point from the clusters comprising the design points, running a statistical simulation to determine a yield of the design using the representative design point and a statistical model of manufacturing process variations, generating statistical corners in accordance with results of the statistical simulation, and optimizing the design in accordance with the statistical corners using an iterative process.