Yield Prediction System for IC Design Layouts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing yield prediction software assumes random defect distribution across integrated circuit design elements, failing to account for systematic defects that are specific to design layout or equipment, leading to inaccurate yield predictions in new technology nodes.
Innovation Solution
A method and apparatus that differentiate between systematic and random defects during the design stage by using inspection data from previous wafer designs, predicting yield separately for each type and calculating a defect criticality factor to determine the likelihood of defects in new design layouts, thereby selecting an optimal design layout.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing yield prediction software uses statistical critical area calculation with random defect distribution assumption, then the software is simple to operate and compatible with older technology nodes, but the yield prediction accuracy deteriorates in new technology nodes where systematic defects dominate
Solution Approach 1:
The patent segments defects into two distinct categories: systematic defects and random defects. This segmentation allows the yield prediction system to handle each type differently, applying appropriate analysis methods to each category rather than treating all defects uniformly, thereby improving prediction accuracy without overwhelming complexity
Solution Approach 2:
The patent applies local quality by treating systematic and random defects with different analysis approaches tailored to their specific characteristics. Systematic defects receive specialized handling through design element association and conditional probability calculation, while random defects use traditional statistical methods, optimizing the analysis for each defect type's local properties
2Measurement precision
If the Kill Ratio method is used to predict yield loss due to visual defects, then empirical defect classification can be performed, but significant manual classification effort is required and the method requires repeated learning phases when new defect classes arise
Solution Approach 1:
The patent performs preliminary action by pre-establishing the relationship between design elements and systematic defects during the design stage. This preliminary classification framework is built before production, allowing rapid yield prediction without requiring manual reclassification when new defect classes emerge, as the system can map new defects to existing design element categories
Solution Approach 2:
The patent uses copying by applying the established systematic defect patterns and design element relationships from previous designs to new designs. Once the classification framework is created for a design family, it can be copied and applied to subsequent designs, eliminating the need to repeat the entire learning phase
Data Source
AI summary
In one embodiment, a method for predicting yield during the design stage includes receiving defectivity data identifying defects associated with previous wafer designs, and dividing the defects into systematic defects and random defects. For each design layout of a new wafer design, yield is predicted separately for the systematic defects and the random defects. A combined yield is then calculated based on the yield predicted for the systematic defects and the yield predicted for the random defects.