Semiconductor Yield Prediction Using Neural Network Regression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In the field of semiconductor integrated circuit yield analysis, it is challenging to evaluate rare defect events due to a large number of process variables, making it difficult to predict yields accurately, especially for high-sigma defects that significantly impact production.
Innovation Solution
A yield prediction apparatus using neural network regression analysis and advanced Monte Carlo simulations to determine parametric and defect loss yields by simulating operating and layout characteristics, allowing for the prediction of semiconductor integrated circuit yields based on these analyses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional simulation methods are used to evaluate rare defect events, then measurement precision can be maintained, but productivity decreases due to the large number of process variables requiring evaluation
Solution Approach 1:
The patent creates a neural network model that copies the complex relationship between process variables and yield outcomes. Instead of performing repeated traditional simulations, the trained neural network model serves as a simplified copy that can rapidly predict yield for different process conditions, maintaining accuracy while dramatically improving evaluation efficiency
Solution Approach 2:
The patent performs preliminary training of the neural network model using historical simulation data before actual yield evaluation. This preliminary action pre-computes the complex relationships between process variables and yield, enabling rapid predictions during production without requiring repeated full simulations
2Productivity
If the number of process variables is reduced to improve productivity, then evaluation efficiency increases, but measurement precision deteriorates due to inability to capture rare defect events
Solution Approach 1:
The neural network model captures the complete relationship between all process variables and yield outcomes in a compressed computational form. This copying approach preserves the effects of all variables including rare defect events, while enabling rapid evaluation without requiring all variables to be explicitly processed in each prediction
3Measurement precision
If traditional Monte Carlo simulation is used for yield prediction, then measurement precision is maintained, but loss of time increases due to computational complexity
Solution Approach 1:
The patent performs preliminary training of the neural network model using Monte Carlo simulation data to capture accurate yield relationships. Once trained, the model can rapidly predict yield for new process conditions without requiring repeated time-consuming Monte Carlo simulations, thus maintaining precision while reducing computation time
Solution Approach 2:
The neural network model serves as a computational copy of the Monte Carlo simulation process, pre-learning the complex probability distributions and relationships. This copying enables rapid predictions that replicate Monte Carlo accuracy without the iterative computational overhead
Data Source
AI summary
A yield prediction apparatus is provided. The yield prediction apparatus may include at least one processor coupled to at least one non-transitory computer-readable medium. The at least one processor may be configured to receive a first variable associated with operating characteristics of a semiconductor device, perform a simulation for the operating characteristics of the semiconductor device, perform a neural network regression analysis using a result of the simulation to determine a first function for the first variable, and predict a yield of the semiconductor integrated circuit based on an advanced Monte Carlo simulation. An input of the advanced Monte Carlo simulation may include the determined first function.