Weighted Relative Bias Estimation for APC Sampling Delays
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In semiconductor manufacturing, advanced process control (APC) strategies face challenges due to limited sampling rates and measurement data uncertainty, leading to aged and unreliable data for process tool biases, which reduces control efficiency and increases the risk of process deviations.
Innovation Solution
The method involves determining and weighting relative biases in a state estimation matrix based on the age and uncertainty of measurement data to enhance control accuracy, using a model-based control system that calculates updated manipulated variables for process tools, thereby accounting for the delay and noise in measurement data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If measurement data is collected at a limited sampling rate, then the measurement process is simplified and less time-consuming, but the measurement data becomes aged and less reliable for determining current tool biases
Solution Approach 1:
The system performs preliminary actions by collecting and storing measurement data over time at a limited sampling rate, then uses predictive modeling to estimate current tool biases before actual measurement is needed. This allows the system to maintain reliable bias estimates without requiring frequent measurements, thus resolving the contradiction between measurement throughput and data reliability.
Solution Approach 2:
The patent introduces an intermediary element - a predictive model - that bridges the gap between limited sampling rate measurements and current process control needs. The model acts as a mediator that translates historical measurement data into current bias estimates, allowing the system to maintain reliability without increasing measurement frequency.
2Reliability
If measurement data is collected frequently to maintain reliability, then current tool biases are accurately determined, but the measurement time and resource consumption increase significantly
Solution Approach 1:
The system performs measurements at a reduced frequency in advance, storing the data for later use. The predictive model then extrapolates this historical data to determine current biases, eliminating the need for frequent time-consuming measurements while maintaining accuracy.
Solution Approach 2:
The patent replaces the mechanical measurement system with a computational prediction system. Instead of physically measuring tool biases frequently, the system uses mathematical models to predict current biases from historical data, significantly reducing measurement time and resource consumption.
3Quantity of substance
If a limited number of substrates are subjected to measurement, then measurement resources are conserved, but the uncertainty in determining tool-specific systematic deviations increases
Solution Approach 1:
The predictive model serves as an intermediary that compensates for the information loss caused by limited substrate measurements. By using historical data and process knowledge, the model infers tool biases even when direct measurement data is scarce, thus preventing information loss while conserving measurement resources.
Solution Approach 2:
The system changes the parameter of measurement frequency and sample size dynamically. When measurement resources are constrained, the system relies more on predictive modeling with fewer measurements. When resources are abundant, it performs more frequent measurements to update the model, thus optimizing the balance between resource consumption and information quality.
Data Source
AI summary
By directly using relative biases, contained in the relative bias date matrix, and by appropriately weighting the components thereof, sampling rate limitations in an APC control scheme may be efficiently compensated for. In particular embodiments, an age-based weighting factor is established that scales measurement data uncertainty according to the delay with which the corresponding measurement data for a specific control thread are obtained.