Semiconductor KPI Residual Analysis for Process Drift Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In semiconductor manufacturing, existing monitoring systems fail to accurately detect process drifts and outlier substrates that may be masked by meeting key performance indicator (KPI) specifications, leading to inefficiencies and potential defects in high-volume manufacturing.
Innovation Solution
A method is introduced to decompose and filter out contributions from tools and reticles towards KPI variations, improving the signal-to-noise ratio by using statistical models like ANOVA or ANCOVA to isolate residual KPI variations, which indicate process drifts or outliers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If monitoring systems use KPI specifications to evaluate semiconductor manufacturing processes, then manufacturing precision is maintained within specification limits, but process drifts and outlier substrates are masked and fail to be detected
Solution Approach 1:
The patent segments the total KPI variation into distinct components: systematic variations (attributable to specific tools, reticles, or process conditions) and residual variations (unexplained deviations). By decomposing the variation sources, the system can monitor residual variations separately to detect process drifts while maintaining overall KPI compliance, thus resolving the contradiction between maintaining manufacturing precision and detecting reliability issues.
Solution Approach 2:
The patent introduces an intermediary statistical model (ANOVA or ANCOVA) that acts as a mediator between the monitored substrates and the evaluation criteria. This model separates the influence of known factors from unknown variations, allowing the system to maintain KPI specifications while simultaneously detecting process drifts through residual analysis, thereby resolving the detection masking problem.
2Productivity
If multiple tools and reticles are used in high-volume manufacturing, then productivity increases, but KPI variation increases making it difficult to detect process drifts
Solution Approach 1:
The patent segments the sources of KPI variation by attributing specific portions of variation to individual tools and reticles using statistical models. This segmentation allows the system to account for expected variations from multiple equipment sources while isolating residual variations that indicate genuine process drifts, thus maintaining measurement precision despite high productivity requirements.
Solution Approach 2:
The patent changes the monitoring parameter from raw KPI values to residual KPI variations (after removing systematic components). This parameter transformation allows the system to maintain high productivity with multiple tools and reticles while improving detection accuracy by focusing on the residual variations that truly indicate process drifts rather than normal equipment-to-equipment variability.
3Manufacturing precision
If separate specification limits are applied to different tools and reticles, then manufacturing precision is maintained for each tool, but the complexity of monitoring and control increases
Solution Approach 1:
The patent merges the monitoring of multiple tools and reticles into a unified statistical framework. Instead of maintaining separate specification limits and monitoring systems for each tool, the patent combines all tool and reticle data into a single ANOVA or ANCOVA model that systematically accounts for tool-specific and reticle-specific variations. This merging reduces monitoring complexity while maintaining manufacturing precision through the residual variation analysis.
Data Source
AI summary
Described herein is a method for determining process drifts or outlier wafers over time in semiconductor manufacturing. The method involves obtaining a key performance indicator (KPI) variation (e.g., LCDU) characterizing a performance of a semiconductor process over time, and data associated with a set of factors associated with the semiconductor process. A model of the KPI and the data is used to determine contributions of a first set of factors toward the KPI variation, the first set of factors breaching a statistical threshold. The contributions from the first set of factors toward the KPI variation is removed from the model to obtain a residual KPI variation. Based on the residual KPI variation, a residual value breaching a residual threshold is determined. The residual value indicates process drifts in the semiconductor process over time or an outlier substrate corresponding to the residual value at a certain time.