Two-Phase Golden Path Search for Multistage Manufacturing Yield
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional yield enhancement methods struggle to identify the root causes of yield loss in multistage manufacturing processes due to the complexity of production-related data and the inability to detect manufacturing paths with significant yield influence, especially when devices of the same type perform differently at each stage.
Innovation Solution
A golden path search method and system utilizing a two-phase algorithm, including a Group LASSO and Group OGA, to identify key process stages and predict yield rates, considering interactions between devices, thereby selecting a golden path with high reliability and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If conventional big data analysis is used to collect all production-related data to find root causes of yield loss, then comprehensive data coverage is achieved, but the complexity of data analysis becomes enormous and it is very difficult to search for root causes
Solution Approach 1:
The patent segments the enormous production data into two categories: final inspection data (Y) and in-line metrology data (y). By separating these data types and analyzing them through different phases (Phase-I for key process stage identification using final inspection data, Phase-II for yield rate prediction using in-line metrology data), the system reduces analysis complexity while maintaining comprehensive data coverage.
Solution Approach 2:
The patent extracts only the most relevant data elements for yield analysis: final inspection values (Y) that indicate yield loss and in-line metrology values (y) that correlate with yield. By extracting and focusing on these specific data points rather than analyzing all production data, the system achieves root cause identification with reduced complexity.
2Measurement precision
If Key-variable Search Algorithm (KSA) is used to identify univariate variables influence on yield, then single device impact is detected, but the manufacturing paths with significant influence on yield cannot be detected in multistage manufacturing process
Solution Approach 1:
The patent transitions from univariate analysis (single device impact) to multivariate path analysis by introducing a new dimension: the manufacturing path itself. Instead of analyzing individual devices in isolation, the system analyzes complete manufacturing paths (sequences of devices) and their collective impact on yield, enabling detection of path-level influences that single-device analysis cannot capture.
Solution Approach 2:
The patent merges multiple in-line metrology measurements (y1, y2, ..., yn) from different process stages into a unified yield rate prediction model for each manufacturing path. By combining these measurements along the entire path, the system captures the cumulative effect of multiple devices on final yield, enabling comprehensive path analysis.
3Reliability
If all possible manufacturing paths are analyzed to find the golden path, then the best yield rate is achieved, but the number of paths to be confirmed becomes extremely large in multistage manufacturing process
Solution Approach 1:
The patent performs preliminary action by first identifying key process stages and constructing yield prediction models for all possible paths before actual production. This allows the system to pre-calculate and rank paths based on predicted yield rates, so that during production, only the top-ranked paths need to be confirmed, dramatically reducing path confirmation time while maintaining reliability.
Solution Approach 2:
The patent applies partial action by analyzing and ranking all possible paths using prediction models, but only confirming and implementing the top-ranked paths (e.g., top 1-5 paths) in actual production. This partial confirmation approach achieves near-optimal yield rates without the time cost of confirming all possible paths.
Data Source
AI summary
A golden path search method for manufacturing process provides a two-phase process to search for a golden path. A first phase step of the two-phase process includes preparing a search model based on a search algorithm, and selecting a plurality of key process stages of a plurality of process stages by feeding sets of final inspection values and the production paths of the workpieces into the searching model, and then generating a plurality of key paths according to the key process stages. A second phase step of the two-phase process includes building a plurality of prediction models of the key paths according to the production paths and the sets of final inspection values, and predicting a plurality of yield rates corresponding to the key paths according to the prediction models, and then searching for the golden path of the key paths according to the yield rates.