Semiconductor Defect Classification Monitoring via Temporal Correlation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current semiconductor defect classification methodologies fail to leverage historical information and time-dependent trends in classifier performance, leading to inadequate quantification of performance changes and potential misinterpretation of defect inspection data, resulting in under-correction or over-correction in classification maintenance.

Innovation Solution

A system and method for monitoring time-varying classification performance by receiving signals from scanning inspection tools, determining defect type populations using classification rules, and computing correlations with high-resolution inspection tools to identify stability and population variations, enabling quantitative analysis and corrective actions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If ad hoc snapshots of classifier performance are observed and qualitative information is used for maintenance, then the process is simple and quick, but the ability to quantify performance changes and identify root causes is insufficient

Engineering Contradiction:

Improveclassifier maintenance processVSAvoidclassifier performance quantification

Core Design Contradiction:

Ease of operationVSMeasurement precision

Solution Approach 1:

The patent implements a feedback mechanism by continuously monitoring classifier performance metrics and comparing them against baseline values. The system automatically detects performance degradation, quantifies the extent of degradation, and triggers retraining processes based on measured thresholds, transforming ad hoc qualitative maintenance into a data-driven quantitative feedback loop

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent replaces manual qualitative assessment with automated statistical analysis systems. Instead of relying on operator judgment of classifier performance, the system uses automated computation of performance metrics, correlation analysis, and statistical testing to objectively measure and quantify classifier degradation, substituting human intuition with computational measurement

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Device complexity

If classifier performance is monitored without considering time-dependency trends, then the monitoring process is simpler, but the ability to identify whether performance change is due to instability or population changes is lost

Engineering Contradiction:

Improvemonitoring systemVSAvoidtime-dependency information

Core Design Contradiction:

Device complexityVSLoss of information

Solution Approach 1:

The patent applies preliminary action by establishing baseline classifier performance metrics before production runs and storing historical performance data for future comparison. The system pre-configures performance thresholds and correlation criteria, so when performance monitoring is needed, the comparison against pre-established baselines and historical trends can immediately identify whether changes represent true degradation or normal variation

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent adds the time dimension to performance monitoring by tracking classifier performance across multiple production runs and comparing current performance against historical baselines. This temporal dimension enables the system to distinguish between transient performance fluctuations and sustained degradation, transforming single-point performance checks into time-series analysis

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

3Ease of manufacture

If ad hoc metrics are used without considering inspector tool hardware and sensitivity, then the metrics are easier to compute, but the ability to account for tool-specific variations and misread significance is reduced

Engineering Contradiction:

Improvemetrics computationVSAvoiddefect inspection data interpretation

Core Design Contradiction:

Ease of manufactureVSReliability

Solution Approach 1:

The patent applies local quality by customizing performance metrics and baseline values for each specific inspector tool and classification algorithm combination. Instead of using universal thresholds, the system establishes tool-specific performance baselines that account for hardware characteristics, sensitivity levels, and detection capabilities of individual inspection systems, ensuring that performance evaluation is tailored to local tool characteristics

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS8537349B2Monitoring of time-varying defect classification performance

Publication Date: 2013.09.17 KLA CORP

AI summary

Systems and methods for monitoring time-varying classification performance are disclosed. A method may include, but is not limited to: receiving one or more signals indicative of one or more properties of one or more samples from one or more scanning inspection tools; determining populations of one or more defect types for the one or more samples according an application of one or more classification rules to the one or more signals received from the one or more scanning inspection tools; determining populations of the one or more defect types for the one or more samples using one or more high-resolution inspection tools; and computing one or more correlations between populations of one or more defect types for one or more samples determined from application of one or more classification rules applied to one or more signals received from the one or more scanning inspection tools and populations of the one or more defect types for the one or more samples determined using the one or more high-resolution inspection tools.