- Excursion
- An event in which defects or process performance deviate from the baseline
- Golden Reference
- An approved reference with a defined applicable scope
- Matching
- Evaluate matching against agreed metrics and tolerances
- Drift
- Gradual baseline deviation relative to golden reference over time
First Clarify Which Engineering Question the Investigation Addresses
| Comparison | Production Defect Investigation | Equipment Matching Diagnosis |
|---|---|---|
| Starting Point | Anomalies in defect count, type, or distribution | New equipment, return to service after maintenance, or consistency evaluation |
| Reference | Progressively search for comparable normal operating conditions | Usually an approved tool or reference distribution agreed in advance |
| Objective | Narrow the Possible Sources of the Anomaly | Quantify deviations, stability, and chamber-to-chamber matching |
| Deliver | Candidate sources, evidence, gaps, and check recommendations | Matching assessment, deviation regions, and verification recommendations |
- 01AnomalyProduction Defect Investigation: Start with the Abnormal Group
References may be identified as the investigation proceeds, such as contemporaneous normal groups or batches before the anomaly.
- 02SourceFind Common Differences in Processing Histories
Use reference groups, time, and process conditions to screen candidate sources.
- 03ReferenceTool Matching: Define an Approved Baseline First
New-tool, return-to-service, and cross-chamber comparisons usually have references and tolerances agreed in advance.
- 04DeliveryQuantify Deviations and Identify Verification Directions
Check means, distributions, spatial patterns, and stability, and explain which metrics remain unmatched.
More Abnormal Wafers Passing Through a Tool Is Not Enough
Also ask how many normal wafers passed through it. If the equipment handles most production, a high hit rate may simply reflect scheduling. Check whether product, recipe, time, and maintenance state vary together with the equipment.
Before-and-after inspection can help identify newly detected defects. However, sensitivity, masks, coordinates, and material changes affect detection; differences must not all be called defects newly generated by this process step.
Tool Matching Extends Far Beyond Comparing Means
Two chambers with identical average film thickness can exhibit completely different edge gradients and wafer-to-wafer variances. Evaluations must compare means, standard deviations, radial/azimuthal distributions, and lot-to-lot stability against process specifications, logging measurement repeatability.
Advanced: Difference Testing vs Equivalence Testing
ANOVA tests differences in group means; distribution tests examine other distribution features. Failure to reject a sameness hypothesis may simply reflect insufficient samples and does not establish engineering equivalence. Define acceptable deviations, samples, and uncertainty in advance; equivalence or non-inferiority designs may be appropriate. Defect counts and spatially correlated data also require suitable statistical methods.
Equipment Drift: Introducing the Temporal Dimension into Baselines
Current tool performance can be compared against its historical baseline or parallel peer tools, but baselines must specify product code, recipe version, and maintenance status. A baseline shift following preventative maintenance clean is not necessarily a fault; long-term sensor trends do not automatically convict component wear. New tool qualification requires broader verification than matching, encompassing long-term stability and process flow compatibility.
Two Distinct Tasks Demand Tailored Deliverables
Defect investigations must specify how far suspect sources were narrowed and document competing candidate mechanisms. Tool matching evaluations must quantify metric deviations and determine conformance to acceptance thresholds. When evidence is insufficient, recommend supplementary checks; specific recipe tuning requires validated models or controlled splits, rather than recommending arbitrary gas flow adjustments based solely on defect images.
WAFERPILOT / APPLICATION TO THIS TOPIC
Transforming Comparisons into Engineering Recommendations
When underlying data and methods are connected, WaferPilot structures comparative analyses and executes agreed workflows to generate source-attributed diagnostic or matching dossiers. It does not unilaterally authorize production release, tool holds, or APC updates; current interactive demos are distinguished from end-to-end production scenarios.
Deliverables for Engineers:Defect investigations deliver candidate root causes and localized tiers; matching tasks deliver deviation metrics, baseline applicability, and verification split plans.
View the Current Demo and Capability Scope →Related Reading & References
Below are verifiable industry references; specific processes, permissions, and acceptance criteria are governed by fab site specifications.