Engineering Team Get Started — Daily Fab-Duty Use of fab spc drift sync portal

Educational engineering purpose only. This is a software architecture exercise and not process-release advice.

Purpose

help PE / PIE / EE / Metrology / Yield engineers understand why the portal exists, what each screen means, how to use it during daily fab duty, and how to turn CD-SEM SPC checking into a production-synchronized drift defense loop.

---

1. What problem does this portal solve?

In normal fab operation, CD-SEM tools often pass routine SPC checks every 8 hours, 24 hours, or once per shift. However, production wafers continue to be measured and released between those checks. If a CD-SEM drifts during this blind window, the tool can still look “green” from the last SPC run while product lots are already being measured with biased or noisy data.

This creates a dangerous chain:

CD-SEM hidden drift
  ↓
Wrong CD measurement / CD-SEM index shift
  ↓
APC receives biased feedback
  ↓
Etch / litho correction moves in the wrong direction
  ↓
Electrical parametric shift / WAT shift
  ↓
Sort yield drop
  ↓
Yield-loss watch

---

2. Design for What?

The portal is designed to close that blind window by combining:

● routine SPC / golden wafer data;

● product-lot CD data;

● CD-SEM FDC hardware health signals;

● dynamic tool matching metrics;

● fleet-wide virtual comparison;

● yield-loss triage logic;

● human-reviewed dispatch / hold / route-limit recommendations.

The page is advisory only. It does not autonomously command equipment, change tool state, modify recipe, or bypass fab approval.

---

3. Who should use this portal?

3.1 Metrology engineer / CD-SEM owner

Use the portal to:

● monitor CD-SEM drift between scheduled SPC checks;

● check tool matching health;

● identify whether abnormal CD is caused by metrology drift or real process shift;

● decide whether to run golden wafer, recalibrate, restrict layer, or call equipment support.

3.2 Equipment engineer / EE

Use the portal to:

● connect FDC hardware signals to metrology impact;

● investigate vacuum, emission current, deflector DAC, vibration, column contamination, stage issue, focus/stigmator issue;

● provide hardware evidence before tool release.

3.3 Process engineer / PE and PIE

Use the portal to:

● avoid changing etch/litho recipe based on suspicious metrology data;

● check ADI/AEI CD-SEM route history;

● understand whether APC feedback should be guarded or paused;

● prioritize critical layers such as Gate / Fin / Contact / Via.

3.4 Yield engineer / YEE

Use the portal to:

● backtrace yield-loss lots to CD-SEM route and FDC state;

● separate real process excursion from metrology false alarm;

● prepare an evidence pack for root-cause review.

---

4. Quick-start: how to open and use the HTML page

4.1 Open locally

● Save fab_spc_drift_sync_portal.html on your PC.

● Double-click it, or right-click → open with Chrome / Edge.

● No server is required. The demo data is inside the HTML file.

4.2 Important demo limitation

The portal currently uses built-in sample tool data such as CDSEM-01 to CDSEM-10. In a real fab, these must be replaced by validated source data from:

● CD-SEM metrology database;

● SPC / golden wafer system;

● FDC historian;

● MES route history;

● APC feedback log;

● YMS / yield-loss watch system;

● equipment event and maintenance log.

4.3 What you should do first as a learner

Open the HTML page and click each top navigation tab in this order:

● OVERVIEW — understand the control-loop concept.

● LIVE RISK BOARD — learn how to read daily tool risk.

● FDC HEALTH-LINK — learn hardware → metrology → yield mapping.

● DYNAMIC MATCHING — learn fleet mean and tool matching rules.

● YIELD TRIAGE — learn what to do when a lot is under yield-loss watch.

● RUNBOOK — learn the POC-to-production implementation path.

---

5. Key concepts from the study PDF

5.1 CD-SEM

CD-SEM means Critical Dimension Scanning Electron Microscope. In fab language, it is the high-resolution metrology tool used to measure critical dimensions such as:

● gate width;

● line / space CD;

● contact / via diameter;

● dense and isolated feature CD;

● LER / LWR related indicators.

It is commonly used at:

● ADI — After Develop Inspection: after lithography develop. If CD is wrong, rework may still be possible.

● AEI — After Etch Inspection: after etch. Data is used for feedback and APC, but the wafer pattern is already fixed.

5.2 Why CD-SEM drift is dangerous

A CD-SEM can drift because of:

● electron gun aging or emission instability;

● gun chamber vacuum degradation;

● beam current fluctuation;

● column / aperture contamination;

● image blur;

● astigmatism drift;

● lens charging / hysteresis;

● stage vibration or stage micro-creep;

● laser interferometer thermal disturbance;

● local wafer charging;

● edge-detection algorithm sensitivity;

● recipe threshold mismatch;

● standard wafer aging or carbon deposition.

When this occurs, the CD result may shift even if the actual wafer did not change.

5.3 SPC blind window

A blind window is the time gap between the last valid SPC / golden wafer check and the current production measurement.

Example:

08:00  Golden wafer SPC passes
10:00  Emission current starts drifting
11:00  Product lots measured and released
14:00  APC receives biased CD feedback
16:00  Next SPC detects issue

By the time SPC detects it, lots may already be affected.

5.4 Inline drift vs lot drift

5.5 Offset drift vs widening drift

This is one of the most important learner points.

Offset drift

Before: mean = 25.00 nm, 3σ = 0.24 nm
After:  mean = 25.30 nm, 3σ = 0.25 nm

This is mainly a mean shift. It may be correctable by matching offset if the tool is otherwise stable.

Widening drift

Before: mean = 24.98 nm, 3σ = 0.24 nm
After:  mean = 25.05 nm, 3σ = 0.72 nm

This is more dangerous. The mean may look acceptable, but uncertainty/noise has expanded. APC and lot disposition become unreliable.

---

6. Core control limits used by the portal

Use these limits as the default learning baseline. Actual production limits must be approved per node, layer, product, customer, and module owner.

---

7. How to read the portal header and KPI strip

At the top, the portal shows:

● Tools Watched — number of CD-SEM tools in the current dashboard.

● SPC Blind Window — average time since last trusted SPC/golden wafer point.

● Fleet OOC Risk — number of high-risk tools in the current demo data.

● FDC Health Links — number of hardware-to-metrology health-link categories.

● Dynamic Limits — shows dynamic control logic is ON conceptually.

● Yield Watch Lots — lots or cases under yield-risk monitoring.

Daily duty interpretation:

● If blind window is high and FDC warning appears, do not wait for next physical SPC automatically.

● If fleet OOC risk increases, check whether one tool is separating from fleet mean.

● If yield-watch lots were measured by one suspect CD-SEM, prioritize metrology triage before process recipe change.

---

8. Tab-by-tab learner guide

8.1 OVERVIEW tab

What it teaches

The overview explains the full control loop:

Detect blind-window drift
  ↓
Convert FDC signals into metrology risk
  ↓
Protect dispatch, lot release, and APC

What learners should notice

● Routine SPC can remain green while production lots are already exposed to drift.

● FDC hardware signals are not just equipment data; they can predict CD metrology instability.

● Dispatch and APC should be protected if measurement credibility is questionable.

AWS data-plane concept shown in the page

The HTML page describes a reference-style data architecture:

● AWS IoT SiteWise for collecting tool/FDC time-series.

● Amazon Timestream for high-frequency sensor and metrology time-series storage.

● AWS Glue and AWS Lake Formation for preparing and governing datasets.

● AWS DataZone for publishing curated data products for PE/PIE/EE/Yield teams.

● Amazon SageMaker AI for predicted TMP/TMG and anomaly models.

● Amazon Bedrock / AgentCore conceptually for evidence summaries.

Learning point: the dashboard is not only a chart page. It represents a governed data workflow from raw tool health to fab decision.

---

8.2 LIVE RISK BOARD tab

What it shows

This is the daily-duty main board. Each row represents one CD-SEM tool and layer context.

Columns:

Search and sort

Use the search box to filter by:

● tool name, e.g. CDSEM-03;

● layer, e.g. Gate, Fin, Contact, Via;

● symptom, e.g. Emission, Vacuum, Vibration, Deflector.

Sort buttons:

● RISK — use for normal shift priority.

● BLIND WINDOW — use when SPC interval is long or tool missed qualification.

● TMG — use during tool matching review.

● SLOPE — use for dense/isolated mismatch or multi-CD linearity issue.

● FLEET σ — use for virtual matching / fleet outlier review.

● RESIDUAL — use when random noise, vibration, or false alarms are suspected.

Expanding a row

Click the action button, such as WATCH, HOLD REVIEW, GOLDEN WAFER, APC GUARD, or ROUTE LIMIT.

The expanded panel shows:

● virtual SPC / production-lot ΔMean trajectory;

● TMG vs UCL;

● slope health;

● fleet σ;

● residual 3σ;

● blind-window age;

● recommended action;

● root-cause hint.

Recommended interpretation of action labels

---

8.3 FDC HEALTH-LINK tab

What it teaches

This tab maps:

Hardware FDC signal → metrology impact → yield-loss risk → recommended defense

Four key FDC relationships

Dynamic interlock concept

When FDC is warning, the portal conceptually predicts metrology risk:

if FDC_state == "warning":
    predicted_TMG = model(gun_vacuum, emission_current, deflector_dac, vibration)
    tighten_TMG_UCL_by = 20%
    block_critical_gate_layer = predicted_TMG > tightened_UCL
    recommend = "downgrade layer / run physical golden wafer / engineer review"

Daily duty interpretation:

● Do not wait until TMG officially fails if FDC shows correlated hardware drift.

● If vacuum is still within equipment spec but predicted residual rises, treat it as metrology risk.

● If emission current ramps across lots, check for mean offset step jump.

● If deflector DAC changes, look at Mandel slope and multi-CD linearity.

---

8.4 DYNAMIC MATCHING tab

What it teaches

Traditional tool matching may compare one tool with one reference tool on a fixed schedule. The portal promotes dynamic matching:

● compare every tool to fleet mean;

● use master tools for critical layer anchors;

● use multi-CD standard wafers;

● monitor slope, residual, site-to-site range, and fleet σ;

● increase matching frequency when risk rises.

Fleet mean guardrail

Instead of only comparing Tool A to Tool B, compare each CD-SEM to a virtual fleet mean:

ΔMean(tool) = Tool mean - Fleet mean

Rules:

● Fleet deviation > 2σ → watch.

● Fleet deviation > 3σ → hold candidate.

Master-slave anchor

Select 1–2 most stable and best-resolution CD-SEM tools as master anchors for critical layers.

Master tools should have:

● stricter maintenance control;

● restricted hardware change;

● stable recipe baseline;

● clear event log for aperture change, venting, beam-on recovery, PM, and calibration.

Multi-CD linearity

Do not match tools using only one CD size. Golden/standard wafers should include multiple line widths because a tool may be correct at 20 nm but wrong at 40 nm.

Mandel slope checks this behavior.

---

8.5 YIELD TRIAGE tab

What it teaches

When a product lot enters yield-loss watch, the portal helps separate:

Real process problem
vs
Metrology false alarm
vs
APC feedback problem caused by bad CD data

Three-step metrology triage

Step 1 — TMG matrix check

Check ADI/AEI route history:

● Which CD-SEM measured ADI?

● Which CD-SEM measured AEI?

● Were those tools cross-tool compatible on that day?

● Was TMG within Target CD × 10%?

If TMG is worse than limit, suspect metrology false alarm before changing etch time.

Step 2 — Waveform profile review

Pull secondary-electron waveform / line profile.

Warning signs:

● peak-to-base ratio lower than baseline;

● slope softer than baseline;

● baseline rising across repeat shots;

● edge signal unstable;

● profile shifted between repeated scans.

Likely causes:

● resolution degradation;

● local charging;

● contamination;

● beam current instability;

● focus/stigmator problem.

Step 3 — Fleet backtrace

If routine SPC is still green, compare production-lot means against the whole fleet.

Look for:

● suspect tool product lots shifted vs other tools;

● fleet deviation >2σ;

● tool/layer/product interaction;

● ADI/AEI pairing issue;

● specific shift or maintenance event.

Evidence pack generated for review

For each yield-loss case, collect:

request_id
lot_id / product / layer / route history
ADI + AEI CD-SEM tool ID
SPC point age and blind-window duration
FDC sensor deltas vs baseline
TMG / TMP
Mandel slope
Site-to-site delta
Residual 3σ
Fleet deviation σ and peer distribution
APC feedback status
Recommended hold / re-route / APC guard action

---

8.6 RUNBOOK tab

What it teaches

This tab gives an implementation path from concept to production.

Suggested POC timeline

Promotion gate

Do not promote the portal to production unless:

● all time-series joins align lot timestamp, SPC timestamp, FDC timestamp;

● TMG, slope, residual, site range, and fleet σ are reproducible from source data;

● warning/hold recommendations are reviewed by owners;

● model output includes reason code, confidence, and safe fallback;

● dashboard is advisory-only unless MES interlock is formally approved.

---

9. Daily-duty operating procedure

9.1 Start-of-shift checklist

● Open fab_spc_drift_sync_portal.html.

● Review KPI strip:

● Tools watched;

● SPC blind window;

● fleet OOC risk;

● yield-watch lots.

● Go to LIVE RISK BOARD.

● Sort by RISK.

● Expand all tools with:

● risk ≥ 75;

● fleet σ > 2;

● TMG over limit;

● slope outside 0.98–1.02;

● residual 3σ > 0.18 nm;

● blind window > shift rule threshold or local rule.

● Cross-check if any yield-watch lot was measured by high-risk tool.

● Decide if action is needed: release, watch, run SPC, golden wafer, route limit, APC guard, or hold review.

9.2 During-shift monitoring

Every 1–2 hours, or before releasing critical lots:

● sort by BLIND WINDOW;

● check tools measuring Gate/Fin/Contact/Via layers;

● review FDC warning tools;

● ensure high-risk tools are not feeding APC without review;

● compare suspicious tool against fleet mean;

● check if any action changed from WATCH to HOLD REVIEW.

9.3 Before releasing critical-layer lots

For Gate / Fin / tight Contact/Via layers:

● Confirm CD-SEM tool is not hold candidate.

● Confirm last SPC/golden wafer is fresh enough.

● Confirm TMG is within rule.

● Confirm Mandel slope is inside 0.98–1.02.

● Confirm residual 3σ and site-to-site delta are within limit.

● Confirm FDC sensor fingerprint is not drifting.

● If uncertain, route to master tool or run golden wafer.

9.4 If a tool shows HOLD REVIEW

Do not immediately assume process problem. Follow this order:

● Freeze further critical-layer release on that tool pending review.

● Check FDC signal history: vacuum, emission, deflector, vibration.

● Run golden wafer / standard wafer.

● Compare with master and fleet mean.

● Review waveform / line profile.

● Check maintenance/event log: aperture change, venting, beam-on, PM, facility event.

● Decide with PE/PIE/EE/Yield:

● release after verification;

● downgrade to non-critical layers;

● hold tool for calibration;

● guard APC feedback;

● backtrace lots already measured.

9.5 End-of-shift handover

Include this in daily passdown:

Date / shift:
Engineer:
High-risk CD-SEM tools:
Actions taken:
Lots held / watched:
Golden wafer result:
FDC abnormal sensors:
Fleet deviation summary:
APC guard status:
Open issues for next shift:

---

10. Practical troubleshooting decision tree

Abnormal CD / yield-loss watch appears
│
├─ Is golden wafer / standard wafer also abnormal?
│    ├─ Yes → likely CD-SEM tool issue
│    │        Check FDC, repeatability, waveform, stage, column, recipe baseline
│    └─ No
│
├─ Is abnormality only on one recipe or one pattern type?
│    ├─ Yes → recipe / edge detection / charging / model issue
│    └─ No
│
├─ Is abnormality only on one product or layer?
│    ├─ Yes → pattern dependency or process-product interaction
│    └─ No
│
├─ Is abnormality tied to one CD-SEM route?
│    ├─ Yes → tool matching / fleet deviation / FDC issue
│    └─ No
│
├─ Is abnormality fab-wide or across many tools?
│    ├─ Yes → upstream process, facility, material, or environment issue
│    └─ No → continue lot/chamber/shift backtrace

---

11. Daily check repeatability troubleshooting

If daily check repeatability becomes worse, remember: the standard wafer is stable and the measurement sites are fixed. If repeatability fails there, suspect tool, environment, recipe, or standard wafer aging before blaming process.

Common causes

Recommended check sequence

● Review waveform:

● rising baseline → charging or leakage;

● peak left/right movement → vibration or beam flicker;

● peak-to-base degradation → resolution/contamination.

● Run beam stability test for 5–10 minutes.

● Check gun chamber vacuum and facility logs.

● Check TCU/chiller temperature stability.

● Move daily-check recipe to a fresh standard-wafer area.

● Re-baseline only after confirming no hardware issue.

---

12. Thermal drift and charging compensation notes

Thermal drift compensation

Common recipe/tool-side actions:

● increase laser interferometer alignment update frequency;

● use global alignment plus die-to-die correction for sensitive layers;

● enable auto target tracking / image shift;

● add column thermal stabilization wait time after venting, PM, or beam-on;

● trigger periodic autofocus and astigmatism calibration.

Local charging compensation

Common recipe/tool-side actions:

● optimize landing energy through acceleration voltage and retarding voltage;

● reduce dwell time;

● use fast scan / TV scan and frame averaging;

● use random or interlaced scan if available;

● apply pre-charging / pre-irradiation when stable saturation helps;

● use flood gun / charge compensation hardware if equipped;

● switch edge detection from fixed threshold to derivative/model-based method if charging causes edge blooming.

---

13. How to replace demo data with fab data

The HTML file has JavaScript arrays near the bottom:

● tools = [...]

● fdc = [...]

● runbook = [...]

For real deployment, do not manually edit daily. Instead, generate this data from governed sources.

13.1 Minimum required tool data schema

Each tool row needs:

{
  "name": "CDSEM-03",
  "layer": "Gate ADI",
  "symptom": "Deflector DAC wobble, slope mismatch",
  "risk": 91,
  "blind": 7.4,
  "tmg": 0.23,
  "slope": 1.031,
  "fleet": 3.2,
  "resid": 0.16,
  "action": "HOLD REVIEW",
  "series": [0.02, 0.03, 0.06, 0.05, 0.08, 0.12, 0.16, 0.20, 0.22, 0.24]
}

13.2 Field meaning

13.3 Recommended source joins

Join by:

● tool ID;

● recipe ID;

● layer;

● product;

● lot ID;

● wafer ID;

● measurement timestamp;

● SPC/golden wafer timestamp;

● FDC time window;

● route step;

● APC feedback event.

Time alignment rule:

For each product CD event:
  link FDC features from prior N minutes / hours
  link latest valid SPC/golden wafer baseline
  link route history and APC feedback
  calculate blind-window duration

---

14. Suggested risk scoring logic

The demo risk score is sample data. In production, calculate risk transparently.

Example scoring design:

Risk = 0
+ 20 if blind window > 6 hours
+ 25 if TMG > Target CD × 10%
+ 25 if Mandel slope outside 0.98–1.02
+ 20 if fleet deviation > 2σ
+ 35 if fleet deviation > 3σ
+ 20 if residual 3σ > 0.18 nm
+ 15 if FDC warning linked to metrology metric
+ 20 if yield-watch lots routed through this tool
Cap at 100

Each high-risk score must show reason codes. Never use a black-box score alone for hold decisions.

---

15. Example daily case walkthrough

Case: CDSEM-03 shows high risk

Portal signals:

Risk: 91
Blind window: 7.4h
TMG: 0.23
Slope: 1.031
Fleet σ: 3.2
Action: HOLD REVIEW
Symptom: Deflector DAC wobble, slope mismatch

Interpretation:

● slope is above 1.02 → nonlinear matching issue;

● fleet σ > 3 → hold candidate;

● blind window > 7h → production exposure risk;

● deflector DAC symptom suggests scan linearity issue;

● critical layers should not be released without review.

Daily action:

● Hold critical Gate/Fin measurement on CDSEM-03.

● Run golden wafer / multi-CD standard wafer.

● Compare dense and isolated CD behavior.

● Check deflector DAC FDC trend.

● Check if yield-watch lots were measured by CDSEM-03.

● Guard APC feedback from suspect lots.

● Decide calibration / route downgrade / release after owner review.

---

16. Common mistakes to avoid

● Only checking mean offset. 3σ widening can be more dangerous than mean drift.

● Waiting for 3σ SPC failure. Trend and shift rules should trigger earlier review.

● Using one CD size for matching. Multi-CD linearity is required to catch slope mismatch.

● Ignoring FDC because CD is still green. FDC may detect hardware drift earlier than SPC.

● Changing etch recipe too quickly. First confirm metrology credibility.

● Trusting one master forever. Master tools also need strict health control and event review.

● Not checking ADI/AEI route pairing. Different CD-SEM pairings can create false deltas.

● Not guarding APC. Bad CD feedback can cause real process shifts.

● Not tracking standard wafer aging. Golden wafer sites can degrade from repeated beam exposure.

● No human review. Portal actions are advisory. Final decisions need fab approval.

---

17. Glossary

---

18. Production safety note

This portal should be treated as a decision-support dashboard. Before linking to MES, dispatcher, APC, or equipment control:

● validate metrics against source systems;

● define owner-approved rules;

● require human approval for hold/release decisions;

● implement audit trail;

● maintain model and rule versioning;

● create fallback mode if data feed is stale or model confidence is low.

---

19. Recommended learner exercise

Use the demo page and answer these questions:

● Which tool has the highest risk?

● Which tool has the largest blind window?

● Which tool has slope outside 0.98–1.02?

● Which tool has fleet deviation above 3σ?

● Which tool needs APC guard?

● Which FDC signal would you check first for vacuum-related residual noise?

● If a yield-loss lot was measured by a high-risk CD-SEM, what evidence pack would you collect?

● If golden wafer passes but one product fails, what does that suggest?

● If daily check repeatability widens but mean is stable, why is that dangerous?

● What should be included in shift handover?

---

20. One-page fab daily-duty summary

1. Start shift → open portal → sort LIVE RISK BOARD by RISK.
2. Check high-risk tools and long SPC blind windows.
3. Expand suspect rows → review TMG, slope, fleet σ, residual 3σ, FDC symptom.
4. If critical layer + high risk → do not release blindly.
5. Run golden wafer or route to master tool if needed.
6. If FDC warning links to TMG/residual/slope → tighten decision boundary.
7. If yield-loss watch appears → perform TMG matrix, waveform review, fleet backtrace.
8. Guard APC if CD data credibility is questionable.
9. Record action and owner decision in shift handover.
10. Remember: portal is advisory; final hold/release requires human fab approval.