Root Cause Analysis Beyond 'Operator Error'

When a defect or incident happens, someone writes a report. And many reports end the same way: "Root cause: operator error. Action: operator retrained." It closes quickly and nobody objects. A few months later the same problem shows up with a different operator.

Root cause analysis isn't only the quality department's job. What a new hire looks at first when they meet their first defect sets how well the organization prevents repeats. The semiconductor fab and construction site maps in First Run include several quests about the moments right after a problem is found. This article uses them to set out the order of root cause analysis.

"Operator error" is a symptom, not a cause

This is a quality and yield quest in the fab. A customer field failure has been traced to a particle defect, and the customer wants an 8D report. The team suggests closing it fast with "root cause: operator error, action: retraining".

• The record shows the operator skipped a wipe step, so write "operator error, retrained" and close it: Poor
• Hold off until our responsibility is proven, and first ask the customer for the failed part and all their data: Poor
• Send containment first and leave the root cause "under investigation" until an engineer is free: Risky
• Follow the 8D: contain, find the occurrence and escape causes with evidence, then put in permanent actions and verify they work: Excellent

The feedback makes the key point in one line: "operator error" is usually a symptom, not a cause. If the gaps in procedure, equipment and inspection that made the error possible are still there, it will happen again. Whether you use 5 Whys or an 8D, the starting point is not stopping the questions at a person's name.

Two terms in the Excellent answer matter: the occurrence cause and the escape cause. Why the defect happened and why it wasn't caught before reaching the customer are different questions, and both have to be closed. The reason "under investigation" is Risky is also clear: an open-ended investigation promises the customer nothing. It needs a timeline and interim updates.

Step 1: set the containment scope

Before finding the cause, stop the problem from spreading. The question is how far.

In the fab's process engineer quest, a lot comes off one lithography scanner with overlay shifted to one side, and that scanner was calibrated yesterday.

• The calibration was approved, so hold only the lot that failed measurement and run the rest: Poor
• Stop every scanner in the fab until the cause is known: Risky
• Keep the scanner running and double overlay sampling from the next lot: Risky
• Inhibit the scanner and use MES to find and hold every lot that went through it since its last confirmed good point: Excellent

The containment scope is not "what was caught" but "everything that went through the same conditions". Measurement is only sampling, so holding just the one lot lets others with the same problem move on. Stopping the whole fab without evidence halts healthy tools and lots too. Contain where the evidence points first, and widen it only if the data says so.

The construction site's pressure test quest has the same shape. Before the ceilings close, the 7th-floor sprinkler line has no pressure test record. Postponing finishes on floors 5 to 8 and retesting all four is Good: safe, but it stops floors whose records are fine. Excellent holds the 7th floor only, retests with someone witnessing and photos as the record, and tells the finishing crew about the changed sequence straight away.

Step 2: stop irreversible work and check the facts

In the construction quality quest, the 7-day compressive strength of the 8th-floor slab cylinders comes back lower than expected. The 28-day results aren't in, and the 8th-floor shoring is due to be removed tomorrow.

• The 7-day result is only indicative, so remove the shoring as planned: Poor
• Cast new cylinders from another floor, retest and record the better result: Poor
• Tell the whole site the 8th-floor slab is defective and demand a demolition review: Risky
• Report straight away to the quality and construction leads and hold the shoring removal; check curing temperature records and delivery tickets, and decide with supervision and the structural engineer whether more testing is needed, alongside the 28-day results: Excellent

The two Poor answers show root cause analysis's two enemies: ignoring a signal and swapping the evidence. Cylinders from different concrete say nothing about the 8th-floor slab. The Risky answer errs the other way: it's too early to call the slab defective on a 7-day result, since temperature and curing conditions can make early strength low. Excellent stops only the irreversible work, checks the candidate causes (curing temperature, the delivered concrete) against records, and makes the call with the experts.

Step 3: check that the measurement can be trusted

Finding a cause from data requires data you can trust. In the fab's quality quest, a new film-thickness gauge has been installed, and the line wants to use it to disposition product next week. Gauge R&R hasn't been done.

• Same maker, same model, signed install report: use it for disposition now: Poor
• Widen the spec limits slightly to absorb the measurement spread: Poor
• Correlate it with the existing gauge on the same wafers and use it for monitoring only until the full study is done: Good
• Bring the Gauge R&R and reference-gauge matching forward, and release it for disposition once it passes: Excellent

An unvalidated gauge scraps good lots or passes bad ones, and analyzing causes with its data sends you looking in the wrong place. The Good answer falls short because correlation alone doesn't show repeatability and reproducibility. A measurement system analysis (MSA) has to show that measurement variation is small enough relative to the tolerance before the gauge's calls can be trusted.

Step 4: segregate by source and keep the record

Rebar arrives on site, but the mill certificate's heat number doesn't match the tags and some bars have flaking rust. The rebar crew wants to start cutting today.

• Use it only for secondary bars that look less critical: Poor
• A little rust is normal, so start cutting today and get the right certificate later: Risky
• Refuse it at the gate, send the truck back and order a replacement by phone without recording anything: Risky
• Tag the lot "on hold", stack it separately, request the correct certificate and check its condition with supervision; use it only once it passes: Excellent

The third option is interesting. Quality was protected because the steel wasn't used. But without a record, the same supplier repeats the same problem and the schedule slip can't be explained. Root cause analysis isn't only for after an incident: recording nonconformances is what feeds the next analysis.

The co-op map shows the same principle. When pesticide residue is found in a spinach sample, the Excellent answer stops shipments, runs the recall and notice, and uses the spraying records to find the cause. In the minced-meat complaint, throwing the foreign object away is Risky, because the only clue to the cause disappears.

The order of a root cause report

The Excellent answers line up into an order a new hire can follow.

1. Contain: find and stop everything that went through the same conditions since the last confirmed good point.
2. Check the facts: stop irreversible work and compare records with reality.
3. Check the measurement: confirm the measurements and tests behind your judgment are trustworthy.
4. Find the causes: look for the occurrence cause and the escape cause separately; don't stop at a name.
5. Act and verify: put in permanent actions and confirm they work.
6. Share: give stakeholders a timeline, interim updates and the results.

Checklist for root cause analysis training

• Is closing with "operator error, retrained" explained as a wrong answer?
• Is there a quest that separates the occurrence cause from the escape cause?
• Are containment scopes that are too narrow and too wide both wrong answers?
• Are ignoring a signal and swapping the evidence both graded Poor?
• Do the quests cover judgments made before the measurement system is validated?
• Do they ask both for not using nonconforming material and for recording it?
• Do they cover that "under investigation" needs a timeline and interim updates?

All the fab quests are covered in Semiconductor new hire training scenarios, and a reporting culture that doesn't blame people after an incident in Teaching new hires to report near-misses.

Try workplace scenarios as onboarding quests

Pick the quality and yield engineer on the semiconductor fab map to play the 8D quest yourself. Play the Onboarding