What is root cause analysis, and how do 8D and the 5 Whys work?

Root cause analysis means that after something goes wrong, you don't stop at the first cause you see. You keep digging until you reach the layer where, once it's fixed, the same thing won't happen again, and you back that cause up with evidence. Fabs often write the whole process up as an incident report, and a common format is called 8D. Two tools often used to find the cause are asking "why" five times (the 5 Whys) and the fishbone diagram.

How is a root cause different from the cause you can see?

Think of a headache. You take a painkiller and the pain goes away. But if it comes from too little sleep every night, it'll be back next week. The painkiller treats the symptom; too little sleep is the root cause. Tools are the same: if a part wears out fast for a reason, a new one will wear out in a few days too. So you look for why it broke, not just what broke.

What is 8D?

8D is short for "eight disciplines", which here means eight steps. As a problem-solving format, it spread through the auto industry from the 1980s, and many kinds of manufacturing have used it since; when a customer asks a supplier to answer for a quality problem, they often ask for an 8D report. The exact wording varies from place to place.

An example: one etch tool, the same particles three weeks running

The example below is made up.

An etch tool runs a monitor wafer every day to count how many particles it adds (see "What are monitor wafers and dummy wafers?"). Three weeks in a row, the added particles go over the limit, and they're all the same kind, about the same size and shape. The first two times, people cleaned the chamber and replaced a part. It was fine for a few days, then came back the next week. The third time, the chief says: let's do an 8D.

Step 1 (D1): Form a small team

Bring together people who can see different parts of the problem: process, equipment, defect inspection and QA. That's because each person sees only their own piece: process reads the data, equipment knows what was last touched inside the tool, and the inspection people know what the particles look like. One person alone only looks where they already know their way around, which is what happened the first two times.

Step 2 (D2): Describe the problem clearly

Write it down in facts and numbers: what (the same kind of particle), where (only this tool, mostly near the wafer edge), when (it started three weeks ago and comes back a few days after each clean), how much (dozens more per wafer, and how many lots ran in that time). Once it's clear, everyone is looking for the same thing, and later you can tell whether things got better. Write only "lots of particles lately", and everyone has a different problem in mind.

Step 3 (D3): Contain it first

Before you know the cause, stop the damage. This is called containment: put the lots that ran in the last three weeks on hold and check them wafer by wafer, and until the cause is found, run an extra monitor wafer with every lot. Skip this, and lots with problems keep moving down the line during the days it takes to find the cause. Containment is temporary and never counts as closing the case (for how a lot on hold gets decided, see "What is a lot hold?").

Step 4 (D4): Find the root cause

First, use a fishbone diagram to lay out the possible causes. The problem goes at the fish's head, and the bones split into six groups, often remembered as people, machine, material, method, environment and measurement:

• People: how were parts put back after maintenance?
• Machine: buildup on the chamber walls, worn parts?
• Material: the gases, or wafers dirty to begin with?
• Method: a changed recipe or maintenance routine?
• Environment: the cleanroom air, the wafer pods?
• Measurement: is the particle counter counting wrong?

A fishbone diagram only lists possibilities. It can't prove which one is right, so next you rule groups out with evidence. The particle counter is checked with a reference wafer and is fine: cross out measurement. Monitor wafers from the same box come out clean on another etch tool: cross out material and environment. The particles show up only on this tool, and there are more the longer it's been since the last clean: that points to machine and method. Then the particles go for analysis of what they're made of, with a cross-section if needed, and they match the buildup on the chamber walls (see "What is failure analysis (FA)?").

Then, following that thread, ask "why" five times:

1. Why are there particles? Buildup on the chamber walls is flaking off.
2. Why is it flaking? It grew too thick before the scheduled clean.
3. Why is it building up faster than before? Last month the tool switched to a new recipe that leaves more on the walls per wafer than the old one.
4. Why didn't the cleaning move earlier? The maintenance schedule was still set for the old recipe.
5. Why did nobody change it? The form for changing a recipe has no box for "does the maintenance schedule need recalculating?"

Here you hit a process you can change, and that's the root cause. Also ask the other direction: why did it take three weeks to notice? Because the first two times, the case was closed after the clean, and nobody connected the three. "Five" is only a habit: sometimes three gets you there, sometimes it takes more. Different people asking can also end up on different paths, so every answer needs evidence. Stop at the first cause you see, "the chamber is dirty", and you'll just keep cleaning, and it'll come back next week.

Step 5 (D5): Choose a fix and try it small first

The fix has to target the root cause: recalculate and shorten the cleaning interval for this recipe, and add that box to the form. Try it on this one tool for a while first, to see whether the particles go back to normal and how much running time the tool loses to more frequent cleans. Roll it out everywhere untested, and it may not work, or it may cause new trouble.

Step 6 (D6): Put it in place and confirm it really works

After the change goes in for real, watch the monitor wafers and the control chart for several weeks (see "What is an SPC control chart?"). Only if the numbers stay in their usual range does it count as working, and only then do the temporary checks from step 3 come off. That's because "done" and "working" are two different things: you look at the evidence, not just at someone saying it's done. Close the case without checking, and the same thing comes back a while later.

Step 7 (D7): Prevent recurrence and spread the fix

Write the new way into the SOP and the form, so the next person who changes a recipe can't miss it, even without remembering. Then do horizontal deployment: check tools of the same model, and tools running the same recipe, for the same gap, and fix them too. Fix only this one tool, and a few months later the same model next door starts shedding particles, and it all starts over.

Step 8 (D8): Close it out and thank the team

Write up the whole thing and file it, and formally thank the people who worked on it: this was done together. With it on file, the next person who runs into something similar can find how this one was traced and fixed. Without it, they start from zero.

Is finding the cause about finding who's to blame?

No. Root cause analysis looks for gaps in the process, not for a person.

A common mistake with the 5 Whys is stopping at "the operator was careless". People always make mistakes. Write only "human error", and the only fix left is "be more careful next time"; the gap is still there, and the next person falls right in. Ask one more time: why would a careful person make a mistake here? The answer is often a screen that's easy to get wrong or a missing check, and those can be fixed, for example with mistake-proofing. Public occupational safety guidance says the same: concluding that the cause was carelessness or not following a procedure is often misleading. For more examples, see "What does MO mean in a chip fab?".

In fab24hr

Chapter 8 of "The life of a chip", "In a Hurry", is an incident report. You're the etch process engineer at the production fab. On the afternoon shift a manual rework etch goes wrong, and at 15:00 you write a report marked 8D: what happened comes from the tool's log, and containment, impact and root cause each have to come from a different person. When you hand it to the chief, Zhou thinks the root cause should say "careless"; whether you change it his way or write why a careful person still gets it wrong is up to you. At the 16:30 review meeting, the chief asks "Is the same manual operation used anywhere else?", which is horizontal deployment, and each of the three countermeasures needs an owner and a date. The etch process engineer's afternoon shift in "A day in the fab" is the same afternoon.

In chapter 9, "Almost Out the Door", the report has to shrink to one page for upper management; keep the line saying who ran it, and the chief asks you to take the name out. As the quality engineer, on the evening shift the chief asks: on each countermeasure's due date, how will you know it's done? The companies, people, tools and screens in the game are fictional, and so is the etch tool shedding particles above; how 8D, the 5 Whys and the fishbone diagram are used comes from public, general sources, and formats differ from company to company. For other jargon, see Chip fab jargon in plain English.