What does QA (quality assurance) do in a chip fab?

In a chip fab, QA (short for quality assurance) makes sure what leaves the fab can be trusted: before a lot goes out, someone has looked at it and said it can go; when something goes wrong, the cause can be traced and the countermeasures are followed until they're really done. Process engineers look after whether each step comes out right, and equipment engineers look after whether the tools run (see What's the difference between a process engineer and an equipment engineer?). QA looks at something else: was the path that should have been followed actually followed, and was it enough?

How is QA different from QC?

Think of a shop that sells lunch boxes. Glancing at each box before it goes out, checking the food is right and nothing has fallen in, is quality control (QC). Running the whole shop's way of working, how ingredients are received, how cold the fridge is, how a problem gets traced, so that every box is hard to get wrong, is quality assurance.

The international quality management standards (the ISO 9000 family) draw the same line: QC is about meeting the quality requirements, and QA is about giving people confidence that those requirements will be met.

	QC (quality control)	QA (quality assurance)
Looks at	What was made	The whole way of working
Asks	Does this lot pass?	Why should anyone trust every lot passes?
Typical work	Measuring, inspecting, sampling	Procedures, audits, following up incidents, explaining to customers

The quality expert W. Edwards Deming has an often-quoted line: inspection comes too late; whether a product is good or bad was settled when it was made. So QA tries to make bad output hard to produce, not just to pick it out afterwards. QC is often treated as one part of QA, and the two sometimes sit in one department; the names and the split differ between companies.

What does QA check before a lot ships?

Before wafers leave the fab for test and assembly, there's one last check: outgoing inspection. It looks at two things:

• The records: whether every step was finished, whether anything was put on hold and never resolved, and whether the WAT results (electrical measurements taken when the wafers are finished) are in range.
• The lot itself: wafer count, labels and appearance, for example scratches on the surface.

There's too much product to look at every wafer closely, so a set number are picked using a sampling table agreed in advance, and what's found in the sample decides whether the lot can go.

Sampling can miss things. Say a box of 25 wafers has 3 bad ones and you look at 5. Work it out and there's about a one-in-two chance that all 5 you picked are good. So once a sample does find a problem, the whole lot often gets checked again; removing only the wafer that was caught isn't enough.

Until the checks are done, the lot isn't released, and the release record has to show who said it could go. Product with a problem is marked and kept apart, so it can't ship by mistake.

Sometimes a lot didn't follow the rules exactly, for example it waited a little past the limit between two steps (see What is Q-time, and why can't wafers wait?), and an engineer judges it still usable. Then a deviation is written: what was different, why it can still be used, and who signed. In some cases the customer has to agree too. QA checks that the form is clear and that the right people signed it.

When a customer says a chip has a problem, how does QA investigate?

When a customer finds chips that don't work, or a lot whose yield is worse than usual (see What does "yield" mean in chip news?), QA is often one of the teams that takes it on. The first step is to ask clearly: which lot, what happens, and at which step the customer found it.

Returned parts usually start with a return ticket, called an RMA (return material authorization). Then comes failure analysis, and the checks that don't damage the part come first, then it's opened up step by step: look at the outside, retest it electrically, X-ray it, and only at the end open the package or make a cross-section. Once it's opened, there's no going back.

QA also traces back through the records: the MES records which tools each lot went through, when, and with which recipe, so you can find which other lots went through the same tool at the same time.

Finally a report goes back to the customer. A common format is the 8D, which the fab's own incident reports often use too: what happened, how it was contained, what the cause was, and how to stop it happening again. Customers for car chips are stricter, often counting "how many bad parts per million", and the closer to zero, the better.

When something goes wrong in the fab, what does QA follow up?

When a tool throws an alarm or someone makes an operating error (see What does MO mean in a chip fab?), fixing the tool and finding the cause is mostly Process and Equipment's job. QA follows whether the whole loop gets closed:

1. Containment first: stop the lots that might be affected, then check whether other lots were hit too.
2. Find the root cause: ask not only "why did it break?" but also "why wasn't it caught sooner?". Common tools are asking "why?" five times in a row, or drawing a fishbone diagram that sorts the possible causes into groups.
3. Set countermeasures: each one needs an owner and a date.
4. Horizontal deployment: check whether other tools and other products have the same gap.
5. Confirm it worked: on the due date, look at the evidence; hearing "it's done" doesn't count.

QA also goes back over the everyday records: whether anyone is looking at the control charts (see What is an SPC control chart?), and whether people followed the OCAP when something went wrong (see What does OCAP mean in a chip fab?). Where things don't match, the point isn't who missed something, but where the procedure makes it easy to miss.

What is an audit, and who audits whom?

An audit is a planned check, against requirements agreed in advance, of whether "what's written down" and "what's done" match, with a record kept. There are three common kinds:

• Internal audits: the fab checks itself.
• Customer audits: a customer checks a supplier's fab and processes.
• Third-party audits: an independent certification body checks, for example, whether the fab meets ISO 9001 (the international standard for quality management systems), and issues a certificate if it does.

Audits often look at whether documents are the latest version, whether records were kept, and whether last time's findings were fixed. Findings are handled like incidents: someone fixes each one, by a date, with evidence when it's done.

Who does QA work with?

• Process and equipment engineers: they investigate incidents together; they fix, and QA follows it through to done.
• Process integration: whether a wafer is kept (disposition) and whether a deviation can be signed are often decided together (see What does a process integration engineer (PIE) do?).
• Manufacturing: outgoing inspection, lots on hold, handovers (see What does the manufacturing department do in a chip fab?).
• Production control: when a lot is blocked and delivery slips, they replan it together (see What does production control do?).
• The test-and-assembly house: the wafers' next stop, which has its own QA (see What is an OSAT?). In some fabs QA also looks after reliability, for example taking a sample of chips for a life test such as HTOL (see What is burn-in?).

In fab24hr

Tap "Clock in", choose "A day in the fab", then on the "Your role today" card tap "Production fab" and swipe to "Quality engineer". Which of the four shifts you get follows the time you open the game:

• Late-night shift, "Release or Block": you're QA on-call this week. A bit after one in the morning, Hao on the night shift calls: a lot of 25 wafers is due on the truck at 06:30, he sampled 5, and 2 have scratches near the edge, one of which runs into the area where the chips are. Check all 25 again, block the whole lot, or pull just that one wafer: your call.
• Morning shift, "By the Plan": at 10:00 you go to the etch area and check the night ETCH-03 went down against the OCAP, step by step. Three items don't match. Do you record them as "the night shift didn't do it", or as "gaps in the form and the plan"?
• Afternoon shift, "How to Record It": at 14:05, a manual run on ETCH-05 goes wrong. You preserve the scene first, then match what the tool actually ran against what the MES recorded, wafer by wafer.
• Evening shift, "Write to the Customer": you write the customer a statement, then turn the three countermeasures from the review meeting into a tracking table: who, by which day, and what counts as done.

In chapter 12 of the main story, The life of a chip, "The Night Before Shipping", you're the customer liaison at a test-and-assembly house, and its quality engineer Mei-Hua goes through the pre-shipment checklist with you item by item: "One record missing, and it doesn't ship."

The companies, people, lot numbers, tools and figures in the game are fictional; what QA does comes from the general wording of quality management standards and public material on statistical quality control, and the real split of work, the names and the forms differ between companies. For other jargon, see Chip fab jargon in plain English.