What is change control in a chip fab, and why does one change need so many sign-offs?

Change control is the procedure a chip fab must follow before it changes any process condition: write down what will change and why, assess what it could affect, try it on a small scale and compare, get the right people to review and sign, tell customers first when needed, and only then make the change; after that, keep watching. A few degrees hotter, a few seconds longer, a different gas, a new tool, a different material supplier: all of these count. The change notice sent to customers is usually called a PCN.

Why not just change it if it looks better?

Because a chip is built up through hundreds of steps, and a change often doesn't show itself at the step where it was made.

Here's an example. Soak the wafers a few seconds longer during cleaning, and they may come out cleaner at this step; but the chemical also eats a little more of the layer underneath. Nothing looks wrong at this step. Only when the wafer is finished and its electrical numbers are measured (WAT, see "What is WAT?") does one number turn out to be off, dozens of steps later. Some effects only appear after a chip has been in use for a long time, so a process change often means redoing reliability tests, such as HTOL, where chips run hot for about six weeks straight (see "What is burn-in?").

The other reason is the customer. Customers design their own products around the chip and qualify it, and what they qualified is "the chip made this way". Only when every lot is made the same way can they keep using it. The standard process that production follows is called the POR, and that's why it's locked down so tightly.

What counts as a "change"?

Manufacturing often sorts the things that can change into four groups, called the 4Ms: man (the people), machine, material and method. In a chip fab, the last three look like this:

• Machine: running the step on a tool that hasn't run it before, or switching to a part with a different specification. The same numbers on a different tool can give a different result.
• Material: a gas or chemical from a different supplier, or a supplier changing its own formula.
• Method: changing the numbers in a recipe (temperature, time, gas flow, power), or adding or removing a step.

Fitting a new part of the same specification during maintenance is usually treated as maintenance; but the tool still needs a qual run afterwards.

What are the steps of a change?

Here are seven common steps. Names and details differ from fab to fab.

1. Propose the change: write down what will change and why. Everyone after you starts their judgment from that "why". Leave it vague, and others can only guess, and later no one can trace why it was changed.
2. Assess the impact: find which products use this step, which later steps could be affected, and how big the risk is; bigger risks need more checking. One step usually serves more than one product, so looking only at the product in front of you can break another.
3. Trial it small, measure, compare: use test wafers, or a split lot that sends a few wafers through the new condition and keeps an unchanged group as a baseline, often over several lots. With a baseline you can tell whether the new condition caused a difference or that lot just happened to be different. Try it on a whole customer lot from the start, and a mistake costs the whole lot.
4. The right people review and sign: depending on what the change touches, people from Process, Integration, QA and Equipment look at the data and sign. Each of them looks at something different: this step, the steps after it, the quality of what ships, and whether the tool can take it. If only one person says yes, it has only been seen from one angle, and problems anywhere else go unseen.
5. Tell customers when needed, and wait for their reply: if the change can affect customers, a PCN goes out first (next section). Customers need time to work out whether their own products are affected. Change it without telling them, and when they find out later, the trust is gone.
6. Make it official: the recipe, the documents, the training: the system switches to the new version and the old one is retired (see "What is a recipe in semiconductor manufacturing?"), the SOP is updated, and the people who follow it learn it before the work starts. The tool, the documents and the people all need to follow the same version. If the system changes but the paper doesn't, the night shift may do it the old way.
7. Keep watching after the change: once the new condition is live, the measurements still go on the control chart, and the first few lots after the change are watched more closely (see "What is an SPC control chart?"). The trial lots can't cover every situation production will run into. Without someone watching, the problem may not show up until it reaches the customer.

What is a PCN, and when do customers have to be told?

PCN stands for product (or process) change notice: a document a supplier sends its customers that says what changed, why it changed, and what impact is expected.

Not every change needs one. Public industry standards sort changes into two kinds: a change that could affect form, fit, function, quality or reliability is a major change, and customers must be told; whether a minor change is reported depends on what the two sides agreed in advance. Moving production to another fab, changing the process and changing materials are often major changes.

The notice has to come early, too. A common industry practice is to give notice at least 90 days before the first shipment of the changed product; shipping it earlier takes the customer's approval or a separate agreement. For products used for many years, such as cars, the requirements are stricter, and a change often means qualifying again to a standard (see The news says "automotive chips").

What are temporary changes and deviations, and how are they different from APC's automatic corrections?

	How long	Example
Change	From now on, the new way	Changing one number in the POR
Temporary change	Has an end time; then it goes back	Running a different set of conditions for the few days a tool waits for a part
Deviation	Only the lot (or lots) on the form	A lot waited past its time limit, and an engineer judged it still usable

A temporary change is still requested and signed, and it also states when it ends; the worst case is getting through the rough patch and nobody remembering to change it back. A deviation stays with that lot, so if a problem turns up later, it can be traced.

APC's corrections are something else: after each lot is measured, the system follows rules designed in advance and adjusts only the few numbers agreed on, so the next lot lands closer to target (see "What is APC?"). On the other hand, a person reaching in on the spot to change a locked number is a change that skipped the procedure, even if the result looks fine.

What does an audit check?

An audit is someone checking, against requirements agreed in advance, whether "what's written" and "what's done" match (see "What does QA (quality assurance) do in a chip fab?"). There are internal audits, where the fab checks itself, and customer audits, where a customer sends people to check.

For changes, two things get checked often: whether the documents and recipes in use on the floor are the latest approved versions; and whether every change has a record of what changed, why, where the data is and who signed. The international standard for quality management also requires keeping the results of reviewing a production change and a record of who authorized it. The records aren't just for auditors: if something goes wrong one day, they are how you find which change it was and from which lot it started.

In fab24hr

In fab24hr, tap "Clock in", choose "A day in the fab", then on the "Your role today" card tap "Production fab", swipe to the process engineer, and tap "Wet clean" under the name: you're the wet clean process engineer. On the morning shift, "Climbing Particles", you can lengthen WET-05's soak time to get by and move the acid change to the afternoon; choose that, and Chief Ho says: "Lengthening the soak time is a temporary change, and I've signed the form." On the evening shift, "The Stronger Bath", the freshly changed bath is stronger, and if the morning's longer time hasn't been changed back, the wafers get over-cleaned.

Swipe to the equipment engineer and tap "Wet clean", and the evening shift, "Too strong", is more direct: the amount a monitor wafer lost is right at the upper limit. You can ask Process to put the soak time back and confirm with a monitor wafer, or turn the heater down a few degrees yourself and get tonight's lots through. The second is faster, but there's no entry for it on the process condition sheet.

In the main story, "The life of a chip", Chapter 4, "Two Weeks Early", the R&D line's transfer checklist has the POR on it, and the customer's sign-off too; in Chapter 5, "The Last Box", you fill in the POR's last empty box, and the production fab will use that number for years.

The companies, people, tool numbers and figures in the game are fictional; how change control works comes from general descriptions in quality management and industry standards, and the real names, forms and who has to sign differ from company to company. For other jargon, see Chip fab jargon in plain English.