A Chip Fab Doesn’t Fear Blackouts. It Fears the Two Seconds Before One.

The transfer test went perfectly. Both generators cranked, synced, and took the building in eleven seconds — well inside the target. The facilities manager smiled. Then a technician ran in from the cleanroom: three process tools had faulted. The wafers inside them were scrap. Total interruption of power? About four-tenths of a second.

That was the day I learned what semiconductor fabs really fear. It’s not the blackout. It’s the blink.

The sag, not the outage, is what kills the wafers

Here’s the thing about fabs. A warehouse can lose power for an hour and lose nothing but time. A chip fab can lose power for a fifth of a second and lose more money than the generator building cost to build.

The grid almost never fails cleanly. Before the lights go out, the voltage sags — a lightning strike twenty miles away, a fault on a parallel feeder, a big utility switching operation. The voltage dips to 70%, maybe 50% of nominal, for a few cycles. Your eyes wouldn’t even see it. The process tools feel it like a punch.

Lithography, etch, deposition — these tools live on hair-trigger power monitoring. A dip below their threshold and they trip, mid-wafer. And a scrapped batch in a modern fab isn’t a rounding error. It’s the kind of number that makes a facilities director go quiet in a meeting.

So the backup-power question in a fab is upside down compared to everywhere else I’ve worked. Nobody is asking “can the generator start?” Of course it can. They’re asking: “can anything on this site feel it start?”

What SEMI F47 quietly asks of your generator

The semiconductor industry wrote its own answer to this problem. It’s called SEMI F47 — a voltage-sag immunity standard for process equipment. Roughly speaking, it says: tools must ride through sags down to about half of nominal voltage for around 200 milliseconds, with shallower dips tolerated longer. (Check the current standard text for the exact curve; it gets revised, and your spec should reference the edition, not my memory.)

Now read that from the generator’s side. The standard draws a line in the sand: the tools will survive a dip this deep, this long. Everything below that line is your problem.

And a diesel generator, on its own, lives below that line for the first several seconds of every start. Cranking, firing, ramping to speed, the AVR hunting for voltage, the governor hunting for frequency, the first load step dragging everything down again — the first ten seconds of a generator start are electrically ugly. Anyone who believes “generator equals clean power” has never watched a frequency meter during a cold transfer.

This is the part that surprises newcomers to fab power design. The diesel generator is the muscle of the whole scheme — it runs for hours, days if needed — but it is not, and cannot be, the precision instrument. Expecting a generator to deliver SEMI F47-grade power during its first seconds is like expecting a fire truck to do surgery. Right vehicle, wrong job.

Three layers, one job

Every fab power scheme that actually works is a sandwich. Three layers, each covering what the others can’t:

• The diesel generator — the long-haul muscle. It carries the site for hours or days, and it feeds everything downstream. Without it, the other two layers are just countdown timers.
• The static UPS (double-conversion) — the bodyguard. It sits between the generator and the critical tools, delivering perfect sine waves no matter what the generator or the grid is doing. It covers the transfer gap and the generator’s rough first seconds. Its battery time is the most expensive line item in the whole scheme, per minute.
• Flywheel or DRUPS (diesel rotary UPS) — the bridge. A flywheel stores kinetic energy and covers short sags that are too small to start the diesel but too big to ignore. A DRUPS takes it further: diesel engine and generator on the same shaft as the flywheel, so there’s no transfer break at all for the protected bus. Some fabs bet the whole plant on DRUPS sets. It’s elegant. It’s also one very expensive machine with one very expensive maintenance schedule.

Honestly? If someone made me pick one layer to oversize, it wouldn’t be the generator. It would be the UPS battery time. Generators rarely fail to start in a well-maintained fab. What fails is the arithmetic: transfer time plus governor settling time exceeds battery hold-up time, by four seconds, once, at the worst moment. I’ve seen that math drawn up on whiteboards by very smart people who still got it wrong, because they used the brochure’s transfer time instead of the measured one.

What a diesel generator must actually do in a fab

A fab isn’t just a cleanroom. The facility loads around it are some of the nastiest in industry, and the generator has to carry them — or at least the ones that matter.

Start with the chillers. A fab’s process cooling load is enormous, and chillers are induction-motor monsters with starting currents that make a generator’s voltage sag like a tired boxer. Then the cleanroom air handlers — fans the size of small cars, and they can’t stop, because the cleanroom’s particle count starts climbing the minute airflow drops. Then ultrapure water pumps, vacuum pumps, bulk gas systems.

And then there’s the load nobody puts on the nice diagram: abatement. The scrubbers that treat toxic process exhaust. Those cannot stop. Ever. A scrubber going down during an outage isn’t a production problem — it’s a safety problem, and the local environmental authority will treat it that way. In one project I was involved with, we found during commissioning that the abatement system wasn’t on the UPS-protected bus at all. It had been drawn on the one-line diagram, value-engineered off it in a late revision, and nobody noticed until we walked the as-builts. That discovery, made two weeks before handover, is the reason I now walk every protected bus myself with a highlighter.

This is also where motor starting math stops being theoretical. The generator isn’t sized by adding up the kW list. It’s sized by the first three seconds of the biggest chiller start, with everything else already running. Get that wrong and the generator starts fine, takes the first load step, and then folds when the second chiller tries to start — which is exactly what happened on a site I visited years ago, where the “400 kVA of spare capacity” turned out to be spare only on paper.

Clean power has a maintenance bill

People price the generator. They forget to price everything that keeps the power clean.

A double-conversion UPS big enough for a fab’s critical bus burns a few percent of everything passing through it, all day, every day — that’s real electricity money, year after year. Its batteries need replacement on a schedule, not when they feel tired; a five-year-old VRLA string in a warm room is a promise, not a battery. Flywheels need bearings and vacuum pumps. DRUPS sets need the kind of maintenance attention that only a dedicated technician provides — and the technician has to actually exist, on site or on contract, not just on an org chart.

Look, I’ve watched a fab spend seven figures on a gorgeous DRUPS installation and then argue for six months about the maintenance contract. The machine was perfect. The maintenance was a line item somebody kept “deferring.” That’s like buying a racehorse and feeding it leftovers. The most expensive backup power scheme in the world is the one nobody maintains.

My rule for any fab generator spec

If I were writing the generator spec for a fab tomorrow — and I have, more than once — it would have four non-negotiable lines.

One: witnessed step-load testing to ISO 8528 G3 performance class, with the actual load profile of the fab, not a generic load bank script. A load bank is polite. A chiller start is not.

Two: an excitation system that can take the beating — PMG or auxiliary-winding excitation on the alternator, not a shunt-excited afterthought. When the voltage dips during motor starting, the AVR needs its own independent power source to force the field back up. This is the difference between a generator that recovers in a second and one that takes five.

Three: UPS battery hold-up time specified against measured transfer-plus-settle time, with margin. Not the brochure number. The number from the last witnessed test. Plus twenty percent, because the next test will be worse.

Four: the protected-bus audit I mentioned. Walk it. Every load that must not blink — tools, scrubbers, controls, the UPS control power itself — gets highlighted on the one-line, then verified in the field. If it isn’t on the drawing and on the bus, it doesn’t exist.

And one thing I’d cross out of every fab spec I’ve ever reviewed: the sentence “process tools are on UPS, so generator power quality is not critical.” That sentence has cost more wafers than any generator failure I’ve ever seen. The generator feeds the UPS. The UPS feeds the tools. Garbage in doesn’t always mean garbage out — a good UPS cleans a lot — but it means shorter battery life, hotter UPS components, and a thinner margin every single time. The generator’s job is to give the UPS something decent to work with. That’s the deal.

Fabs are the most demanding backup-power customer in the world, and honestly, they’re my favorite kind of project — because there’s no room for hand-waving. Either the tools stay up through the sag or they don’t, and the wafers keep score. If you’re speccing backup power for a fab and want to argue about step loads, excitation systems, or protected-bus audits before the concrete is poured, talk to us — we’d rather fight about chiller starting current now than about scrapped wafers later.