What is PM (preventive maintenance), and why do fab tools stop on a schedule?

In a chip fab, PM is short for preventive maintenance: a tool that hasn't broken yet is stopped on a schedule, opened up, cleaned, has the parts near the end of their life replaced, gets checked over, and is put back together. When someone in the fab says "this tool is in PM today", that's what they mean. It has nothing to do with the project managers (or product managers) people talk about in offices.

It's a bit like servicing a car: every so often, or after so many miles, you take it in for an oil change instead of waiting for the engine to die on the highway. The difference is that fab tools cost far more, and every hour one stands still gets counted (why tools can't sit idle: Why do chip fabs run 24 hours a day?).

Why stop a tool that isn't broken?

There are two main reasons.

One: deposits slowly build up inside the chamber. In many tools, the place where the wafer is actually processed is a sealed chamber. When a tool etches or deposits a film in there, what comes off doesn't all land on the wafer; some of it sticks to the chamber walls and parts. Wafer after wafer, it builds up layer by layer, and that causes two problems:

• It flakes off: the bits that fall are particles, and a particle landing on a wafer is a defect.
• It makes results drift: as the walls change, the same settings slowly give different results (why: What is etching?).

That's why some university cleanrooms ask users of their thin-film tools to log how much film has built up in the chamber, and to clean it before going on once it reaches a limit.

Two: parts wear out. Some parts spend all day next to plasma, heat and reactive gases. An O-ring, the rubber ring squeezed between two parts to keep gas in, is slowly eaten away and ages in plasma, seals less well, and can shed bits of itself. A vacuum pump that keeps pumping corrosive gases also wears down bit by bit. So these get replaced before they get too bad.

Wouldn't it be cheaper to fix things when they break?

It sounds cheaper. The math often says otherwise.

• Breakdowns don't pick their moment: it might be the middle of the night, or the busiest hour of the week, and the part may not be in stock. When a tool fails at night, the people on the floor first follow the procedure and then call an engineer (see What does OCAP mean in a chip fab?). A scheduled PM is different: the people, the parts and the time are all lined up in advance.
• Not broken, but not good either: the tool is still running lots, yet the wafers coming out have already drifted. By the time a measurement catches it, several lots may have gone through, and yield has already dropped. A yellow light that keeps coming back is sometimes the early sign of exactly this (see What do the red, yellow and green lights on a tool mean?).

So PM trades a stop that is planned, with a fairly predictable length for the kind that may come later, at a bad time, for who knows how long.

When should a tool get PM? By the calendar, by usage, by condition

There are roughly three ways to set the timing:

Method	What it looks at	For a car
Calendar	A fixed number of days or weeks	Every few months
Usage	Wafers run, hours of plasma, how much film has built up	Miles driven
Condition	Sensor readings starting to change	How much tread is left on the tires

The first two are easy to plan and manage, but they don't look at how the tool is actually doing: set them conservatively and some PMs aren't needed yet; set them too loosely and the tool can still break between PMs. The third uses the numbers a tool records all the time, such as the pressure, power and temperature that FDC watches, and it can cut out unneeded downtime. It only works if the data is reliable enough to tell "normal" from "starting to change". Some research tracks the condition of vacuum pumps on purpose, to estimate how long each one has left.

A PM usually isn't nailed to one exact day, either. There's a window it can move in: a little earlier or later is fine, but it can't be pushed back forever. Which day it lands on often depends on how many lots are on the line and how many other tools at the same step are still running.

Why can't production start right after PM?

Because "put back together" doesn't mean "the same as before". The chamber has been open, so the inside has changed, and the PM itself can go wrong: a part not quite seated, a step skipped. So the tool goes through a series of checks:

1. Pump down and leak check: close the chamber, pump the air out, and make sure every joint that was just reassembled is tight.
2. Season: freshly cleaned walls make the first few wafers come out differently from usual. A few dummy wafers (not customer product) are run first so the walls settle down. The season run follows its own recipe too (see What is a recipe in semiconductor manufacturing?).
3. Qual run: test wafers measure what the tool now produces, for example how fast it etches and how many particles it adds. The results are checked against the spec, and also compared with the tool's usual numbers.
4. Tool handback and tool release: the split varies between fabs. A common one is that the equipment engineer finishes the hardware and hands the tool to the process engineer, who confirms the results. Only when both pass does the tool go back to production.

The details of each step are in What does an equipment engineer do in a chip fab?, and how the two engineers divide the work is in What's the difference between a process engineer and an equipment engineer?.

How do fabs balance PM against availability?

A tool's availability is about how much of its time goes to production. During PM the tool can't run lots, so PM counts as down time too, just the scheduled kind, and it's recorded separately from breakdowns: when people work out "how long between failures", scheduled PM isn't counted.

That's where the trade-off sits:

• Too much PM: more downtime, parts replaced while they still had life in them, and availability looks bad.
• Too little PM: when the tool does break, it's down longer and costs more to fix, and bad wafers may come first.

One study, using an ion implant area as its example, starts from exactly this point: PM done too early or too late can both add cost. What you're looking for is the point that's just right. How long a PM takes also varies from one time to the next, and research counts PM among the main sources of ups and downs in tool availability.

In fab24hr

Tap "Clock in", choose "A day in the fab", then on the "Your role today" card tap "Production fab", swipe to the equipment engineer and tap "Etch" under the name: you're the etch equipment engineer. Two of this role's shifts are about maintenance:

• Morning shift: Chamber A of ETCH-03 threw an alarm in the night, and you do the chamber open to replace a part. That's a repair after a failure, not scheduled PM. Later, Kai, your senior, brings up Chamber B of the same tool: "Lin came by this morning to ask about Chamber B's PM. I checked the hours and pushed it to tomorrow." It's still within range on the hours, but it has to be done at 10:00 tomorrow: "We do PM so we don't have to bet on 'maybe.'"
• Afternoon shift: the major PM on the thin-film tool CVD-02 is done, but it fails the particle test: 18 particles, and the spec is 10 or fewer. Kai explains that a major PM is scheduled by wafers run and plasma hours. Wiping the chamber doesn't get it to pass; the cause is the showerhead, where film around the little holes is flaking off. The spare won't arrive until tomorrow. Borrow one from another section, let it run with a monitor wafer on every lot, or keep it down until tomorrow: your call. Before you leave you still have to write the PM record, and Kai says: "A PM record is written for whoever does the next PM." On some days, the afternoon shift is about the litho area's coater/developer, which came back from PM that morning, instead.

As the etch process engineer (swipe to the process engineer and tap "Etch" under the name), the morning handover sheet already has the line "ETCH-03 Chamber B: PM scheduled for 10:00 today", and you're the one who asks whether it can be pushed back: Chamber A is already down, and with both chambers stopped the tool has no capacity at all. Chapter 7 of The life of a chip, "Firefighting", is the same morning.

The companies, people, tool numbers and figures in the game are fictional; what PM is, why it's done and how it's scheduled come from public, general sources. What actually gets done in a PM, and how often, differs for every kind of tool and every company. For other jargon, see Chip fab jargon in plain English.