What is an SPC control chart, and how are OOC and OOS different?

SPC stands for statistical process control: using statistics to check whether a process is stable. Every measurement is plotted on a control chart to see whether it still looks the way it normally does. The chart has two kinds of limits, and crossing each one has its own name: OOC (out of control, outside the control limits) means "not normal anymore", and OOS (out of spec) means "no longer good". The first is a warning; the second is a problem.

What does a control chart look at? Think of your commute

Say you take the bus to work, and it usually takes about 30 minutes: sometimes 27, sometimes 33, depending on the traffic lights. You have to be in by 9:00, and you leave at 8:00 every day.

• One day it takes 45 minutes. You arrive at 8:45, not late, but that's far from normal. Something happened on the way.
• For 8 days in a row, it takes a little more than 30 minutes: 31, 32, 31, 33... No single day is strange, and you're never late, but that many days on the same side probably means roadwork has started somewhere.
• One day it takes 65 minutes and you walk in at 9:05. You're late.

The first two are "not normal anymore", like OOC. The last one is "no longer good", like OOS. A control chart lines up each day's number as a row of dots so you can tell at a glance which is which.

What are the three lines on a control chart?

Each time a fab measures a lot, for example the post-etch line width, it adds a point to the chart. The points run left to right in time order, and three horizontal lines cross the chart:

• Center line: the process's usual average.
• Upper and lower control limits: usually drawn three standard deviations above and below the center line. A standard deviation is the statistician's ruler for "how much does it normally wobble". While the process hasn't changed, and its numbers roughly follow the familiar bell curve (the normal distribution), almost every point lands between the two limits; fewer than three in a thousand fall outside.

The key point: these lines are calculated from the process's own data, from a stretch of records taken while it was stable. They describe "this is how this step, on this tool, normally behaves". What the customer wants doesn't enter into it.

In a fab, almost every important number has its own control chart: film thickness, line width, how well two layers line up, and the particle count on monitor wafers. For how those numbers are measured, see How does a chip fab know each step went right?. The control chart is an old idea: Walter Shewhart in the United States proposed it in the 1920s, and factories still use it today.

What are spec limits, and how are they different from control limits?

Spec limits are the range the customer or the design requires: for example, line width must be between so many and so many nanometers for the product to pass. They answer "can this be used?", and they have nothing to do with how much the process normally wobbles.

	Control limits	Spec limits
Where they come from	Calculated from the process's own data	Set by the customer or the design
The question they answer	Is this normal?	Is this good?
Crossing them is called	OOC	OOS
On your commute	Much slower than usual	Late

When a process's normal wobble is small enough, its control limits sit inside the spec limits, so OOC often shows up before OOS: you see the process changing before anything has been made wrong. That is exactly what a control chart is for.

If it's the other way round, and the calculated control limits are wider than the spec limits, the process wobbles too much even on a normal day. Every point can be "normal" and it will still make some parts that fail. Then it's the process itself that needs to change; in the industry, its "process capability" is too low.

Why can a point inside the limits still be OOC?

If nothing has changed and the numbers wobble evenly up and down, each point is about equally likely to land above or below the center line, like a coin toss. The chance of tails 8 times in a row is 1 in 256. So when you really do see 8 points in a row below the center line, the more likely explanation is that something has changed.

The patterns that count are called run rules. Common ones:

• One point outside the control limits.
• Several points in a row on the same side of the center line (8 is common; some use 9).
• 6 points in a row going steadily up, or steadily down.

The last two catch slow change: a part wearing a little at a time, a chemical bath slowly getting used up. Every point is still inside the limits, but the whole string keeps moving one way. Each fab decides which rules to use and how many points count. The more rules you use, the sooner you notice, and the more false alarms you get: with only the first rule, by the same bell-curve arithmetic, about one point in 370 is a false alarm.

Why not adjust the tool as soon as a number drifts a little?

The other half of a control chart's job is to tell you when to leave things alone. Every process has a natural wobble made of many small causes added together, and that's just how it is. Treat that wobble as a problem and tweak the tool each time a number moves, and the results get more scattered, not less. The well-known funnel experiment in quality textbooks shows exactly this.

So SPC splits variation into two kinds. Small ups and downs inside the limits: leave them. Points outside the limits, or lined up in one direction: find the cause.

If a lot is OOC, is it no good?

Not necessarily. OOC says "the process has changed"; the lot itself may still be in spec. It still has to be dealt with, though. A common first step is to put the lot on hold and measure again, because a changed number can also mean the metrology tool itself has drifted. Then the people on the line follow a plan written in advance, called the OCAP (see What does OCAP mean in a chip fab?). In the end an engineer decides: release the lot, or keep it back and look further. The person reading the control charts and deciding on OOC lots is usually that step's process engineer (see What's the difference between a process engineer and an equipment engineer?).

And if a lot is OOS?

OOS means it has already failed. The lot usually goes on hold, and an engineer decides what to do: if the step can be redone, it's rework; if it can't, the wafers may have to be scrapped. Then comes the question of what else is affected: since the last time the numbers looked normal, which other lots went through the same tool? They may have the same problem. Not every lot is measured, so the longer the gap between measurements, the more lots have to be traced.

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 "Etch" under the name: you're the etch process engineer. The morning shift shows you both kinds of OOC:

• On metrology tool MET-01, eight post-etch line width points in a row sit below the center line, each lower than the last, yet none is outside the control limits. Trace the dates back, and the first low point was three days ago.
• At 10:40 Ya-Ting calls about the OOC lot in Metrology: its line width is a little above the upper control limit, still in spec, and it's still on hold. Release it with a note, or keep it back and let it wait longer.

After the 16:30 review meeting, you can go back to the metrology tool for another look: with the part replaced, is the line width back around the center line? On the morning and afternoon shifts a random event can also turn up: the film thickness chart for thin-film tool CVD-02 has 8 points in a row above the center line. "Still within spec, keep running", or follow the OCAP and ask Thin Film to check the tool first? The process integration engineer's evening shift has a contact resistance chart creeping upward too, and you look through the earlier steps to find which number moved first. The company, people, tool numbers and figures in the game are fictional; "8 points in a row" follows one common rule, and fabs use different numbers of points. What SPC, OOC and OOS mean follows public, general sources. For other jargon, see Chip fab jargon in plain English.