What is a wafer start, and why does a fab count from that day?

A wafer start is putting a lot of wafers onto the production line and starting the first step. When people talk about how many wafers a fab starts in a month, they say wafer starts per month. From the day of its wafer start, the lot works through hundreds of steps, one station at a time, and many of a fab's clocks and numbers are counted from that day.

What happens on the day of a wafer start?

Work in a fab travels in lots. In a 12-inch fab, a lot usually rides in one sealed wafer pod, a FOUP, which holds up to 25 wafers (how the pods get around: How do wafers move around a fab?).

Two things have to be ready before a lot can start:

• Photomasks: the "negatives" used to print the circuit pattern. Only once the chip design has been handed over and made into masks does the fab have something to print.
• A route: which stations the lot will visit and what conditions each one uses (its recipe), laid out ahead of time like a very long itinerary (more on recipes: What is a recipe in semiconductor manufacturing?).

At the moment of the start, the system gives the lot a lot number, and the manufacturing execution system, MES, starts keeping track: which station the lot is at and where it goes next. At every station after that, the system issues a work order that tells the tool and the people what to do with this lot.

At this point the wafers are still bare, shiny silicon. In many process flows the first station is a clean: it washes away invisible particles and traces of metal, so the next station can grow a clean first layer.

Why count from the day of the wafer start?

Think of planting rice. The day the seedlings go in more or less decides when you harvest and how much. If you want more partway through, all you can do is plant another batch and wait for it to grow. Wafers are the same, and three things are counted from that day:

1. Cycle time: the time from wafer start until the lot is finished and leaves the fab. The common industry figure is about three months on average, and the more advanced the process, with more layers, the longer it takes. Most of that time isn't spent being processed; it's spent waiting in line at tools.
2. Due dates: once you know the average, you can count forward from the start date to estimate when the lot will be done and when it can go to the customer. After the wafers are finished, they still go on to packaging and testing before they become chips anyone can use.
3. How much ships: how much a fab can ship a few months from now is largely decided on the day the wafers start. Adding more partway means the extra wafers start again from the very first station, which takes another several weeks to months.

So how many wafers to start, and when, is usually planned by production control from orders and capacity. A lot that a customer is in a special hurry for can be marked as a hot lot, which goes to the front of the line at every station.

What does "wafer starts per month" mean?

When the news or industry reports describe how big a fab is, they often give its monthly capacity in wafers: the most wafers the fab can start in a month. That's wafer starts per month. How many it actually starts depends on orders, so it isn't always running full.

Wafers come in different sizes, so totals are often converted to one size before they are added up. A common form is "200 mm (8-inch) equivalent". A 12-inch (300 mm) wafer is 1.5 times the diameter of an 8-inch one, so its area is 1.5 × 1.5, about 2.25 times, and one 12-inch wafer counts as a little more than two.

Keep in mind that wafer starts count wafers, not chips. On the same wafer, bigger chips mean fewer of them, and a lower yield means even fewer that work (What does "yield" mean in chip news?).

Is starting more wafers always better?

No. Tools can only do so much in a day. If you start more than the line can finish, the extra doesn't turn into output. It just waits in line: the unfinished work on the line (WIP) piles up, and every lot waits longer. There's a simple relationship behind this, explained in What does the manufacturing department do in a chip fab?. That's why production-planning research often recommends checking how much work is already on the line and how full the busiest tools are before deciding how much to start.

What is a risk start?

For a new product, the normal order goes like this. First comes a pilot run on the R&D line, to find the problems and make the process stable. Then the reliability tests have to pass (for example, running chips hot for a long time to see whether any fail early). Only then does the product go to the production fab for wafer starts and volume production.

A risk start means starting wafers in the production fab before everything is confirmed, typically while the reliability tests are still running. Why take the risk? Because it takes months to get from wafer start to finished wafers. Start early, and by the time the tests are done, the wafers are well along, so the customer gets chips sooner.

The risk is just as clear: if the tests fail, the lots started early may have to be scrapped whole. So whether to do a risk start, and who pays for what, is agreed between the fab and the customer in advance.

How is tape-out different from a wafer start?

The two terms often come up together, but they mean different things:

	Tape-out	Wafer start
Who hands over what	The chip designers hand over the final design	The fab puts a lot of wafers onto the line
What comes next	Photomasks are made from the design	The wafers go station by station and finish months later
How often	Once per design version; changing the design means new masks	Every lot has its own start date

The order is: design → tape-out → photomasks → wafer start → hundreds of steps in the fab → packaging and testing. The name tape-out is usually traced to the early days, when the final design files went out on paper or magnetic tape. Masks can be used again and again, and one set can print many lots, so a design version is taped out once while wafer starts come one lot after another. When design and manufacturing happen at different companies, see how the handover works in The news says "foundry". What is it?.

In fab24hr

The main story, The life of a chip, follows a made-up hearing-aid chip, "Glowmere". In the prologue, "Tape-out", Glowmere's design goes to the wafer fab and is made into a set of photomasks. Chapter 1, "How to Split 12", is the day before wafer start: you are the R&D integration engineer, and at the split meeting you decide how to split the first 12 wafers. In chapter 2, "Station by Station", Day 1 is called "Wafer start": the 12 wafers begin on the R&D line with a clean and then go into a furnace, and over the first five days you go through the seven front-end process areas, doing one step by hand in each. In chapter 4, "Two Weeks Early", about four months after wafer start, the customer wants the chips two weeks early, but the reliability test still has three weeks to go: you recommend either waiting for it or a risk start.

You can also meet it in "A day in the fab": tap "Clock in", choose "A day in the fab", then on the "Your role today" card tap "R&D line" and swipe to "R&D integration engineer". Its four shifts are different days on the same timeline: the day before wafer start, the sixth day after it (the 12 wafers have just been cleaned, the next furnace has broken down, and their Q-time keeps counting down), a month or two later, and about four months later. The companies, people, lot numbers, chip and timeline in the game are fictional; what wafer starts, tape-outs and risk starts are comes from public, general sources. For other jargon, see Chip fab jargon in plain English.