How long does a wafer take to make, and why does cycle time run to months?

Going from wafer start to a finished wafer that leaves the fab takes several weeks to several months. The usual industry figure is about three months on average (around twelve weeks), and the more advanced the process, with more layers, the longer it takes. This stretch is called cycle time. The odd part is that, of those months, the time a wafer actually spends inside a tool being processed is only a small share. Most of the time, it is waiting.

Where does cycle time start and end?

It starts at wafer start: a lot of wafers goes onto the production line and the first step begins (why a fab counts from that day is in "What is a wafer start, and why does a fab count from that day?"). It ends when the wafers finish the last step and leave the fab.

A finished wafer still goes on to assembly and test, which takes a few more weeks. The "lead time" you see in the news runs from a customer's order to chips in hand: before the wafers start there may be a wait for capacity, and afterwards come assembly, test and shipping. So it is often much longer than the cycle time.

Why does it take so long?

Because there are so many steps. A chip is built up layer by layer, and for every layer the wafer makes a round of litho, etch, thin film and other areas (what each area does is in "What do the seven process areas of a chip fab do?"). A chip has dozens of layers, which adds up to hundreds of steps, and over a thousand for complex processes.

The more advanced the process, the more layers it has and the more photomasks in a full set, so the longer it takes. People who study fabs often compare them by "how many days, on average, each mask layer takes", and the answer is on the order of one to a few days. A made-up example: a process with 40 mask layers at 2 days a layer comes to 80 days, close to three months.

What is a wafer doing all those months?

Think of a health checkup at a hospital. The blood test, the X-ray and the eye test each take a few minutes; yet the whole morning goes by, and you spend most of it sitting in the hallway, waiting for the next station to call your number.

A wafer is the same. It doesn't spend much time inside tools being processed; the rest of the time it waits, and there are several kinds of waiting:

1. Waiting in line for a tool: other lots are ahead. This is usually the biggest part.
2. Waiting for transport: in a 12-inch fab, wafers travel in a FOUP, carried between tools by the OHT running along the ceiling ("How do wafers move around a fab?"). If the vehicle hasn't come, the pod waits on the shelf.
3. Waiting for measurement: after many steps, the wafers are measured first, and only go on once they check out ("How does a chip fab know each step went right?"). The more you measure, the longer the line at the metrology tools.
4. Waiting to fill a load: tools that take many wafers at once, like a furnace, often wait until enough lots have gathered before they start.
5. Waiting for a decision: a lot is put on hold while an engineer decides whether it can go on.

Research papers on fabs often put it this way: a lot spends most of its time not being processed, but waiting.

What does "so many times the theoretical time" mean?

People who study cycle time first work out a "theoretical time": suppose that at every step the tool happens to be free and the vehicle happens to arrive, so the lot never waits a minute. How long would it take from the first step to the last? That time counts only the processing itself, and is also called the raw process time.

Then you divide the actual time by it and get a multiple, often called the X-factor, or sometimes the flow factor.

• 1x: no waiting at all. No real fab gets there.
• 3x: as a made-up example, a lot that could in theory be done in 20 days actually takes 60. Only a third of the time is processing; two thirds is waiting.

Different products have different numbers of steps, so comparing days isn't fair; the multiple shows who waits more. Real fabs run well above 1x, and the closer to 1x, the better.

Why are the lines so long?

The fuller the tools, the longer the wait

This is one of the most basic reasons, and the wait doesn't grow in proportion (how production control balances the two is in "What does production control do in a chip fab, and how are due dates set?"). The simplest queueing model in the textbooks gives a rough picture. Say a tool takes 1 hour per lot on average, and both when lots arrive and how long each one takes are quite random:

How busy the tool is	Average total time a lot spends there
Half the time	2 hours
80%	5 hours
90%	10 hours

The processing is 1 hour on average every time; everything beyond that is waiting in line. A real fab is more complicated than this model, but the shape is the same: when the tools are nearly full, a little more work makes the line grow fast. That's why a fab runs 24 hours a day but doesn't book its busiest tools to 100% ("Why do chip fabs run 24 hours a day?").

The same wafer keeps coming back

For every layer, the wafer goes back through litho, etch and the other areas; some lots return to the same group of tools dozens of times, and each time they queue again (why lining up lots is hard is in "What does the manufacturing department do in a chip fab, and how are lots scheduled?").

Hot lots cut in

A hot lot usually goes first at every step. The waiting it skips is waiting the other lots take on; research has found that the more hot lots on the line, the longer the average cycle time of regular lots, and the harder it is to predict ("What is a hot lot, and why doesn't it always go first?").

Tools stop

While a tool is down or in maintenance, it can't run lots, and lots pile up in front of it. A stop doesn't only take away capacity; it also makes "when will it be my turn" harder to predict, and research has found that both make the average cycle time longer ("What is PM (preventive maintenance), and why do fab tools stop on a schedule?").

What does a shorter cycle time buy?

1. Faster, more reliable delivery dates: lots finish sooner, and extra lots added later arrive sooner too; how long each lot waits is steadier, so the dates Production control gives can be trusted.
2. Less work tied up on the line: with the same daily output, if each lot stays a shorter time, there are fewer lots on the line, and less money tied up in them (the relationship is in "What does the manufacturing department do in a chip fab?").
3. Problems come back sooner: much of whether a chip's electrical behavior is right can only be measured once the wafer is completely finished ("What is WAT, and how does a wafer map show what went wrong?"). A new process raises its yield one round at a time: find a problem, change the settings, check again. With a shorter cycle time, each round comes sooner. Research papers on fabs have discussed how yield relates to cycle time for a long time.
4. Q-time is harder to overrun: the Q-time countdown includes the time spent waiting in line. Shorter waits between two steps mean fewer lots near their limit, and fewer to redo ("What is Q-time, and why can't wafers wait?").

The price: for lots to wait less, the tools need some spare time, and not every tool can be booked full. So a fab is always balancing "don't leave tools idle" against "don't keep lots waiting too long".

In fab24hr

In fab24hr, the role that deals with cycle time most is Production control. Tap "Clock in", choose "A day in the fab", then on the "Your role today" card tap "Production fab" and swipe to Production control. On the late-night shift it's 05:30, and you're at home reading the overnight WIP report on your phone: the etch tool ETCH-03 went down at 02:47, with nine lots queued behind it, and the report says "Cycle time at the etch step will be longer than usual today." The hot lot H-2417 is at the etch of this layer, with several more layers to go; it leaves the fab in a little over four weeks. You go through the rest of its route step by step: which dates have moved, and whether the buffer in the schedule can absorb it. Play Manufacturing, and the same morning you split the nine lots behind ETCH-03 across the tools that can still run. The main story, "The life of a chip", says so too: at the start of Act three, "Test-and-assembly house", you learn that after that day H-2417 spends a little over four more weeks in the production fab before it is sent to the test-and-assembly house. The company, people, tool numbers, lot numbers and dates in the game are fictional; what cycle time is and where the time goes follow public, general sources and research papers. For other jargon, see Chip fab jargon in plain English.