How is capacity utilization calculated?

[robertjm]

Hey SemiWiki, sorry for the basic question.

I was going over Hua Hong's recent 2Q21 earnings report, and as you can see from the report they reported the following:

8" Wafer/month in 2Q21
8 inch fab #1: 65K 8" wafers/month
8 inch fab #2: 60K 8" wafers/month
8 inch fab #3: 53K 8" wafers/month
12 inch fab: 90K 8" equivalent wafers/month (40K 12" wafers)
Total 8'' equiv. wafers: 268K 8'' equivalent wafers/month

So as you can see, Hua Hong says they can produce 268K 8'' wafers per month at 100% capacity utilization. In a quarter, that means they can produce 804K 8" equiv. wafers.
However, on their report there was conflicting information. They reported:

Capacity utilization: 109.5%, which translates into 8'' equiv. wafer output of 880K in 2Q21;
Actual wafer shipments: 730K 8'' equiv. wafers!!

Even if we assume that they produced using 1Q21 capacity (which was 241K/month), they should have still produced 792K wafers at 109.5% capacity!

Can anyone explain why such a gap exists? Am I doing the calculation wrong? I'm very confused. Thanks

 

[count]

Nameplate capacity doesn't exactly equal actual capacity. There may be adjustments for maintenance, downtime, yield, ect. So the confusion might be in one place they are talking about nameplate capacity, and in another place they are talking about actual capacity after whatever adjustments they make.

 

[Jert]

So, in this case, actual utilization is 730/804 = 90.7% utilization rate. For fabs, that's pretty much about the maximum utilization one can achieve, on average.
804 wafers/quarter is the installed capacity. No factories of any kind can achieve 100% utilization because of equipment maintenance, scheduled down time and unscheduled down time, factory bottlenecks, factory flow issues, etc.
109.5% utilization only means they received more business from their customers than their factory capacity can handle. There is not such a thing of >100% utilization.

 

[Daniel Nenni]

In my understanding yield is not a factor in utilization but maintenance and scheduled downtime is. So 100% utilization includes scheduled down time. If the fab skips scheduled downtime they can go above 100% utilization. TSMCs utilization dropped last year due to COVID (delayed wafer orders) but this year it should be 100% or more.

 

[count]

So, in this case, actual utilization is 730/804 = 90.7% utilization rate. For fabs, that's pretty much about the maximum utilization one can achieve, on average.
804 wafers/quarter is the installed capacity. No factories of any kind can achieve 100% utilization because of equipment maintenance, scheduled down time and unscheduled down time, factory bottlenecks, factory flow issues, etc.
109.5% utilization only means they received more business from their customers than their factory capacity can handle. There is not such a thing of >100% utilization.

I disagree.

Nameplate capacity is capacity in name only. It's based on the rated capability of equipment, staffing levels, maintenance and downtime schedules. It's possible to exceed 100% of nameplate capacity by deferring maintenance, temporarily increasing staff in certain areas, temporarily pushing equipment past rated capacity. This is not something a company can do for an extended period of time, but there absolutely is such a thing as >100% capacity.

As an example, I once worked in a production operation and we were operating at something like 120% capacity during a peak period. However there were serious safety concerns, because a critical safety system of the facility was only designed for operation at nameplate capacity. We did some risk assessment and decided we would redesign the safety system, but there would be a couple of months where we were taking some risk. So we were running at over 100% capacity for a few months while we redesigned the safety system and then changed the nameplate capacity of the facility after the project was complete.

 

[robertjm]

Thanks for the answers from everybody. So I checked the theory suggested by the first three commenters to find-out what the actual internal capacity is at 100%. I only did it for their 8'' fabs, since their 12'' fab is still ramping up, so I didn't want to introduce more room for error.

It was basic division, just divided the output from the quarter by the reported overcapacity. Here's the result for data going back to 3Q 2018:

I think the result is a bit curious, it indicates that their internal 100% capacity is unstable QoQ.

Mean implied capacity was 520.8K wafers/quarter.
Standard deviation of capacity was 19.4K wafers.

Thanks again for the input from all, looking forward to hearing more ideas.

 

[benb]

Say your fab can produce 100K wafers per month. One month the fab produces 106K. How is this possible?
Simple, you have a 65 day fab cycle time and started 106K wafers 1 month in advance, instead of 100K. The fab cycle time extended to 75 days. But you got more wafers out at a cost to the cycle time.

The fab 100K average production rate is a net figure. It includes all the practical things mentioned: Tools don’t operate 24/7, they have scheduled downs; tools don’t stay occupied all the time, they have some idle time (caused by bottlenecks elsewhere); there is some variation in cycle time of the tool caused by loading, where a steady state load cycle time is longer than the cycle time starting from no load; and 3 or 4 other buckets of time I can’t think of at the moment. These are somewhat universal in factory operations, it’s the same for Tesla or Samsung or TSMC or Hua Hong.

You don’t learn about these 6-7 buckets in school or from books, unfortunately, but Sematech or SEMI have standards that define the buckets and define cycle time, lead time and other important words that different organizations use differently (making it frustrating). I’m certain there must be good ISO or IEC standards which would be even more universal.

90% utilization is pretty typical and includes 3 or 4 of the buckets, depending on the organization, and you wouldn’t be able to compare Tesla with Hua Hong without knowing which buckets each uses in the utilization calculation. 106% utlization has some meaning, I’m sure, at some organization, but it doesn’t mean magic, it either isn’t an average, or includes something unusual in the numerator or denominator. And it definitely doesn’t imply maintenance being deferred or other shortcuts taken. My intuition from statements like that is the utilization over one period, like a month, is being compared to the longer-term average, and the cycle time is growing.

 

[3ertram]

In my opinion, there is no reasonable formula out there to somewhat accurately calculate general Fab capacity (and even less, to compare capacity between different Fabs). The impact of product mix and start rates of different products will result for most Fabs easily in a +/- 15…20% total capacity, since different bottlenecks will limit different products (exception to that are possibly leading edge Fabs, where EUV might be the common bottleneck) Second large influence comes from cycle time , as mentioned above. Depending on market situation different cycle times are acceptable / offered and the very same factory might have 50k capacity at 2dpml , 45k at 1.8dpml , 42k at 1.5dpml. I would be very cautious to use any general Fab capacity statement, without more detailed context, or at least assume the number you see is only +/- 25% accurate