Wafer area relevance?, Chapter 3

September 8, 2006

View from Silicon Valley- "Wafer area" relevance, Chapter 3

(c) copyright View from Silicon Valley. All rights reserved.

On August 19, we wrote, "Does wafer growth impact profits?"   (http://www.viewfromsiliconvalley.com/id248.html)
Reader correspondence led on September 5 to, "'Wafer area' relevance?, Chapter 1"   (http://www.viewfromsiliconvalley.com/id259.html)
and on September 6 to, "'Wafer area' relevance?, Chapter 2"   (http://www.viewfromsiliconvalley.com/id259.html).

Today we're going to dissect the critique and share some thoughts on a few of the critic's assertions.

In short our critic's claims included:
1) the preferred predictor for semiconductor demand is unit volume rather than revenue
2) it is "more meaningful" to use SICAS than SEMI (since SICAS includes fab utilization).

Until next quarter's wafer area shipped is released, we hope to be done with this topic but, for now, some "serious 'splainin" seems appropriate.

Here goes:
1) the preferred predictor for semiconductor demand is unit volume rather than revenue.
We were shown a chart "proving" wafer output and unit demand correlated perfectly from 2000 to 2006.   The chart's creator is to be congratulated --for proving absolutely nothing.

He then slips in a clever shift in terminology by equating "unit demand" with "units shipped." We submit these are not the same thing. (Anybody ever heard of inventory? And how impossible it is to calculate inventory in today's global supply chain?) In the last recession, the difference between the two was only recognized in hindsight.

Even so, units shipped may be a reasonable predictor --EXCEPT when prices are falling. When unit prices fall, unit demand has to grow faster than prices fall in order to keep revenue (and profits and stock prices) up. When wafer area shipped grows faster than dollar sales, this is a clear indicator units shipped are NOT growing faster than prices are falling. Revenue and profits (and stock prices) can be expected to follow.

Said another way, how does units shipped correlate with semiconductor stock prices? Or the NASDAQ? If the chart "proves" the premise, we should have bought up semiconductor stocks in 2000. After all, wafer area and unit demand grew significantly, and correlated almost perfectly with one another, across the entire 2000 -2006 period.

For those of us dealing with the real world, charting variables which happen to correlate but tell us nothing about attendant prices or profits is a meaningless exercise. We can think of a number of insulting, sarcastic metaphors to add here but we decline. Perhaps it will set an example?

2) it is "more meaningful" to use SICAS than SEMI (since SICAS includes fab utilization).
As the critic concedes, SICAS runs even later than SEMI. Since SEMI data is already pretty late (April -June, 2006 wafer area shipped data wasn't available until mid-August). Waiting even longer for SICAS data risks using outdated information.

Conveniently, this highlights a point: Despite the semiconductor industry's revolutionary history of simultaneously enabling smaller, faster, low power and cheaper products, during our 26 years in the business, fab throughput time has changed very little. It still takes about 12 -13 weeks from wafer start to wafer shipment. "Back end" processing such as probe, scribe and break, mount, wirebond, assemble, test, package and ship typically accounts for another several weeks. Some interpretations of accounting rules can delay the recording of an actual sale until a net30 (or longer) invoice is paid.

In other words, "front end" wafer starts turn into wafer area shipped roughly 90 days later. It takes another several weeks after wafers are shipped before you see actual SIA sales dollars. If the goal is to gain insight into semiconductor profitability and stock prices, switching from wafer area shipped to wafer starts decreases your chances of catching a change in trend.

As an aside, with a few hours to kill it might be informative to time-shift graphs of wafer area shipped (or wafer start) history data on top of SIA sales dollars and the SOXX (or NASDAQ or whatever). Accounting the time domain difference might might isolate some circumstances which portended a change in trend and so become a useful model going forward. Remember to cut us in for royalties if you have the time and inclination to set this up.

But isn't utilization an even more-useful number than wafer starts? No, fab capacity utilization data is another "front end" statistic. When sales falter, it takes some time for companies to work backward through their supply chains and turn off wafer starts. It then takes another quarter, or longer, to show up in wafer starts. Using wafer area shipped gives you a snapshot further down the supply chain.

Just for fun, since 2Q06 SICAS wafer starts were finally released, let's add them to our August 19 table:

      Sales$  y-o-y   Area y-o-y   Starts y-o-y
2Q05  $53.9M  +0.7% 1,606   -0.8%   1,530   +8.8%
3Q05  $58.7M  +5.6% 1,748   +7.3%   1,580   +8.7%
4Q05  $59.6M  +8.6% 1,826 +22.9%   1,635 +11.5%
1Q06  $59.1M  +7.3% 1,884 +28.6%   1,703 +12.9%
2Q06  $58.9M  +9.4% 1,966 +22.4%   1,744 +14.0%

So what do you know? Wafer start data is also growing faster than underlying IC sales. "Only" +52% faster, as opposed to wafer area at +138%, but still faster.

You should also note this wafer start data may actually understate the divergence since SICAS changed how they calculate the numbers the last two quarters. (Maybe it's just a coincidence?)

Conclusion:
Our premise remains that SIA and SICAS and SEMI and all the other industry-generated statistics are not well-suited to catch a change in trend. Our goal is to add value by looking beyond the headlines and checking to see if the underlying facts support the cheerleading.

Contrary to the sarcastic insults, nothing has disproved our assertion:
"SEMI's y-o-y wafer area growth in 2Q06 was +22.4% while SIA's y-o-y IC dollars sales increased only +9.4%.  We read this as a warning sign for semiconductor profits."

We will now return to our regular programming, If we get another round of complaints, we will spare you from it...

* * * *
The above is not intended as advice to buy, sell or hold any stock, bond, real estate nor any other financial product or service. Buy and sell at your own risk (just like we do.)