Wafer area relevance?, Chapter 1

September 5, 2006

"Wafer area" relevance?, Chapter 1

A knowledgeable reader who makes a living as a technology analyst recently wrote:

"I read your recent article, “Does wafer growth impact sales?” and wondered whether you have factored in the recent spurt in solar energy demand for silicon wafers and may be that’s why Wafer growth has been outpacing semi sales."

The short answer was:

"I was involved in discussions last summer (2005) with a Shenzhen-based company trying to acquire wafer material for their solar power product. However, they were only interested in scrap or "excess" wafers they could acquire cheaply from the back door of various fabs. Purchases would have been made by weight, rather than wafer unit. (FWIW, we learned any fab which would even admit to producing any excess or scrap also insisted they were already contracted out for such material for the foreseeable future.)

In short, no, it does not seem likely the various solar power products trying to acquire wafers are contributing to the SEMI's reported growth in wafer area shipped."

The reader wrote back with:

"I am also hearing from knowledgeable people that you may be under estimating solar demand for wafers (huge) and that you may be forgetting the transition we just made over the last 3 qtrs from 200mm wafers to 300 mm wafers which increases wafer area per die."

Thinking this would be of interest to our readers, we are including the rest of this "conversation."

It would not be hard to imagine solar market is huge. The open question is if this wafer area is part of the SEMI figures. It seems like they would have mentioned this effect if it was happening but, other than that, & instinct, it is difficult to state conclusively SEMI is or is not counting solar wafers. (If that is even the correct term.)

Since it would seem to be in SEMI's interest to trumpet any new source of demand, the lack of such news makes it hard to imagine solar power demand is already included in their reports. Inquiries to SEMI for clarification have not been answered.

The 300mm vs 200mm argument is a red herring for several reasons:

1) Area is area. A chip buyer mostly doesn't care if it was built on 300mm or 200mm. If it has the same specs and fits in the same package, the price is not dependent on its ancestry.

Admittedly, a savvy buyer might try to extract price concessions if they know a part was built on 300mm since these fabs are supposed to be more efficient. In our direct experience, that very same buyer will later argue 200mm-built chips must be discounted because the fab equipment is already depreciated and so the seller has lower costs. ;=)

Sorry, back to the point at hand:

2) 300mm fabs tend to run more-advanced process geometries than 200mm operations. Obviously, a 300mm wafer built using a 65nm process puts out more die per mm^2 wafer area than a 200mm (or whatever) fab building on .90nm or larger processes. (We tried to normalize for this impact in an earlier piece by adding 1% or 2% more die per mm^2 wafer area per quarter starting maybe ~2Q04 but omitted such a step in the latest missive.)

(Editor's note: Even without adjusting for more die per wafer due to tighter process geometries, y-o-y wafer area was +22.4% while SIA sales were only up +9.3%.)

In other words, 300mm fabs make the "problem" of wafer area shipped expanding faster than IC sales "worse," not "better." Truly "knowledgeable people" would understand this HELPS our argument.

3) Unlike the "old days" when a new fab often literally "replaced" an old one, older 200mm wafer capacity is not going into mothballs or being sold off for scrap. Instead, the gear is being shipped to China where cheap money hopes to use it to carve out a niche. There is a good possibility SEMI is missing some of these and therefore under reporting wafer area shipped.

4) A related issue is last quarter's ICs sold were not necessarily built with last quarter's wafers. Wafer area shipped probably lags SIA-recorded IC sales by ~90 days. Despite the productivity "miracle" of the semiconductor business, it still takes the same 10 -13 weeks to process a wafer through a fab that it took 25 years ago. It then takes another several weeks to probe, scribe and break, assemble, package, mark, test and ship finished IC's to customers. Add another 30 days to account for payment terms before some firms can record an actual "sale." (Often longer then 30 days in today's environment of heavyweight customers demanding, and getting, extended terms.)

(Editor's note: Going back and using 1Q06 wafers to track 2Q06 IC sales, the "problem" is made "worse." 2Q06 wafer area growth was still +22.4% y-o-y but 1Q06 IC sales growth was only +7.2%. Could the differential be accelerating?)

5) Keep in mind SIA admits to tracking only ~85% of the IC market. It would be no surprise to learn SEMI misses a similar percentage of wafer area shipped. It would also not be surprising to learn SEMI and SIA miss more sales/shipments as the business grows inside China with non-western firms.

In other words, on top of the above time lag between wafer area and IC sales in 4), the published wafer area shipments (and IC sales) numbers may under-report the differential and under-report it a bit more each quarter. ( Details on SEMI's wafer area methodology are not clear but, on the equipment side, they use anonymously-reported numbers sent to a 3rd party accountant.)

Thanks for asking the questions! This will probably become another missive. (Maybe it already is?)

Feel free to point out flaws in the above?

* * * * *

From there, the "conversation" got really "interesting." A separate expose' on solar power hype will also be forthcoming. Stay tuned for future chapters!

* * * * *

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.)