Hi Konrad, thought I'd share with you some of my comments on
FPGAs, here I am on a Saturday, working on them as usual. (
Employment times are good in engineering right now...)
The last time (I think) I commented on ALTR was back on Sept 16:
exchange2000.com
Incidentally, ALTR closed at $53.00 per share that day, Of course I
can no longer make the statement Can anybody really be buying
a this stock at $50+ when they only sell about $6 per year? as
they no longer are buying at those levels :). Closing yesterday at
$44.94, they've lost 15 percent. I always try to buy cheaply,
and ALTR is still nowhere near a cheap stock yet, in my opinion.
But prices do fluctuate, so who knows.
The comments I made then about the limits to the size of the
market niches enjoyed by ALTR and XLNX then (and more
recently on the Ask Michael Burke thread) still hold. If you
want some easier math, here it is:
Niche (Sales/year) = E D N C V
where
"E" is the number of digital engineers using EPLDs, a number
bounded by the total number of electronic engineers. This
number could increase as more engineers use EPLDs, but not by
much, they are almost universally used to prototype difficult
to simulate systems already, and for manufacturing low volume
equipment.
"D" is the number of designs one engineer can complete per year,
one that appears to me to be in decline as systems get more
complicated faster than the tools advance, and the low
unemployment rate brings less competent engineers into the
business.
"N" is the number of programmable parts used in a design,
typically 1-5.
"C" is the cost of programmable parts (for prototypes, this is
usually less than $1000; competition keeps margins finite).
"V" is the number of units produced for a given design. As
I have stated many times before, (and can provide numerous
hyperlinks to articles saying the same thing), EPLDs get cost
reduced out when volumes get above a certain amount.
That amount appears to me to be in decline, as chip houses
continue to reduce NREs and turn around times, and tools
become more standard between processes (i.e. you transfer
the design in VHDL or Verilog. It can go right from a PC
workstation running Altera tools to a Sun Unix box at
LSI-Logic.)
Since each of these contributing numbers has a limit, and
since the current state of the industrial art is a large fraction
of that limit, the opportunity to expand programmable logic
sales (especially at the high ends) is limited. Consequently,
I analyzed the total market niche (for XLNX) over on the
Ask Michael Burke thread. XLNX sales data has the
advantage of being beautifully smooth, while Altera data
includes some years where their sales actually decreased.
So I used the XLNX data. I believe the decline was due to
when XLNX decided to protect market share. But the
constraints are on the total market, for both Altera and Xilinx
(and those other competitors).
Remember that the programmble logic market consists of
digital designers like myself, not members of the general
public who accidentally happen to buy a product with a
programmable logic part in it. We are sophisticated
purchasers with relatively more or less uniform, limited
requirements, and we are very very very price sensitive.
That is why the market is brutally competitive.
I Once did a design that ended up using about 100,000 of
an FPGA (the XC2064), but that was due to the company I
worked for running out of money and halting the gate array
designed to replace it with. Our manufacturing purchase
agent used the threat of the gate array design to force Xilinx
to supply us with XC2064s for about 1/3 the price charged
other users. I am sure they were just using up extra foundry
capacity at very low margin. Eventually the company did
allow the gate array to go through, it provided much more
capability at about half the (reduced) Xilinx cost. Altera parts
have the same stories, but I've never put one in a mass-
produced design, at least partly because their parts are
somewhat (say 50%) more expensive for the same bang.
But they are easier to design with, particularly for beginners.
To truly understand this business you have to see it from the
trenches as a customer. Search for cheaper alternatives in
order to save a buck; Have lunch paid for by the sales reps
when they want you to at least look at their parts; Use the
tools to create designs of great beauty and utility; Transfer
designs to gate arrays and full custom; Sit around tables
with marketing trying to figure out how you're going to get
a product together cheap enough to make money; Feel
fear when a competitor announces a product at a lower
price or better performance than yours.
Some of the sages of wall street claim you should only invest
in things you understand. I've been buying and using these
things in quantity for 15 years, I understand them, and I
wouldn't buy the stock at the current price. Should tell you
something. On the other hand, I don't think it's high enough to
short, though it's close. All investments have to be made
relative to the nearly infinite number of other things you can
do with your money, I just think there are better opportunities
out there (preferably out of semiconductors).
Only time will tell the final size of the programmable logic
market, but my sense of the market is that it is about 1/3
saturated right now. New designs will come around, but
if you look at it from the point of view of total sales of
programmable parts per engineer, a substantial fraction
of the market niche has already been filled. To get a larger
percentage, you will have to find some way to make
programmable logic use silicon more efficiently than the 2%
to 10% efficiency it currently achieves. Or you will have to
see the cost of silicon drop a lot closer to zero (which would
not be a pretty thing for anybody in the industry, certainly
not ALTR). I don't see that happening, not in the past,
not now, and not in the future. Other possibilities for growth
are even harder for me to expect.
-- Carl |
