Chuck, Atiq maybe the reason:) **A MUST READ**
Mind games
electronicsweekly.co.uk
In the second of a series of high profile interviews with the chief executives of
the electronics industry David Manners speaks to the president of AMD, Atiq
Raza about profits, prizes and property
Atiq Raza, president of AMD, is a courteous man engaged in battle with one of the most formidable
corporations on earth - Intel.
He has a brain as big as the proverbial planet which may explain why he savours the world's most
intensive industrial mind-game.
Another explanation is that the game has one of the world's richest prizes: "Intel takes more than 50 per
cent of the profit of the PC industry," says Raza.
That profit is routinely sucked out of the computer industry because Intel supplies most of the
microprocessors used in PCs and nearly all the high-priced ones.
In Raza's office, a PC is displaying a spreadsheet. "Intel's earnings per share are $4.81. Only 10 cents
per share of that comes from the sales of sub-$1,000 dollar PCs. They need to get $91 for the
microprocessor to achieve that. In fact my guess is that they get only $80. So Intel could be losing
money in the sub-$1,000 segment. What does that suggest about our business? It suggests we could be
losing money too."
"Intel doesn't make its profits in sub-$1,000 PCs," adds Raza, "it earns its profits in higher performing
segments of the market. In the $1,800 PC arena, Intel makes profits of $2 plus per share. That's where
we want to be."
Surely that depends on producing microprocessors that outperform Intel's?
"K7 has already overtaken Intel's microprocessors in performance," responds Raza, "we have a
microprocessor development team that has the talent and the design methodology to produce the best
microprocessors in the world, and we'll keep getting better at it."
"Microprocessors are changing,"asserts Raza."The purpose of today's microprocessor is to process
content. If the nature of content is going to change, the processor is going to have to understand the kind
of data coming down the pipe. It has to recognise different types of data stream and execute them
faster."
"The first example of that was the K6 3DNow! concept," continues Raza, "the K7 3DNow! concept is
an extension of that intended to accelerate all forms of streaming data particularly in the form of Java.
K7 has the features to execute it very fast."
Having got ahead technically, AMD has to keep running faster than Intel. "I would say we will
accelerate our rate of bringing out new products by between 50 per cent to two times. We will do it by
hiring more designers among other ways," said Raza, adding "we haven't yet got ahead in the market.
Intel should try to kill us before we get ahead because, if we got ahead, life would get more difficult for
them."
Has Intel conceded technical superiority in recognition that the market does not demand higher
performance? "There is no question of Intel saying that we don't need more performance. That would
be the end of Intel," replies Raza, "we hope people will buy AMD's processor because it is a higher
performance part, but Intel is a formidable competitor."
Although AMD's main effort is now on the high end of the PC microprocessor market, it still wants to
retain its position in the sub-$1,000 PC. Has that been helped by the exit of Cyrix? "The exit of Cyrix
from the PC CPU business well help us in the long term, but in the short term it has left a huge amount
of inexpensive inventory," replies Raza.
Is AMD hampered by having fewer factories than Intel? "Our comparative lack of capacity compared
to Intel would only be a problem if we were booked out. I would be delighted to face that problem," he
rejoinds.
"Every quarter we have to compete with Intel on the number of microprocessor units shipped. The PC
guys can give us the numbers of PC units shipped, but how many are going to be shipped with an Intel
microprocessor and how many with an AMD microprocessor is decided on a quarterly basis."
This gives him a headache over when to build new capacity. "There's an eighteen months lead time on
building a fab, which is why I would rather build one later than earlier," says Raza, "the capital cost of
building fabs is so considerable that the timing of the investment has to be planned so that the availability
comes on line with a lag time relative to when they're needed rather than ahead of when they're needed.
I'd rather be sold out and lack capacity than have excess capacity that is paid for."
Why not make them in foundries? "For our x86 microprocessors to be manufactured in foundries
requires a degree of planning and execution that is extensive," explains Raza, "it can be done, and it can
be done extensively, but we don't anticipate that model for the bulk of of our microprocessor line-up."
Is the vertically integrated semiconductor company model phasing out as the industry compartmentalises
into IP creation, design and manufacturing?
"There are some areas, for instance networking and communications, where the products are more
compatible with the compartmentalised industry modes - where the chip is just one part of the whole
consisting of system, software and manufacturing," explains Raza.
"Other products are not compatible with the compartmentalised model and are better made by vertically
integrated companies because what is contained in them is very tightly tied up with the precise capability
of the manufacturing technology. Like microprocessors which have schizophrenic complexity - the most
complex IP in the world - which have to be designed in a way which is very attuned to the
manufacturing process."
"AMD has three businesses," adds Raza, "in flash it requires the vertically integrated model because the
design requires cognisance of manufacturing technology. The same thing applies to microprocessors.
But in communication chips it is better to go for the foundry model."
"There are three considerations in producing a chip: time to market; IP - both silicon and software; and
cost.
With IP, greater is better;
With cost, greater is worse but it's a secondary consideration;
With time to market, greater is very very bad.
If I am a designer of a communications product, my number one focus is having the right IP. My second
concern is to get it done fast. I don't want it costly but that's my third priority," explains Raza.
"If I'm a communications company I can't afford to be late. And since IP is so valuable I can't leave the
manufacturer to do it for me. So I will do as much of the design as I can and then put it out to various
manufacturers to get the best cost," he continues. An example could be AMD's planned play in xDSL.
Raza sees an opportunity there: "The more dynamic ISPs are trying to bypass the traditional telecos."
"But where products are settled in terms of the silicon and software and where constant change is less
important, and where there is very high volume, devices tend to be better suited to vertically integrated
companies."
Such a product is flash memory - could it suffer the same fate as DRAM in becoming commoditised
and, seemingly, chronically unprofitable?
"Flash is susceptible to what happened in DRAMs," replies Raza, "but that is not the fear that we have
for the near future. The barriers to entering the flash business are higher than DRAM because of the
design of flash memory and how it is being utilised in applications. In the higher applications, flash is
customised for user flash plus Asic. 30 per cent of AMD's flash output is customised in this way."
With flash moving into shortage and communications chips on a long term growth path, two legs to the
AMD stool look secure. The third leg, microprocessors, is a death or glory affair. Can he do it? "We'll
do our best," says Raza quietly.
|
