Slashdot Mirror
Upside interviews Jerry Sanders of AMD
An Anonymous reader writes "Titled The Last Man Standing, this Upside interview offered an inside view of the bloody war between the two CPU makers from Sanders' point of view. He also talks about upcoming Hammer, flash memory, Transmeta and telecomm bubbles. Somehow I get a feeling that both companies are living under the heavy cloud of Microsoft. Pretty lengthy, but an interesting reading.""
After 33 years, the unthinkable is happening: W.J. "Jerry" Sanders III is leaving his position as CEO of Advanced Micro Devices (AMD), the company he co-founded after working at Fairchild Semiconductor (FCS), where he had already established himself as a legendary seller of semiconductor products.
The "last man standing" appellation is deserved, because, as you'll see in our story, AMD is the last remaining microprocessor manufacturer among the 15 companies that Intel (INTC) licensed as second sources for the Intel 8080 at the beginning of the PC boom in the 1980s and the only real challenger to Intel's total domination of the microprocessor industry.
Upside: As you look at AMD on your 33rd anniversary, what do you see?
Sanders: One thing constant through our history is people first; products and profits will follow. All companies pay lip service to that, but it's honored more in the breach than in the action. When AMD started out, circuit design was pretty simple: We had bipolar technology, then we moved to MOS technology, and then VLSI [very large scale integration]. Today, the semiconductor industry is based on process-technology
leadership.
We put together a team in 1990 when we brought in William Siegle from IBM (IBM) as our chief scientist. Since that time, AMD has made dramatic improvements. In 2001, we had more than a thousand AMD patents granted, many on process technology. I think that made us number 13 of all the companies in the world in patents [held]. If you consider the fact that AMD is a company of $4 [billion] to $5 billion in revenue, that's awesome. 2000 was our first year of a thousand patents. At the IEDM [International Electron Devices Meeting] in Washington, D.C., we introduced the world's fastest transistor, a 3.3THz transistor. That's faster than Intel's 1THz or IBM's 2THz [transistors].
See how AMD's stock price has changed over three decades.
What changes do you see in semiconductor companies generally?
At one time, you had to have your own fabs. Now you've got this Taiwan and China syndrome. The fab is determined more by the equipment maker and money than it is by the original device manufacturer. The processes are largely determined by the equipment maker--Applied Materials (AMAT) or Novellus [Systems] (NVLS)--which works with the large user of equipment to give them the process they need. That large user was IBM, TI [Texas Instruments] (TXN), Hitachi (HIT), Fujitsu, NEC (NIPNY), or Intel. The big spenders going forward are the foundry guys, TSMC [Taiwan Semiconductor Manufacturing Company] (TSM), UMC [United Microelectronics Corporation], and maybe Grace Semiconductor [Manufacturing Corporation], a mainland China company. So process technology has now become available to everyone through a foundry, except, of course, for memory and high-performance microprocessors.
Do you foresee more alliances?
AMD's got the design and process technology base that enables us to make alliances with foundries, or nations, like when China says, "We want to be in this game."
You recently announced an agreement with UMC of Taiwan, for example. Correct. AMD and UMC have just made a tripartite agreement. The first aspect of the announcement is a straightforward foundry relationship. We are going to qualify a UMC fab to make 130-nanometer [AMD] Athlon processors, with output beginning at the end of this year. That will increase our output by 12 million pieces per year.
The second part of the agreement is the collaboration with UMC in the development of the logic for process technology, which will allow UMC to produce 65-nanometer technology on 300 mm wafers in mid-2005. The third part of the agreement provides for a fifty-fifty sharing of the ownership of this fab between UMC and AMD, but the intellectual property will belong to AMD. That will give us additional production of 300 mm wafers in the near future, without going to the expense of investing in new fabs.
So why is Intel building multiple 300 mm fabs?
Because their die is so goddamned big, they use that space up. If we have a competitive offering, we'll be able to undersell Intel. We'll be the lowest-cost provider, and we can beat them. But right now, what they're beating me with is their treasury, their market-development funds, their sweetheart deals, and their advertising. That's what I've got to overcome.
What is the future of AMD's business?
AMD has focused its business into two major areas, with a third one coming, where our technology can make a difference. [First] PC- or Windows-compatible microprocessors. In the future, and even in the present, they're moving into servers and workstations. I'm sure they'll be in games. Intel is the dominant player, with AMD as the only real alternative. [Second] in flash memory, where the same thing is true. You want to have the smallest cell; you want to have the highest performance and the lowest cost. AMD has a mantra: We're going to use process technology to drive the lowest costs. That means simple processes, but, most importantly, it means small dies. When you talk small dies, everybody says, "Yeah," but everybody's based on the international road map, going from 130 nanometers to 90 nanometers to 65 nanometers, and of course they are.
So how do you differentiate?
If our metamorphosis in the '90s was to process-technology leadership, and our metamorphosis in the mid-'90s was to become a megafab leader with our Fab 25 in [Austin] Texas and our Fab 30 in Dresden [Germany] with copper interconnects and flip-chip technology, then our new metamorphosis is emphasis on design leadership. Example number one: flash memory. We're doing something called [AMD] MirrorBit next year to double the capacity of a cell without compromising the integrity, reliability, or endurance. Intel has something called [Intel] StrataFlash [memory]. They're basically dividing the charge in half, so there's a compromise in the endurance and the reliability because there's only half as much charge to give the information. MirrorBit is an example of a superior design. All of our flash memories have smaller cell sizes--design excellence like the Athlon processors. We have a design that is only 80 square mm in a 130-nanometer technology, compared with maybe 136 square mm for the [Intel] Pentium 4 [processor]. So we outperform the Pentium 4, and our die size is much smaller.
Intel's approach is threefold: a PC line of processors, a server line of processors, and a multimedia line of processors. They're a processor company.
They have over a billion dollars in communications-related circuitry, but they lose over a billion dollars on it. The only place they make money is in microprocessors. In the processor business, only one market matters, and that is the PC market. The PC market is over 135 million units a year and growing--and growing.
Aren't embedded processors outgrowing PC usage?
There is no market that uses a single operating system that can compare with 130 million PCs a year. The next closest is games. You can call those embedded, but there's nothing bigger. Automotive? Forget about it. There aren't that many cars built in a year.
My point is, Microsoft (MSFT) rules. They won. In case you missed it, their operating system drives all of the volume in PCs and is now moving into network servers. If they have their way--and my guess is they will over time--they're even going to move their Windows NT64 into the high-end Unix-Linux domain. So if you can't make it in the PC-processor business, you're screwed, because the volume elsewhere won't enable you to generate enough revenue to support your design effort, your infrastructure-support effort, and your manufacturing.
Can gate-array or semicustom technology create a successful processor?
Not a chance. Not in the PC space and not with the performance. If a guy builds a server, he wants performance. And what gives him performance? One of the things is integrated memory on a chip--a cache. What? They're going to put an L2 cache or an L3 cache on a gate array? The chip would be huge, so it wouldn't be cost-effective.
So what about telecommunications, the cellular market?
[That's a] whole different deal. We believe that the PC, in mobile and desktop form, will continue to be the hub of the digital universe. Only Intel and AMD have the resources and the technology to succeed. Intel is trying to continue their monopoly, because otherwise their business model doesn't work. They're going to make all their own chips, and they're going to invest in factories, and as they invest, the only way they can get it back is by charging high prices.
As opposed to AMD?
[We have] eroded their margin from 48 percent to 24 percent. And, at AMD, we see that the answer is not more factories, but better design.
Speaking of which, where are next-generation fabs going to be built?
They're going to Taiwan and China, because those countries want to be in the business and because capital is low-cost there. Wafer fabs today are like steel and automobile factories in the past. Every nation feels that if it's going to be a world player, it's got to have an indigenous semiconductor capability.
So what happened in Europe, which wanted an indigenous industry?
They've got it. Siemens [Semiconductors] (SI)--or Infineon [Technologies AG] (IFX), I should say--and ST [STMicroelectronics] (STM). And then AMD in Dresden. Intel is in Ireland and Israel, if you consider Israel [part of] Europe. When you look at the economies, the reality is that very few companies can afford to have their own fabs. In a study by the Goldman Sachs [Group], they determined that unless you're doing about $8 billion of revenue, you can't afford to have your own fab. Very few [semiconductor] companies are doing $8 billion in revenue.
And what about generations of technology?
Everyone except Intel will have to be using foundry relationships or joint ventures. AMD is working with its own fab, because we developed the technology. Or [like AMD] you have to develop your own transistors, which is why a foundry like TSMC, UMC, or Grace Semiconductor would want to have a relationship with AMD. We can provide that extra performance in the transistor through our technology development. Collaborations between foundries and leaders in process technology, like AMD, will mean we don't have to make the massive investments in plants. That is an incredible changing of power, because it [means] Intel can't beat me to death with their capital. That era has passed.
We've got a foundry that's currently supplying us our first samples of a 130-nanometer Athlon processor. We're producing 130-nanometer Athlon processors now, in Dresden. I don't need to be a 100 percent owner of a megafab that costs billions of dollars. Intel has committed $7.5 billion this year to megafabs, because their die size is too big. What they're trying to do is offset the cost disadvantage of a large die by accelerating the move to 300 mm. Even then, their die size is so large that they won't be able to offset my cost advantage. My 300 mm will be done in partnership with someone who values my technology enough to give me an advantageous position. I've done this before in our joint venture with Fujitsu.
Banking analysts raise the issue that AMD is limited by capacity.
We'll produce 32 million processors this year. We can produce 50 million when my Fab 30 is at full ramp, which [will be] by the end of 2002. Our issue isn't with producing; it's getting orders away from a monopolist who makes very aggressive deals.
Talk about your processor lines.
The desktop processor is the [AMD] Athlon XP [processor]. It's what we are selling against the Pentium 4 on the desktop. We have the [mobile AMD] Athlon 4 [processor], which is the same core but has a dynamic-feedback capability that adjusts the speed at which it runs depending upon the performance required by the application. The higher the speed, the more power it burns, and therefore, the more battery drain. The [AMD] Athlon MP [processor] has a dual processor for servers and dual-processor workstations. We've got the little brother of the Athlon, the [AMD] Duron [processor], for mobile or for low-cost servers. At the end of next year, there will be a 64-bit processor called the "Hammer." That's the internal code name, [and it has] a remarkable capability in that it is based on a Microsoft-supported instruction set developed by AMD.
I thought Intel dominated the Microsoft relationship.
We call it x86-64 [architecture]; it supports all of the x86 instructions. We've added 64-bit capability and instructions that Windows NT64 from Microsoft will support. This is unprecedented in history--Microsoft supporting x86 instructions other than those developed by Intel. This means anybody can run existing 32-bit applications with higher performance and move to 64-bit [applications] seamlessly. This is in marked contrast to the Intel approach, which requires developers to go to a whole new instruction set and rewrite all their software. Or, if they want to run their 32-bit software, it will run on an [Intel] Itanium [processor], but at a degraded performance. When we start shipping in 2003, my life's work will have come to fruition: an independent platform supported by Microsoft that will compete with the Intel monopoly.
Explain that.
I started the company in '69, but we made our first fixed-instruction set processor in '75. It was the 9080A, which we renamed 8080A because it was a plug-in replacement for Intel's 8080, an unauthorized second source. As a result of that, we negotiated a technology cross-license and patent cross-license with Intel in 1975. Roger Borovoy was [general counsel] at Intel at the time. Leo Dwork was [the director of contracts and licenses], and he worked out the details. We negotiated a 5-year patent cross-license, which was renewed in 1981 on an expanded basis for 12 years. The ill-fated technology exchange threw me into arbitration with Intel for close to six years.
AMD's first microprocessor was developed in a Chinese wall environment, with a compatible-code instruction set?
Yes. There wasn't microcode in those days, so it was just knocking off their chip. We took their chip, looked at it, and redid the logic on a smaller die. I remember [Intel Co-Founder and Chairman Emeritus] Gordon Moore said to me, "So what?" The reason he said that was because their price was so high. Intel has always based their products on the fact that they'll be able to command a high price. That's why they couldn't be successful in memory; they've never been successful in any business where cost was a criterion. That's why we're going to win.
What was the order of events?
1975 was our first 8080A, an unauthorized second source that resulted in a patent cross-license agreement with Intel and their request that we second source the 8085, because they were losing business to Zilog's Z80. The 8085 was just a second generation of the 8080A. Then they said they didn't want us to do the 8086, and that really annoyed me, because the 8086 was the next generation. So we went to market with a Zilog Z8000 deal.
In those days, Intel needed a second source for the 8086 to satisfy IBM. In 1981, we renegotiated a deal with Intel that would make us an authorized alternate source with full legal rights. We did that with the 8086, the 8087, and then with the 286 and 287. Then they went to the 386, and there were two versions, and it was at that time that Intel said, "No more. We're not going to give you any more technology."