Keeping Up With Moore’s Law
By Ed Sperling -- Electronic News, 10/20/2006
Chia Song Hwee, president and CEO of Chartered Semiconductor Manufacturing, sat down with Electronic News to talk about the recent inventory glut, the Common Platform alliance and future generations of chips. What follows are excerpts of that conversation.
Electronic News: How’s business?
Chia: It could be better. We were really expecting a very strong second half in the early part of the year because we have new business we are ramping. What we did not anticipate was the inventory situation that is slowing down the foundry business.
Electronic News: Isn’t that inventory problem limited?
Chia: If you look at the semiconductor chip companies, clearly inventory is up.
Electronic News: Is it a specific sector that’s causing the slowdown, or is it across the board?
Chia: It started with the PC, but we didn’t have much exposure so we were fine. Then it spread to the communication side—phy, routers, that kind of stuff—and the so-called second-tier handset space. The top tier like Motorola, they were okay. It was the second-tier China market.
Electronic News: Is that partly because they’re young companies that are not experienced managing inventory, or is it because their market has dried up?
Chia: It’s more the latter. But I don’t think the extent of the correction will be what we saw in 2004 and 2005.
Electronic News: So it will be a few months?
Chia: There’s no such thing as a few months in any type of correction.
Electronic News: But the problem in 2004 only lasted a quarter.
Chia: Yes, but from a foundry perspective it was two quarters. It will probably be shorter than the one in 2004 and 2005, and probably not as deep. Typically what happens is customers have inventory issue with a slow quarter, like in Q3. They cut back on wafer starts. As their business recovers, they deplete inventory. So there will be no wafer starts for the time the inventory returns to normal.
Electronic News: By then you’re into the first quarter, which is typically slow, though.
Chia: Agreed. But the inherent issues that caused the problem will be gone.
Electronic News: So then this whole idea of lean manufacturing hasn’t really stopped the market gyrations, right?
Chia: Not for everyone. Broadcom will see a revenue drop in Q3. But they expect their revenue growth to rebound in Q4. I am a supplier to Broadcom. When their Q3 is down, my Q3 is down, but when their Q4 is up that doesn’t mean my Q4 is up because they have to burn off inventory. There is easily a quarter delay.
Electronic News: Is there any technology that is so compelling that it will pick up the slack?
Chia: All the new stuff we say we are doing that will provide the growth is happening. We started volume shipment in Q3 for AMD and I expect the volume to increase into the fourth quarter. So we do have diversification and activity to grow faster, and we are also expecting our volume for exports to increase. So net-net, we should be growing. But what I did not expect was the base drop, which was more widespread.
Electronic News: But still, isn’t the technology all evolutionary rather than a brand new segment of products?
Chia: The only boom we enjoyed was in the 2005 when the Xbox was launching. That’s gone.
Electronic News: Will you be building all three of the major gaming platforms?
Chia: No, on the CPU side we are only building the Xbox. We hope one day that the volume of Sony is so big that we become a natural candidate to be a supplier of that, as well. Nintendo’s market share is small so IBM is handling that.
Electronic News: So far, though, there are not any applications that take advantage of multicore chips, particularly for those video games. Is this hype or reality?
Chia: Maybe the issue was that when the first batch of chips came out, they were sent to developers, and some games were developed even before the hardware was developed. The new batch of games will not be able to have the full range of effects, though, if they do not take advantage of multiple cores.
Electronic News: Let’s swap topics. Right now there seems to be a love-fest going on between IBM, Chartered, Samsung and Infineon over the Common Platform. What happens when the competition side kicks in?
Chia: Today, IBM outsources 20 percent of its silicon production. I am the only one doing that. In the next three years, they expect that will be 40 percent.
Electronic News: It doesn’t look like IBM is going to be building more fabs, because that’s not their core business. But Samsung is a different story, right?
Chia: Samsung is in the foundry business and they want to grow that business. Today, we are working together rather than against each other because we need to win designs for the platform. We cannot undersell each other. But once we get into production—we are beginning to do so—the customer will decide how they want to split the rulings. To me, the criteria that will decide it is performance. The price is the second thing. Who does the best job will get the majority of business—or maybe the entire business.
Electronic News: But assuming you are sharing a common platform, will there be any difference in the output?
Chia: No, there should not be. If there is, it didn’t work. There’s a lot of work today to make sure the fabs are synchronized.
Electronic News: That’s what Intel does with its Copy Exactly program, isn’t it?
Chia: Yes. But it’s harder to manage with different companies. We have to have a process and a methodology for everyone to follow.
Electronic News: So what differentiates one foundry from the next if everyone is doing the same thing?
Chia: Unfortunately, the customers do not want us to be different. If there was only one design rule, customers would love it.
Electronic News: Then haven’t you, in essence, just split one company into a number of companies?
Chia: No, we are combining three our four companies and making the offering more competitive. I don’t see it that way because my first priority was closing the technology gap. Going to 45 nanometers and below is going to be even more challenging, and the depth of the research we can draw upon—which we have no resources to do on our own, and neither, in my opinion, do the Taiwanese—is going to make a difference.
Electronic News: We’re beginning to see governments making a difference there. In the United States, there is government involvement with universities in New York, Texas and California. Singapore’s government is involved, and so are Japan’s and Korea’s. Isn’t it basically becoming a nationalistic battle?
Chia: The universities can do some work, but you need more than that. For those technologies, you need the entire supply chain to support it. You need the latest tools and processes. We have a very clear line of sight. We have the potential of differentiating on the process because the infrastructure has been created.
Electronic News: Then what you’re saying is you can’t survive without these alliances, but at the same time you can’t say what the future looks like. Is that a fair assessment?
Chia: Yes. But I would add that we certainly cannot be worse off. You just have a totally different business model.
Electronic News: How far down the Moore’s Law road map will this collaboration go? Will it involve wet lithography and multiple passes?
Chia: Yes, that’s what our 45 [nanometer] process is. The low-power will be in production by December 2007. We are still working on 32 nanometer. I’m not sure when that will be ready.
Electronic News: Are the design kits that you are developing going to be competitive with the EDA players?
Chia: In the initial phase, because of time constraints, it’s true we are not giving customers a lot of choice. But that will change over time.
Electronic News: At what point will the entire chip manufacturing change with technology such as atomic layer deposition?
Chia: It will not happen at 45 nanometers and not even at 32 [nanometers]. It’s beyond that. We do not have much visibility beyond 32.
Electronic News: Is there any change in the competitive landscape at future process nodes?
Chia: So far, we don’t see any change. While we compete for TSMC’s customer base and market share, because they’re the largest, customers are never going to exit TSMC completely—especially large customers. We are trying to compete and gain business, but we are not actually competing with TSMC. We’re competing more with UMC. TSMC is the largest.
Electronic News: So then it’s a battle for the second source?
Chia: The customer doesn’t use that term anymore, though. They use dual source or multi-source. The reason is that second source, in the past, meant you didn’t change your design. The libraries are different.
Electronic News: So only within the Common Platform there can be a true second source?
Chia: Yes, that is correct. I see myself as an alternative supplier at the leading edge. The customer has to make a decision about who that will be. Once they make that decision, they actually have to invest resources.
Electronic News: Does that mean you have to invest more in new fab capacity?
Chia: We have to continue to expand. With the pipeline of business we have now, we expect to continue to expand Fab 7 all through next year.
Electronic News: Will all of your capacity remain in Singapore?
Chia: We always talk about China, but that’s more for our 8-inch needs. On 12-inch, our challenge will be at 45 and 32. With that, the issue is, do you want to add a new uncertainty, which is a new location? In China, you have a lot of import and customs issues.
Electronic News: Let’s go back to the market. Is there any indication of an overall slowdown?
Chia: It’s not very strong, but it’s not bad, either.
Electronic News: What caused the softness?
Chia: A while back there was uncertainty about high oil prices, the Middle East and interest rates, but each of those three elements has changed. Oil prices are dropping, the Middle East is stabilizing and interest rates are not rising.
Electronic News: If the market is to continue evolving, it will require lots of new technology. Are you confident the low-k and high-k materials will be available and work at future process nodes?
Chia: Yes. It is coming up from the 32-nanometer development. IBM is seeing very good results from the high-k work they’re doing. I have been pushing IBM to put some of this learning into 45 nanometers.















