Economic Outlook: Kicking into High Gear
John Baliga, Associate Editor -- Semiconductor International, 1/1/2000
The semiconductor industry continues to follow the capacity and pricing cycles it has followed for the past decade (Fig. 1), and right now that cycle is in the start of the strong growth phase. Most analysts have the 1999 growth rate in the high-single-digit to low-double-digit range, with growth projections in the 20% to 30% range for the next two to three years. In addition to this capacity buying cycle, technology buying promises to strengthen the upswing.
This cycle is particularly evident with commodity ICs, such as DRAMs. Tracking DRAM prices gives a good indication of the industry's economic condition. Cheap DRAMs are good news to college students looking to upgrade their computers, but not to college students looking to find work in the semiconductor industry.
This simple cycle continues to be the main driver of the industry's economic condition. The "Asian crisis" is typically blamed for the depth of the downcycle in 1998, but at least half of the downturn effect was due to overcapacity. The "Asian crisis" hit at exactly the wrong time, and thankfully both it and the overcapacity condition are over.
Consolidation
Consolidation took place in the equipment supply industry, as well as the semiconductor manufacturing industry, as is expected during a downcycle. Foundries continued to gain strength, gaining technology capabilities on par with those of the top integrated device manufacturers (IDMs). The number of competitors selling ICs to the market has gone down, though the total number of IC manufacturers hasn't really changed.
|
| 1. The semiconductor industry follows predictable economic cycles. (Source: IC Insights) |
The increasing cost of fabs, though, can limit the number of people who can jump in. As the ante for entering the market goes up, fewer companies can afford to do so. "If you figure that a 300 mm full-scale production fab is going to cost north of $2 billion, and you start mapping that against the revenue of chip companies, you see very quickly that not a lot of them can afford to spend enough to build a new fab. This points to further consolidation to the foundry model," said Susan Billat of Robertson Stephens.
Foundries have grown stronger as more top semiconductor companies outsource their production. Billat noted: "There is truly a make or buy choice open to almost everyone other than the extreme leading edge. You would expect a top manufacturer to make, not buy, its most leading-edge microprocessors; but the fact is that those manufacturers have many products, and they now have the choice of making or buying those products. The foundry model is making more and more sense, because it works; it's efficient and cost effective for all concerned."
Though overheating has been prevalent in the DRAM sector in past cycles, that may not be the case this time. According to Gunnar Miller of Goldman Sachs: "The old assumption is that when pricing goes up, everyone would try to dogpile on the market and take advantage of it. Then when pricing goes down, everyone would slash capital spending, and it would be all over. Earlier last year, when spot pricing on DRAMs went below $5, there was really nothing left to cut." Miller added that the "dogpile" didn't happen when spot pricing spiked at $20 a few months ago. Clark Fuhs of Dataquest warns it is too early in the cycle to declare that fewer players will jump in. He interprets current spending by DRAM manufacturers as caution, saying, "We expect DRAM spending to come back and be more dominant in the late 2000-2001 time frame."
Technology effects
Technology, of course, has its effect on the economic landscape, and vice versa. One area where technology can affect economics is lithography. With recent advances in optical phase correction (OPC) and phase shift mask (PSM) technology, the usefulness of 248 nm exposure tools can be extended down to the 100 nm technology node. According to the 1999 ITRS Roadmap, that node is expected to arrive in 2005, which is roughly when the next economic upturn would be expected. The effort and expense of lithography doesn't disappear, though; it will be shifted to the mask-making and photoresist arenas. It's possible more pressure will be exerted on electronic design automation (EDA) vendors to ensure that their design tools can produce a final design on the first pass, since masks will be more expensive.
In the packaging arena, chip scale packages (CSPs) and ball grid arrays (BGAs) have been the big news in the last few years, and multichip packages also have started to emerge. Flip-chip interconnection to a package substrate has become more prevalent in BGAs, and wafer-level packaging is a growing method for CSP manufacturing. Wafer-level packaging has the potential to bring packaging into the fab, and multichip package technology is being spearheaded by contract packaging companies.
Another trend is that on-chip interconnect is starting to be designed together with off-chip interconnects. The 1999 Roadmap states that it is becoming more important to consider the chip, package and circuit board together, and that the interface between chip and package is blurring. In the near term, this means companies in all three sectors will need to cooperate more closely. In the long term, companies likely will develop capabilities in all three areas. According to Clark Fuhs, Dataquest is on record predicting wafer foundries, contract assembly companies and electronics manufacturing services will be competing in the same market 10 years from now. One indication of this change occurred two years ago when Amkor started a foundry fab in Korea, licensing technology from Texas Instruments.
The big story in the past year or two has been copper interconnects and low-k dielectrics. What has not been publicized so much is the development of high-k dielectric materials for memories and gate dielectrics. Also, new electrode materials will be required for use with high-k dielectrics. The rate of new material adoption will be faster than it has been. This could have an interesting effect on the system-on-a-chip (SOC) trend. Leading-edge logic and memory processes are becoming even more different than before, and integrating them in a cost-effective manner may be extremely difficult. This is another area where packaging and assembly may have to provide innovative solutions.
On the other side of the ledger, where the market is affecting technology, a more diverse set of end products is helping a lot of "technologies of the future" find a solid niche. Gallium arsenide, for instance, has long been proposed as a replacement for silicon. It will not replace silicon, but it has found a home in RF chips in cellular phones. Silicon germanium is in a similar position. Silicon on insulator (SOI) may find a home in high-speed processors eventually, but it has one now in low-voltage mobile applications. Smart cards do not generate a large portion of semiconductor revenue, but they are the most prevalent semiconductor-containing product in terms of units sold, and they haven't even taken hold yet in the United States. Wafer-thinning and chip-scale packaging technologies are required for those. So, in addition to more wafer processing materials and processes, more substrate materials and technologies will come into play, all driven by specific end markets.
One more important shift is that the end product mix has a higher percentage of applications for individual consumers. Fuhs says: "The semiconductor industry is changing from one driven by the needs of corporations and businesses to one driven by the individual consumer. The unit volumes for consumer products are significantly higher, and the requirements for different consumer applications are different enough that different technologies are required to meet them."
Conclusion
The semiconductor industry is in the start of an economic upcycle. The capacity-driven cycle has always been the dominant effect in the economics of semiconductor manufacturing. Capacity must be created in large chunks to be cost effective, so it is likely that capacity cycling will remain the dominant effect. Adding that capacity in the form of a new fab is becoming much more expensive, which means fewer companies can afford to build fabs, and this leads to more consolidation and partnering.
The foundry model is here to stay, with foundries providing leading-edge capabilities and a "safety valve" for more semiconductor manufacturers. With the cost of new fabs continuing to increase past the $2 B mark, more companies are likely to outsource to the foundries. Suppliers also are seeing them more as strategic customers.
The general consensus is that the next three years will be very healthy economically in the semiconductor industry. These next few years should be a time for companies to regain economic strength.
VLSI Research http://www.vlsiresearch.com/, risto@vlsir.comCahners-InStat http://www.instat.com, Grantj@InStat.com
IC Insights http://www.icinsights.com/, bill@icinsights.com
Goldman Sachs http://www.gs.com/, gunnar.miller@gs.com
Strategic Marketing Associates http://www.scfab.com/, sma@scfab.com
Robertson Stephens http://www.rsco.com/, sue_billat@rsco.com
Advest http://www.advest.com, Timothy.Summers@advest.com
Dataquest http://www.dataquest.com, clark.fuhs@gartner.com
Advanced Forecasting, Inc. http://www.adv-forecast.com/afi/, mosheh@adv-forecast.com
|
Risto Puhakka  This year is going to be a very happy
year. For chips, we see growth rates over 20% for 2000 and 2001. Chip
manufacturing capacity is going to be tight, which will drive prices up.
We think capacity will continue to be tight for a couple years because we
haven't seen new fab announcements to the level we would expect. The chip
equipment will have a strong three-year upcycle. This is driven by the
underinvestment in the past couple years. Industry should see the
beginning of the next downcycle around 2003. In addition to capacity
cycling, we see the value of semiconductors in electronics getting too
high, which should drive chip prices down as well.
Grant Johnson
Bill McClean
Gunnar Miller
George Burns
Susan Billat
Timothy Summers
Clark Fuhs
Moshe Handelsman
|
In 1998, $14.5 B worth of new fab activity
got underway. In 1999, the figure was over $30B. Some was in the form of
upgrades, but there also was a lot of greenfield activity. That money is
going to be spent over the next two to three years and buoy the recovery
in semiconductor capital equipment spending over the coming cycle. DRAM
manufacturers and foundries account for about half of that spending, with
9 or 10 foundries coming on line this year. There are fewer DRAM players
this time, which should yield a simpler competitive situation and more
rational spending. However, given the strong rivalries between DRAM
competitors, rational spending still could be sacrificed for pride of
place. There is strong potential for an upside in this cycle because we
have some strong end market drivers we haven't had before, particularly
for the Internet, information appliances and communications.
I'd say the capital equipment outlook is very
bullish indeed. Capacity is very, very tight, and new fabs are on the
drawing boards, as well as major expansions and upgrades. There are quite
a few shells out there that can be built out, which should save time
bringing capacity on line. Even 300 mm is being talked about, although it
probably will be mostly pilot lines until about 2002 or 2003. This upturn
seems to be a confluence of technology and capacity buying cycles, and
that double effect can create a really big upturn. It seems everyone's
"secret" strategy is to outsource production to foundries to handle excess
demand; it's a cost-effective strategy for all concerned.
We are looking for a broad-based increase in
the semiconductor equipment industry this year, which will be fueled by
several factors. One, the semiconductor industry is running at roughly
100% capacity utilization, which means we're seeing significant increases
in fabrication equipment purchases not only for technology but also for
capacity. Also, DRAM pricing is at a point where manufacturers can make a
significant return on their investment, thus spurring investment. Demand
for Internet and communication semiconductors, principally manufactured by
Taiwanese foundries, are helping push the cyclical upturn, though the PC
industry still seems to be the key driver at the margin. Capital spending
is expected to rise about 15% in 1999 over 1998, and we're looking for it
to rise roughly 25% per year between 2000 and 2002.
We were looking for flat to low-single-digit
percentage growth for 1999, and it ended with mid- to high-single-digit
growth. That has postured the industry well for having 30-plus percent
growth in 2000. We are still early in the cycle, at about the same point
we were in mid-1993 of the last cycle. Things didn't start going crazy in
the last cycle until 1994-1995. Purchasing by foundries and logic
manufacturers is happening first, much like it did in early to mid-1993.
The DRAM manufacturers are actually still cautious. They are increasing
spending, but they're not getting overly excited yet. We expect DRAM
spending to be more dominant in late 2000 to 2001, and we are expecting a
mid-40s growth rate for 2001.
Though the semiconductor industry is
currently in an upswing, our models show the possibility of a change in
direction. To know for sure if a change in direction will occur, we need
another two months of measurements of the economic factors that are the
sole input to our quantitative forecasting models. Right now, we are not
going to ring any alarm bells. If a turn does occur, the growth rate for
2000 may be a double-digit number as many people are predicting, but it
would include a downward elbow. At this point, I would not assume that the
industry is going to have three years of 20% growth just because the last
few quarters have gone up. If there is a downward turn, it would be
similar to the one in 1996, as opposed to the one we had in
1998.
