GS US SEMI EQUIPMENT: TAKEAWAYS
FROM SILICON VALLEY MEETINGS
SUMMARY: We met with several semi and semi equipment companies over the last
several days in Silicon Valley. Overall tone remains upbeat but companies were unwilling
to discuss near-term order trends. Our key takeaways from our meetings are as follows:
(1) Strategic focus of equipment companies is reducing the cyclicality of business models
as managements recognize slowing long-term growth rates, (2) Increased acquisition
activity is being used as a means to reduce cyclicality and/or supplement existing
expertise, (3) Foundries are still reluctant to commit additional capital for capacity, (4)
The semi test Open Architecture platform continues to gain traction, (5) Low-k still
presenting yield challenges, and (6) Sarbanes-Oxley 404 implementation is costly. No
change to our view that it is too late in the fundamental cycle to be overweight the stocks.
We met with several semi and semi equipment companies in Silicon Valley over the last
several days. Overall management tone remains upbeat with companies highlighting
continued expected growth into 2005. We would note, however, that the semi equipment
companies with which we met were unwilling to discuss near-term order trends. We
provide greater detail behind our key takeaways following our meetings, which include:
(1) Equipment companies are beginning to recognize slowing long-term growth rates and
are therefore directing attention to reducing the cyclicality inherent in their business
models, (2) Acquisitions are being used as a means to reduce cyclicality and/or
supplement existing expertise, (3) Foundries are still reluctant to commit additional
capital for capacity, (4) The semi test Open Architecture platform continues to gain
traction, (5) Low-k still presenting yield challenges, and (6) Sarbanes-Oxley 404
implementation is costly.
COMPANIES BEGINNING TO RECOGNIZE SLOWING LONG-TERM GROWTH
RATES AND ARE THEREFORE TURNING ATTENTION TO REDUCING THE
CYCLICALITY INHERENT IN THEIR BUSINESS MODELS. Despite the fact that the
industry is experiencing a strong cyclical upturn, during several of our meetings, we have
perceived the first signs of recognition by management teams that long-term growth rates
in the industry are slowing. To that end, several semi equipment management teams are
therefore turning their attention to reducing the cyclicality inherent in their business
models. As we have highlighted on numerous occasions, we believe that a driver of
slowing long-term growth rates is the lower capex as a percentage of sales ratio that is
being driven by a consolidation in capital spenders. We believe this consolidation is
being driven by: 1) the foundry model, 2) fewer new entrants into the semiconductor
market, and 3) the numerous joint- ventures/partnerships that have been formed in order
to fund the high cost of building and equipping 300mm fabs. One consequence of this
consolidation in the customer base is that the semi equipment companies do not have as
much pricing leverage as they have seen in previous cycles. For example, we learned
during our meetings that a PVD tool that would have sold for approximately $8 million in
the last cycle now sells for about $5.5 million.
One semi equipment company highlighted during our meetings that they believe the
normalized industry-wide semiconductor capex as a percentage of sales ratio is likely to
decline to 16% to 18% over time (down from over 20% historically), which we believe is a driver
of slowing long-term semi equipment growth rates. We believe that the lower capex as a
percentage of sales ratio is, among the other issues that we highlighted above, a reflection of fewer
new entrants into the semiconductor market this cycle than last. One of the long-term negative
secular trends impacting the semi equipment industry is the saturation of the semiconductor market
from a competitive standpoint, as new entrants into the market tend to be the most irrational
spenders of capital (which of course is good for the semi equipment companies!). While in
previous cycles there were many new entrants into the semiconductor market (including TSMC,
UMC, Chartered, and a host of DRAM companies) that spent a great deal of capital on semi
equipment in an effort to gain manufacturing capacity and hence a piece of the semiconductor
market, there are significantly fewer new entrants into the market this cycle. SMIC and Dongbu
Anam (the only true buyers of "new" (as opposed to used) equipment this cycle) are spending the
most on capex in 2004 as a percentage of their sales, while many of the industry's major spenders
are showing significantly lower ratios (particularly those companies that have already far along in
the 300mm transition). We believe that as the semiconductor market continues to mature there will
be fewer new entrants into the market. This, in turn, will lead to lower industry- wide capex as a
percentage of sales.
We also believe that the increased prevalence of the foundry model this cycle is driving more
efficient capital spending which in turn is also leading to a lower capex as a percentage of sales
ratio this cycle. During the 2000 cycle, we estimate that Motorola, Texas Instruments, Philips, and
STMicro spent about $10 billion in capex in aggregate. These four chipmakers are outsourcing at
least some of their manufacturing needs (Motorola more so than the others) this cycle and as a
consequence we are estimating that they are spending in aggregate about $4.5 billion in capex in
2004, about 56% less than they spent in the 2000 cycle. While one would expect that the foundries
would be making up some of this difference by spending more on capital this cycle, we are
estimating that TSMC will spend about $2.5 billion in capex in 2004 (note that we are modeling
capex above the company's $2.0 billion guided budget) vs. $3.2 billion in 2000 and UMC will
spend $2.2 billion in 2004 vs. $2.8 billion in 2000. We believe that these numbers demonstrate that
the economies of scale recognized by the foundries allow them to spend capital more efficiently
than any of their individual customers would alone.
In addition to the foundry model driving more efficient capital spending, we believe that the
300mm technology transition is driving a consolidation in the customer base that is also leading to
lower capex as a percentage of sales this cycle. With the cost of building a 300mm fab approaching
$3.0 billion, few semiconductor companies can afford the expense of providing all of their own
300mm manufacturing capacity. Moreover, few semiconductor companies can afford to remain
only at 200mm, because they are very likely to be driven out of business by other semiconductor
companies that will be able to offer lower cost products because they have achieved lower
manufacturing expenses by transitioning to 300mm. This phenomenon is leading to a consolidation
of the customer base as many chipmakers have either partnered or formed joint-ventures in order to
fund 300mm fabs. One need only review a partial list of partnerships that have been formed,
including a) Sony, Toshiba, and IBM, b) Nanya and Infineon, c) STMicroelectronics, Motorola,
and Philips, d) Hitachi and Mitsubishi, e) Toshiba and Sandisk, and f) NEC and Hitachi to
understand that many semiconductor companies have turned to sharing facilities to resolve the
issue of the expense of 300mm.
The increased capital efficiency in the semi market that is being driven by fewer new entrants into
the market, the foundry model, and 300mm implies that upturns won't be as robust as they were in
the past. Semi equipment companies will therefore need to focus on achieving more consistent
returns throughout the course of an entire cycle and will not be able to lose as much money during
downturns as they have lost in previous cycles. As we learned in our meetings this week, several
semi equipment companies are therefore pursuing ways to reduce the cyclicality inherent in their
business models.
(2) ACQUISITIONS ARE BEING USED AS A MEANS TO REDUCE CYCLICALITY AND
ALSO ADD EARLY ALD EXPERTISE. One way in which the semi equipment companies are
attempting to reduce the cyclicality inherent in their business models is via acquisitions. Several
companies with whom we spoke are searching for acquisitions to augment the service segments of
their business as services are likely to hold up significantly better during the next downturn.
Another, albeit much smaller, acquisition done with the attempt at mitigating cyclicality is
Novellus' recent acquisition of Peter Wolters AG in order to expand upon the industrial polishing
business (separate from the semi CMP business) it acquired when it bought Speedfam-IPEC. This
acquisition is intended to help Novellus expand its polishing exper
tise into non-semiconductor related businesses such as optics, mechanical membranes, plastics,
etc., which in turn is expected to aid in reducing the cyclicality inherent in the company's
semiconductor-focused business model.
Another area of M&A focus is Atomic Layer Deposition (ALD). Novellus aims to expand and
extend its expertise in the PVD market with the acquisition of Angstrom's ion-induced Atomic
Layer Deposition (ALD) technology as ALD is expected to replace PVD at 32-nm and below
geometries. Not to be outdone, our checks during the week suggest that Applied Materials has also
made an acquisition in the ALD segment, buying Torrex, a small private company which provides
Applied with a platform for surface-induced ALD. While ALD is a small segment today, it is
clearly an area that will come into greater focus as the industry continues to follow Moore's law
down to 32-nm.
(3) FOUNDRIES STILL RELUCTANT TO SPEND CAPITAL. Another theme that emerged
during our meetings was that the foundries remain reluctant to release significant capital
expenditures beyond what has already been announced. A significant customer of TSMC indicated
that the foundry remains adamantly focused on achieving a 20% return on equity over the course of
an entire cycle and is therefore spending capital significantly more rationally than it has in
previous cycles despite continuing to run at high utilization rates. A foundry customer of IBM also
indicated that IBM does not intend to expand its capacity significantly as it does not believe that
spending on wafer fab equipment for its foundry business is a "good use of capital." We would
note that while our checks indicate that IBM's yields have increased significantly in recent months,
the company has not and does not intend to significantly increase its wafer starts despite being
capacity constrained. We would note that the foundries continued restraint relating to capacity
additions argues for a much less severe next downturn but it also implies that investors who are
waiting for a flood of H2'04 foundry orders to drive a second leg of order growth for the semi
equipment companies may be disappointed.
(4) OPEN ARCHITECTURE SEMI TEST PLATFORM CONTINUES TO GAIN TRACTION.
Recall that the Semiconductor Test Consortium (STC) is an independent consortium of companies
led by Advantest that are focused on providing an open architecture SOC test platform. The
difference between the STC and a traditional tester company (i.e. Teradyne, Credence, LTX, etc.)
is that the testers built by the STC all include a mainframe supplied by Advantest (called the
OPENSTAR platform) and then any member of the consortium can offer off-the-shelf modules to
modify a given tester based on customer requirements for price, performance, etc. Advantest
believes the platform can be adapted to meet RF, digital, mixed-signal or high-end SOC
requirements.
The open architecture platform was created to address the increasing prevalence of consumer chips
and the corresponding explosion of different types of devices for use in products with shorter life
cycles. These shorter life cycles are driving the cost of test significantly higher and forcing the
semiconductor companies to be creative in their pursuit of reducing cost of test. We continue to
believe that the open architecture platform is a disruptive technology in the back-end that may
potentially cause the legacy ATE companies to lose SOC market share or at the very least face
greater pricing pressure in the semi test segment. The STC is currently comprised of 37 members
including Intel (whose traditional ATE suppliers include Teradyne and NPTest (now part of
Goldman Sachs Global Investment Research 3
Analyst Comment June 9, 2004
Credence)), Motorola (Agilent), Renesas (LTX), Philips (Teradyne), Fujitsu (Agilent), Toshiba,
Tokyo Electron, and Advantest. We believe that several other semi companies are likely to join the
consortium in the near-term including NEC, STMicro, and Infineon. We believe that the
semiconductor companies that have joined the consortium are demonstrating an obvious interest in
the technology, which may bode poorly for the legacy test providers at these chipmakers over time.
Of note, however, is that Texas Instruments has demonstrated only limited interest in joining the
consortium which should help to somewhat (but not completely) insulate LTX from the potential
disruption caused by the open architecture platform as TI is a significant customer of LTX. We
note that Teradyne also provides testers to TI but TI isn't anywhere near the same percentage of
revenue of Teradyne as they are for LTX.
Advantest's T2000 is the first commercial tester produced by the STC and is targeted at high-end
logic devices. We believe that the T2000 is gaining traction as Advantest shipped around 15 T2000
units in FY2003 and is expected to ship around 100 T2000 units in FY2004. The open architecture
platform has been used thus far primarily by Intel (which has highlighted a 60% cost reduction by
using the platform) and is therefore criticized as being primarily applicable to microprocessor
chips. However, the platform was designed to address various segments including chipsets, MCUs,
and ASICs. Advantest has indicated that it expects the open architecture platform to gain even
greater traction in the next 12 months with Intel potentially receiving less than 50% of systems
shipped after that time (thus mitigating the criticism that the STC is only providing testers to Intel).
While we do not expect that the open-architecture initiative will have any significant impact on the
current strong cyclicality that is driving the semi test suppliers, we believe that investors with
anything beyond a 6 month time horizon should seriously consider what we believe to the very
negative implications from this disruptive technology.
(5) LOW-K CONTINUES TO PRESENT SIGNIFICANT YIELD CHALLENGES. Several
semiconductor companies with which we met indicated that there remain significant yield problems
with low-k despite several semi equipment companies highlighting that yields have improved
substantially. A major customer of TSMC (that uses the foundry for mostly non-low-k chips)
indicated that the performance enhancements visible in low-k chips are not substantial enough to
offset what continues to be a significant loss in yields. We believe that the continued yield
problems at low-k threaten its ubiquity at the 90-nm node unless the yield problems are addressed
quickly. Some customers seem content to wait to use low-k at the 65-nm node when the yield
issues are expected to be fixed.
(6) SARBANES-OXLEY 404 IS PROVING TO BE EXPENSIVE TO IMPLEMENT. Recall that
The Sarbanes-Oxley Act of 2002 required public companies to personally certify that their
financial statements fairly present all material aspects of the company's operations and financial
condition. A provision of the Sarbanes-Oxley act, Section 404, further requires that public
companies assess the effectiveness of their internal control structures and procedures for financial
reporting. Section 404 also requires each company's auditor to attest to and issue an opinion on
management's internal control assessment. This auditor's opinion is different from, and in addition
to, the financial statement opinion that was required by auditors in prior years. In fact, it is possible
for companies to receive a clean or unqualified financial opinion and a qualified internal control
opinion. We believe that the implementation of Section 404, which goes into effect at the end of
this year, is no small task and our meetings in Silicon Valley reinforced this notion. Several
companies indicated that they have had to devote significant time and resources to formally assess
their internal control structures to comply with Section 404, which is resulting in incremental
expenses.
In addition to the key takeaways we discussed above, we garnered several
other data points from our meetings including: (1) We believe that Novellus, which has
traditionally had a strong presence in the Front-End-Of- Line (FEOL) dry strip market, is likely to
introduce a tool intended to compete against Mattson and Axcelis in the Back-End-Of-Line
(BEOL) dry strip market in the next month or so. (2) TSMC is still providing at least 80% of
Nvidia's wafers while IBM is providing the other approximately 20%. (3) The industry is not
expected to transition to 450mm wafers until 2010- 2012. And (4) LSI Logic has recently qualified
SMIC as a foundry partner and is currently comfortable with the ability of all of its foundry
suppliers to meet its demands.
I, Jim Covello, hereby certify that all of the views expressed in this report accurately.. |
