GS US Semi Equip Asia Tour Notebook Part
2 ? We wish you could have been there
We completed the second leg of our Asia tour on Friday with visits to Taiwanese
foundries and DRAM makers including TSMC, UMC, Powerchip, and Nanya/Inotera. (1)
There is a startling mismatch between what DRAM makers expect industry-wide supply
bit growth to be in 2005 compared to what actual supply bit growth is likely to be based
on a bottom-up analysis, which will likely lead to continued tremendous pressure in the
DRAM segment in the coming quarters. (2) Foundry utilization rates are set to decline
significantly in CQ2, likely preventing meaningful incremental equipment orders in
H2?05. (3) As a result of excess supply in both the DRAM and foundry segments, we
expect H2?05 orders to decline significantly after a slight sequential decline in CQ2. And
(4) Customers expect continued declining capital intensity due to 300mm efficiencies and
price discounts on equipment.
We completed the second leg of our tour of Asia last Friday with visits to Taiwanese
foundries and DRAM makers including TSMC, UMC, Powerchip, and Nanya/Inotera
(please see our note out on March 31st for our key takeaways from the first part of our
Asia tour, including details from our visits with Japanese semi, semi equipment, and
supply chain companies).
We wish that all of our clients could have attended our meetings in Asia last week
(especially in Taiwan on Friday), as we believe that it was essentially impossible to come
away from the meetings believing anything other than that there is a significant excess
supply problem emerging in the DRAM industry. We say this because each of the
individual DRAM companies believes that they will be the only company to add
significant supply in 2005. Each individual DRAM company thus believes that overall
DRAM bit supply growth will only be about 40% - 45% in 2005, while each individual
company is simultaneously projecting at least 65% and as much as 130% bit supply
growth for their own business. Further, we believe that it was extremely difficult to walk
away from the meetings with the foundry customers believing that the foundries will place
significant orders with the equipment companies in H2'05.
We provide our key takeaways below followed by details from our individual company
meetings.
WE EXPECT CQ2 ORDERS TO DECLINE SLIGHTLY SEQUENTIALLY AND FOR
H2'05 ORDERS TO DECLINE SIGNIFICANTLY SEQUENTIALLY. With regard to
sequential order patterns for the remainder of 2005, we believe that there will be a small
sequential decline in orders in CQ2 (while most on the Street are calling for a meaningful
sequential increase in CQ2 orders, we believe DRAM orders will disappoint as prices
continue to fall vs. DRAM company expectations that prices will stabilize) followed by a
much more significant sequential decline in orders in H2'05.
We expect the slight
sequential decline in CQ2 orders to be driven by declining orders from the foundries,
several DRAM customers, and some of the Japanese chipmakers. We expect these
declines to be only partially be offset in CQ2 by increases in orders from Samsung (for
both flash and DRAM), what we believe will be the last slug of capacity additions from a
few DRAM customers (including Powerchip), and potentially Intel.
In H2'05, we believe orders will decrease sequentially by double digit percentages. We expect
significant sequential order declines to be driven by the DRAM makers having completed their
planned capacity expansion and the foundries having ordered the vast majority of their 2005 capex
in H1'05, with the very low likelihood (according to the foundries during our meetings) of front-half
loaded capex budgets in 2006 that would necessitate placing orders in H2'05.
CHIPMAKERS EXPECT CAPITAL INTENSITY TO CONTINUE TO DECLINE FOR AT
LEAST THE NEXT SEVERAL YEARS UNTIL 300MM BECOMES A LARGER PART OF THE
OVERALL INSTALLED MANUFACTURING BASE. In addition to the shorter-term cyclical
takeaways from our trip, customer commentary regarding longer-term capital intensity was also
notable. Several customers commented during our meetings that the increasing number of process
steps (and in turn, increasing number of tools) required for 90-nm and 65-nm will continue to be
more than offset by both the capital efficiency of 300mm as well as what the chipmakers expect will
be continued pricing concessions from the equipment suppliers.
The prevailing notion amongst the chip companies is that "declining capital intensity will still be a
benefit in the coming years." While many of our competitors are calling for increasing capital
intensity in the next few years based on commentary from Intel, we believe that Intel is unique
because it has substantially completed its transition to 300mm manufacturing while almost every
other chip company only began the 300mm transition in 2004. We believe capital intensity will
continue to decline due to 300mm efficiencies until the industry has about 50% of its capacity on
300mm, which we don't expect will happen before 2008.
Further, we always believe that it is critical to recognize the starting point growth rate from which
any potential future increases in capital intensity may arise. In other words, the semi equipment
industry has grown at a 1% CAGR for the last ten years, so increasing capital intensity in the future
is necessary to even bring the semi equipment industry growth rates in-line with GDP growth rates.
DRAM INDUSTRY VERY LIKELY FACING A SEVERE EXCESS SUPPLY PROBLEM IN
H2'05/2006, AS EACH INDIVIDUAL COMPANY IS GROWING BIT GROWTH
SIGNIFICANTLY. The excess supply problem that we believe is facing the DRAM industry is
being driven by the fact that every single one of the DRAM makers with which we met this week
believes that DRAM industry-wide bit supply growth in 2005 will only be about 40% - 45%, but
each individual company is planning to add at least 65% and as much as 130% bit supply growth by
the end of the year. Amazingly, each DRAM company acknowledges the problems that would/will
exist in the DRAM industry if indeed bit supply exceeds 65% in 2005, which we believe is a near
certainty given that each individual company is projecting at least 65% bit growth. Further, many of
the DRAM companies are already acknowledging that 2006 is likely to be one of the toughest years
in the history of the DRAM segment. That said, each of the companies with which we met intends
to complete the capacity expansion that began at the end of 2003, as each company believes that it
will need the lowest cost and most efficient capacity in place to survive the upcoming downturn in
the DRAM segment.
Company Projected industry bit supply growth Company bit supply growth Powerchip 40-45%
65%+ Elpida 40-45% 65-85% Nanya/Inotera 50% 100%+
Source: Company data, Goldman Sachs Research.
TAIWAN DRAM AND FOUNDRY CAPEX BUDGETS ARE FRONT-HALF LOADED AND
WE EXPECT THAT THE FOUNDRIES WILL NOT PLACE ANY SIGNIFICANT ORDERS TO
THE EQUIPMENT SUPPLIERS IN H2'05. We believe the most important takeaway from the entire
week is that every single DRAM company, as well as the foundries, have front-half loaded capex
budgets in 2005. All of these companies are acknowledging that without a significant pick-up in
demand as we move throughout 2005, they will not make significant incremental orders to the
equipment suppliers over the next several quarters.
In addition to the significant issues that we believe are imminently facing the DRAM industry, the
problems plaguing the foundry segment persist. We now believe that UMC's capacity utilization rate
will be 50% - 55% in CQ2, down from 60% in CQ1, and TSMC's capacity utilization rate will be
approximately 70% in CQ2, down from 78% in CQ1. While both foundries highlighted that future
capex plans will be dictated by leading edge utilization rates (which are significantly higher than
overall utilization rates), both also highlighted that very low overall utilization rates are driving
significantly lower profitability (or in the case of UMC potentially losses over the next several
quarters). As a result of this lower profitability, the foundries are now forced to be more cautious in
their forward looking capex plans.
The bottom-line, in our view, is that absent a very significant pick-up in end demand over the next
several quarters, the foundries will make no new significant orders to the equipment suppliers in
H2'05. To that end, UMC is decreasing its order rates to the equipment suppliers in CQ2, as the
company is done rounding out its complementary equipment set to match the capacity expansion
that it started in 2004.
EQUIPMENT VENDORS APPEAR TO BE OFFERING MORE ATTRACTIVE TERMS TO
THEIR CUSTOMERS. Many of the chip companies with which we met last week noted that they
are getting attractive terms from the equipment suppliers, with the concessions ranging from price
discounts to extended payable agreements to sig
nificant service and spare part bundling along with tool purchases. We would expect both the
discounting and terms to get even more favorable for the chipmakers over the next several quarters
if we are correct that the semi equipment cycle will worsen before it gets better.
We provide data points from our individual company meetings below:
UMC: (1) Recall that UMC's top five customers include Texas Instruments, STMicro, Xilinx,
Mediatek, and Infineon. The company indicated that its IDM customers have in-sourced their
capacity needs given current low in-house utilization rates. Utilization rates will need to pick-up at
the individual IDM customers before they begin to outsource significant capacity again. (2) On the
pricing environment, UMC continues to face pricing pressure from its customers. The company
expects prices to decline 5% sequentially in Q2 after an anticipated 10% sequential decline in Q1.
Price declines are most severe for the company on the lagging edge. (3) Q1 effective utilization rates
will be around 55% (official utilization rates will be around 60% in Q1 but this includes a 5%
utilization rate benefit related to annual maintenance that will not occur in Q2). Management
expects utilization rates to be in the low 50% range in Q2. 300mm and advanced technology (i.e.
90-nm) utilization rates are significantly higher than overall utilization rates. (4) The company's
capex budget for 2005 is $1.0 billion to $1.5 billion. The orders for the 2005 capex budget have
already been placed with equipment delivery having already occurred in Q1 with more expected
again in Q2. The capex budget is being dedicated to 300mm and 90-nm capacity, with more than
80% of the budget related to 300mm (primarily 90-nm) tools. The company has already ordered
equipment totaling the low-end of its capex budget in 2005. Capex decisions are being based upon
the demand for leading edge capacity. UMC is adding capacity in 2005 to bring 90-nm to 15% of its
total installed base. If 90-nm demand is greater than expected in H2'05, the company may add
another 5% in additional 90-nm capacity. This would cost approximately $500 million and would
likely drive the company to place orders in Q3 and spend closer to the high end of its capex budget
in 2005. (5) While 90-nm yields are still low, the company's yield curve at 90-nm has been easier
than its yield curve at 130-nm, primarily driven by the fact that the industry was implementing
several new materials at 130-nm (including low k and copper) that are not being changed at 90-nm.
Note that UMC indicated that it uses Novellus for low-k. The yield curve is anticipated to be tough
again at 65-nm due to the introduction of high-k metal gate dielectric materials. (6) The company
expects capital efficiencies to continue to improve due to the 300mm transition. (7) UMC is seeing
better terms from its equipment suppliers today than it did several months ago.
Nanya/Inotera: (1) On DRAM prices, the company expects spot prices to continue to decrease in
April and May (to an approximate $2.00 spot price). Management believes that spot prices will
Goldman Sachs Global Investment Research 3
Analyst Comment April 4, 2005
rebound after Q2. Note that the company sells about 30% of its output to the spot market. (2) Nanya
believes that industry bit supply growth in 2005 will be approximately 50% while Nanya/Inotera are
growing their own bit supply by more than 100%. Management's bit supply growth assumptions are
based on 8% - 10% PC unit growth in 2005. The company also believes that bit supply growth
increased 25% in Q1. (3) Regarding capex for both Inotera and Nanya, Inotera is expected to spend
$900 million in 2005 after spending $1.2 billion in 2004. Nanya is expected to spend $119 million
in 2005 (the company's maintenance capex level). Management indicated that Inotera has already
placed significant orders with its equipment vendors and that orders in H2'05 will be lower than in
H1'05. The company expects to take linear shipments of tools at a rate of about 3k wafer starts per
month in capacity through September for the Inotera 300mm capacity expansion. (4) On Nanya and
Inotera's capacity, Nanya's has two 8-inch fabs with total capacity of 73k wafer starts per month.
Inotera has a 12-inch fab with current capacity of 30k wafer starts per month, which management
expects to increase to 34k wafer starts per month by the end of 2005. Management intends to
increase 300mm capacity at Inotera to 62k wafer starts per month by the end of 2006. Inotera
intends to build a second 300mm fab (called "Fab 2"), with ground breaking expected in May of
2005. Infineon and Nanya are each expected to contribute 25% toward the total $2.1 billion cost of
the new fab while Inotera is expected to finance the remaining 50% of the cost. First silicon at
Inotera's Fab 2 is expected in Q3'06, so orders for the new facility should be placed in late 2005.
Nanya is also in the final stages of deciding whether or not to pursue the construction of its own
300mm fab. Whether or not the company plans to move forward with the new fab will be decided
shortly, and management implied that the company is very likely to pursue the project. (5) Nanya
indicated that profit margins at DDR2 are currently lower than at DDR1 because back-end costs are
higher for larger die size DDR2 chips (where back-end costs are about $1.20 per chip across the
industry according to management). (6) The company believes that 2006 could be a very difficult
year for the DRAM industry due to excess supply.
TSMC: (1) TSMC reiterated its guidance for CQ1 with shipments expected to decline in the single
digit percentage range, ASPs expected to be flat (although management indicated that flat ASPs are
driven by positive product mix as the company continues to see pricing pressure from its
customers), and the overall utilization rate should be about 78% (including annual maintenance
which takes some capacity offline and hence drives a higher utilization rate). Management said all
major markets will decline sequentially in Q1. (2) Regarding CQ2, the company has limited
visibility but it does expect wafer shipments to grow slightly sequentially driven primarily by
wireline and MP3 applications. (3) The company's 2005 capex budget is approximately $2.6 billion
(mid-point of guidance). TSMC had an approximate $1 billion accounts payable at the end of 2004,
which represents equipment that has already been ordered by and shipped to TSMC from its
equipment suppliers. That payable is expected to decline in Q1 as the company transfers cash
payments to its suppliers, which then becomes capex. Management highlighted again that its 2005
capex will be significantly front-half loaded. 80% of TSMC's 2005 capex will be for the company's
300mm capacity expansion, including Fab 12 Phase 2 and Fab 14.
Most of the capacity in Fab 14
will be 130-nm while 90-nm will be a large part of the Fab 12 expansion (management indicated
that 90-nm yields are ahead of schedule). The company's 2006 capex budget will depend on demand
for advanced technology capacity according to management. (4) TSMC expects capacity to increase
24% year-over-year in 2005, with an approximate 5% increase in capacity expected in each of Q3
and Q4 after a 10% seq. increase in CQ2. (5) TSMC is seeing better terms from its equipment
suppliers this year as opposed to last year. (6) 90-nm products are ramping to 10% of sales in H2'05
driven by baseband and graphics applications as well as consumer related applications in the US
and Japan. (7) TSMC introduced copper and low-k at 130-nm but the company will not introduce
many new materials at 90-nm. For 65-nm, the company is considering the use of SOI technology,
but it is not expected to become mainstream until 45-nm. The company is also starting to use high k
gate dielectrics as it makes additional linewidth shrinks.
Powerchip: (1) Powerchip's 2005 capex budget is approximately $1.2 billion, up from $800 million
in 2004. Management indicated that the majority (about 70%) of its
2005 capex has already been ordered. (2) Powerchip's 300mm expansion plans are as follows: in
2004, total 300mm capacity reached about 40k+ wafer starts per month. In 2005, the company
hopes to bring this capacity to 60k - 70k wafer starts per month and in 2006, management intends to
increase 300mm capacity to 85k wafer starts per month. The company has two 300mm fabs, Fab
12A and 12B. Fab 12A is near full capacity (40k wafer starts per month out of total capacity of 45k
wafer starts per month). The company is currently moving-in 15k wafer starts per month in capacity
to Fab 12B out of total planned capacity of 45k wafer starts per month. Phase 2 for Fab 12B's
capacity expansion will take place in Q4 and is expected to add about another 10k wafer starts per
month in capacity. The orders for the second phase of Fab 12B have not yet been placed and will be
approximately $400 million when placed in approximately the June timeframe. The company
intends to break ground on a third 300mm fab, Fab 12C, in Q4. Powerchip believes that it will be
able to increase its market share by increasing its 300mm capacity. The company also views the
lower 300mm costs as a competitive necessity. (3) Management estimates that industry-wide
DRAM bit supply growth in 2005 will be about 40% - 45% while the company estimates that its
own bit supply growth will be approximately 65% or more in 2005. Management is assuming about
7% PC unit growth in 2005. (4) The company also indicated that back-end DDR2 costs are currently
approximately 2x back-end DDR1 costs (with back-end DDR1 costs estimated to be $0.70 to $0.80
per chip). (5) On overall DRAM pricing, management indicated that the Chinese Lunar New Year,
seasonality, Intel chipset shortages, and increasing yields are all driving spot prices to fall off since
February. The company expects spot prices to improve in about 4 weeks. (6) Powerchip is another
chipmaker that is seeing better payment terms from its equipment suppliers, including more spare
parts and services bundling.
Last week's news, events and price performance:
Last week
Tuesday 29 March (1) Veeco Instruments announced the introduction of the NEXUS 420 Ion
Source designed to enhance the performance of Veeco's NEXUS Ion Beam Etch systems. (2)
Credence Systems said that Silicon Storage Technology has chosen the Personal Kalos 2 test system
to test its 89 Series of FlashFlex51 Microcontrollers. (3) Teradyne announced the election of Paul
Tufano to its Board of Directors. Mr. Tufano most recently served as the President, CEO and
member of the Board of Directors of Maxtor Corporation and has also held several management
positions at IBM.
Thursday 31 March (1) Tegal Corporation announced that a Silicon Valley-based technology
company has placed an order for Tegal's Endeavor AT PVD cluster tool. The Endeavor system will
be used in the development of RRAM devices. (2) Cymer shipped its first XLA 200 to a lithography
system solution provider. (3) Cymer also announced that due to a correction in the method of
accounting for its consumables refurbishment activities, it has updated its operating results for the
fourth quarter of 2004, which were originally reported on January 25, 2005. As a result of the
change, Cymer's revenues increased by $28.5 million, cost of product sales increased by $25.6
million, net income increased by $2.0 million, 2004 diluted earnings per share increased by $0.05
and the year-end inventory balance increased by $2.9 million.
Friday 1 April (1) Nikon Instruments announced the integration of two high precision industrial
metrology business units, Semiconductor Inspection, and Nexiv Vision Measuring Systems under
the SITECH Division located in Tempe, Arizona.
For Price Performance, please refer to Exhibit 1.
I, Jim Covello, hereby certify that |
