October 13, 2011 - Litho tool juggernaut ASML's business is viewed as a bellwether for semiconductor demand, so its quarterly results are watched closely by everyone from toolmakers to end-market analysts. A deep dig inside the company's 3Q11 results, and executives' subsequent commentaries, unearths insights into 4Q11 and 2012 projections, an updated EUV timeline, and a curious disconnect about the industry's 450mm readiness.

-- The numbers: 3Q11 sales were down about 4% sequentially to €1.46B, net income slid 21% to €338M, and gross margins fell from 45.1% to 42.1%. Excluding EUV, net bookings sunk 39% to €514M for 23 units vs. 34 in 2Q11, (split 41% foundry, 30% NAND, 18% DRAM and 11% IDM). The backlog was pared from €2.76B/105 units to €1.99B/74 units, also skewed heavily to foundry (42%) and NAND (25%), with 82% of the backlog planned to ship within six months. That includes two 2nd-gen EUV systems representing €80M in sales but "zero profit margin," with five NXE:3100 systems delivered and a sixth being shipped. The company still has more than a dozen EUV orders in the pipeline (four for its NXE:3100 and all ten of its NXE:3300s, slated to begin deliveries in 1H12) -- and thanks to EUV its overall system ASPs rose 9% during the quarter to €23.2M (the NXE:3100 alone sells for €40M).

Despite another anticipated sales dip in 4Q11 (-25% to €1.1B, including another 2nd-gen EUV tool, with bookings "at a level above" 3Q11), ASML still forecasts record sales of €5.5B for all of 2011. Beyond year's end, though, visibility into semiconductor end-demand remains cloudy, admitted CEO Eric Meurice, though he believes increased 4Q11 bookings herald the start of new technology upgrades (2Xnm for logic and NAND, 3Xnm for DRAM). Those tool purchases should start showing up in 4Q11, agreed company CFO Peter Wennick.

ASML's results weren't much of a surprise, if anything just slightly better than expected by Wall Street watchers. (Put another way by Credit Suisse's Satya Kumar: "The results will probably not satisfy the bulls nor the bears.") Barclays' CJ Muse suggests where the 3Q11 orders probably came from (GlobalFoundries, Samsung, "incremental TSMC, Samsung's Line 16, and Toshiba), and who likely will drive the better 4Q11 orders (Samsung's Line 16, TSMC, and Intel).

The update and forecast offers "some confidence that ASML isn't falling off a cliff," says Victor Bareno from SMS Securities, quoted by Reuters. Others, like Nomura's Richard Windsor, think the 2012 outlook is a "worst-case scenario" hedging against uncertainty, but he's hoping for a stronger 4Q11 order book.

-- 2012 depends on foundries: At first blush 2012 looks like it'll be "a good technology transition wave," Maurice said. In DRAM and in logic, customers are improving and reducing their usual cycles -- DRAM now every year, logic down to 18 months. DRAM has been awful lately, so the industry's been relying on NAND and foundry customers. Even if DRAM wafer starts don't improve much in 2012, it is still "a big user of lithography," noted Meurice. And NAND demand should stay strong thanks to tablets and phones and even some PC inroads (SSDs).

The real wild-card for 2012 is foundry demand. "Foundries underinvested for a period of time so 2011 includes a bit of catch-up," noted Meurice. If enough foundry customers push for the accelerated node migration and production ramps, "then we are going to see a sustained foundry business," he said, "and if it is sustained in 2012, indeed [it will be] a very good year."

Barclays' Muse calculates that foundries need to add ~59K of 28nm capacity, and ~43K of 32nm capacity, in the next year, which translates to a ~$27.5B WFE spend -- well up from the current ~$22B WFE order run-rate. And Credit Suisse's Kumar agrees that "logic/foundry capex could surprise positively given rising capital intensity."

-- EUV update: Long-maligned source power issues have continued to take their toll on EUV progress -- now cumulatively at a "one-and-a-half year delay" in development, Meurice confirmed. Dreams of 60 wafers/hour or more by the end of 2011 have long since fallen by the wayside. (≥100 WPH is the generally understood baseline for EUV's high-volume feasibility.)

Nevertheless, work in source-power scaling has progressed enough that there's a "separation now of complex stuff from less complex things, giving us a confidence level that we're much more under control of development," Meurice said, pointing out that customers' total EUV exposed wafers nearly doubled in 3Q to >2500 wafers. He declared the 60 WPH mark with demo power levels will now be reached by next summer (2012), and then gradually scaled to ~125 WPH by the end of 2012, thanks to improvements in pre-pulse, optimized filtering, and duty cycles.

Each segment of the industry has their own baseline requirements where they'll have little choice but switch to EUV: ~18nm for NAND, 22-20nm for DRAM for the topmost two critical layers, logic probably closer to 14nm but with far more intensity (~10-15 critical layers). The contact layer is "extremely difficult to manage without EUV," ASML CEO Eric Maurice said during the conference call Q&A. "If we delay this, there will be a significantly difficult process" i.e. multipatterning. In a presentation slide, ASML currently plots production start for EUV at 2012 for NAND, 2014 for DRAM, and late 2016 for logic. With EUV tool lead-times (for the next-gen 3300s) stretching up to 18 months, look for orders to "come in clearly in 1H12" (note that the first 10 tools have already been ordered). Wennick said ASML has capacity for 18 units till end of 2013 (i.e. churning out roughly one per month from July 2012 onwards.

Note that even if EUV encounters further delays, that only means the industry has to use more complicated lithography techniques in the meantime, e.g. double (and more) patterning and other tricks. And guess who has the lion's share in installed systems for such litho tools? "Lithography intensity is sustaining the business model," Meurice said. (He confirmed ASML currently has ~70% share, likely rising to ~80% in the EUV era.) To that end, ASML also announced new milestones for its NXT:1950i immersion system: 200 WPH throughput at 125 shots (230WPH/96 shots) for a system being shipped, and >4000 wafers in a single day on a tool at an unidentified customer's site.

-- A 450mm disconnect: No discussion with semiconductor firms these days is complete without someone asking about the 450mm wafer-size transition, which recently picked up steam with the formation of the "Global 450 Consortium" centered at the U. of Albany's CNSE. While many vendors have been actively discussing and showing their 450mm tools (or at least compatibility), one key piece of tooling remains silent: lithography. "The elephant in the living room is still litho," Gartner research VP Bob Johnson recently told SST. Fellow Gartner research VP Dean Freeman agrees: "Lithography is the key to 450mm success and it needs to be demonstrated that it is economical."

It's interesting, then, that the top exec of the biggest litho supplier sees a significantly different 450mm timeline: "I would think at this moment the whole industry would agree [...] the best-case: serious prototyping beyond 2016, serious production beyond 2018," Meurice said during the conference call Q&A. Anything before that point, he said, will be "minimum investment" and " more demonstration than it is industrialization." Note how much later that timeline is than the Global 450mm consortium's collective goals of an operational pilot line sometime around 2013, and actual production by sometime around 2016, which are reaffirmed by its individual participants (TSMC and Intel). "If INTC's roadmap holds and if ASML really believes its own timing, this could potentially make lithography a major gating item for the 450mm transition," notes Citi's Tim Arcuri.

To be fair, ASML's focus is understandably on getting EUV's throughput production-ready (see above), as it is with SEMATECH. And unlike some tools that can be made 450mm compatible with a simple staging tweak, "the G forces are considerable when moving the stage and optics system," Freeman said, requiring significant resources (time and money) to reengineer the source to either considerably boost the power or the number of shots to get higher throughput. Freeman summed: "Why am I going to worry about 450mm if I don’t have EUV for 300mm at 120 WPH, and can't get a 450mm 193i system over 100 WPH because of my source power?" And the consortium may find in its feasibility studies that 450mm doesn't make economic sense anyway -- while EUV seems to be clearly on the roadmaps of memory and logic makers in the near future.

So what happens if ASML doesn't get in line with everyone else's 450mm timeline? "If the consortia wants to get lithography on a 2012-1014 prototype timeline, money will need to be involved," Freeman declared. "There is no other way for the lithography companies to make it happen with the amount of work on their plate at this time." So if ASML drags its feet, could that be an opportunity for another supplier to contribute? "Potentially [it will] open the door for Nikon as an alternative for 450mm lithography," Arcuri muses. Freeman also acknowledges this as a possibility, "but it will also take some money to move Nikon's development forward." Meanwhile, he added, "Canon is out of the market at this moment."