Editor's notes from MRS Spring Meeting: Symposium C

Veronica Bermudez of NEXCIS presented paper #C1.9 (“In-line quality control of chacolpyrite-based solar cells by advance Raman spectroscopy”). The work was driven by the fact that CIGS’ competitiveness is compromised by relatively low efficiencies achieved at the module level (9-13%). The presence of inhomogeneities that degrade the optoelectronic properties of the layers is the culprit. The researchers wanted to determine the presence of such inhomogeneities at an early stage in the fabrication process and Raman spectroscopy is well-suited for the application – specifically, use of quasi-resonant Raman scattering measurements. Bermudez said that Raman measurements at quasi-resonant conditions are a fast and sensitive tool for assessing alloy composition.

Yanyan Cao of DuPont presented paper #C2.2 (“Cu2ZnSn(S,Se)4 thin film solar cells from binary and ternary chalcogenide nanoparticles”). One reason researchers are interested in CZTS for solar PV applications is because it comprises earth-abundant elements/compounds – a definite cost consideration. Cost considerations also come into play by the choice of using nanoparticles; because they are dispersable in various solvents, they enable multiple low-cost coating techniques such as R2R processing on flexible substrates.

In the coming days, watch for podcast interviews with Daniel Josell of NIST and Fred Seymour of PrimeStar Solar.

(Debra Vogler)

AMD: The good old days of fabs & servers

AMD's 1Q11 sales were in line with estimates and sees 2Q a bit better, but while longtime rival Intel shone in its 1Q11 breakout, AMD still is weighed down by chip price declines and share losses. Credit Suisse's John Pitzer compares AMD's latest results (1Q11) vs. five years ago (1Q06) judging that sales haven't increased much ($1.6B vs. $1.3B), and margins are noticeably lower (GM 44.6% vs. 58.5%, OM 5.7% vs. 19.4%) as is EPS ($0.08 vs. $0.37).

But the biggest concern is AMD's share of the server market has evaporated, from 22% down to 7% (and its PC share is down a few notches to 18%), points out Pitzer. "The lesson: Owning server market share is more important than not owning a fab," he writes.

True, fabless AMD no longer has the hassles associated with funding next-gen process technologies and new fab capacity, nor the same debt/interest/legal issues, points out FBR Research's Craig Berger. But he agrees that AMD has a big problem in server share, he's skeptical about AMD's proclaimed ability to ramp 32nm this summer, and questions whether partner GlobalFoundries will be properly motivated to push hard on 22nm:

[S]ome investors may get excited that AMD's 32nm Llano Fusion processors are now shipping, that AMD is profitable even though it only has 6% market share in server chips, that 2H11 seasonality should improve, and that possible market share gains could drive meaningful leverage in the model[.]

[W]e, on the other hand, continue to see AMD serving a niche market not occupied by Intel, with Intel's massive process leadership and R&D investment scale continuing to hamper AMD's ability to get any real traction over time. [...] AMD appears to us as having a lack of strategy and leadership, given that 32nm production yield visibility remains low, and given that Intel's much-improved Sandy Bridge HD Graphics 3000 could potentially address more of the consumer notebook market for discrete GPUs, thus possibly hitting discrete GPU attach rates.

The board's unexpected CEO dismissal leaves AMD without a clear leader or strategy at a very important time for AMD as it ramps integrated APU Fusion processors. [...] [W]e await the hiring of a big-name outsider, and hope that AMD does not become the umpteenth chip firm to target the tablet market with a "me too" ARM-based processor that drains AMD of the vital resources it needs to compete against Intel.

Berger also takes a stab at the impact of tablets on PC usage, calculating that 2.5 tablets sold cannibalize one PC. Assuming 2011 projections from Apple (35M-40M tablets) and new entrants (20M-30M), as many as 70M tablets could ship in 2011, and thus ~25M PCs cannibalized. Given a 375M PC unit base, that could shave a precious -6% or -8% off PC unit growth rates, he says.

Analyzing Intel's capex boost, PC disconnect

Two key takeaways from Intel's 1Q11 blowout financial performance and execs' comments in the follow-up conference call are resonating across the industry: why bearish analysts were so far off on PC demand, and why Intel is already hiking up its 2011 capex.

Trust our market visibility, not theirs. Despite analysts' assertions and concerns over perceived sluggishness in PC demand, Intel saw 17% Y/Y growth (12% Q/Q) in its PC client business. It may not have direct inroads into chips for mobile/tablet devices, but those devices rely greatly on connectivity, which means accessing backend/server systems, where Intel does have solid business. And as far as the PC market, Intel's Paul Otellini didn't mince words:

Like many of you, I noted that some of the third-party research firms issued reduced forecasts for PCs in 2011. I want to be clear that our views differ from some of theirs. The PC business has evolved into a global industry that is approaching 400 million units this year.

While some channels like PCs sold through consumer retail outlets and mature markets have deep visibility, other channels, especially in emerging markets, are not well reflected in the forecast of third-party firms until shipments from Intel and its competitors have been reconciled.

Over the last 5 years, we have put considerable effort into improving our visibility with systems like just-in-time inventory hubs for our major customers, as well as realtime metrics to monitor sales through all of our worldwide channels. As a result, we were able to call the inflection in our business in Q1 of '09, as well as predicting 2010 growth to within 1 point of accuracy.

Our projections for PC segment growth in 2011 remain in the low double-digit range based on early sell-through strength we are seeing as we begin 2011 and the great reception to Sandy Bridge in both Consumer and Enterprise segments. And while it's too early to call 2012 with an improving global economy, we see no reason for growth to be materially different from what we see in 2011.

Mea culpa, admitted a few analysts. FBR Research's Craig Berger upgraded INTC to "outperform" from "market perform," acknowledging that while PC checks have been weak, "investors (including us) have been overly bearish on Intel," "PCs and tablets/smartphones can co-exist and PC units can still grow," and "more time [needs to] be spent assessing the smartphone/tablet cannibalization impacts on PC growth."

Deutsche Bank's Ross Seymore noted that the disconnect between market analysts and Intel's results is largely explained by "Intel-specific drivers" including emerging markets, Sandy Bridge adoption (and overworries about chipset glitches), and pricing. "The Street (GLCH included) missed the magnitude of burn through 3Q10 and 4Q10 in the channel ahead of Sandy Bridge launch," which led to undershipments for 2010 PC demand, added Gleacher's Doug Freedman. Also a contributing factor: an extra week (14th) of revenue in the quarter.

Better performance needs better silicon. Three months after mapping out a $9B (±$300M) capex plan for 2011, Intel has now raised that ceiling to $10.2B (±$400M), citing a need to support 22nm and 14nm (both production and R&D). Intel says it spent $2.72B of that in 1Q11 alone (up 46% Q/Q). "We think that Intel has placed orders for equipment deliveries through mid-3Q, and we think there should be continued Intel orders in 2Q/3Q to ensure the company spends its capex," writes Credit Suisse's Satya Kumar.

Explained Intel CFO Stacy Smith during the conference call Q&A:

Probably the biggest single chunk that's happening inside of this increase in CapEx is the fact that we've made the decision that for the development fab for 14 nanometer, we're going to make that fab bigger. That gives us the ability to actually, at the early stage of the ramp, move more products to 14 nanometer, take advantage of that process technology leadership, ramp it faster. So we're going to spend some construction dollars today to have that capability in place of 14 nanometer. But over the 14-nanometer life, it should save us money by going faster on that first factory.

What Intel is seeing earlier than anyone is the effect of rising capital intensity, argues Kumar. "The rest of logic is only spending on 28/46nm," he notes, and should start feeling the higher capital intensity pressure next year. Recent pushouts by Samsung and TSMC are not an industrywide weakness trend, he says. Barclays' CJ Muse echoes what Intel execs said in the call: that the needs of emerging mobile devices (notebooks, tablets, phones) require more platform features integrated into the processor, meaning more leading-edge silicon capability for power management and performance in smaller formfactors.

Of course rising capital intensity is good for equipment suppliers, and those with best visibility to new business from Intel (particularly R&D) include ASML and KLAC, Kumar says (adding that LRCX has no Intel exposure). Others, like UBS' Stephen Chin, see others taking advantage, including NVLS and VSEA. Muse adds CYMI, ASMI, and TEL to the list as well.

Intel's capex boost also means overall 2011 industry capex should grow 20% (vs. 15%), notes Muse, with "aggressive spend still on bricks and mortar," and WFE spending loaded into 2H11. Intel's WFE spending also should stay strong in 2012, he notes.

One area Intel isn't really concerned about is foundry. Intel reportedly fabbed a 22nm chip for FPGA maker Achronix, and "we are interested in talking to some very specialized companies in terms of doing foundry things," Smith said, but "we are not building a broad-based foundry business and it's not driving our CapEx number."

CA's 33% renewables target: Promise or paper tiger?

California Gov. Jerry Brown has signed a new bill that hikes up the state's commitment to 33% renewable energy of overall use, up from 20%.

Bill "SBX1 2" revises some terms within the state's Renewable Energy Resources Program, signed as an executive order in 2009 by then-Gov. Arnold Schwarzenegger. But the bottom line: The new target for renewable energy usage by 2020 is fully one-third, up from 20%.

In a statement, the Governor said the increased threshold would stimulate investment in green technologies, create "tens of thousands of new jobs," and promote energy independence. But making 33% of the state's energy overall portfolio come from renewable sources -- the highest in the nation -- "is really just a starting point, a floor, not a ceiling," he said, suggesting that as prices drop and more RE sources come online, 40% "at reasonable cost is well within our grasp in the near future."

The measure, announced at a SunPower-Flextronics plant dedication, is supported within the renewables sector and by major power producer SoCal Edison, which is already close to the current 20% benchmark. But it's not without controversy; some suggest that residents' utility bills will spike greatly.

What do you think? Is this a significant stake-in-the-ground for renewable energy adoption, a catalyst for industry and jobs that paves the way for others to follow? Or is it a partisan paper tiger that will cause more financial problems than the ones it aims to solve: power generation/availability, environmental concerns, and eventually cost/W?

AMD's carrot for GloFo, patience for CEO

AMD has revised its wafer supply agreement with Globalfoundries, apparently to help light a fire under the foundry to keep its leading-edge process technology up to snuff -- but perhaps there's another reason.

The two firms' original WFA signed in 2009 involved a "cost-plus" payment plan (fixed regardless of utilization or yields). The amended one changes those terms for 2011 to fixed prices for 45nm wafers, but based on "good die" for 32nm wafers. AMD anticipates payments to GF will total $1.1B-$1.5B in 2011 (vs. $1.2B in 2010). The "cost-plus" terms will revert again in 2012, bumping up AMD's annual payments to $1.5B-$1.9B, with bonuses if GF meets goals for 32nm capacity.

In a post-PR presentation, AMD execs acknowledged that the changing terms are in response to what were "challenges relating to 32nm yield ramp at GF" (which caused a chip-launch shuffle) but that those yields are now "in-line with our expectations." (Now those the terms just offer a bit more of a carrot for GF to keep its 32nm yields up.) Other reasons for the change were to tighten pricing options and provide better cost visibility for its 32nm ramp.

In addition to yield insurance, AMD also pledges to give GF a little extra in 2012 if 32nm capacity benchmarks are met -- so this rearrangement of pricing terms is not just about making up for past slipups, but rewarding future availability.

Ultimately there could be a higher goal here -- all this talk about 32nm yield and incentives is really just a way to smooth out the company's margins over the next year or two, concludes Charlie Demerjian at SemiAccurate. John Pitzer with Credit Suisse notes that AMD's gross margins are basically unchanged at 4%-48%, but sees long-term targets rising to 50% due to decreasing cost/die and favorable mix.

Meanwhile, AMD's search for a new CEO "seems stalled," says MarketWatch's Therese Poletti. Why? It's a unique exec and personality who'd be willing to square up to longtime dominant rival Intel, plus emerging fabless heavyweights Nvidia and Qualcomm as well as ARM, notes Raymond James analyst Hans Mosesmann. Both AMD and Intel are still weighted heavily to PCs (currently a slumping sector), and largely criticized for (so far) missing the boat to mobile devices, smartphones, and tablets and new Apple and Google devices utilize ARM's architecture, a battle front for both chip companies. Look for clarity (or at least investor inquiries) at AMD's quarterly earnings call in a couple of weeks. Mosesmann speculates that the matchmaking efforts have "been tougher than [...] they thought," but that it's still too soon to judge. "You don't want them to rush," he notes.

Calling all photographers/moviemakers: Nikon Small World Photomicrography competition

I know there are a lot of photographers and movie makers in our ElectroIQ audience. Nikon's 37th annual Nkon Small World Photomicrography Competition is accepting movies (new this year) or digital time-lapse photography taken through the microscope. The new moviemaking category will be judged separately. The call for entries deadline for both images and movies is April 30, 2011. Rules and entry forms are at http://www.nikonsmallworld.com. Good luck and have fun! (Debra Vogler)

Demands on the Cloud, data center; Wally Rhines' 3D IC roadmap

Sharon Holt, SVP/GM, Semiconductor Business Group at Rambus, cited some interesting information during her presentation at the GSA Memory Conference (3/31/11, San Jose, CA). For example: the cost of a 22nm logic IC design is >$140M – for each new design. And from Cisco’s Global Mobile Data Traffic Forecast update – Holt presented the following data for consideration on the demands on the Cloud: 1) In 2010, global mobile data traffic was up 159%; 2) smart phones are 13% of the global installed base, yet drive 78% of the total traffic; 3) so far, video represents >50% of all mobile traffic in 2011; 4) and the average tablet traffic is 5X that of smart phones, which are 24X that of feature phones. Watch for my upcoming podcast interview with Holt on the search for a unified memory solution (i.e., one that works for PCs/servers, and for smart phones/tablets).


And in his presentation at the GSA Memory Conference, Jim Elliot, VP, Memory Marketing & Product Planning at Samsung Semiconductor, tackled the power consumption challenges of the data center. He said a Web 3.0 impact on traffic load study indicates that by 2015, 1266PB of memory will be required by data centers. 1PB = 13.3 years of HD video. And data center power consumption is growing. Elliot noted that data centers account for 23% of global ICT power consumption (about 1% of total worldwide power) and growing. SSDs can help address the challenge because of course, they have no moving parts, thus, use less power. “One SSD can replace up to 20 15k HDDs,” said Elliot. “An SSD outperforms an HDD by 47X in IOPS.” (Debra Vogler)


The 3D IC roadmap according to Mentor Graphics
Wally Rhines, chairman & CEO of Mentor Graphics, outlined what he considers to be a realistic 3D IC roadmap at the GSA Memory Conference (3/31/11, San Jose, CA). Today, there are sensors on logic, limited volume stacked memories, and PoP and flip-chip memories on processors. In the next 2-3 years, the technology will move to what Rhines calls 2.5D+. It comprises a rapidly increasing use of interposers, the integration of logic and memory with flip-chip and interposers, and mixed analog, RF, logic and memory in multi-die stacks, and TSVs outside the active circuitry. In 5 or more years from now, Rhines sees the industry at full 3D with embedded TSVs in leading edge logic chips. (Debra Vogler)