Applied Materials (AMAT) took the wind out of most people's sails before Thanksgiving when CFO George Davis said that Applied sees the market for wafer fab equipment ranging from $26 billion to $30 billion, down from a range of $30 billion to $32 billion in 2012. He notes that the expected 5%-15% drop is consistent with most other market forecasts.
Part of Applied Material's hypothesis is that 20nm will be 25,000 or 30,000 wafer starts in 2013 and they note they have no visibility for the second half of 2013.
Our contention is that 20nm will be 40,000 or 45,000 wafer starts in 2013, and the semiconductor equipment market can grow as much as 10% instead of decreasing as Applied Materials suggests.
Our Proprietary Leading Indicators (PLIs), which we have utilized and fine tuned since 1995 and correlate with semiconductor equipment billings supplied by SEMI, started diverging in June 2012. In other words, our PLIs continued to show an increase since June while SEMI's billings have shown a decrease, and a pretty significant decrease at that. This correlation is shown in the graph below.
Calculating the difference in reported billings since June, we see that 13.4% more revenues should have been generated in 2012 than were recorded, as shown in the chart below. Again, these were calculated against our time-proven PLIs. Since all the parameters for our PLIs have been consistent since 1995, then equipment sales should have been consistently increasing since June 2012.
While this anomaly can be interpreted as fear, uncertainty, and doubt (FUD factor) associated with the eurozone fiscal crisis, the U.S. election, and the upcoming Fiscal Cliff, historically there have been numerous comparable periods throughout the past 17 years and yet equipment correlated with the macroeconomic forces represented by our PLIs.
One factor causing this anomaly could be delays in equipment purchases due to uncertainty in what leading-edge semiconductor manufacturers see as the next "node" in device design and processes. For example, Advanced Microdevices (AMD) , Altera (ALTR), Nvidia (NVDA), Qualcomm (QCOM), and Xilinx (XLNX), likely will make the traditional progression from 28nm to the 20nm planar node before moving to 3D FinFETs designs. We see these moves by 2015, meaning huge purchases of equipment in 2014.
However, some foundries could skip the 20nm design and move directly from 28nm to 14nm, or at least accelerate their 14nm schedules, due to competitive factors not only among themselves but against Intel (INTC), which plans to move to 14nm in 2013.
In that case we should see pent-up demand for an additional 13.4% of fab equipment bought in 2013, which equates to $4 billion on top of planned purchases by other semiconductor manufacturers.
These purchase delays may be a stroke of luck for the equipment suppliers, because it gave leading-edge semiconductor manufacturers six months to re-evaluate their technology roadmaps in light of a competitive posturing by Intel in 22nm and 14nm chip designs. The 13.4% equates to an additional $4 billion in revenue pushed out to 2013.
Why the move to smaller dimensions on a chip? "IC makers that moved from 40nm to 28nm have experienced a 35% average increase in speed and a 40% power reduction," said Jack Sun, vice president of R&D and chief technology officer at Taiwan Semiconductor (TSM). "In comparison, IC vendors that will move from 28nm to 20nm planar are expected to see a 15% increase in speed and 20% less power."
For example, the chart below shows Intel's Atom Roadmap, and they will are already making Ivy Bridge processors at 22nm and plan on Silvermont at 22nm in 2013 and Airmont and others at 12nm in 2014.
For the company's desktop processor business, Sandy Bridge-based 32nm processors currently still account for 60% of shipments, while Ivy Bridge-based 22nm processors, launched in April 2012, account for 34%, Atom series 4% and Sandy Bridge E series 2%. In the first half of 2013, with the launch of the new Pentium and Celeron processors, Ivy Bridge-based processors' proportion will rise to 75-78%, Sandy Bridge and Atom will together have less than 20% and Haswell, which is set to launch in the second quarter of 2013, will have 4%.
In the second half of 2013, Ivy Bridge E, Sandy Bridge E and Sandy Bridge-based processors will together account for 4%, Atom will also have 4%, while Haswell will have 20%, leaving Ivy Bridge-based products still the mainstream processor in the market with 72%.
The time for Intel to purchase equipment for new 14nm production capacity is 2013. Intel has four 22nm fabs online. Our opinion is that Intel has already purchased sufficient equipment for 22nm production, particularly in light of delays in construction of the Ireland fab.
CLSA Asia-Pacific Markets's Srini Pajjuri offered up a think-piece on the question of whether Intel will offer chip manufacturing services in its foundries to Apple (AAPL), a subject of ongoing speculation. However, the main issue, as he sees it, is a lack of excess manufacturing capacity until 2015 at the earliest:
"Assuming Intel and Apple both agree on ARM (ARMH), we estimate that Apple would need a total of 85k wafer starts per month on 22nm in 2014. Our analysis shows that Intel's own CPUs will need 58% of its 22nm capacity in 2014 even if we assume that half of Intel's units are on 14nm. That leaves 39k 22nm wafers for others. Clearly that won't be enough though Apple could start using Intel for some of its volume. Intel should have its 2 new 14nm fabs fully ramped by late 2014, and we estimate that Intel will need 87% of the 14nm capacity in 2015. However, Intel will still have five 22nm fabs and converting two of them to 14nm should be enough to address most of the Apple volume."
Other semiconductor manufacturers are moving toward 20nm and 14nm.
United Microelectronics Corp. (UMC)
The state-of-the-art Fab 12A phases 5 and 6 will provide advanced 28nm, 20nm, and 14nm capacity, and is scheduled for equipment move-in during the second half of 2013. Total capacity is projected to be 50 thousand 300mm wafers per month, which will bring total monthly design capacity for Fab 12A to 130 thousand wafers.
UMC on last week it would aggressively develop its 14nm FinFET process technology using the same methodology as adopted by its rival Globalfoundries with the 14nm-XM process, which utilizes back-end elements of the 20nm manufacturing technology.
"Based on our IBM FinFET licensing, UMC has decided to aggressively develop 14nm FinFET technology on 20nm metal. 14nm FinFET will deliver the most optimal low-power and high-performance solution to offset the cost impact from using double patterning lithography," said Shih-Wei Sun, chief executive officer of UMC, during a conference with financial analysts.
Globalfoundries' leading-edge Fab 1 capacities in Dresden, Germany, and Fab 8 capacities in the U.S. are fully ramped by early 2014, the company will be able to process 140 thousand of 300mm wafers using 20nm and 28nm process technologies per month.
The 14nm-XM offering from Globalfoundries is based on a modular technology architecture that uses a 14nm FinFET devices combined with 20nm-LPM process back-end-of-line (BEOL) interconnect flow. Even though UMC provided no details about its development, it looks like it will take a very similar approach as Globalfoundries. What is important here is that UMC's 20nm manufacturing technology will use FinFET transistors.
Globalfoundries intends to mass produce chips using its 14XM fabrication process in 2014, it is unknown when UMC intends to initiate manufacturing at its 14nm node.
Taiwan Semiconductor Manufacturing Corp.
TSMC's new Phase 5 expansion at Fab 14 will be the company's second 20nm-capable fab following its current 20nm capacity at Fab 12 Phase 6 in Hsinchu. According to the company's plans, Fab 12 Phase 6 in Hsinchu will begin 20nm volume production sometime in 2013, while the new Fab 14 Phase 5 expansion in Tainan will begin 20nm volume production in Q1 2014.
TSMC's 20nm pilot line is still on track for 2013. But recently TSMC accelerated its FinFET risk production schedule from February 2014 to November 2013. Mass production is slated a year after its 20nm planar process.
Samsung Electronics (GM:SSNLF)
Samsung commenced work on a new plant in mid-2012 , where it plans to make smartphone and tablet processors with the groundbreaking 20nm and even 14nm processes.
Samsung Electronics has started producing the industry's first 4Gb, low power double-data-rate 2 (LPDDR2) memory utilizing 20nm technology. The mobile DRAM chip went into mass production in March 2012, and is expected to help in the delivery of advanced devices that are faster, lighter and provide longer battery life than today's mobile devices. Samsung is looking to ramp up its 20nm process, making the 4Gb DRAM line its mainstream product.
Samsung makes most of the silicon in Apple's gadgets and its own handsets and tablets. Ever since Intel announced plans to offer foundry services at 22nm speculation has run rampant on whether Apple will move from Samsung to Intel or TSMC.
One thing to remember in the semiconductor industry is that manufacturers historically follow the "herd mentality", which has been responsible for periods of excessive purchases in the past, as illustrated in the first chart, and will continue to do so in the future.
The semiconductor equipment market stands to grow 10% in 2013 based on pushed-out sales from 2012 and on new fabs by leading semiconductor foundries and manufacturers. Each new fab can result in $2 billion in sales generated.
The global economy may be ready to turn the corner, according to recent economic reports coming out from China, Germany, and the U.S. We just have to convince the semiconductor manufacturers.