As a highly volatile sector, how will memory fare in light of the onslaught of tablets and smart phones? Can manufacturers address high demand for specialized memory and overcome pricing, unit volume and bit growth needs?
The disruptive effect of tablets on the wider semiconductor supply chain is an everyday news item lately. Rightly so. The adoption rate and penetration level of these new smart wireless devices (SWDs) is highly significant and set to climb dramatically through 2014, according to analysts across the board, including those within companies such as Intel and Samsung. The question explored in this article is how are SWDs, and the exciting disruptive design changes, affecting the memory sector?
Our companion consumer electronics (CE) article in this issue of MarketWatch Quarterly considers the wider, disruptive changes happening in the semiconductor and electronics industries based on the penetration of the SWDs, especially tablets and smart phones. Importantly, these two articles leave us with a clear message that the disruptive change is significant and fortuitous for many. In other words, the industry data point solidly to vigorous growth as a result of the new, demand-rich, directions – and yes, even for the volatile memory sector.
Memory cycles and forecasts – new, divergent trending?
Among the serious concerns in the memory sector presently is the outlook for DRAM, having come off of an above-average ASP second-quarter (2Q10) with supply shortages and strong demand, to move to what many have feared was a negative third-quarter (3Q10). While earnings downgrades plagued the news for many weeks at the end of 3Q10, Intel and Samsung's mid-October 3Q10 earnings reports surprised many by showing jumps and good visibility for demand through 1Q11 (cf. this summary by WSJ.com). While these are only two companies, they contribute greatly to setting the tone and pace for the semiconductor industry.
With the wider industry excitement over the record adoption rates of tablets by consumers and some corporations and health care firms, there are a notable few semi sectors that are not set to benefit from this phenomenal demand opportunity. Is DRAM one of those sectors (cf. this article from WSJ.com as an example of the current industry news)? With the internal design of tablets based not on the laptop model (favoring DRAM and HDDs) but mobile phones (favoring NAND flash, specialty DRAM, and new, integrated MPU and GPU chipsets (e.g., processors such as Intel's Sandy Bridge, Apple's A4, etc.)), the demand driver for 'traditional' DRAM is reduced.
DRAM – understanding the numbers
Interestingly, what we are also seeing in the DRAM sector is a shift away from contract-based ordering and towards demand-driven ordering, according to inSpectrum, as cited in DigiTimes, 10-1-10. This shift represents an important change for forecast models for both manufacturers and the entire DRAM supply chain.
Delving more closely into the DRAM market numbers, we find a healthy dose of cyclicity, although not following the usual quarterly patterns. In the case of DRAM, this unfortunately means that the ASP gains found in 1H10, notably the high 2Q10, have dissipated based on lower demand in 3Q10, and conservative outlooks for 2H10 (cf. DigiTimes, ibid.; and here from WSJ.com).
However, there are new reports of positive forecasts by leading manufacturers of a sector rebound during or after 1Q11, depending on the source. Kwon Oh-hyun, President of Samsung Electronics' chip business provided the following forecast here by WSJ.com, "memory chip prices are expected to recover from the mid-second quarter next year after they continue to weaken through the first quarter."
The reason for the healthy turn in the first half of 2011 (1H11) is found in the anticipated heating up of the tablet and smart phone competition with the release of the numerous OS systems from leading OEMs. While these devices are not as DRAM-rich as notebooks and netbooks, smart phones do have mobile DRAM chips that carry ASPs of four- to five-times that of PC DRAM, according to industry analysts (cf. Barclays Capital "Semiconductors Global Memory Outlook for 2H10 and 2011", 9-29-10, pp. 14, ff.). With significant growth expectations for the mobile handset sector, particularly for smart phones, analysts at Barclays Capital, "expect handset and other communication DRAM demand may grow 65% Y/Y in CY10 and 102% in CY11. […] Smartphone [sic] units are estimated to be up another 45% Y/Y to 274 million units in CY10 and up 40% Y/Y to 384 million units in CY11." (ibid., p.14)
Some questions still linger as to what exactly happened during September, including the pre-estimates that may mean some DRAM pricing rebound is likely. Specifically, Yukio Sakamoto, CEO of Elpida was recently cited here in ElectronicsWeekly, offering that "late September showed a 'big increase' in orders and that he expected firmer DRAM pricing for October." For the most part though, DRAM manufacturers have entered declining estimates for September sales.
While not directly DRAM related data, the mass production of Intel's second generation Core processors, Sandy Bridge and Oak Trail, set to go into production 4Q10, meaning that 1Q11 new tablet competitors to the iPad as well as next generation smart phones will be able to incorporate these combined MPU and GPU on a single chip into their releases. The advantages to these chipsets? They were specifically designed to meet the specifications of these sophisticated mobile devices, tablets and smart phones. As a result of this timing, diversified manufacturers, such as Intel, Samsung, Micron, Hynix and others, who are agile enough to support the component needs of the smart wireless device (SWD) growth (e.g., with MPU+GPU ICs and specialty, mobile DRAM) are forecasted to see solid growth beginning already in 4Q10 and continuing through 2011.
An important check to these forecasts, particularly from executives at leading DRAM manufacturers, requires looking to the analysts' reports. What we find is that there is significant variability in DRAM demand and forecasts depending on the type of DRAM. These differences in forecasts are based on (1) the end product demand driver forecasts for DRAM content percentage of Bill of Materials (BOM), and (2) leading versus lagging geometries.
Considering the ASP drops for both DDR2 and DDR3 1Gb, and the margins inherent in the production costs based on geometry, it is unlikely that lagging technology vendors (primarily Tier 2 and smaller manufacturers) will be able to make a profit at the present contract prices hovering at US $2.03-$2.09 (cf. Barclays Capital "Semiconductors Global Memory Outlook for 2H10 and 2011", 9-29-10, pp. 15-19). Critical to manufacturers will be the ability to successfully (and quickly) transition to the new process nodes, architecture conversions, and additional capacity in order to improve yield and meet the demand for appropriate DRAM. "[T]he power targets of low-power DRAM for smartphones/tablets could also impact yields in 2011." (cf. Credit Suisse "Semiconductors – Global Memory Industry Update" 9-28-10, p. 5)
One of the events that contributed to this demand was not necessarily a direct 'cannibalization' of notebooks from tablets, but rather from the 2Q10 higher ASPs that were then pushing OEMs to de-spec their devices because of the relatively higher percentage of total BOM that DRAM components were holding per device. With consumers sensitive to price points still from the volatile economic situation, unemployment concerns and consumer spending weakness in mature economies, rather than increase prices, OEMs opted to begin lowering the amount of DRAM in the end market products to maintain competitive pricing to attract buyers – both consumer and enterprise/corporate.
However, with present enterprise and corporate PC and notebook refresh data showing positive demand forecast to hold, a critical market sector for DRAM looks to be able to support estimates for 4Q10 through 1Q11. The refresh data are strongly supported by the positive upgrade to Windows 7, and the necessity of 64-bit cycle capability (cf. Credit Suisse, "Semiconductors – Global Memory Industry Update" 9-28-10, p. 13). Coupling the Windows 7 refresh with the growing data handling needs of corporations coupled with the increase in cloud computing for businesses (both those providing the service and those accessing the services), the growth forecasts for servers, storage and enterprise hardware is solid. The connection to IC manufacturers again arises here. According to analysts at Credit Suisse, Intel's next generation Westmere and Nehalem based servers and integrated memory controller, "should support another year of DRAM quantity growth in servers. Our base model calls for average DRAM in a server to grow from 18.5GB in 2010 to 25.8GB (+39% y/y) in 2011 and an 11% increase in units." (cf. Credit Suisse, "Semiconductors – Global Memory Industry Update" 9-28-10, pp. 7, 15-19) These refresh cycles represent a critical support base to offset any possible consumer spending lulls during the upcoming holiday periods.
Importantly, smart phones (and now tablets, which are based on smart phone designs, see our companion piece here) are both a high demand driver and are increasing in data handling features, particularly for the popular multi-tasking capabilities, HD video, and gaming, DRAM bit levels are also on the rise. This presents a new opportunity for specialty DRAM in addition to traditional NAND requirements of these devices (cf. Credit Suisse, "Semiconductors – Global Memory Industry Update" 9-28-10, p. 7).
Coupling the extremely strong growth forecasts for smart phones and tablets and their increasing DRAM bit content with the strong enterprise and corporate refresh cycles, the recent weaknesses are increasingly seen as dissipating with moderating and improving growth, ASPs and demand beginning 4Q10, and a new balance and healthy DRAM cyclicity in place by 2Q11.
NAND – ready to face the surge?
While DRAM finds a critical component of its market share in the enterprise and corporate sectors, NAND is more heavily impacted by consumer end-market products and consumer demand volatility (e.g., cameras, music and video storage, handsets, flash data storage, and the latest phenomenon, tablets). (cf. Barclays Capital "Semiconductors Global Memory Outlook for 2H10 and 2011", 9-29-10, pp. 20)
An interesting question is how will the shift in smart wireless devices (SWDs) alter the market share of DRAM and NAND. With NAND presently covering roughly one-third of total memory based on revenue, the dramatic growth forecasted for SWDs, principally tablets and smart phones, could completely alter this traditional division. According to many analysts, financial and industry alike, the impact of tablets will be phenomenal. As iSuppli forecasted here in September, "Shipments of NAND flash for tablet devices are projected to reach 1.7 billion Gigabytes (GB) next year up a phenomenal 296.1 percent from 428 million GB in 2010. The shipments will continue to climb steadily over the next few years, hitting 8.8 billion GB by 2014."
Source: iSuppli, Shipments of NAND Flash in Tablets to Triple in 2011
Although one might think that the 'good fortune' of the tremendous growth in tablets, and the ensuing demand for NAND to support these devices, was reason enough to be excited about NAND's forecasts both short- and long-term, there's an equally phenomenal end-product driver in NAND's court: mobile phones.
While smart phones are critical to the SWD discussion, particularly their pairing with tablets (as a companion consumer electronics (CE) device and in some cases a tethered device), the more basic mobile phone is important to the overall NAND forecast. Smart phones bridge DRAM and NAND in critical ways, most notably their focus on the user interface (UI) and user experience as the design driver (e.g., response rates, multi-tasking, power management, and heat dissipation to name a few component design features). As such, smart phones act as an important driver for both of these memory sectors, albeit differently. Mobile phones, it must be remembered, are also dependent on both NAND flash and specialty DRAM components.
While smart phones are quickly moving from high-end CE to a watershed moment of wider penetration (cf. this MarketWatch Quarterly's issue companion article on the CE phenomena), the broader category of mobile phones in general, are reaching significant volume numbers globally. If we consider the subscription base for wireless services, we are offered a glimpse into the incredible unit volumes encompassed by the array of handsets, from ultra low-cost to high-end smart phones:
After crossing the 5 billion threshold in September, global wireless subscriptions will crease by another 100 million and hit 5.1 billion at the end of 2010, amounting to 74.5 percent of the world's population. […] the global installed base of wireless devices will amount to 4.9 billion at the end of 2010. […]
In terms of components, the wireless communications semiconductor market is set to swell to [US] $80.2 billion in 2014, rising at a Compound Annual Growth Rate (CAGR) of 7.1 percent from [US] $60.9 billion in 2010." (cf. this market analysis from iSuppli)
With only five major companies providing nearly the entire global supply, and noting that there are additional end-market devices driving NAND demand, what is the impact of these phenomenal growth forecasts on the NAND sector? Most of the major NAND manufacturers are forecasting, on average, 83% year-over-year (YoY) bit growth for 2011, up from 76% YoY for 2010 (cf. Barclays Capital "Semiconductors Global Memory Outlook for 2H10 and 2011", 9-29-10, pp. 20, 24). For mobile phones alone, NAND content is forecasted "to double from 7GB today to 14GB exiting next year", according to Credit Suisse (cf. Credit Suisse, "Semiconductors – Global Memory Industry Update" 9-28-10, p. 27). Notably, today's leading smart phones are already shipping with more than double that NAND content, underscoring the watershed moment approaching with increased NAND content penetrating the wider mobile phone base (ibid., pp. 27-28).
Before becoming overly concerned about the ability for NAND manufacturers to meet these bit-demand forecasts, an historical footnote needs to be offered: these estimates are only moderate levels historically. The preset 76-83% forecasts represent roughly half the percentage bit growth that NAND experienced in 2007 and 2008. Certainly, there is little risk of excess, and the strong demand brings with it very solid forecasts for NAND in terms of bit growth, ASPs, and overall unit and revenue growth.
The critical question is, have the necessary CAPEX investments been made to ensure meeting both capacity and next generation NAND demands going forward? The simple answer is, yes. The focus of these CAPEX investments has been underway and it has been geared towards shrinks in order to keep an eye on improved margins through cost reductions. Notably, the estimated 2010 CAPEX levels are also roughly half that of the 2007 peak, (US $7.25 billion for 2010, compared to US $13.7 billion in 2007, according to the same Barclays Capital report (ibid., p.21); Credit Suisse estimates lower CAPEX spends, but Barclays' seem more realistic given the companies' liquidity, deferment of DRAM CAPEX investment, need to remain competitive, and overall market conditions).
What about NAND ASPs then? If the demand in both unit volume and bit-growth is as significant as the device demand forecasts for SWDs and mobile, more generally, indicate, and if manufacturers are moderate in their CAPEX investments and production forecasts, how are ASPs most likely to react? While one might expect the heavy demand levels coupled with moderate production and capacity forecasts to push ASPs up, the forecasts are for the 1H10 relatively firm levels to continue to be followed by less than normal decline cyclicity, at least for contract prices. What that means is that ASPs will continue to decline, just not as quickly as they typically have. The slowing in ASP decline is directly related to the demand, each acting positively upon the other with lower prices supporting increased unit volumes.
The demand is not seen as posing any significant disruptive supply chain events. As Credit Suisse succinctly outlines, the increases in demand for NAND are forecasted to be met thusly:
For 2010 we expect 77% bit supply growth to be driven by 38% increase in bits/wafer from 3xnm shrinks and 28% from higher wafer starts. The wafer start increase is being driving by a 10% increase in wafer capacity and a 16% increase in aggregate utilization from 86% in 2009 to 100% in 2010. (cf. Credit Suisse, "Semiconductors – Global Memory Industry Update" 9-28-10, pp. 20-21, 25-26).
The greatest boost to the NAND sector as a result of the above, staggering growth forecasts is the overall increase in higher NAND adoption and at higher densities across devices. Beyond the demand for tablets, smart phones and mobile phones, more generally, demand for NAND, and the related ASP moderating, is forecasted to boost the adoption of SSDs. SSDs, as an end-market product, have experienced a relatively slow penetration rate, owing primarily to the cost differential with DRAM-based HDDs. With NAND ASPs moderating and increased competition in the NAND manufacturing sector, analysts are forecasting a market opening for the adoption of SSDs (finally), especially for high-end PCs and notebooks (cf. Barclays Capital "Semiconductors Global Memory Outlook for 2H10 and 2011", 9-29-10, p. 30).
Same but different
There are a few critical take-aways from this deep dive into understanding the critical news from the DRAM and NAND memory sectors in light of the positive disruptions that SWDs are having on the semiconductor and electronics industries:
- While the phenomenal demand for SWDs are based on mobile designs as opposed to NB designs (i.e., favoring NAND and specialty DRAM over traditional DRAM and lighter NAND content levels), there is no overall forecasted cannibalization of memory as a result.
- DRAM will undergo shifts in focus favoring DDR3 and specialty DRAM driven by the significant demand of SWDs and more generally the proliferation of mobile handsets into emerging economies.
- NAND will undergo moderate bit-demand growth, further promoting moderate ASP declines. However, the significant unit volume increases will ensure solid margins and growth for this memory sector.
- CAPEX investments will continue to downscale in DRAM in response to softening in end-market device demand and broader, macro-economic outlook, globally. For those companies with diversified manufacturing portfolios, the DRAM CAPEX downscale will likely be coupled with NAND CAPEX investments to support the necessary yield improvements in light of ASP declines and volume increases.
In sum, the overall outlook for the memory sector is solid with good visibility through 1H11, with reason to expect long-term forecasts through 2014 to hold, baring unforeseen macro economic events. The question of cannibalization of memory by the phenomenal popularity of SWDs this year, by tablets and smart phones, is not borne out by the detailed end-market product forecasts, nor by detailed component analyses.
Memory has traditionally been a highly cyclical sector and experiences notable volatility. While DRAM is seen as moving toward a more demand-sensitive sector, in general, memory will continue to behave in cyclical patterns, but there is some question as to the long-term impact of contract pricing structures in light of increases in demand-sensitivity. Internally, the memory manufacturers appear to be in a good position to engage the high demand from SWDs and fundamentals are solid for increased growth in short- and long-term models for pricing, unit volume and bit growth.