GLG News by Jim Handy

Implications: What is going on here?  Samsung is working in their shareholders' best interests, as is SanDisk.  In the end, the separation between these two interests is thwarting meaningful progress in negotiations.  Objective Analysis, in its Alerts, has outlined the impact this should have on each company.  We do not anticipate the consummation of this deal before the memory market returns to profitability in mid-2009.  See: http://www.Objective-Analysis/Reports

Analysis: The ongoing melodrama has been the subject of a number of Objective Analysis Alerts posted HERE.  After Samsung's initial open letter to SanDisk's board, with an offer to purchase SanDisk shares at nearly double the price, SanDisk communicated their opinion that this did not reflect a fair valuation for the company.  After a subsequent move by SanDisk that raised several questions, Samsung withdrew their offer, sending SanDisk's shares from over $20 to below $8.00.

SanDisk's strong IP position, coupled with the fact that the company is extremely well managed, has best-of-class manufacturing, is a predominant force in consumer electronics, and has a visionary outlook, combine to give Objective Analysis every reason to expect that the company will still be a major force in flash by the time the NAND flash market returns to profits in the second half of 2009

Implications: Large scale solar energy plants are not new, but can be useful if fossil fuels maintain high prices. They can help put a cap on high energy costs, but since they are large-scale projects they take years to install. The impact of these on photovoltaics should be minimal.

Analysis: The project described in this article is a follow-on to the Solar One project in California's Mojave Desert constructed in the 1970s, and operated from 1981-1986.  This plant was later upgraded and named Solar Two.

The point of such large operations is to generate steam and drive the same kind of turbine that is used in fossil fuel plants, a well-proven means of generating power.  This is only feasible in large-scale operations, and this 50MW plant is indeed large compared to a typical photovoltaic array, which is typically built to supply a couple of kilowatts of power.

It is interesting that the article focuses on the international nature of the project.  The contractor is procuring technology from an Israeli firm, so it is only natural that the initial tests would be carried out in the developer's facility in Israel.  The world economy is causing all large projects to have a high international element.  Enough said!

As for the impact of this generator, it will provide a minor portion of California's overall power, and is more a continuing feasibility study than an important provider of power.  Power rates will still track fossil fuel prices.  Should fossil fuel prices continue to rise, though, or even if they stay at today's rates for a protracted period, many similar plants may be constructed, and within a decade this kind of plant may help stop spiralling oil prices.

In the mean time, the impact that this kind of facility will have on photovoltaics will be immeasurably small, since the impact this plant will have on California's power rates will be immeasurably small.

Although Objective Analysis has not yet published any reports on photovoltaics, we encourage GLG members to scan our site periodically for upcoming reports on this important new technology.  Please visit www.Objective-Analysis.com.

Implications: The semiconductor market (excluding NAND and DRAM) tends to behave relatively well, with mild soft spots and less remarkable highs.  Add the schizophrenia endemic to memories and the whole complexion of the market changes to one of wild mood swings. Objective Analysis finds that the semiconductor market without DRAM and NAND is a more sedate haven for investors, but those with a flair for adventure may well choose to try and out-guess the DRAM and NAND markets to make a killing overnight.

Analysis: Every month, when the Semiconductor Industry Association (the US lobby for the semiconductor industry) announces the World Semiconductor Trade Statistics (WSTS) a side mention is always made about the DRAM market.  Lately NAND has also taken a place alongside DRAM.  These mentions necessarily tell what the market would have done had DRAM and NAND not been included - Why is that???

To put it simply, DRAM and NAND are a commodity's commodity.  There is so little differentiation between one company's product and the next that all decisions are based upon price and availability.  When availability is high, prices collapse.  When availability is scarce, prices stabilize, and sometimes even increase.

It is no big surprise, then, that the same companies that make DRAM also make NAND.  Samsung, Hynix, and Micron are leading NAND and DRAM makers.  Toshiba used to make DRAM, but is now focused on NAND with their partner SanDisk.  Intel is in NAND with Micron, but this is a strategic play since Intel believes that the PC of the future relies upon NAND to achieve necessary performance improvements.  SanDisk is in NAND as a captive source for its cards.  This leaves a few DRAM-only companies, like Qimonda (Infineon's spin-off) Elpida, Powerchip, ProMOS, Nanya, and certain smaller firms as DRAM-only companies.

All of these companies focus enormous resources on cost management, and sell their products at market prices, hoping that their cost structure teams up with the vagaries of market pricing to leave them some scant margin at the end of the day.

Interestingly enough, demand for these products rarely slackens.  The last two demand-driven slowdowns were in 1986 and 2001.  Given that demand is so consistent, it is clear that the amazing upswings and downturns in the DRAM and NAND markets come solely from the fact that prices either flatten or crash, based simply upon the existence of a shortage or a glut.

To gain a deep understanding of the market we recommend an Objective Analysis report: Understanding the NAND Market, which can be purchased at http://www.Objective-Analysis.com/Reports

Implications: The semiconductor industry is undergoing a very predictable down cycle based upon excessive capital spending in 2006.  As it always does, demand will catch up to this spending in 2009, and the market should see another upswing similar to those in the past. In the mean time, all semiconductor companies will suffer from low earnings, and will cut capital spending.  These cuts will result in a late 2009 shortage that should last through 2010.  This shortage will drive higher prices and improved profitability.

Analysis: The cited article for this piece does not seem to account for supply side dynamics, yet Objective Analysis has found that the supply side is the single most important factor governing the revenues and earnings of any semiconductor company - When supply is tight, prices stabilize while costs continually declines.  This drives improving margins and, in turn, higher levels of capital spending.  Once this capital spending begets higher manufacturing levels, an overcapacity results, and prices rapidly crash to cost.

There a forecasting method called the "Boo-Ray" approach: When the market's down, make your forecast call for eternal bad times, but when the market is good, then change the forecast to show that no wrong can ever happen.  This appears to be the case with the forecast in question: A bad start to 2008 implies that the market has changed, that there is nothing to drive demand, and that semiconductors will never again see the kind of growth they experienced only a few months before.  When one looks at the underlying numbers like unit shipments or DRAM or NAND gigabyte shipments it becomes quite clear that these have not slackened, and that demand is relatively stable.  Semiconductor revenues suffer from price crashes even during times of stable demand, and such price crashes are the result of excessive capital spending made during profitable times.

All this is spelled out in detain in a very brief and simple analysis entitled: 2008: Tough Year Ahead, which can be ordered from http://www.Objective-Analysis.com/Reports. This 4-page report explains how and why capital spending impacts the fate of semiconductor revenues and proves the point with two decades of statistics that illustrate the tight coupling between semiconductor capital spending and price collapses.

Implications: Solid-State Drives (SSDs) may be ahead of their time, but they are certainly generating a lot of press!  Companies who don't want to get left in the dust need both to understand this technology, and to understand why their customers would or would not like to purchase it.

Analysis:  In late March Avian Securities claimed that there were significant returns of SSD-based PCs at a major PC OEM.  According to information an Avian analyst heard in Asia, the un-named OEM was seeing returns of 20-30% based upon a combination of hard failures and these systems’ inability to perform to user expectations.  Industry reporters later observed that Dell Computer was the only major PC OEM to ship SSDs in volume.<br>Lionel Menchaca, chief blogger at Dell Computer, was quick to point out that the report’s claims “don't even vaguely resemble what's happening in our business” and that the 20-30% figure was off by an order of magnitude from typical SSD-related returns.  Menchaca also said that “global reliability data shows that SSD drives are equal to or better than traditional hard disk drives we've shipped.” <br>A few days later, Matt Kohut from Dell competitor Lenovo chimed in, saying that “the article circling the Internet is fear mongering meant to drive readership. On the other hand I do doubt that return rates (for SSDs) are as low as traditional hard disk drives.” <br>Without knowing more about Avian’s research model we will not step into the fray about the firm’s reliability claims.  But what about basic SSD end-user satisfaction?  After all, PCs that ship with SSDs tend to be priced about $1,000 higher than their HDD-based counterparts, and the SSDs they use are typically smaller in capacity than their HDD counterparts.  In other words, since their price/performance essentially makes SSDs luxury items, do customers find the SSD experience luxurious?<br>Although we can’t provide any hard data for this argument, there are lots of indications that users may not be gaining performance benefits by switching to an SSD.<br>The original article said that SSD-based notebooks are slower than HDD notebooks when running applications like video streaming.  This agrees with other data that we have heard from HDD analysts who point out that sequential reading and especially writing are intrinsically fast with a hard disk drive, perhaps even faster than SSDs that tap into their inherent parallelism to accelerate throughput. <br>There also may be some issues relating to the drivers currently used, since the technology is new, and these drivers may not be fully optimized this early in the game. In fact, Dell themselves have said that today’s SSDs are sometimes slower than HDDs when executing applications that exchange data in small packet sizes, one example being Microsoft Outlook.<br>SSD companies sell their products based upon promises of better reliability, lower power (extending battery life), faster throughput, and quieter operation.  In truth, few of these promises materialize when consumers switch from an HDD to an SSD.  This seems to stem from a combination of immature SSD controller technology and poorly-tuned software, both of which should improve over time.  Although Objective Analysis has faith that SSDs will eventually find widespread use in the PC, we don’t expect this to occur until substantial benefits can be realized for a reasonable cost.  Today we are very far from that point.<br>This works against the very strong SSD campaign being mounted by Samsung and Toshiba, the more modest campaigns of SanDisk, Micron Technology, STEC and SMART Modular Technologies, and the very modest campaigns of companies like Intel, Silicon Systems, BIT Micro, and others.  Certain controller makers like Silicon Motion and Phison will also see little growth from SSDs until their time comes.<br>A very thoough examination of the outlook for SSDs can be found in a report: The Solid State Disk Market: A Rigorous Look, available at: http://www.Objective-Analysis.com/Reports.

Implications: SanDisk has done very well with its captive chip supply to work a business model that is, for the most part, vertically integrated.  As the company has grown, they are now looking at improvements on this model that will further enhance both their cost structure and their flexibility.  The new model should help SanDisk keep their advantageous cost structure while providing more flexibility than prior agreements have done.

Analysis: During SanDisk’s 2/25/08 Analyst Meeting a lot of time was devoted to explaining what the new Toshiba/SanDisk MOU was really all about.  SanDisk touted that Toshiba was interested in this model from Toshiba’s high esteem for SanDisk’s technology, and SanDisk equally esteemed Toshiba’s value as a manufacturer and a partner.   <br/>

In past JVs, however, Toshiba and SanDisk shared tooling costs and wafer production 50/50.  With this new venture, Toshiba is paying for 75% of the tooling, and SanDisk gets only 25% of the wafers with an option on the other 25%.   <br/>

The real crux of the deal is that this $7-8 billion fab will be devoted to nothing but NAND flash production.  The question is whether or not SanDisk believes that they can find homes for a full 50% of the output of this behemoth after spending $3-6 billion to bring up the prior mega-fabs – Fab 3 and Fab 4.   <br/>

In this deal SanDisk gets 25% of the total wafers of this fab at cost, and has the option to purchase (at cost plus) a portion of another 25% of the plant’s output.  At a later date, should the company find an advantage to this move, SanDisk can reimburse Toshiba for 25% of the capital outlay, and gain rights to a full 50% of the fab’s output at cost, rather than at cost plus.  Presumably, anything that SanDisk purchases from Toshiba’s 50% would be done at market prices.   <br/>

The investment plays upon the uncertainty that the NAND flash market will continue to grow at its current phenomenal rate.  Every year consumption of NAND gigabytes has doubled – something never before seen in the history of semiconductors.  SanDisk is hedging their bets.  If the market slows, then the company needs only take what it can use.  Should consumption continue to grow at its current rate then SanDisk can purchase product at a cost-plus rate which will most likely be lower than market prices.  If growth is very high, then SanDisk can buy out the other 25% of the fab and lower this portion back to cost, rather than cost plus.   <br/>

In the future we should expect to see more SanDisk deals of this nature.  The company continues to make good decisions that are likely to pose difficulties for Samsung, Hynix, Intel, and Micron, the four other key players in the NAND flash market.<br/>

Investors who would like a better understanding of the market may want to read Understanding the NAND Market, a report that clairifies the business model for NAND and the factors required to succeed in this trying market.

Implications: The semiconductor cycle is here to stay.  It is driven by factors beyond the control of any single semiconductor maker.  Industry consolidation will occur, but it will be driven by the spiralling cost of manufacturing plants.  Industry cycles will not abate by the consilidation of the market unless only one supplier remains.

Analysis: This is in response to another GLG News article by Mr. John Berg - Chief Technical Officer of American Semiconductor entitled "Consolidation: A Neccessity in the Semiconductor Memory Supplier Market."

In the article, Mr. Berg presents a well conceived argument, but there are factors which he missed that heavily impact this market.

The first is that DRAM capacity cannot be turned on and off at will.  The overcapacity that is in place today, and which is discussed in depth in the report at http://www.objective-analysis.com/Reports.html#2008_Forecast, was committed to in 2006.  DRAM suppliers at this time had a choice: spend earnings on capital expansion or pay taxes to retain them as cash.  Clearly the first choice is the more appealing, despite the fact that it would cause a DRAM/NAND glut.  Memory makers have to use their capacity at full tilt or it will simply depreciate unused, bleeding off any profits the manufacturers might make.

The second is the belief that the Micron & Nanya alliance indicates a change in the way that the industry is moving.  In fact, this kind of joint venture is anything but new, and the Inotera joint venture between Qimonda and Nanya is very similar to the Micron Nanya Joint venture this post mentions.  If we turn the time machine back to the early 1990s we see a number of such alliances, the most notable being that of IBM, Siemens (later Infineon, and now Qimonda), Toshiba (whose DRAM business is now a part of Nanya), and Micron.  These four companies collaborated to develop DRAM technology that was beyond the R&D budget of any one of these companies.  This particular alliance didn’t work out as planned, with IBM dropping out of the DRAM market, Toshiba selling their DRAM business to Elpida, Siemens divesting itself of DRAM, and Micron selling the Dominion fab in Manassas, VA, USA to SanDisk and Toshiba for NAND, before buying it back to return it to DRAM production, then converting it back to NAND production with its IMFT NAND flash joint venture with Intel.

The most important factor, though, is that DRAM makers (and NAND flash makers as well) have no control over their market.  Mr. Berg mentions “predatory pricing”.  On the one side, should any manufacturer in any commodity market (NAND and DRAM included) attempt to price in a predatory manner, the simple result would be lower earnings than their competition, and that would put shareholder pressure on the company’s officers.  On the other side, if they set their prices too high, the market opens up for their competition, and the higher-priced company loses market share.  Although the argument indicates that such price decreases would stop if there were only 2-3 competitors, it would, in fact, only stop if there were a single legitimate source for such products.

Mr. Berg's argument also mentions that NAND is starting to substitute for DRAM in DRAM’s traditional markets.  Although this will happen in the next few years, the argument is premature for now.  DRAM continues to be the dominant memory of choice.  There are indeed some interesting breakthroughs being made by Spansion, EMC, and STEC.  Even Dell and Apple are looking into the use of Solid State Drives (SSDs discussed in http://www.objective-analysis.com/Reports.html#SSD) and Intel, Seagate, and Hitachi are exploring the use of NAND in hybrid hard drives (HHDs) to improve PC performance through the use of NAND (as is explored in a report at http://www.objective-analysis.com/Reports.html#Hybrid).  Interestingly enough, there is interaction between the NAND and DRAM market since Hynix and Samsung, two leading DRAM suppliers, have production capacity that can alternate between NAND and DRAM.

In brief, the Micron/Nanya deal is a rational extension of existing industry protocol, and these companies are no more likely to consolidate than are any others in the market.

Consolidation will indeed occur as the cost of manufacturing plants continues to spiral, but this will not bring an end to the price collapses endemic to the semiconductor industry.  Rest assured that there will be semiconductor price cycles as long as the industry exists, or until tax laws change, a subject that I will leave to another post.

Implications: An obscure new semiconductor memory technology called PRAM (or PCM) promises to eventually change the nature of the flash memory market.  Companies who are on top of PRAM, mainly Intel, STMicroelectronics (and their Numponyx spin-off) and Samsung are well positioned to capture this technology's success when it develops.  Energy Conversion Devices, the owner of the phase-change technology, will also benefit from royalties over the long term.  Spansion's not talking, so we don't know how this will play out for them, but BAE, Elpida, Hynix, and Qimonda are licensees, so they are likely to move in a similar direction.

Analysis: Intel and STMicroelectronics this week announced that they had found a way to bring MLC (multilevel cell) technology to their phase-change memory (also known as PRAM or PCM).  Although this announcement was made at a very technical conference, it has business implications that would not normally be encountered at such a gathering.

We note that these two companies are collaborating on PRAM technology and that it will soon be transferred to their flash memory spin-off named Numonyx.

Why is this announcement important?  MLC allows a memory chip to shrink to roughly half the size of its non-MLC counterpart.  This means that the manufacturing cost is cut by more than half.

The memory market is cost-driven - very little else matters.  In rare circumstances improved functionality will justify higher prices, but for the most part cost is the overriding concern.

PRAM, once it gains some acceptance, will only win an important share of the memory market if its cost becomes competitive with NOR.  Since MLC NOR has been shipping in volume for most of this decade, and since PRAM chip sizes are about the same as their non-MLC NOR counterpart (called SLC for single-level cell), the lack of an MLC PRAM has prevented this technology from making much progress.

With MLC, PRAM should lose its die size disadvantage against NOR flash.  Does this mean that PRAM can now start to take a portion of the $7 billion NOR market?  Well, not quite yet.  Something else is standing in the way: Wafer cost.

Any new semiconductor material faces an issue with the economies of scale: A wafer with a new material is bound to be much more costly than a pure silicon wafer until it ships in high volumes, and the volume won't increase until the price is competitive with flash.  It's a Chicken And Egg problem.  NOR uses pure silicon, but PRAM needs a silicon wafer with a new phase-change material added.  This will keep PRAM wafer costs high until high volumes are reached, and a high wafer cost translates to a high chip cost.

Right now you're probably wondering why Intel and STMicroelectronics would embark on such a hopeless journey.  What is driving these companies to invest research in a technology whose cost continues to lag behind that of NOR?

It is widely accepted that NOR will hit a scaling limit after which the chips will no longer shrink over time, so costs will level off.  Researchers originally expected to hit this brick wall in 2006 at 65nm.  Some breakthroughs were made and the process has been coaxed to go to 25nm which should allow costs to continue to shrink into 2012.  There is some evidence that NOR could be coaxed to go down to 10nm and last until around 2016.

PRAM has no such barrier, so it will continue to shrink long after NOR has hit its limit.  Once NOR reaches its scaling limit PRAM costs should speed right past it.  The big problem today is that the NOR brick wall keeps moving ahead of us.   Until then PRAM should have trouble competing on cost, preventing it from reaching the necessary volume to get the wafer costs down. 

There's one other consideration that works in favor of PRAM: It doesn't have to perfectly match NOR's cost in order to take share away.  PRAM has a faster write than NOR, so in many cases a PRAM chip can replace a NOR and a RAM.  This means that the cost target is something higher that accounts for the price of the NOR and the price of the RAM.

PRAM is a technology for the future, and any company that can make its costs low enough to compete against NOR will win in the long run.  It looks like the Intel/ST Numonyx JV will be an early provider of PRAM, leading to possible success in that market.

Implications: DRAM companies will see faltering growth. OEMs with NAND strategies will prosper while others will lag behind. NAND companies are poised to tap into PC market growth.

Analysis: On November 13th flash card maker SanDisk announced the Vaulter Disk, a NAND flash based PCI Express module which the company says “tag-teams with the hard drive to provide enhanced performance.” Similar to Intel’s proprietary Turbo Memory modules (formerly known as Robson) the 8-16GB Vaulter device harnesses the speed of embedded flash memory for frequently-accessed files while continuing to use a standard hard drive for bulk storage of less frequently accessed data. Vaulter Disk will debut at CES (Jan. 7-10) in Las Vegas.Our Objective Analysis

Our Objective Analysis

Ten years from now we will look at PC and server memory architectures from 1981 to 2007 and say: “Did we really do that?” It will seem odd to us that the entire main memory of systems was made of DRAM.

This is not the first such change to main memories – roll the clock back to the 1970s and early 1980s and you see computers using SRAM as main memory. That seems unbelievable today, with DRAM dominating main memory and SRAM relegated to “small” 2MB on-processor caches that, incidentally, are 4,000 times larger than some of those SRAM PC main memories of yesteryear.

Today, NAND flash is poised to roar into the PC market, and it’s not coming in the form of SSDs.  In fact, SanDisk’s Vaulter Disk and Intel’s Turbo Memory even challenge the need for hybrid HDDs. Compared to HDDs, NAND flash is an extraordinarily expensive form of storage which should maintain its current 20:1 cost/GB ratio with HDD for the foreseeable future. But although it’s costly, NAND is also fast so it fits well within the memory hierarchy of computing platforms.

Why now? What has changed over the last year or two to make NAND more attractive for PCs than it has been in the past? Well, NAND crossed over DRAM’s price per gigabyte in 2005, and is cascading down the cost curve at a rate that is causing the DRAM/NAND price ratio to double from today’s 3:1 to 6:1 by 2013.

Now NAND may not be as fast as DRAM, but it’s significantly cheaper. On the other hand NAND is more costly than HDD, but it is also faster. This is exactly the argument that makes any memory fit into the memory hierarchy: Faster than the downstream technology but cheaper than the upstream.

By 2013 we expect for users or OEMs to look at their memory budget and observe: “For $100 I could either buy 75GB of DRAM or 450GB of NAND.” With this in mind, they are likely to refer to some benchmark to determine which should get them a bigger performance boost, and go with that solution. In many or even most cases, that solution is quite likely to be NAND.

Where does this leave DRAM? While we expect to see the average size of DRAM in PCs continue to grow, this growth will slow eventually. Historically this growth rate has averaged 45% per annum. In the future the growth rate could slow to something like 10%.  As a result, the DRAM market, whose gigabyte growth has recently slowed from its historical average of 75% down to around 55%, will see further slowing to something in the 20-30% range, driving continued consolidation in this business, and forcing most participants to adopt a flash strategy.

DRAM will follow the path of SRAM, becoming the smaller faster expensive bit of memory that augments the speed of the larger NAND memory. Although today only two companies (Intel and SanDisk) offer such solutions, we expect flash-based memory topologies to become widespread over time.

Companies that will be impacted by this change will be all NAND flash makers (Samsung, Toshiba, Intel, Micron, Intel, STMicroelectronics, and Renesas), DRAM makers (Elpida, Qimonda, Powerchip, Winbond, and Etron), and all those in the SSD market (including SanDisk, STEC, SMART Modular, Adtron, Mtron, SiliconSystems, BiTMICRO Networks, Targa Systems, Apacer, Transcend Information, Power Quotient International (PQI), Super Talent Technology AKA Ma Labs), Advanced Media (Ritek), Hagiwara Sys-Com, Delkin Devices, Texas Memory, Solid Data Systems, and many more.)

Implications: NOR flash is threatening the DRAM business in servers.  This will further harm the already-reeling DRAM makers.  Companies who make NOR flash will see larger growth over the long term than will companies with DRAM and no flash.

Analysis: On November 14th Spansion, a NOR flash manufacturer, announced an agreement with power-zealot Virident to develop and market memory solutions designed to dramatically reduce power consumption and improve performance in Internet data centers. Citing that Spansion’s NOR flash consumes as little as one-tenth the power per GB of conventional DRAM the companies plan to co-develop a “breakthrough” memory product that enables a significant expansion of main memory in data center servers, with fast read performance and high capacity but that consumes much less energy than DRAM.

Our Objective Analysis

Ten years from now we will look at server memory architectures from 1981 to 2007 and say: “Did we really do that?” It will seem odd to us that a computer's entire main memory was made of DRAM. NOR flash will become a dominant part of server storage, and DRAM will be relegated to the task of accelerating a portion of the main storage, which will largely be flash-based.

This is not the first such change to main memories – roll the clock back to the 1970s and early 1980s and you see computers using SRAM as main memory. That seems unbelievable today, with DRAM dominating main memory and SRAM relegated to “small” 2MB on-processor caches that, incidentally, are 4,000 times larger than some of those SRAM main memories of yesteryear.

NOR flash is on a steeper price decline than DRAM, and has the added advantage of consuming significantly less power.  Just to sweeten the deal, NOR is about 30% cheaper than DRAM.

Where does this leave DRAM? While we expect to see the average size of DRAM in servers continue to grow, this growth will slow eventually. Historically this growth rate has averaged 45% per annum. In the future the growth rate could slow to something like 10%. As a result, the DRAM market, whose gigabyte growth has recently slowed from its historical average of 75% down to around 55%, will see further slowing to something in the 20-30% range, driving continued consolidation in this business, and forcing most participants to adopt a flash strategy.

DRAM will follow the path of SRAM, becoming the smaller faster expensive bit of memory that augments the speed of the larger NOR memory. Although today only one company, Spansion, offers such a solution, we expect flash-based memory topologies to become widespread over time.

This change will impact all of the DRAM firms: Samsung, Qimonda, Elpida, Powerchip, Hynix, Micron, Etron, and Winbond.  It will also influence companies who make NOR like Spansion, Intel, STmicroelectronics, Sharp, Toshiba, Macronix, SST, Fujitsu, and Renesas.

Implications: The Samsung 64Gb NAND at the 30nm node is not only 2 process generations ahead of today's NAND production, but it appears to be a far larger chip than the company is likely to put into production.  It is fair to question how soon such a challenging device will find its way into volume production. Samsung's 30nm node process, with its charge-trapping technology, is likely to sprout a new royalty stream for Spansion.

Analysis: Samsung's recently-announced 64Gb NAND chip disclosed on October 24 is certainly a head-turner, but the chip shown in the photo appears to be a whopping 370mm²! This is based upon the fact that the die in the press photograph is longer than ten of the bonding pads on its package. Most NAND makers don't put a part into production until it hits 140mm², which would require this device to migrate to 22nm before being put into mass production.  This would make mass production unlikely until 2013.

Samsung, Hynix, Toshiba, SanDisk, Intel, Micron (through IMFT), and STMicroelectronics (ST) are all migrating from 70nm to 50nm at this time, but the 30nm node is two process generations ahead of that, so it is likely to be four years off.

Even so, this is a technological feat.  Samsung has had to make significant changes to get this far, and has expressed a need to move to a charge-trapping technology to manufacture a 35nm device.  Since most charge-trapping IP was owned by Saifun, which has been acquired by Spansion, it is quite likely that Spansion will realize sizeable royalty stream from Samsung as they ramp their 35nm process.

Implications: SanDisk is taking a more active approach in enforcing their IP.  This could signal the beginning of an increased royalty stream for the company.  it is such a broad offensive move that it could trigger retalitory lawsuits.  This is a gamble.

Analysis: On October 24, the same day that Samsung held their Analyst Conference, SanDisk anounced a lawsuit seeking damages and a permanaent injunction against 25 companies, most of whom are likely to be Samsung customers.  These companies are:

ACP-EP Memory
A-Data
Apacer
Behavior Computer (d/b/a Emprex)
Buffalo
Chipsbank
Corsair Memory
Dane-Elec
Edge
Imation/Memorex
Interactive Media (d/b/aKanguru)
Kaser
Kingston
LG Electronics
Phison Electronics
PNY
PQI
Silicon Motion
Skymedi
Transcend
TSR (d/b/a T.One)
USBest
Verbatim
Welldone Company
Zotek/Zodata (d/b/a Huke)

It has been a surprise to us in the past that SanDisk has not taken more companies to court, since SanDisk has IP that covers the bulk of the removeable flash business.  The current list of companies is probably limited only to companies with appreciable business in the US, which would be the low-hanging fruit for SanDisk.

We see this as a means for SanDisk to profit from their extensive patent portfolio, but are concerned about posible retaliation, perhaps like the anticompetitive practices suit filed by DRAM makers against Rambus, or like the suit filed by ST against SanDisk asking for the rights to SanDisk's patent portfolio.

Implications: Semi sales are trending up significantly. This year will see significant growth over 2006. Current statistics dwell on difficulties at start of 2007 Bodes well for semi makers.  Less well for users.

Analysis:

• The SIA's relatively pessimistic 2% growth number stated ijn the press release referenced in this piece is based on 3-month moving averages which include all of the April-June downslope in DRAM (and other pricing and none of the recent up-slope
• Looking at the year-to-year sales, even with the early 2007 price declines (40% in DRAM, and about 50% in NAND) overall semiconductor revenues are up almost 1% over last year, which indicates that demand is strong.  Even the SIA release illustrates this with their mention of a 10% annual increase in unit shipments.
• OEM inventories that were causing difficulties at the beginning of the year have now been burned off.
• Since the beginning of February there has been a mild shortage in many markets which could turn into full-fleged allocation by the end of the year. This will drive prices higher.
• These shortages are likely to be worsened by Samsung's recent closure of their Kieheung complex of six fab lines. Samsung tells us that the lines have been returned to operation after only a few hours, but they still expect to suffer a loss of W40B (about $50M US)
• Objective Analysis stands by its forecast of 2007 revenues growing 6.3% to $268B

Implications: Any disruption to the supply of a device can impact prices and availability.  This will have a short-term negative impact on DRAM and NAND purchasers who are the major PC, MP3, and module makers, and a positive pricing impact on other memory chip makers.

Analysis: Most fabs have procedures in place to accommodate such disruptions, and Samsung is most likely now implementing the disaster recovery plan. This plan will involve the disposition of wafers according to their stage of production and the difficulty of re-inserting them into the production line.

A power loss, no matter how brief, may force the company to scrap some of the wafers that were undergoing a high-temperature process at the time of the outage. Other wafers, for example those going through photographic processes, will simply need to be re-worked by backing up and starting at some prior step of the process. Short power losses tend to cause more minor difficulties than do longer interruptions. A power loss of 20 minutes or longer may require a longer period of downtime to allow furnaces to be brought back to a stable temperature. The Giheung power outage lasted four hours.

Samsung has issued a statement that they believe all facilities will return to full operation within a day or two, resulting in losses of less than 50 billion Korean Won (about $50M US). To put this in perspective: Samsung’s 2007 semiconductor revenues were 19.1 Trillion Korean Won, so the 50 Billion figure would account for less than three tenths of one percent of annual revenues. Worldwide memory revenues for 2006 were $60 Billion, 1,200 times the size of Samsung’s maximum anticipated loss.

In a follow-up release early Saturday Samsung announced that full operations had resumed at noon, less than 22 hours after power was initially interrupted.  Lines 6 and 7 were upand running as early as 4:30 AM.  The company subsequently lowred its loss estimate to less than $40 million.

All in all this event should not have a dramatic impact upon semiconductor prices, but there will be a shock wave that will ripple through the markets over the coming week.

Companies most likely to be impacted are those that consume significant volumes of Samsung NAND or DRAM, including most PC OEMs like Dell, HP, and Lenovo, and MP3 player makers like Apple.  Other memory makers Hynix, Elpida, Micron, Intel, STmicroelectronics, Powerchip, and module makers like PNY, SMART modular, STEC, Sanmina SCI will also see the ripple and its effect.

Implications: The odd spelling of Numonyx' name may imply the company's direction. Numonyx is in a strong position with the PRAM, a technology poised to displace existing flash technology over the coming decade. Their PRAM technology is licensed from Ovonyx, a subsidiary of Energy Conversion Devices (ECD).

Analysis:  

On July 19 Intel and ST announced that they had finally chosen a name for their flash memory spin-off: Numonyx. Most will conclude that this is a "cute" misspelling of the word "mnemonics" which many mispronounce anyway. But what of the rest of that unique spelling?

Those who listened to the companies' conference calls at the disclosure of the spin-off will easily understand why. While Intel was quite typically guarded in their comments, ST was almost gushing about the two companies' great progress in the introduction of PRAM technology - a technology that is poised to displace flash in the future. PRAM stands for their phase-change memory licensed from Ovonyx, a subsidiary of ECD (Energy Conversion Devices). OUM is an Ovonic Unified Memory - Ovonyx' claim to fame.

It appears that Numonyx is grooming itself to eventually stop being a flash supplier and become exclusively a provider of Ovonyx’ devices. Perhaps Ovonyx and Numonyx will become even closer in the future.

Implications: NAND-based notebooks are a hot topic of conversation today.  Palm's Foleo offers most of the required features of a notebook PC with NAND instead of an HDD.

Analysis: Sometimes innovation comes from an unexpected direction.  Right when everyone was expecting to see PCs adapt an HDD-less format, substituting NAND solid-state drives (SSDs) for HDDs, and right when the industry's focus was on notebooks and ultramobile PCs from established PC suppliers like Dell, Lenovo, and HP, along comes PC-outsider Palm with just what the industry expected.

The Palm Foleo, announced on May 30, is billed as a "Smartphone Companion Product", but the press release goes on to give us a list of features that obviate any need for a companion smartphone.

With autonomous WiFi capability, a web browser, email, and the ability to create, edit, and read pdfs and most Microsoft Office documents (Word, Excel, PowerPoint), the need to couple this device with a smartphone is not immediately clear.

What is clear is that this Linux-based notebook offers 5-hour battery life in a small size that makes it very attractive as a substitute for business travelers.  It weighs 2.5 pounds and is slate to retail for only $499.  Especially attractive is the device's instant-on feature, which is a feature often touted by promoters of NAND-based PCs, but will not be a reality until Windows can shed a lot of the legacy that is the actual root of the PC's slow-boot problems.

It will be interesting to see if Palm's Foleo has an unintended impact on the business notebook market, or if instead PC makers ignore the device and continue in the direction they are already following.

Implications: Intel and ST on Tuesday 5/22 finally announced the NOR flash spin-off that had been rumored for months. Intel's NOR business and ST's NAND and NOR will be combined into a new company jointly owned by ST, Intel, and Francisco Partners. This move will impact these companies' competitors and customers in different ways taht are detailed in the commentary of this posting, along with notes on the details of the deal, and mentions of what the companies did and did not disclose about the deal. One big question remains: These two companies plan to finance a very necessary 300mm expansion through profits they hope to start generating by 1Q08. It was not clear how this was to be made to happen, nor what their plan was should they be unable to finance this expansion because of market conditions beyond their control. There are still some pretty significant open issues with this deal, but investors in the parent companies should be pleased that a step has been taken to stop the bleeding.

Analysis:

Intel & ST Create New Flash Company

STMicroelectronics and Intel finally announced a joint venture that had been rumored for weeks. With Francisco Partners, a private equity firm, the two companies will create a new jointly-owned company in the second half that combines Intel’s NOR business and ST’s NAND and NOR businesses.

What they Said

• The new company will be owned 48.6% by ST, 45.1% by Intel, and 6.3% by Francisco Partners.
• ST will receive $468M and Intel $432M in addition to their equity stake for their flash businesses. Francisco will invest $150M.
• The company will purchase both ST’s NAND and NOR businesses, but only Intel’s NOR business.
• The relationship with Hynix will continue, with the new company assuming ST’s option to buy back their full 33% share in the Hynix/ST Wuxi JV fab. They intend to exercise this option as profits return to the company. (ST recently sold a portion of this back to Hynix leaving ST with only a 16.6% share.)
• Their hope is to see profits return by the first quarter of 2008.
• Loans have been arranged to total $1.55B, partly for working capital, and in part to pay Intel & ST for their businesses.
• The firm will be incorporated in the Netherlands with headquarters in Switzerland.

What they DIDN’T Say

• How the business plans to go about manufacturing 300mm
• Any name for the new company.
• Whether Hynix would be used to manufacture NOR.

Observations

Intel’s compensation for their assets is smaller than ST’s although ST’s flash business is smaller than Intel’s. Last year ST sold only $1.57B in combined NAND and NOR flash compared to our estimate of $2.0B of Intel NOR revenues. ST’s stock and cash compensation are both 8% higher than Intel’s. When asked about this the companies stressed that this was a negotiated deal and that all parties were pleased with the results. All assets and “resources” were figured into the split, and the companies frequently mentioned their combined patent portfolio of 2,500 issued and 1,000 pending patents. This is likely to have been a key factor in determining this valuation, which was stated to be approximately $3B.

ST’s shuttered M6 facility in Catania will be part of the deal, giving the new company a shell into which they can put a badly-needed 300mm wafer fab. Neither ST nor Intel would project when the shell would be equipped – they only said that they expect synergies to generate cash hopefully starting in 1Q08, and that this would be used to invest in additional 300mm capacity. We can only guess that this new-found profitability would stem from scaling their products from 90nm to the current volume 65nm process, then to 45nm in the near future.

In response to investor questions about the rationale of Intel providing NAND through two entities, Intel replied that the new company’s NAND will only be sold into the cell phone market and other embedded applications, and will be provided as stacked products (NAND/NOR, NAND/DRAM, etc). The new company will not enter the mass market for NAND.

ST frequently mentioned the importance of both companies’ PRAM development efforts. It appears that they expect PRAM to start to become an important business sooner, rather than later, even though Intel mentioned that they are in volume production of 65nm NOR and already have working 45nm prototypes. (Incidentally: We have heard a rumor that Nokia is evaluating PRAMs, and these would most likely be from ST, Intel, and Samsung, all of whom appear to be sampling devices.)

What this means to the Industry

Spansion pulled ahead of Intel into a leading position in NOR in 1Q06, a position that it will have to yield to the new firm. Spansion plans to return to profitability via its migration to 45nm and its new SP1 300mm fab which has already started processing wafers and is expected to ship revenue units in the second half of this year. A 300mm fab usually gives about a 30% cost reduction over 200mm. So although Spansion may be losing its positions as both the first-ranked NOR manufacturer and the largest pure-play flash maker, the company is quite likely to be the first NOR maker to return to profitability. This will make it easier for Spansion to invest early and perhaps pull ahead of the new company. This will be an interesting race to watch.

Samsung has set a goal of becoming number one in NOR revenues by 2009 or 2010. Had ST and Intel remained separate entities this would have been easier – both companies were in the $1.5-2.0B range in sales last year, and Spansion had $2.6B in NOR revenues. With the new company already at a revenue level of $3.6B, (most of which was NOR), Samsung will need to grow revenues by another $1B to reach its first-place goal.

Hynix is in an enviable position, moving from a single flash partnership with ST to now having joint ventures with ST, Intel, and SanDisk, all in the space of a few months. Hynix will be managing the Wuxi China fabs and other manufacturing plants in a relationship that should give it very large volumes that will help the company take advantage of economies of scale.

Intel and ST will be able to distance themselves from the difficulties posed by their NOR businesses in the past, yet they can profit from the larger scale of the operation and its narrowed focus. Once the new company becomes profitable these two will have the option of making a capital gain on sales of its stock.

OEMs will now have no second source for the products that Intel and ST introduced 2 years ago, under their agreement to second-source each others’ NOR chips. This was a strategic initiative that will be a casualty of the deal. Other companies may use this to try and convert Intel/ST designs to their own products. The new company believes that their synergies in providing a complete solution will outweigh any disadvantages OEMs see in the lack of an alternate source.

Jim Handy
Objective Analysis
www.Objective-Analysis.com