Newly introduced and available for pre-availability orders, the privately branded 2.5-inch SSDs from established Apple Mac component vendor OWC (Other World Computing) are a sign of the rapidly changing SSD landscape with respect to mass marketing of solid-state drives to end users. The Mercury Extreme Pro series of SSDs range in capacity from 50 to 480 Gbytes and they physically look pretty much like every other SSD out there (milled, blue-anodized aluminum cases notwithstanding). These drives are based on SandForce’s SF-1500 SF-1200 controller chip and what’s significant about the introduction of these drives is that OWC is passing through a significant chunk of SandForce’s technology and terminology but reframing that information and passing some if it along at a less technical level. In other words, OWC has substantially transformed SandForce’s tech speak features through words and pictures into important benefits easily understood by end users. This mass marketing approach opens a new front in the SSD wars.

First, OWC offers two versions of the Mercury Extreme Pro SSD. There’s an RE version (RE stands for RAID-ready enhanced) and a non-RE version. The RE versions have capacities of 50 to 400 Gbytes and cost $230 to $1600. The non-RE versions have capacities of 60 to 480 Gbytes and cost $220 to $1580. OWC’s marketing materials make it very clear that the difference between these drives is SSD controller firmware and that the RAID enhancements are based on SandForce’s RAISE (redundant array of independent silicon elements) technology. RAISE, one component of SandForce’s so-called DuraClass technology, distributes data across the various NAND Flash chips within the SSD using RAID-like algorithms. SandForce claims that this technique coupled with extended ECC algorithms reduces the chance of read errors by 100x. OWC highlights quotes from the SF-1500 data sheet to bring the complexity level of the claim down a notch or three and to push a benefit that’s easily understood by client-level users:

“Best in class error correction (ECC) and SandForce RAISE™ (Redundant Array of Independent Silicon Elements) technology provides RAID like data protection and reliability without loss of transfer speed due to parity.”

Another fascinating facet of the differentiation between the RE and non-RE versions of OWC’s Mercury Extreme Pro is the explicit discussion of capacity overprovisioning. The non-RE SSDs are 7% overprovisioned and the RE versions are 28% overprovisioned, which partially helps to explain the difference in drive capacities and the different warranty periods (3 years for the non-RE drives versus 5 years for the RE drives).

However, the convincer for most client-level SSD buyers isn’t going to be all the text claims. It’s more likely to be the marketing graph OWC has posted that shows how the write speed of Mercury Extreme Pro SSDs doesn’t degrade over time as do unnamed “competitive” SSDs. Here’s the graph:

Who in their right mind would pick an SSD that rapidly approaches a zero Mbytes/sec write throughput as shown in the graph above? No one, that’s who. Yet this is clearly a “marketing” graph, lacking numeric scales for both the X and Y axes. Even without numbers however, you have to admit that the graph does its job.

OWC has also posted a boot video similar to the one we blogged earlier (See “Corsair Video vividly shows SSD speedup on laptop”) but for a MacBook Pro booting up with SSD assistance instead of a PC. Videos like these from Corsair and OWC make it tangibly clear why an end user would want in SSD in their laptop of PC. They're selling the sizzle, not speeds and feeds.

These ongoing changes in the way SSDs are marketed point toward the beginning of a maturation in the SSD market. As SSD prices continue to fall and as capacities continue to rise, SSDs become increasingly attractive to a larger number of end users, and changes in the way SSDs are marketed to these prospects reflect that evolution.

Most SSDs are designed to be interface- and form-factor-compatible with existing rotating mass storage devices (aka: hard drives or HDDs). However, that need not be the case. Many SSD vendors are making significant sales by offering SSDs with alternate interfaces. Fusionio’s SSDs are based on PCIe for example. Perhaps the SSDs with the highest unit volume are lowly USB sticks, although their performance leaves a lot to be desired—even with the USB 2.0 interface’s maximum transfer rate of 480 Mbits/sec. However, USB 2.0 is no longer the zenith of USB performance. Now we have USB 3.0 with its 4.7 Gbits/sec maximum transfer rate starting to appear on some of the most advanced PCs and high-performance PC component vendor OCZ aims to capitalize on the USB 3.0 speed boost. OCZ has announced that it plans to sell a USB 3.0 SSD. The product’s name is Enyo and it’s housed in a futuristic milled aluminum package. Beyond the fashionable stylistic flash however are some impressive Flash memory performance specs.

First, the specs vary depending on drive capacity, which indicates how adding flash devices to the SSD improves parallelism and therefore boosts performance. OCZ’s Enyo drive will be available in capacities of 64, 128, and 256 Gbytes. The 64-Gbyte Enyo SSD’s maximum read performance is 225 Mbytes/sec while the maximum read performance for the 128- and 256-Gbyte drives is 260 Mbytes/sec. Maximum write performance for the 64- and 128/256-Gbyte drives is 135 and 200 Mbytes/sec respectively. For sustained writes, the maximum write performance for the 64- and 128/256-Gbyte drives is 40 and 150 Mbytes/sec respectively. Most of these transfer rates are well beyond the abilities of the USB 2.0 spec (and somewhat beyond the ability of the original 1.5-Gbits/sec SATA interface), hence the need for a USB 3.0 interface on the external Enyo drive.

OCZ’s Enyo SSD is an early indication of the kind of ingenuity that will be unleashed by the USB 3.0 SuperSpeed interface. You should expect rapid adoption of the USB 3.0 interface in laptop, notebook, netbook, and desktop PCs. In fact, PC World’s prediction is that next year’s CES will be crawling with products sporting USB 3.0 devices. No doubt, OCZ’s Enyo will be one of those.

Numonyx has announced or reannounced two 128-Mbit non-volatilve memory devices based on the company’s 90nm PCM (phase change memory) process technology. These two devices target existing NOR Flash memory sockets and the company’s press release claims that both products are available now in production quantities. The two announced PCM memory devices are the Omneo P8P PCM parallel-I/O memory and the Omneo P5Q PCM serial-I/O memory. The Omneo P5Q PCM device is configured as a 16Mx8-bit memory and employs an SPI interface and supports dual and quad SPI protocols. The Omneo P8P PCM device is configured as an 8Mx16-bit memory with a 16-bit parallel I/O interface and the device has a secondary SPI channel for issuing block commands to the device and for low-pin-count, in-system programming.

The initial read access time for the parallel PCM device is given as 115 nsec with 25 nsec for subsequent reads during 8-word (the parallel device is configured as a x16-bit memory) , asynchronous page reads. Write cycle time is a short 70 nsec, but the writes are buffered internally by a 64-byte FIFO that masks the actual cell-specific write time. Using the same memory cell, the serial P5Q PCM memory device’s read-access time is constrained by the SPI interface, not the fundamental access time of the memory cell. One of the big advantages of PCM over NOR Flash is the ability to directly write one memory location with no need for an erase cycle.

Two significant attributes of the devices, made significant because of the technical issues surrounding, are the announced write endurance and the operating temperature range of the devices. Numonyx had previously announced a version of the P8P PCM with 100,000-cycle write endurance. The latest announcement jumps the rated write-cycle endurance for both the P8P and the P5Q memory devices to 1M write cycles.

Operating temperature is important because PCM employs a thermally activated storage mechanism. Ones and zeroes are stored in PCM cells based on a heating/cooling cycle. Fast cooling produces an amorphous state in the PCM’s chalcogenide material and slow cooling essentially anneals the cell into a crystalline state. If the ambient temperature is too high, the heat alone will start to anneal PCM cells. Numonyx’ data sheets give both memories’ operating temperature range as 0 to 70° C.

With data sheets on line and announced volume availability, these two PCM devices from Numonyx appear to the first commercially available, commercially viable PCM devices on the market. Numonyx is not currently disclosing pricing except to say they’re currently charging a premium given that the devices are not very far down the learning curve. However, given the activity that the company has put into presentations and articles over the past two years, they likely are not the last.

Vanguard International Semiconductor, Once a DRAM failure, is Now a Successful Junior TSMC

The conundrum that Taiwan DRAM makers find themselves in today has been a topic of this column for many months, along with other commentary that has appeared on the internet, in print media and on the airwaves for far longer. In short, the Faustian agreement struck by Taiwan's chipmakers with the Devil, for DRAM technology and designs, is coming due. Technology-dependent on their 'corporate partners (masters)', with accumulated debt now in the neighborhood of $13-14B US, still losing hundreds of millions of $US a month on DRAM production, and reaching out for assistance whereever they can find it in governments, the courts, as well as asking forgiveness from their creditors and shareholders. In addition, they are still falling further behind on the technology front as they do-not-invest, and, finally, with no solution and still no end in sight and a world forever awash in DRAMs, there's no obvious end to the red ink, and no 'light at the end of the tunnel'. It is a tough situation they are in, to be sure.

But here is a story of true salvation from DRAM hell, as I saw in from inside and outside. And it is right there in Taiwan for all to see for themselves.

My Vanguard Experience in 3Q99-1Q00: I worked for VIS Micro, the US-based design arm of Vanguard Semiconductor, for several months in late 1999 (before they laid us all off in 1Q00). It was a useful and telling experience that has some relevance to Taiwan's DRAM Travails today.

When I got there in August of 1999, the industry was humming and the market bubble of 1999-2000 was well underway; it had another full year to run. For DRAM makers, 1998 had been a disaster, with losses of about $12B on DRAM sales of about $14B (about the same loss rate as this time); TI had tapped out, and sold its DRAM operation to Micron (always looking for a good deal). Japan had backed off DRAM over the prior years, and was set to launch Elpida, then a combination of NEC's and Hitachi's DRAM operations. Toshiba, Fujitsu and Mitsubishi were still forging ahead by themselves, in a cautious and reduced mode, at least for a little while.

But in 1999, things were rosy. In Hsin Chu, they could not get enough talent fast enough. With Coroprate Loyalty a distant second to 'compensation' for many in the mobile workforce, huge 'hiring bonuses' drove the flow of warm bodies from one employer to the one next door (easier than Silicon Valley, even in its heyday, as Hsin Chu is more compact...more "wafer starts/hectare of available land"...and has smaller parking lots to walk across.).

Shut it down!: In 9/99, when I went to Taiwan HQ office, I was told a few things that characterized the Taiwan Culture-de-jure: The most senior staff member on the Vanguard manufacturing line was hired just in July...so had only a few months experience there (big turnover was/is a problem)!. Truth or fiction, it passed for truth. Secondly, Vanguard was struggling to make their home-grown DRAM processes yield, and had 4-5 DRAM designs (their entire portfolio) running on...five separate process flows, running from about 0.18um to 0.30um. No success in any of them.

Vanguard was then run by Rick Tsai (now President and CEO of TSMC), and TSMC held a substantial share of the stock (40%); the close relation existed from Vanguard's inception. But Rick was victim of harsh circumstances, and it is doubtful if anyone could have righted the ship in any other way that how he did it.

December of 1999 was a Corporate 'Moment of Truth'. The Vanguard Board met, it was decided to cease the 'do-it-ourself' approach as unworkable, they then laid off the hundred of so enginers and chip designers who worked in San Jose and many in Taiwan, bought the Mitsubisi DRAM process flow for 0.18um, also bought the proven designs for 64M and 128M DRAMs. In just a few months, they had flushed out their own designs and processes, and brought up the Mitsubishi designs and process, successfully. They took their 1Q writedown of about $100M, including severance for us in California's design team, and moved on. That was all accomplished by about mid-2000.

Fast Forward to 2008/09: Now run the clock forward. Steadily, the world moved on from 0.18um and 64/128M DRAMs; Vanguard bought in one more design and process node from Mitsubishi, and kept making DRAMs. But they also started taking in overflow foundry work from TSMC, building logic in one of their fabs. Within a few years, they had disappeared from the DRAM business (not a cheap departure, with the process and design purchases, plus the miseries of 2001-2-3 enjoyed by all DRAM makers). But, by 2004, they were fully a junior Foundry partner for TSMC (Can you name a better Big Brother?), and off and running on their new and Improved Business Model.

While the production logistics, the tool set and process flow, end markets, 'customer relations', and business practices of Foundry and DRAMs are quite different, the transition is possible. (Many memory makers also take in foundry businesses today...Samsung and Nanya come to mind...not just to absorb spare capacity, but some with the intention of making them on-going successful businesses (esp. Samsung).)

Vanguard's Transitional Years: They were 'a man with a plan' from the start, and initially acquired the TSMC 0.5um, 0.35um and 0.25um logic process flows (again, buying proven processes, not developing their own...but that would come later). 2000 was a mostly-DRAM year. However, by 2001, their revenues were about 35% Foundry (DRAM and Logic) and 65% Vanguard (via Mitsubishi) DRAM. But 2002 was a watershed year, and their product mix was about 50% logic and 50% memory (mostly DRAM, but some flash, eFlash, SRAM and 1T SRAM). They also developed their own high-voltage processes, and steadily shaped their focus to become a specialty foundry. By 2003, memory was slightly less than half of VIS's sales, but by 2004, it was dropped to about 15%, as their logic foundry, CMOS image sensor, analog and mixed signal, and high-voltage business prospered in the industry upturn. They continued to stay close to TSMC process roadmaps, plus added some of their own,home-grown processes that provided a useful adjunct to TSMC's main process flows.

Vanguard Today: Today, Vanguard is a Specialty Foundry, with 2008 sales of about $500M. Their manufacturing capacity is not large by today's standards, with two 200mm fabs running, together, a total of about 110K wafers/month, on installed processes down to 160nm. They are still strongly aligned with TSMC, by history and by strategic choice. They also have some special standing in HV processes, running up to 200V, and other self-developed and licensed technology. (After avoiding all memory fabrication for some time, they added back some memory foundry work in 2008.)

For sure, VIS's relation with TSMC has been invaluable from the beginning.

In VIS's case, transfer of proven proccesses from TSMC (logic) and Mitsubishi (DRAM) was critical: TSMC for the long haul company direction, and Mitsubishi to keep the fab full and get them to the point where they could survive on their logic business. They disproved the aphorism, "Once a DRAM maker, always a DRAM maker"...and that non-lethal exits, and business focus transformations are possible.

Special Relevance of This Story to Today's Chip Business (Maybe): While it may sound naive today to talk about capacity shortages, the issue was raised by IC Insights in July 2008, and has not been satisfactorily resolved. Their premise that huge takedowns in CapEx in 2008 and again in 2009, combined with very fab high utilization rates (until everything went bonkers in 4Q09) make for unusual vulnerabilities for the industry. This, along with today's universally poor cash flow, and both a reluctance and inablilty to invest, join forces with the secular trend to "fablessness" at 65nm & 45nm nodes...means that when prosperity returns, companies may be clamoring for expensive-to-develop advanced process capacity...and it will not be there.

Some analysts see the first glimmerings of this in upward ASP trends taking place today.

Still, there is a huge overhang in capacity in DRAMs and NAND flash, as both were the excessive spenders in 2006-07. ISuppli analyst Nam Hyung Kim warns that much additional capacity has to be decommissioned before the supply and demand will balance, and there is no consensus that demand will bounce back to anything like it was before. But April was a marked improvement from March for all memory makers and foundries, in terms of sales and capacity utilization, putting to rest the most dire of the forecasts.

Whether this improvement will continue is now the open question. But if it does, the scenario that capacity shortfalls will show up has to be taken seriously for 2010. Idle equipment in memory makers might better be put to use in expanding the industry's advanced process foundry capacity, one way or another.

Can Taiwanese DRAM Makers Decide What to do Next...in Time?

We recently saw yet another article, this one in EETimes Asia, suggesting that Taiwan DRAM makers stop bickering with one another, and go ahead and consolidate their operations into a larger, more powerful DRAM force:

What should Taiwan DRAM makers do to survive?
Posted:05 May 2009

Not too far down in the article was this observation, which prompts this mini-BLOG. The author said:

"...But don't count on it. Taiwan's DRAM industry has too much pride to give up and consolidate. They would rather go down fighting..."

I have been going to Taiwan since about 1984, when their Sub-Micron Development Center was perched on a hill outside Hsin Chu. Then, High Tech was starting to push out Low Tech, and my cabdriver had to honk twice to get the horse-drawn cart out of his way on the muddy street that ran next to where we were going. Then, I worked for Dataquest, and we gave an annual Semiconductor Market Update Seminar to the then-nascent Taiwan chip industry. Attendance was reasonably good, exceeding expectations, especially for such a still undeveloped 'local' industry. After the meeting, I collected business cards, and talked with attendees about their companies and the local chip industry.

When I sorted the cards out on the plane home the next day, my seatmate, peering across the arm rest, commented on how large my card collection was. It was, and they were all 'President' and 'Vice President' of their own companies.

What I pieced together later was the following:

At that time, there was ONE significant semiconductor company in Taiwan, General Instrument, once the signal diode market leader. GI produced about $90M in diodes a year in Taiwan (the total worldwide chip industry was only about $20B at that time, and maybe ICs were still a minority share compared to discrete chips). In this market and place of business, GI was surrounded by Taiwanese Diode Minions, making much the same products, who had all worked at GI earlier in their careers. I had all THOSE business cards (maybe I still do). I counted about a dozen 'Presidents' and 'VPs', proudly running their own companies in small buildings around and about, and each of those companies had revenues of $2-3M each. No scale, no capital, no large organization, but lots of pride, talent and energy. (Of course, the world did not have to wait long for Morris Chang, Father of the Foundry Manufacturing movement and TSMC, and Father of the Taiwanese chip industry, to show up in 1986-87, so change was not far behind my visit).

Taiwan has made huge progress since then. After avoiding DRAMs as 'too risky', literally, for nearly two decades, and focusing on SRAMs, ROMs, and simple Logic, they finally bit into the forbidden fruit, licensing DRAM technology from Mitsubishi, Hynix, IBM, Qimonda and applying their significant manufacturing skills to be successful in their markets. And for a while it tasted so good.

But today, the Taiwanese DRAM industry has to decide what it wants to be next, with its capital, its machinery and equipment, its infrastructure, upstream system producers, and its human energy. It can be Detroit, which dallied too long, lamenting its past glory; or it can repurpose its talent and resources, and create something no one has seen before.

DRAMs have always been a seductivly large homogenous market opportunty, and therein lies the risk. Most of today's great chip companies did not displace existing players in large established markets. They built their companies and markets, over a decade or two, from the ground up, with products and technology, strategies, discipline, and close attention to profit.

Intel jettisoned DRAMs in 1985, and went on to MPU greatness; TI jettisoned DRAMs in 1998 and went on to DSP greatness. Japanese companies spun out (or shut down) DRAMs in 1998-99, and their remaining business had a chance to recover without huge financial losses. The many who could not bring themselves to cut DRAMs loose crisply and decisively, or combine operations with players in similar situations...and then agressively rationalize the resulting company...saw their future evaporate. And the longer they waited, the more their options closed in.

Taiwan DRAM makers have to decide what they want to be, next, or they may become a mere shadow of what they could have been.

The SIA reported this week the following memory chip sales results for October:

Sales of both DRAMs and NAND flash declined significantly year-on-year due to pricing pressure. DRAM sales in October were 14 percent lower than one year ago while NAND flash sales were off by nearly 41 percent - over the same time period DRAM 1 Gigabit equivalent units increased 73 percent while sales of 2 Gigabit equivalent NAND units grew 123 percent.

Translating those sales and GB shipment numbers into their respective average selling price (ASP) declines, we calculate that composite DRAM mix suffered a 50% price down over the past twelve months. Similarly, NAND flash prices per GB for the overall NAND mix declined 73.5% over the past year. For some particulars, the leading DRAM...the DDR2 1Gb 128Mx8, 667MHz...dropped from $1.96 to $0.72 from 1 Dec 2007 to 1 Dec 2008 (-63%); the most popular NAND Flash, the 2Gx8 MLC, dropped from $8.89 to $1.67 (-81%) over the same time frame. As usual, the high volume runner took the biggest hit, and you can be sure that those whose product set focused outside the Maximum Danger Zone (such as in LP DRAMs, legacy SDR & DDR1 DRAMs, older NAND Flash, G DRAMs, and SLC NAND) suffered less precipitous overall price declines.

Even in December 2007, the $1.96 price for 1Gb DRAM was estimated to be about 20% under breakeven price for the 1Gb DRAM, thought to be somewhere in the $2.30-2.45 range; vendors were losing money then, and more later as the prices continued downward.

The fact that most memory makers are still standing (albeit, some rather weakly) is testimony to their survival skills and their having made huge strides in bringing down the costs of production for DRAMs over the past 12 months. Probably no die in production in the mainstream parts of the DRAM or NAND product portfolio was in production a year ago; most have been displaced, at least once, by new, smaller cell architectures and use process design rules that are 25-35% smaller than earlier versions. Faster speed distributions come for free for the DRAMs, such as higher yield to -800, which itself is moving into the mainstream in PCs, in advance of volume usage of DDR3, expected to ramp in earnest by the middle of 2009.

Memory chip makers will lose about $5 to $6B in the current 4Q08, on sales of $9-10B, and it will get worse as we head into 2009, at least for a few quarters. After that, no one knows.

Micron Technology buys Qimonda Stake in Inotera: Micron Technology bought Qimonda’s 35.6% stake in Inotera on 15 October 2008, for $400M, and immediately halted upgrading the toolset on its own JV with Nanya fab, MeiYa; analysts enthused over the fact that Micron will get more wafers for far less investment doing it this way, which is true…Inotera is already 300mm and MeiYa was a refit from 200mm to 300mm.

Which begs the question whether Micron needed the DRAM capacity it was adding at MeiYa in the first place. Although they are clearly focused on DRAMs these days, they have delayed finishing up their Singapore IMFS NAND flash fab twice, and in the breath prior to announcing the Inotera buy out, they decided to lay off 3000 over two years at Boise and downsize that 34K wafers/month 200mm NAND fab. (BTW, IMFT partner Intel said they would take a $250M asset impairment charge for that particular Boise downsizing action in its 4Q08 period.) To be sure, IMFT's 34nm NAND technology is compelling, and seems way ahead of the competition; Micron is arguably the lowest cost NAND producer in its best process and its best lines. But, as we have seen in the past 12 months, overall enterprise fab management and execution…having the best product/process running across the entire manufacturing base, is an important key to success…added to fully-loaded fabs, focused product development spread over a broad product line, and sometimes risk separation between one’s technology partners…. Hynix-ProMOS is rather ragged, Elpida-Rexchip-Powerchip is clean and seemingly very effective. We’ll see about Micron-Nanya-Inotera, which has yet to really begin.

For sure, these are not easy investment decisions: Invest/divest in NAND and/or DRAM? Close 200mm lines or expand 300mm lines…which fabs forward and which reverse? And maybe there is yet another shoe to drop in the Micron-Qimonda saga; Qimonda’s G DRAM product line looks pretty good these days, as does its LP/Mobile DRAM position and technology as well as its next-generation 4F2 BWL DRAM technology...but the present deal yielded none of those technology or product gems to Micron, just the Inotera wafer starts, which will be converted to the Micron DRAM process over the next 8 months. For sure, no one knows how long and how deep this price gully will last, and it is safe to say that Micron is probably losing money in DRAMs today…and, if ‘conventional wisdom’ is either ‘conventional’ or ‘wisdom’, in a downturn “Cash is King.”

To continue the discussion with Qimonda, and acquire more of their fabs, Dresden or Virginia, is to believe that the industry is still strongly cyclical and the present malaise will turn back to Green Light in a reasonable time (Micron is cash flow positive, unlike most other memory makers whose financials are publicly open to scrutiny.). This is a bet; otherwise, one could only equate more DRAM capacity with more losses. Not wholly unlike asking whether Samsung’s $26/share offer for SanDisk (then selling at $14/share) was a premium of close to 100%, or a steep discount from SanDisk’s 52-week high or $56. Is the glass full? Or it is twice as full as it will be in a short time? “True believers” in the eternal cyclicality of DRAM (or NAND) markets should load up on memory maker stocks today, which are at historic low prices.

Still, it is a rank gambler, or a brave man, who will place a large bet against the forces of financial anarchy and the uncertainties of these past 60 days.

Micron Chairman Steve Appleton is on record as expecting that recent DRAM capacity takedowns will crimp DRAM bit supply growth to the 35-40% range in 2009; this could make Inotera-Dresden a good bet…short DRAM supplies, prices up, goodness. But you can also get this year’s 70% growth down to 40% by extinguishing demand with a ‘recession of significance’ …which does nothing to raise prices, profits, or anyone’s share price.