Mix liquid nitrogen and Kingston’s HyperX DDR3-2333 SDRAM modules and you get 3068 Mtransfers per second (DDR3-3068). That’s what Benjamin “Benji Tshi” Bioux and Jean-Baptiste “marmot” Gerard demonstrated to a packed room full of gamers on August 21 at the recent Gamescon event held in Cologne, Germany (as reported by Softpedia). Boosting SDRAM transfer rates using liquid nitrogen to cool semiconductors below 77K (−196 °C) isn’t a particularly new stunt but pushing DDR3 SDRAM beyond the 3-Gtransfers/sec barrier is. Most of the liquid nitrogen is needed to fill the cooling tower atop the Intel Core i7-870 processor so that the processor could run at 4347.39MHz—a feat in and of itself. However, in the video (included below) you can see appreciably thick amounts of frost covering the DDR3 SDRAM modules as well. They are cold!

(Note: Although the “mission accomplished” sign atop the cooling tower in the Kingston video below says "3068MHz," the actual DDR memory clock is running at 1532 MHz, as shown in a screen shot within the video. However, DDR SDRAM transfers data on both clock edges, so the transfer rate is indeed 3068 Mtransfers/sec or 3.068 Gtransfers/sec.)

PC add-on vendor OCZ has announced today that its future is in SSDs and high-speed DRAM. The company plans to discontinue low-margin, commodity level DRAM module products in favor of add-ons with higher margins. OCZ's commodity DRAM module products currently represent roughly 70% of the Company's overall DRAM module revenue but over the past six quarters, said the company, the commodity DRAM module product line has operated at less than 3% average gross margins so it’s readily apparent why OCZ would rather not base the company’s future on these products. OCZ plans to cease commodity DRAM module manufacturing by the end of 2010. Presumably, this decision doesn’t include OCZ’s higher-margin DRAM modules such as the company’s water-cooled Flex EX DDR3-2133 modules. Such high-end DRAM modules are still highly prized by high-end users such as gamers and still command good margins.

OCZ’s announcement shows the perpetual shift in the PC add-on market where add-ons and accessories that are initially high-end gradually (sometimes quickly) become commodities as more manufacturers jump on the bandwagon. Currently, SSDs command relatively high prices. However, with more than 200 SSD vendors already in the fray—including behemoths such as Intel, Seagate, and Western Digital as well as long-established SSD vendors such as STEC—the competition in the SSD space seems to get hotter on a daily basis.

The convenience of SSDs that look like HDDs is that they can seamlessly plug and bolt into the same mechanical and interface infrastructure as their mechanical brethren. Many, many embedded designs would happily forego the mechanical compatibility in exchange for a smaller volumetric requirement because many embedded systems, like nearly all mobile devices, are quite short on extra volume. That’s precisely the market SanDisk targets with its new iSSD (integrated SSD), which crams an entire SSD with 4- to 64Gbyte capacities and a 3.0Gbbps SATA II interface into a ball-grid array package with a 16x20mm footprint and 1.85mm height.

SanDisk’s iSSD also foregoes the conventional SATA connector (too big) and instead brings out the standard SATA interface to balls on the BGA. No connector and no SATA cable. Just pc board traces to deal with.

Weight is also important in most volumetrically challenged embedded designs and SanDisk’s iSSD weighs in at a mere 0.83g. Compare that to aluminum-encased 1.8-inch SSDs, which weigh approximately 35g, and 2.5-inch SSDs that can weigh 70 to 90g. There’s as much as a 100x difference in weight!

The SanDisk iSSD’s power consumption is likewise small: 60mW in sleep mode, 1W in active mode, 180mW average. Read and write speeds for the iSSD are “up to” 160 and 100 Mbytes/sec respectively. For PC-centric applications, the iSSDs support the SMART feature common to HDDs and SSDs and the TRIM feature that’s becoming increasingly common for SSDs.

Another figure of merit that concerns embedded designers is SSD reliability. Every storage designer is aware of NAND Flash’s wearout mechanisms. Here, SanDisk provides usable reliability ratings as a long-term data endurance (LDE) specification in the form of a spec called TBW (Terabits writted), which specifies how much data can be written to the SSD over its life. SanDisk rates the LDE of its 4- to 64Gbyte iSSDs at 2.5 to 40 TBW, depending on capacity. (The TBW rating increases with overall SSD capacity.)

Like all SSDs, SanDisk’s iSSDs have no moving parts, thus has no mechanical wearout mechanism, and they are pretty rugged with respect to temperature and shock.

Last week at the Flash Memory Summit, Steve Wozniak gave a keynote presentation where he strolled down memory lane, so to speak, and discussed how important memory decisions were to the development of the many systems he’s created. Memory capacity and cost strongly influence every project’s system-level design decisions and Woz’s projects are no exception. Cadence’s senior manager of technical communications Richard Goering was there and has chronicled this keynote speech in a new blog entry on Cadence’s Community Blog site. Read Richard’s account of the Woz’s speech here.

A couple of weeks ago, I noted the continued disparity between SSD and HDD pricing in a local advertiser’s newspaper ad and got several interesting responses on LinkedIn. (See The differential cost between SSDs and HDDs continue in today’s Fry’s ad. Giant flashing yellow caution light for SSDs.) Here’s a particularly thought-provoking response from Dave Byrne in Swindon, UK:

"On Steve's original question with regards SSD adoption there are a bunch of market dynamics that will shape SSD adoption not just price:

1) Trend to multiple devices. As users adopt more devices in more form factors the data will need to be mostly kept in the cloud or NAS and synched to the devices on a as needed basis. In my house I have two Mac's, a Linux Netbook, Two iPhones and an iPad. I need to access my photo's, music, video and documents on all these devices.

I need 300-400Gbytes to store all my stuff but not in every device. I use a NAS for local storage and web services for media across all these devices. ZumoDrive for pictures and music. For documents I use Box.net. But there are many others. These services will move much data requirement off the client and into the cloud. The HDD's for mass storage will move with that data migrating into the storage infrastructure. Wireless data networks are only going to improve as we move to 4G. Already pricing in Europe is looking quite attractive.

2) As we move to a computing environment of massive parallelism. Multicore Multi threaded CPU's. This is going to result in more software running at the same time on your client. Therefore the storage workload is going to become more and more random. The result to the end user will be that SSD machines will show a wider and wider performance gap with HDD machines. This is already significant to many users but will become overwhelming overtime.

My thoughts are that SSD's will NEVER be as cheap as HDD's on a per-Gbyte basis. But that will not be the key factor driving storage choice in the client. Performance will become key as most data storage moves into the cloud. Client storage requirements will fall well within the range of a reasonably priced SSD. SSD adoption will continue to advance down the client system price points over the next 5-10 years as these trends play out."

I think that Dave has made the case for the eventual obsolescence of HDDs in client PCs quite well. When SSDs are priced “low enough” in “adequate capacities” and when more of us become comfortable with keeping our most important data either in the cloud or in local, networked storage, then we probably will see SSDs replace HDDs.

Yesterday, I blogged about a presentation on embedded SSDs given at the Flash Memory Summit by Viking Modular Solutions during a panel on embedded Flash. Today, I want to discuss the subsequent talk on the same panel, a presentation by Karl F Strauss of NASA’s Jet Propulsion Lab (JPL). Strauss discussed the use of Flash memory for data storage in spacecraft. You might think that shrinking device geometries make newer, more advanced Flash parts increasingly unsuitable for the radiation environments in space—at least I did based on old stuff I learned about semiconductors and radiation back in the 1980s—but the opposite is actually true and Strauss has the data to back up this claim (no pun intended).

First, Strauss discussed spacecraft as embedded systems. Qualitatively, the same project goals taunt spacecraft designers as much as other designers of embedded systems: power, mass, volume, and environment. Power is a problem for cell phone, laptop, and tablet designers who must worry about talk or operating time, standby time, and recharge time. Embedded designers trade off battery capacity and size, overall product size, and weight for power consumption. Power is also a problem for spacecraft designers. Spacecraft power comes from one or more of three sources. Two of those sources are familiar to more earthbound applications: lithium-ion batteries and solar cells. One power source, the plutonium-powered GPHS-RTG (general-purpose heat source, radioactive thermal generator) is unique to spacecraft and early model, time-traveling DeLoreans.

Although they have constantly improving charge/discharge cycle specifications and can last a long time, rechargeable lithium-ion batteries have clear, finite storage capacity and must be recharged when drained or the mission ends. Solar cells can recharge lithium-ion batteries (as on the Mars Rovers) but they only work when the sun shines and they cannot provide sufficient power out beyond the orbit of Mars. In fact, the Mars Rovers don’t really get enough solar power to operate during the Martian winter when the sun is low on the horizon so the rovers must position themselves advantageously, hunker down, and nearly hibernate during the winter.

For each spacecraft and mission, there’s a specification for the power source’s weight and volume and those specifications determine how much power is available to the rest of the system. Plutonium is the answer to continuous power for long-lived missions in space. GPHS-RTGs essentially put out full power—about 300W—for a dozen years and then power slowly tapers off. That’s why they’re used for deep-space missions. Strauss noted that the GPHS-RTGs on the two Voyager spacecraft—which have now entered the termination shock region between the solar system and interstellar space—are still operating at about 3% of their initial power rating, 33 years after launch, and the Voyagers are still phoning home.

The mass of a system is perhaps even more critical for spacecraft than for cell phones and other earthbound embedded applications because it’s expensive to send mass into earth orbit and even more expensive to kick the spacecraft’s mass further out into space. For example, said Strauss, a Minotaur rocket—derived from the Minuteman and Peacekeeper intercontinental ballistic missile weapons delivery systems—can place 1000 kilograms of payload into polar Earth orbit for $13,000/kilogram. (A GPHS-RTG masses 57 kilograms.) So spacecraft designers are always trading mass off against mission specs. They can reduce weight by reducing the number of science experiments (resulting in fewer instruments) or by reducing the mission duration (the space-borne equivalent of talk time). Naturally, they prefer to do neither and they devote a lot of time and effort to doing clever things with less mass. They must perform the same tricks with respect to volume as well, both because volume and mass are closely interrelated and because there’s only so much room under that payload shroud capping the top of the launch vehicle.

Finally, there’s environment. Those who have not designed spacecraft (including me) might think that both temperature and radiation play a role. Not really so, says Strauss. Spacecraft designers can keep electronic components in relatively benign, even balmy thermal environments that humans would not find uncomfortable. They do this with a combination of good thermal insulation and heaters (including electrical heaters, lumps of heat-giving plutonium, and self-heating of the electronic circuitry) to maintain even temperatures for the spacecraft’s electronics systems. The only place that they cannot maintain such benign temperatures, at least not for long, is on Venus where the surface temperature is a uniform 480C—considerably above the melting point of lead (or solder). A spacecraft on Venus would need to use active cooling and, with currently available power sources, would only be able to do so for a limited time before the systems overheat and fail.

That leaves radiation, which finally brings us back to advanced Flash memory for space borne applications. Storage systems based on NAND Flash memory can provide high storage capacity with low mass, volume, and power requirements. Here on earth, we call such storage systems SSDs and SSDs would be ideal for data storage in space (low mass, low volume, low power requirements) except for their susceptibility to radiation. Three types of radiation can damage semiconductors. Gamma radiation causes transistor threshold voltages to shift and can be mitigated with controlled semiconductor processing and special doping profiles. Ion strikes can flip bits (SEFI or single-event functional interrupts) and can actually cause a semiconductor device to go into latchup followed by catastrophic failure from current-induced thermal overload (known technically as a “bad thing”). Neutron radiation can disturb the semiconductor lattice, upset device parameters, and cause faulty operation.

It turns out that NAND Flash devices are most susceptible to ion strikes and that they have been growing less and less susceptible to such strikes as device features shrink. For a Flash cell, radiation susceptibility is merely a matter of mass—the smaller the amount of oxide insulation in the Flash memory cell, the less the ability of an ion or photon to become trapped at a defect site and induce leakage. Because radiation tolerance is inversely proportional to memory-cell volume, Flash memory’s radiation tolerance has been steeply increasing over the last few years to the point where only the on-chip charge pump is really vulnerable to ion strikes. Consequently, Flash memory is becoming a very viable candidate for data storage on spacecraft because it is so attractive with respect to the other critical characteristics. As a result, the projected storage capacity on spacecraft, which has been essentially flat at 1-2 Gbits from 1977 to the present time, is now expected to climb rapidly into the Tbit region. In fact, said Strauss, spacecraft storage capacity will track Moore’s Law now and into the future, riding on the projected increases in Flash capacity.

The detailed presentation is available here to Flash Memory Summit attendees: http://www.flashmemorysummit.com/

Everyone “knows” what an SSD looks like. It looks just like an HDD, usually in a 2.5-inch form factor with a SATA connector. However, that’s not the only possible form factor, not by a long shot. Yesterday, at the Flash Memory Summit, Viking Modular Solution’s Flash Product Marketing Manager Steve Garceau stepped through a series of alternate form factors in a session on NAND Flash SSDs for embedded applications. I found his talk mind-expanding.

Historically, said Garceau, the embedded industry has relied on a very few SSD form factors for board-level SSDs. These existing, adapted form factors include Compact Flash (CF), which is commonly used in high-end digital cameras; embedded CF; and embedded USB. These form factors are all based on PC interface standards originally developed for external PC devices and they do not serve all embedded applications equally well. In fact, if there’s a maxim that doesn’t work across the huge, expansive space of embedded design, it’s “One size fits all.” One size definitely doesn’t fit all embedded applications when it comes to SSDs, processors, or pretty much anything else. The design space is just too big to settle on one or even a few SSD form factors.

Further, performance expectations for all embedded systems are hyperaccelerating. Where embedded systems once relied on 8-bit microcontrollers and 16-bit DSPs, now 32-bit RISC processors (perhaps several such processors) are now the norm rather than the exception and requirements for SSDs in these embedded systems are also expanding rapidly. Embedded applications need “more capacity, more [NAND Flash] channels, and better performance” said Garceau. Ideally, SSDs targeting embedded applications must offer one or more of the following: high performance, increased capacity, enhanced reliability (more ECC), advanced feature sets, easy accessibility, flexible deployment, and easy system scaling. These growing requirements drive SSD controller design and SSD controllers for embedded applications must look increasingly like the controllers used for PC-centric SSDs. Specifically, they must offer high-end SSD features such as automatic wear-leveling because the embedded developer will not be bothered to add such features to the software running on the host processor.

These rising performance and feature expectations have made 2.5-inch, encased SSDs increasingly popular in some embedded designs, but these metal-encased SSDs with (typically) SATA interfaces pose real problems for many, many embedded designs. They’re too big, physically, and they present cooling problems. Consequently, embedded developers are turning to a growing number of alternative SSD form factors for their designs. Garceau showed several offered by Viking Modular Solutions:

• SlimSATA. These SSDs are 70% smaller than 2.5-inch SSDs and are available in capacities to 120Gbytes now, 256Gbytes by the end of the year. They employ a standard SATA drive connector but are not encased and can be bolted to a host board. Transfer rates are 100-200 Mbytes/sec.
• The Cube SSD. This design stacks multiple circuit boards using ball-grid arrays to stack an SSD controller chip on top of one or more Flash memory boards to produce a component with a 1.18x1.3-inch footprint. Interface is through the BGA connections on the bottom of the cube or through a Micro SATA connector, if desired. Capacities to 256Gbytes are currently available with a transfer rate of 250 Mbytes/sec.
• mSATA or SATA mini card. This SSD employs the existing PCIe Mini Card developed for internal PC use but it re-purposes the connector by changing out the PCIe signals for SATA signals. Intel presented this idea at an earlier session at the Flash Memory Summit so it has multi-vendor support. SSDs in this form factor are currently available with capacities to 128Gbytes and with transfer rates of 100-200 Mbytes/sec.
• SSD DIMM. The SDRAM DIMM socket is ever present in most of today’s designs, embedded or otherwise, so why not use such a socket to snap in an SSD? That’s what the SSD DIMM does: it repurposes the SDRAM socket by drawing power from the existing socket and then providing an additional connector on the module’s top for a SATA cable. SSD DIMMs are currently available with capacities to 512Gbytes and with transfer rates of 260 Mbytes/sec. Using four such SSD DIMMs, you can currently fit 2Tbytes of SSD storage in the physical volume of one 2.5-inch SATA drive.

There are other SSD form factors, but this list proves that one size of SSD doesn’t fit all embedded applications.

Just got back from a morning spent at the Flash Memory Summit. The last talk I listened to was the pre-lunch keynote from IBM’s Andy Walls, a Distinguished Engineer who has worked at IBM for 29 years and has lots to say about enterprise storage. Walls started his keynote by discussing the 4-legged stool for a great SSD strategy. The four stool legs are:

1. Enable Enterprise MLC (multi-level cell) NAND Flash. You do this, said Walls, through write-mitigation techniques and large-capacity Flash arrays that allow overprovisioning.
2. Flexible packaging alternatives. No one size Flash package fits all. You need NAND Flash configured in SSDs, in board-level modules, and in shelves. The packaging of choice depends on the application and use case.
3. SSD-optimized infrastructure. We’ve spent 50 years optimizing operating systems, middleware, and applications for HDDs with 5-10 msec response times. It will take some amount of time for us to optimize all of that software for the faster response times we get from various NAND Flash storage devices.
4. Differentiating storage software (what Walls calls “the real differentiator”). Vendors need to help customers figure out what data to put on SSDs either through manual placement methods or through automatic tiering.

SSDs are game changers for servers, said Walls. They provide far more IOPS than HDDs; they reduce access delays; they have reduced I/O wait times; they deliver many more IOPS for fewer Watts of power (30K IOPS at 6W for SSDs versus 300 IOPS for 9W with an HDD).

(Note: Walls cautioned that all SSD IOPS are not the same. Write IOPS are far slower than read IOPS, for example.)

Finally, SSDs have the potential for high reliability said Walls. However, to get that reliability, we need to deal with the wearout failure mechanisms inherent in MLC NAND Flash. “We know how to build reliable electronic assemblies,” said Walls. “We’ve known how for decades.” We just need to figure out how to improve MLC NAND Flash endurance, he said.

When we do figure out how to deal with the endurance problem, there are three primary data-center uses for MLC NAND Flash:

• Put hot data on the SSD using manual placement, assisted placement, or automated placement tools.
• Temporary data placement. Storage applications including data warehousing, paging, and caching all deal with temporary data and all benefit from an SSD’s improved access speeds.
• Fit all of an application’s data onto SSDs. Walls noted that at least half of all databases are smaller than 4Tbytes. In data-center applications, that’s a storage capacity that can be economically achieved with SSDs alone. Workload-optimized servers especially can benefit from SSDs to really maximize performance and in some cases, where processor loading is only 10% or so due to disk-related bottlenecks, a data center can reduce server count by 10x by switching certain servers to all-SSD storage.

As a result of these radical changes in bottlenecks caused by fast SSDs relative to slow HDDs, Walls sees a return to direct-attached storage rather than the currently prevailing, years-old trend towards SANs. All sorts of servers will benefit from this major architectural upheaval. In particular, Walls pointed out that analytics engines—the servers that sift through databases to find trends in financial markets, retail shopping patterns, and criminal activity—can especially benefit from the use of SSDs for storage.