With the major push that Intel and others are making through PCI-SIG for M.2 in the ultrabook market, many industrial embedded OEMs are asking us which form factor they should choose for their next generation designs. The answer is not a simple one – it comes down not only to form factor, but also host computer interface, performance needs and perhaps most important, software drivers.

Both mSATA and M.2 form factors came out of the desire of notebook, netbook and ultrabook manufacturers to integrate a low profile SSD into their products. 2.5” hard drives and SSDs were too thick to accommodate the thin, light design criteria and mSATA (MO-300) and Slim SATA (MO-297) were the first competing form factors. Slim SATA used a standard 2.5” SATA socket and mSATA ran SATA signals to a PCIe miniCard socket – the same high volume connector used for WiFi and other PCIe peripherals. Skipping to the punchline, mSATA won that battle in the notebook/ultrabook space and Slim SATA became the form factor of choice for the networking market.

As NAND flash $/GB continued to decrease, companies like Apple and others demanded more and more capacity for their mobile computing platforms. mSATA accommodated four NAND placements, so SSD capacities were naturally limited. M.2 (formerly known as next generation form factor or NGFF) will be a bit thinner than mSATA and will be available in four different PCB lengths ranging from 42mm to 110mm, to accommodate more NAND components – and thus higher capacities – than mSATA. In addition, M.2 has two different LGA/BGA designs for use in solder-down applications.

With additional SSD capacity comes additional performance and most companies found they could create SSDs that would completely saturate a single SATA 3Gbps (SATA II) or SATA 6Gbps (SATA III) interface. Since there was no SATA IV on the horizon, it made sense to expand performance with multiple channels or lanes and therefore a new socket was needed. The M.2 socket, which is available from Kyocera ,TE Connectivity and others, will accommodate either two SATA 3G/6G ports or four lanes of PCIe for SSDs. (Note: M.2 has multifunction capabilities for WiFi and other technologies that are beyond the scope of this discussion.)

It is Virtium’s hypothesis that M.2 will win the ultrabook battle, especially as SSD vendors and software providers implement NVM express (NVMe) protocols over four-lane PCIe for maximum performance. It will not be an instantaneous transition, however. Ultrabooks for 2013 into 2014 will use mSATA. Next generation designs may use a SATA implementation on M.2 before finally moving to PCIe plus NVMe in 2015/2016.

For industrial embedded systems, it is our belief that both form factors will coexist. mSATA sockets are already designed into most industrial embedded computing platforms from companies like Kontron, Radisys, Emerson and Advantech. Those systems are just now being deployed and will be around for many years. Since most industrial embedded system capacity and performance requirements are modest (there is a much higher premium placed on reliability and consistency), there are few technical reasons to move to M.2 unless customers are willing to live with the market dynamics – read that as constant product changes – driven by the ultrabook space.

Some systems may need PCIe because there are no “free” SATA sockets, but then consideration will need to be given to software.  If the OEM can live with AHCI protocol, PCIe may be an easy choice.  NVMe is focused on ultrabook and enterprise types of operating systems and may not be supported by some real time operating systems or customized versions, but at least it is an industry standard.  That leaves vendor-specific software drivers that may make it more difficult to qualify multiple vendors.

From a Virtium perspective, implementing SATA on either form factor is easy. We already have a complete line of mSATA products and we will support SATA M.2 form factors as demand dictates. Other implementation details are outlined in our product roadmap, which is not available in the public domain. If you are interested in seeing more, please contact the Virtium sales team.

Acknowledgements: drawings courtesy of Intel and PCI-SIG.

As capacity needs get beyond 4GB, networking customers are starting to evaluate MLC based solutions to get more in line with traditional storage budgets. SLC is still good for 4GB and less, but for capacities of 8GB and above, industrial MLC starts looking very attractive since most endurance concerns can be mitigated by higher capacities.

CompactFlash

You can’t start any discussion of what networking OEMs are looking for without starting with CompactFlash.  To be honest, I can’t believe how many industrial CompactFlash sockets are still in the market.  It was truly the one form factor that did it all – fixed or removable, IDE or PCMCIA memory or I/O mode – that in its day was also the flash card of choice for consumer electronic devices.  Many customers are looking for the next thing that does it all, but unfortunately, it’s not out there.  Several manufacturers have come out with new generations of CF controllers, so CF will continue to have legs through the 1xnm and 1ynm NAND process nodes.  Re-qualifications are definitely coming, but to most OEMs, CF re-qualification is much more palatable than system redesign.

USB

We are seeing a lot of volume on embedded USB 2.0.  USB is generally used where there is no easy, “free” SATA port – applications using Cavium or Freescale processors, FPGAs and customer-developed ASICs.  Most implementations have used 10-pin eUSB modules in either standard (2.54mm pitch) or low profile (2.0mm pitch) configurations.  Some customers have used 9-pin as well.  We are also seeing a pick-up in the need for BOM-controlled embedded USB thumb drives either for software load or for service technicians.  OEMs don’t want $5 USB thumb drives to be recognized on a multi-thousand dollar piece of equipment.

We haven’t seen a rush to USB3 yet, but we are keeping our eye on next generation embedded processors.  Even in these applications, capacities for new designs are creeping up to 8GB+, meaning companies are looking at industrial MLC.  Virtium’s soon-to-be-released Gen2 eUSB not only handles NAND down to 1ynm SLC, MLC and TLC, but it is also capable of USB3, meaning it will have a very long product life.  We will certainly have a thumb drive form factor and are looking at several options for standardizing a USB3 based embedded module.

SATA

2.5″ and Slim SATA dominate designs that can use SATA because of easy migration from 2.5″ HDD.  The interesting thing is that most Slim SATA sockets are still SATA 3G (SATA II) where many 2.5″ SATA implementations are moving to SATA 6G (SATA III).  There is interest in 512GB industrial MLC Slim SATA form factor, especially for blade applications because at only 4mm thickness, ATCA designs can get the capacity they need along with the airflow that is required for cooling.  I chuckle at all the hype around 1TB SATA SSD in a 2.5″ form factor.  Virtium gets half that capacity in a Slim SATA form factor that only takes up 15% of the volume.  That’s a trade-off that many industrial embedded OEMs are willing to make.

We are also seeing a lot of interest in solder-down SATA BGA, especially since the 156-ball pin-out has been standardized by SATA-IO.  ATCA and COMs boards with modest capacity points that are looking to save space are finding this interesting.  While there are many “solder-down” SATA solutions available on the market, most are proprietary and therefore not multi-sourced.  The 156-ball pinout from SATA-IO will change that.

PCIe: Since Intel is driving the M.2 spec for ultrabooks, many networking companies are investigating this. Implementing M.2 using PCIe means some software work needs to be done on the host.  The two dominant protocols are NVMe and AHCI with the former likely to be the long-term winner. Implementing SATA interface on M.2 is easy for Virtium.  The one point of caution is that M.2 requires a brand new socket that doesn’t have a lot of traction yet, but should be well developed by 2014.

Minimal re-qualification efforts:

Many SSD suppliers use SSD controller (SSD processor for all you LSI/SandForce fans) for SLC and another for MLC. This makes SLC to MLC transitions difficult. Virtium uses same ASIC and firmware for SLC, MLC and in the future even TLC, making things like compatibility and response to commands easy to qualify. The only things that change are endurance (for sure) and maybe write performance. Everything else is consistent. This makes qualifications as straightforward as possible.