The SATA IO working group announced one new spec and that it has begun work on another: µSSD and SATA Express, respectively. These two specs span the gamut from ultra mobile devices all the way to high end PCs and servers. Let's tackle the smaller one first.

Tablets and smartphones are quickly becoming very powerful. We'll see another 20 - 50% gain in CPU performance in the next 12 months, and chances are we'll see a repeat performance another 12 - 18 months after that. As we've learned from the PC industry, increases in processor performance must be met with faster memory and IO. We're already running into IO bottlenecks on tablets. Copying large videos and audio to the current generation of tablets is a pain, with write speeds averaging less than 20MB/s. Faster storage is necessary.

The µSSD spec simply introduces a standard electrical interface for SATA on BGA devices. Currently if you want to connect a SATA controller to a SATA device you use a physical connector, whether standard SATA, micro or mini SATA. Either way it's a big physical connector. With µSSD the SATA interface is in the ball-out of a BGA chip on a motherboard, there's no physical external connector - just route the traces from your SSD to your SATA controller.

What we'll see going forward is integrated SATA controllers in SoCs, which will now have the option of interfacing with either a standard SATA SSD or a µSSD soldered onto the motherboard. The latter is more desirable for ultra small form factors. SanDisk has already been shipping a µSSD compliant device: its iSSD. Expect others to follow.

The next interface is SATA Express. This one is designed to keep SATA relevant while allowing for the growth of high performance SSDs. It took SandForce all of one year to release a controller that could saturate a 6Gbps SATA interface. Many companies have turned to PCI Express as a solution to offer more bandwidth than a single SAS/SATA port can deliver. SATA Express is designed to fix this issue.

The spec won't be complete until the end of the year, but it will allow for two new SATA speeds: 8Gbps and 16Gbps as well as backwards compatibility with existing SATA devices. SATA Express will leverage PCIe 3.0 for higher operating speeds, perhaps indicating that it'll layer on top of PCIe rather than remain as an independent interface. If the spec is completed at the end of this year, don't expect to see controller support until mid to late next year at the earliest. It'll take a while for SSDs to natively support the standard as well.

POST A COMMENT

15 Comments

View All Comments

  • deed02392 - Sunday, April 22, 2012 - link

    You're not accounting for the fact most drives have cache, which will be read from even at times when the physical drive is seeking (HDD) or an address is having an IO performed on in the flash (SSD). Reply
  • peterfares - Monday, September 24, 2012 - link

    Why not? SSDs can do that just fine or should be able to. Reply
  • Practice - Friday, August 9, 2013 - link

    I don't know where you are making up your information, but SATA isn't and has never been full-duplex. If you want a full-duplex drive you will need to go with SAS. Reply
  • Stormprobe - Saturday, July 5, 2014 - link

    SATA is and always has been half-duplex. SCSI and SAS on the other hand is full-duplex. I can't stand people who assume things and then try to pass that false information along as fact. Do your homework!
    Using your flawed logic, i can connect a full-duplex Ethernet cable to a half-duplex router, and that will make it full-duplex.
    Reply
  • supremelaw - Thursday, April 12, 2012 - link

    Forgive me if readers have already seen this point discussed elsewhere:

    Occam's Razor: the simplest solution is always the best solution.

    Now that the PCIe 3.0 specification calls for 8 GHz clock speeds
    and 128b/130b "jumbo frames" at the chipset level,
    is it not logical to extend this "topology" outwards over
    flexible SATA cables? and SAS cables too?

    A "SATA-IV" standard should support the same features:
    8 GHz clock speeds and 128b/130b "jumbo frames"
    yielding "8G/8b" or 1.0 GB/second raw bandwidth
    over a single SATA cable.

    One key advantage of these 2 features is that
    they allow each x1 PCIe 3.0 lane to "sync"
    with a single SATA cable, greatly simplifying
    capacity planning AND greatly reducing the
    operational overhead in controller firmware.

    With the obsolete 8b/10b "legacy frame",
    SATA firmware is constantly adding and removing
    1 start bit and 1 stop bit from each data byte transmitted!

    That's an extra 20% overhead which the PCIe 3.0 spec wisely eliminated.

    Also, if you look carefully at one of the 2 "SATA Express" connectors,
    there are "keys" which prevent a SATA cable from connecting:
    (cf. "keys that reject the SATA cables");
    so, that mechanical connector is called "SATA Express" but
    it doesn't work with a standard SATA cable or connector.

    In the long run, I also think that we should be designing
    motherboards and add-on controllers that support the
    concept of "overclocking" data storage subsystems,
    just as we now routinely overclock CPU and DRAM
    clocks.

    This can be implemented with a variety of methods
    that are widely in use now, e.g. hardware jumpers,
    Option ROM settings and even auto-detection,
    and hopefully new methods that will be developed
    and proven in the visible future.

    The lack of such features is frankly retarding
    "bleeding edge" experimentation and progress
    feasible with new storage technologies, imho.

    Lastly, I am picking up a distinct bias which
    maintains that the highest performing storage
    technologies must also carry THE highest prices.

    You can observe this trend in Nand flash SSDs:
    highest WRITE speeds require the LARGEST capacities
    (partially due to the number of parallel controllers).

    I believe it's time to challenge that bias openly,
    so that enthusiasts, gamers and DIY builders
    can enjoy very high-speed storage, perhaps at the
    expense of less capacity.

    MRFS
    Reply

Log in

Don't have an account? Sign up now