Since they first started showing up on the market, most smartphones and tablets have used eMMC flash storage. While in some ways similar to the NAND flash used in SSDs – and some devices have even gone so far as to call eMMC storage “SSD storage” – the reality is that eMMC has always been much slower than what we’re used to seeing in SSDs. UFS – Universal Flash Storage – is looking to become the successor to eMMC, and we’re very much looking forward to the host of improvements it brings.

Let’s start with a brief recap of eMMC. Initially designed as an open standard for flash memory cards, MMC (MultiMediaCard) has been around since 1997. eMMC basically takes that standard and embeds the controller and flash memory into a small BGA package for use in devices like smartphones and tablets, or anywhere else where having some flash storage might prove useful. Like most flash memory cards, MMC/eMMC has seen improvements to read and write speeds over the years, and more recently we’ve gone from 104MB/s (eMMC V4.41) to 200MB/s (eMMC V4.5) to 400MB/s (eMMC V5.0). However, eMMC is based on an 8-bit parallel interface, and scaling of interface performance is nearing its limits. UFS looks to move to a new standard.

Instead of a parallel interface, UFS uses a serial interface. Toshiba first announced support for UFS 1.1 and enabled chipset support last year, which has allowed vendor and OS support to mature. Now that they’ve had time to work out the initial kinks that new technologies inevitably bring, Toshiba is taking the standard to mass production with UFS 2.0 controllers and flash, targeting 2Q’14 for the release.

Already there are two options, HS-G2 and HS-G3, with two lanes providing an aggregate 5.8Gbps (~725MB/s) of bandwidth for HS-G2, and HS-G3 doubles that to 11.6Gbps (~1.45GB/s). Effectively that means nearly double and quadruple the interface performance compared to eMMC V5.0. Besides increasing the interface speed, UFS also brings full duplex operation (read and write simultaneously) and command queuing, which will both improve performance.

UFS is a JEDEC defined standard, and as we move into 2014 we should start seeing devices adopt the standard in place of eMMC. Short-term, UFS isn’t necessarily superior in every area compared to eMMC – Toshiba notes that eMMC is likely to be more power efficient and cheaper right now, though the gap should narrow over time. Long-term, however, devices continue to need additional performance and eMMC may not get us there (4K video is one example cited).

In their press release, Toshiba specifically mentions working to enable UFS 2.0 on Qualcomm Snapdragon 805 SoC devices. Qualcomm notes that UFS will enter at the high-end of the smartphone and tablet markets, with eMMC continuing to fill the needs of the midrange and entry-level devices for the time being, but eventually we should see UFS trickle down to all markets. Now let’s just see about improving some of the random read/write struggles we encounter on smartphones and tablets….

POST A COMMENT

8 Comments

View All Comments

  • watersb - Wednesday, January 15, 2014 - link

    Very interesting. I saw "UFS" and wondered why "Unix File System" (an old BSD standard) had been adopted for mobile devices. This is much better news. :-)

    I am amazed, however, that we find ~400 MB/s to be insufficient. You wouldn't be storing uncompressed 4k on a mobile device, you would be streaming VP8 or H264.5, right?

    Is there a simple number for "4k bandwidth" that seemed to be consensus at CES this year? Or is this just hand-waving?

    Thanks!
    Reply
  • Bob Todd - Wednesday, January 15, 2014 - link

    I'm not sure any eMMC v5 devices have even hit the market yet. And while I'm not as immediately concerned about sequential performance, anything they can do to improve random performance, especially writes, would be very welcome. I'm excited to see what this does, and what it does in conjunction with F2FS. Reply
  • MikhailT - Wednesday, January 15, 2014 - link

    It is 400 MBps, one way. So it is more like 200MBps and that is just sequential speed.

    Random is usually the weakest point with these standards. It might as well do 40MBps and still feel slow.
    Reply
  • MikhailT - Wednesday, January 15, 2014 - link

    Also, since majority of devices are still using the 4.4.1 standard, which is ~100MBps one way. Imagine the random speed on these. Reply
  • Nenad - Wednesday, January 15, 2014 - link

    While random vs sequential is usually limit of medium (flash/disk) and not interface, I guess it could be some limit here if you need to transfer address/size request after every block. While I still do not think main random vs sequential limit is interface, I guess queuing and bi-directional transfer could help remove any limitation of interface.

    On a side note, I think integrated flash even in modern top end phones can not reach 100MBs (where I guess they do not need to use eMMC), so it is questionable if we already need >100MBs interfaces ;p
    Reply
  • npz - Wednesday, January 15, 2014 - link

    These are the *interface* speeds right? Because the actual NAND performance behind low end eMMC devices are often slower than mechanical disks, as shown by anandtech's own tests.

    Still, moving to serial interfaces is better as it opens up the road for the future.
    Reply
  • JarredWalton - Wednesday, January 15, 2014 - link

    Correct. The NAND tends to be the big bottleneck. Well, that and slow controllers that can't do random I/O I suppose. Reply
  • azazel1024 - Thursday, January 16, 2014 - link

    Not sure of the bottlenecks in the various devices, if it is NAND or interface limitations.

    My T100 manages 37MB/sec writes and 107MB/sec reads for large block transfers. Its actually faster in smaller block transfers down around 128KB at about 42MB/sec writes and 115MB/sec reads. 4k speed is 25/25MB/sec read and writes. Not amazing, but I am still mildly impressed, especially compared to a lot of tablets and phones on the market which stare numbers in the face more like 3-10MB/sec 4k speeds and 20-30 sequetial writes and 40-80 on reads.

    That's with the (seemingly) more common Sandisk eMMC solution in the T100. There is, apparently, also a Samsung eMMC version that ends up with higher sequential and lower random. I think that hits around 50MB/sec writes and 150MB/sec reads, but down around 8-10MB/sec 4k writes and 20MB/sec 4k reads.

    I am all for anything that'll speed this stuff up. Personally though, I find the storage sub-system on my T100 to be pretty snappy. Its not as nice as my HP Envy 4t with a 32+128GB SSD in it (Samsung mSATA 32GB boot and Tosh Q series 128GB app/storage 2.5" drive), but in a resonable number of applications it is not way, way far off in terms of app loading and similar.

    Vastly, vastly better in a lot of things than my 4t was when it was just 32GB cache drive and 500GB HDD, especially for anything that had to load off the HDD and wasn't cached. Very uneven system performance then.

    Really though, I'll take unlimited speed, but I'd just generally be happy with a rough doubling of performance all around from what I have in my T100. Say around 100MB/sec sequential writes, 200MB/sec sequential reads and around 40-50MB/sec 4k random reads and writes.
    Reply

Log in

Don't have an account? Sign up now