Silicon Motion Displays Upcoming SM2260 PCIe 3.0 x4 NVMe Controller

Silicon Motion has remained relatively quiet about its plans for a PCIe controller. We have known that the company is working on one (well, it would be a major surprise if it wasn’t), but until last week Silicon Motion had not shared any details. The controller is called SM2260 and it’s a PCIe 3.0 x4 design with support for NVMe. Both MLC and TLC NAND are supported at lithographies as small as 15nm, which suggests the use of LDPC error correction similar to the company’s SM2256 controller. Preliminary performance specs come in at up to 2.2GB/s sequential read and 200K random read IOPS, but I wouldn’t be surprised to see these change before the final product ships. 

Silicon Motion is still fairly early in development and the SM2260 won’t be available until sometime next year. For now the company is keeping the additional details close to its chest, but we’ll hear more once Silicon Motion has something concrete to show.

In addition to the SM2260, Silicon Motion showcased its new SM2256 controller that will be shipping in the next couple of months. It’s the first commercially available controller & firmware solution with TLC support and as a part of Computex Silicon Motion announced that it now supports Micron’s 16nm TLC NAND as well, along with all other available TLC NAND dies. ADATA already displayed its SP550 with the SM2256 inside and it will be among the first drives to ship with the new SMI controller. We have some preliminary benchmarks of a 500GB SM2256 reference design sample in our Bench section, so head there if you’re interested in taking an early look.

Another product Silicon Motion had on display was the SM2246XT, which is a DRAM-less version of the popular SM2246EN. There is a slight performance impact from the lack of DRAM cache as only parts of the NAND mapping table can be stored in the controller’s internal SRAM cache, but in return the SM2246XT offers lower cost due to a smaller die (no need for DRAM controller). Obviously the drive requires no separate DRAM either, which further reduces the cost for the assembler of the drive. SM2246XT is already used by SanDisk in the SSD PLUS and Z400s that was released a couple of weeks ago. I believe we will see the low-end market shift more towards DRAM-less designs in the next year as the race to the bottom gets tougher and with more sophisticated controllers it’s possible to have decent performance even without the DRAM cache. Currently Silicon Motion has no plans for an XT version of the SM2256, although I would personally see TLC NAND and DRAM-less controller as the ultimate value play.

ADATA Displays XPG SX930 SSD, Two New TLC SSDs, Several PCIe Drives & USB Type-C Products

After a long week of meetings and over a day worth of traveling, I’m finally back from Taipei. Given our hectic schedule (it was just me and Ian at Computex this year), we didn’t have time to report most of the things we saw last week, so it’s now time to start going through all the new products we saw. First up is ADATA who had numerous new, interesting SSD products to show, including TLC, PCIe and USB 3.1 solutions. 

The first product ADATA had on display was the XPG SX930, which is ADATA’s new high-end SATA SSD based on JMicron’s latest JMF670H controller. If you remember our ADATA booth tour from last year’s Computex, the SX930 was originally supposed to utilize SandForce’s SF-3000 series controller in SATA 6Gbps mode, but as the SF-3000 keeps getting delayed (I heard it’s scheduled for Q4’15 now) ADATA had to look for an alternative controller. JMicron actually sent me JMF670H reference design samples right before I left for Computex and based on my early benchmarking it seems to be a very potent controller. It’s not a Samsung challenger, but it appears to offer a healthy upgrade over the JMF667H and be competitive against other controller on the market. We will have a review of the reference designs up in the next few weeks, so stay tuned for deeper performance analysis. 

In terms of NAND ADATA is using something the company refers to as MLC+. Basically it’s just higher binned MLC NAND because ADATA does NAND binning in-house (which is in fact quite rare for a module house) and it enables the five-year warranty ADATA is offering for the SX930. I can’t remember on top of my head whether ADATA is using Micron’s 16nm or 20nm NAND, but I’ll provide an update as soon as I know more. The SX930 will be available in the next couple of months in capacities of 120GB, 240GB and 480GB.

Aside from the technical aspects, ADATA is trying to revamp its XPG branding to be more appealing to gamers. The SX930 features a new flame logo that extends to the package design and the whole XPG branding now has a new ‘Born to Battle’ slogan. While I think most of our readers buy a drive based on its performance aspects rather than the design, I still think the new branding is a good move because gamers and other PC enthusiasts certainly care about the looks when building a desktop and ADATA’s bird logo perhaps isn’t the number one choice. 

Moving on to TLC SSDs, ADATA didn’t have just one, but two TLC SSDs on display. The first one is SP550, which uses Silicon Motion’s SM2256 controller. The SM2256 is the successor to Silicon Motion’s popular SM2246EN controller and it’s the first commercially available controller platform with TLC NAND support. As a matter of fact Silicon Motion sent me reference design samples a while back and you can find some preliminary benchmarks in our Bench section, but I wanted to wait for shipping drives before doing a full review and drawing the final verdict. 

ADATA’s other TLC SSD is called SP560 and utilizes Marvell’s 88SS1074 controller that is codenamed “Dean”. Similar to the SM2256, the “Dean” also features LDPC error correction, which increases the endurance of TLC NAND by being able to fix a higher number of bit errors, although there is a slight negative impact on performance especially when the controller has to rely on software based LDPC for error correction.

ADATA hasn’t fully finalized the product specifications yet, but the company will be using 16nm TLC NAND from both Micron and SK Hynix. Micron just announced its 16nm TLC NAND last week and is already shipping to its partners, whereas SK Hynix hasn’t made much noise about its 16nm NAND that is now available. The SP550 and SP560 will both be available in capacities of 240GB, 480GB and 960GB with shipping starting in the next few months. Pricing is TBA, but should be below ADATA’s MLC drives, although I was told not to expect dramatic price cuts. 

In addition to SATA 6Gbps drives, ADATA obviously had some upcoming PCIe SSDs to show. To be more specific, ADATA had two drives, of which both were based on Marvell silicon. The SR1020 will be aimed towards the enterprise and features Marvell’s “Eldora” PCIe 3.0 x4 controller. Eldora’s aim is to compete against the SF-3000 platform as it supports both AHCI and NVMe as well as LDPC error correction, similar to the SF-3000, but it’s biggest advantage is the newer PCIe 3.0 interface with twice the available bandwidth. Again, the readers who read our ADATA booth tour a year ago might remember that the SR1020 was supposed to be based on SandForce SF-3739 controller, so it’s starting to look like ADATA has switched solely to Marvell for high-end because of the constant delays of the SF-3000 platform.

Live demo showing SR1020 at 3040MB/s sequential read

The SR1020 will come in 2.5″ form factor and use the SFF-8639 connector, which was actually officially renamed to U.2 by the Small Form Factor Working Group (SFFWG) last week. Given that SFF-8639 is gaining a lot of popularity due to it being the interface for 2.5″ PCIe SSDs, it makes perfect sense to get rid of the number monster and move to U.2, which is more marketable and goes well with the existing M.2 spec. 

As a more consumer-oriented drive, ADATA is offering a PCIe 3.0 x1 NVMe M.2 drive with Marvell’s “Artemis 2” controller. The drive is so new that ADATA doesn’t have a proper marketing name for it yet, but it’s a DRAM-less design for improved cost efficiency. ADATA will specifically target the drive for mainstream users as a SATA replacement. In a live demo the drive reached 641.7MB/s sequential read, so it should still provide a decent upgrade over SATA 6Gbps and most importantly it will be price competitive. Exact availability is still a bit open, but ADATA will be launching full-scale PCIe SSD production in the second half of this year, so we will likely see the PCIe drives enter the market either in the fourth quarter or early next year.

For industrial applications ADATA demonstrated its nano coating technology by operating one of its industrial SSDs inside a block of ice (I even confirmed that it is real ice, not plastic). 

ADATA also had a PCIe 3.0 x1 BGA SSD with a Marvell controller on display. It’s obviously an industrial/OEM SSD, but it’s pretty interesting that so much performance can be fit inside such a small package. I certainly hope high-end tablets will start adopting PCIe designs in the near future given that storage performance is still relatively poor in most devices, which is something that PCIe would fix. 

Last but not least, ADATA had a whole lineup of USB Type-C products. The SE730 is ADATA’s external SSD for USB 3.1 interface and Ganesh already has a sample, but we’ve been having some compatibility issues that have delayed the review. 

ADATA also had a bunch of USB Type-C adapters, which I’m sure the MacBook owners will appreciate given Apple’s pricing, although ADATA didn’t reveal the prices yet. 

All in all, ADATA had a lot to show and I’m eagerly looking forward to getting my hands on some TLC and PCIe SSD samples. I also had a chance to sit down with Kevin Chen, ADATA’s Vice President of SSD Marketing, so stay tuned for a more in-depth article of ADATA’s SSD strategy in the coming weeks!

OCZ Announces Trion 100 TLC SSD & Shows Off an Upcoming NVMe Drive with Toshiba Controller

One of the things I’ve been expecting to see at this year’s Computex is more TLC NAND based SSDs. So far only Samsung and SanDisk have been shipping TLC SSDs in high volume, but OCZ will shortly join the club with its Trion 100 SSD series that was showcased on the Computex showfloor. The Trion 100 is based on a totally in-house designed Toshiba “Alishan” controller, whereas some earlier Toshiba controllers have been modified/renamed third party silicons with the company’s firmware. We discussed the cooperation between the two companies in more detail in our interview with OCZ’s CEO, Ralph Schmitt, but to put it shortly the use of Toshiba IP in an OCZ product is part of the companies’ strategy to consolidate product development to maximize resources as it wouldn’t be efficient to have both companies working on platforms that ultimately have the same end goal. 

The NAND in Trion 100 is Toshiba’s A19nm (i.e. second generation 19nm) TLC, although OCZ is planning on switching to the latest 15nm node later this year once the supply is high enough. For error correction the drive uses Toshiba’s own QSBC (Quadruple Swing-By Correction) technology, which from what I was told is basically in between traditional BCH and LDPC correction. I don’t have too many details of the technology at this point, but I will try to gather more information to better know how the technology stacks up against the competition. Typical to TLC drives, the Trion 100 employs an SLC cache to increase peak performance in smaller capacities and also to boost endurance. 

One of the biggest improvements in the Trion 100 is the support for DevSleep and other power saving modes. Because of the silicon design the Barefoot 3 isn’t capable of offering low idle power consumption, which has been one of main criticisms for a while now. With the Trion 100 OCZ is finally looking to provide a solution that’s also suitable for mobile applications where power draw and battery life are major concerns.

The Trion 100 is now in final validation and the official release will take place later in the summer. Hence performance, endurance and pricing are not final yet, but what I was told is that the performance will be read focused (as most light client workloads are read-centric), endurance goal is 20GB/day and price will be slightly below the ARC 100.

Aside from the Trion 100, OCZ also showed off a new unannounced M.2 NVMe drive that will be released later this year (I’m hearing September-October timeframe) as a new RevoDrive. Unlike the previous designs we have seen, this isn’t based on OCZ’s own JetExpress controller, but on Toshiba’s PCIe 3.0 x4 NVMe controller. The reason OCZ is using Toshiba’s controller is cost because the the JetExpress has been designed mainly with enterprise in mind, meaning that it cannot be implemented into a cost effective client drive. 

Gallery: OCZ Computex 2015

Obviously OCZ also had it’s new Z-Drive 6000 series on display along with the upcoming JetExpress based XD-JX series.