Controllers are the heart and soul of every SSD. Without one, an SSD would be a useless PCB with some components slapped on it. It is responsible for everything from garbage collection and wear leveling to error correction and hardware encryption. In simple terms, all these operations can be quite complicated to implement as well as expensive to develop. This is why many companies who want to get into the SSD game turn to controller manufacturers for solutions. They can offer companies complete reference designs and firmware to get products to market quicker than they could on their own, especially if they don’t have the funds to support in-house solutions.
Over the past few years we’ve mainly seen third-party manufacturers turn to SandForce, however due to long delays for the release of SandForce’s latest controller, they haven’t been a viable option in the ever more competitive SSD market. SSD manufacturers looking to develop new products started to take a look at other SSD controller manufacturers such as Phison and Silicon Motion. As a matter of fact, Silicon Motion has been quite popular over the past year. Their very competitive SM2246EN controller has put Silicon Motion on the map as a force to be reckoned with, not only in terms of price, but performance and reliability as well. Quite a few SSD manufacturers now have an SSD with an SM2246EN controller at its core. Only recently, even Crucial (long time Marvell fan) has chosen Silicon Motion to help create their BX100, an even more value oriented SSD than their MX100 . As well, Mushkin released their 1TB Reactor at a rock bottom price and ADATA, Corsair, and PNY have also manufactured some good performing and low-cost SSDs with it.
As can be seen in today’s market, there is still a strong drive for cheaper and high-capacity SSDs. As a result, we will see increased manufacturing of TLC NAND which is predicted to quickly overtake MLC. With increased TLC memory production, increased demand for controller production will also occur. Presently, Samsung, SK Hynix, SanDisk, Toshiba, and Micron all produce TLC NAND.
Today we have Silicon Motion’s new SM2256 controller on our test bed and this sample is paired up with 256GB of Samsung’s planar 19nm TLC NAND flash. Yes that’s right, Samsung NAND in a non-Samsung SSD! The SM2256 is the one of the first merchant SATA 6Gb/s controller and firmware solutions to support TLC NAND and not only Samsung NAND, but 1x/1y/1z nm TLC from every manufacturer out and extensible to 3D NAND from all major suppliers. This is great news for third-party SSD developers as they can not only create a fast and reliable SSD, but also source out the best deal when it comes to NAND to help drive costs down for the end consumer.
SILICON MOTION SM2256 SPECIFICATIONS
Support for the latest and greatest TLC, MLC and extensible support to 3D NAND is great, but what does the performance look like at this time and what features does this controller support?
PERFORMANCE
For starters, performance is modest as it is a 4 channel controller design. When utilizing Toshiba A19 NAND, sequential reads are rated for 524MB/s and writes for 400MB/s. Random read IOPS are good for up to 90K and write IOPS for 70K.
FEATURES
Silicon Motion also has a proprietary NANDXtend ECC technology that provides comprehensive data protection and enhances the endurance and retention of TLC NAND, delivering more than three times better durability for TLC SSD. It does this through the use of low-density parity checking, RAID data recovery correction technologies, and a three-level error correction algorithm.
The SM2256 controller supports the ATA8 security feature set, data security erase and quick erase, software/hardware write protect, and early weak block retirement. There is an internal data shaping technique to increases data endurance and StaticDataRefresh technology ensures data integrity. As always with NAND flash devices, there is a global wear leveling algorithm that evens the program/erase count and extends SSD lifespan. It also features real-time full drive encryption with AES and is compliant with the TCG Opal protocol, and hardware SHA 256 and TRNG.
ARCHITECTURE
The SM2256 comes in a 323 TFBGA package. The architecture is very similar to the earlier SM2246EN. It is a 32-bit RISC CPU with a 64-bit system bus. It includes automatic sleep and wake-up mechanisms, built-in voltage detectors for power failure protection, built-in power-on reset and voltage regulators, and a built-in temperature sensor. For on-system debugging it also supports JTAG, UART (RS-232), and I2C interfaces.
NAND AND DRAM SUPPORT
THE SM2256 supports ONFI 3.0, Toggle 2.0, and Asynchronous interfaces. In terms of power, it supports 1.8V and 3.3V flash I/O. There is also support for 8KB and 16KB page sizes as well as 1-plane, 2-plane, and 4-plane operation. The 4-channel design also supports up to 32 NAND flash devices. It has a 16-bit wide DRAM interface and supports DDR3 and DDR3L.
HOST INTERFACE
The host interface is industrial standard SATA revision 3.1 compliant as well as ATA/ATAPI-8, and ACS-3 command complaint and the SM2256 also supports 28-bit and 48-bit LBA mode commands. It has a SATA 6Gb/s interface that supports NCQ for up to 32 commands and supports DevSleep along with TRIM, S.M.A.R.T., and PHY Sleep mode (CFast PHYSLP).
> You can see that the write performance drops off soon after the 12GB mark and fluctuates between 25MB/s and 125MB/s.
Wow, thats really bad for a 256GB drive. I mean, even the first generation 840 had write speeds around 250MB/s.
Either Samsung is pulling some wizardry with their controller/firmware (on sustained write, no slc) or this is just a very poorly binned TLC.
Have you notice the samsung sticker is warped and cooking off? I’m not aware of any Samsung chips that uses sticker as its label, actually, I don’t know ONE proper semiconductor company that does that.
The wording underneath the sticker says FORSEE, and that’s the trademark from another company:
https://www.foresee.cc/
So, what the heck are we testing anyways?
I’m gussing foresee is packaging the nand (but bare dies are still from samsung) and building/assambling the board.
If that’s the case, samsung will NOT allow them to mark it as Samsung, because it is NOT packaged and tested by them. Most likely they are unknown flash chip that were rejected as second or third grade stuff, sold as unmarked chip or bare die.
Most of the time those goes into no-name SSD or USB flash drive. Do you see a lot of the cheaper SSD have chips that is almost completely blank? Bingo!
I know because RAM is exactly like that, and I used to help a friend with his RAM wholesale business.
Good job, reading between the lines! Did anyone else notice the FORSEE?
With the rise of TLC nand will come more unreliable SSDs that will expire faster. MLCs and even SLCs could have been made more affordable but the major manufacturers fking blew it. Goodbye SSD. Welcome back HDD.