LSI SandForce controllers were the catalyst that helped SSDs evolve from a niche application to the mainstream market we see today. There were precious few SSD's on the market, and even fewer SSD manufacturers, when the first SF family of processors were released. SandForce took their revolutionary new design and compression capabilities to market with a business model that allowed manufacturers to field a new SSD easily.
OCZ was one of the first to offer SandForce SSD controllers in their products, and soon many companies, such as Corsair, Kingston, and others, followed suit. The key for manufacturers was the solid design of the controllers and firmware provided by SandForce. The reference firmware removed the requirement for teams of expensive firmware engineers. Soon SSD manufacturers began popping up left and right, anyone with access to flash and manufacturing capabilities could easily bring an SSD to market.
Two generations of SandForce processors made their way into everything from consumer SSDs to enterprise and PCIe SSDs, and then LSI purchased SandForce in October of 2011. The acquisition made sense for both sides for several reasons. LSI brought decades of experience in storage processor design and legendary product validation. SandForce provides unique IP with their DuraWrite technology and a series of products that fit well with LSI's existing product stack.
The LSI SandForce team has been working feverishly on their newest controller addition, the SF3700, since the acquisition. With many of the same features, but a more refined architecture and higher performance, the release of the SF3700 is probably one of the most anticipated SSD controller releases in history.
One of the challenges LSI SandForce faced when designing the new SF3700 controller was the need to keep performance and reliability high in the face of a continuing increase in NAND error correction requirements. As NAND lithography shrinks, the error correction requirements rise exponentially. We can see that the error rate is accelerating rapidly from the chart in the lower right of the graphic above.
While the error rate rises, endurance is simultaneously shrinking. Some next-gen NAND is projected to only offer 1,000 P/E cycles. Managing less endurance and higher error rates requires intelligent controllers and error correction techniques. Unfortunately, merely expanding typical ECC requires more computational grunt power, and the additional overhead leads to lower performance.
Another key requirement is to build in forward compatibility with technologies that may not exist yet. The current LSI SandForce family was one of the first controllers compatible with 19nm Toshiba NAND; quite the feat considering its release almost five years prior to the rollout of 19nm. Mixing in the need for forward compatibility with different types of NAND, such as TLC, also raises the complexity level. Having a flexible platform is imperative with 3D NAND also on the horizon.
The LSI SandForce answer to these problems starts with a 14-core behemoth named the SF3700. The Samsung MDX controller was the first three-core SSD controller in the consumer market, and many have credited its dominance to the extra cores. LSI's new controller sports more than four times the number of cores and nine flash channels. We can safely assume each of the nine channels has been assigned its own core. There are four other functions on the back end, along with dual AES-256. It is likely that many of these functions, such as garbage collection, could be assigned their own cores as well.
The controller natively supports SATA 3 and PCIe Gen2 to cover the bases with the new M.2 specification. There is also support for a slew of NAND types built into the backend.
Power management is a huge concern for today's mobile devices. The numerous cores provide a modular platform with tighter power control granularity. All 14 cores can be selectively suspended, or shut down entirely, to conserve power. The new architecture provides a flexible base to build newer controllers with a faster time-to-market in the future.