A typical SSD, whether consumer or enterprise, consists primarily of flash memory chips, an SSD or flash memory controller and, optionally, a small amount of RAM cache. Vendors can build SSDs for a specific market segment by selecting components to meet that customer's performance, cost, reliability or other product objectives. Many consumer-grade SSDs are nothing more than a marketer's assembly of the controller vendor's reference design.
Because the controller is really the heart of SSD, I tend to view controller innovation as key to innovation in the SSD market as a whole. Controller vendors have also been at the center of the many of the acquisitions and deals that have recently occurred in the SSD market. During the past year, several independent controller vendors have been snapped up by larger companies. These companies have taken substantially different approaches to how they will market the technology of their new subsidiaries.
LSI, for example, with a long history as a merchant silicon arms dealer, has continued to sell SandForce controllers on the open market to any SSD vendor interested in using those market-leading controllers. Apple, on the other hand, snapped up Israeli controller startup Anobit, and I don't expect to see Anobit's technology in SSDs on the open market. Instead, I expect to see Apple use Anobit's extended ECC technology to push lower-reliability MLC flash into both consumer products and Macintosh computers. SSD vendor OCZ has taken an especially broad-minded approach, acquiring controller vendor Indilinx but still producing new SSD models based on controllers from SandForce and Marvell.
The past few weeks have also brought an interesting set of announcements from new controller vendors. Longtime storage silicon vendor PMC-Sierra has entered the controller market with a 12-Gbps SAS SSD controller that combines digital signal processing and an extended ECC model that actually adds additional ECC as the error rate of the flash in the SSD increases. The real problem with flash write endurance isn't that an entire flash chip, or even an entire page, fails precipitously as it reaches its 5,000-cycle life, but that over time--or more specifically, over program array cycles--the error rate on the flash increases until it reaches the point where data can no longer be reliably stored. By detecting this increasing error rate and increasing the degree of ECC to accommodate it, PMC-Sierra's controller can extend the life of less-expensive MLC flash to make it competitive in the enterprise market.
Israeli startup DensBits is also specializing in making consumer-grade MLC and even TLC (three-level cell flash that stores three bits per cell with even lower endurance than MLC) suitable for use in enterprise drives. Since it's clear that flash endurance and reliability are going to decrease along with cell sizes, the kind of advanced error correction and data-signal processing that companies like PMC Sierra specialize in is key to maintaining a falling cost-per-gigabyte on SSDs without making them disposable media on the order of floppy disks that can only be written to a small number of times. In fact, Seagate this week announced that it has made an investment in Densbits and will be using denseness technology in a future series of solid-state disks.
It's an exciting time in the SSD business, and customers have to acknowledge that advance controller technology has moved us past the early days of SSDs during which one could say SLC good, MLC bad, TLC garbage.