Brian's Brain

This blog post references my cover story 'Solid-State Storage: Feasible Plan Or Flash In The Pan?' in EDN's September 27, 2007 edition. It's one of a series of web addendums to the print writeup.

Although augmenting (or completely supplanting) a magnetic HDD with flash memory can provide a system with numerous benefits, one key issue validly keeps many potential implementers from considering it. That hinge factor is the so-called 'invisible hand' of the marketplace that constantly calibrates flash memory supply and demand, and translates this association into the price you end up paying at any particular point in time. The week before I filed the print article, for example, a power outage at Samsung's (the world's largest NAND flash memory manufacturer) semiconductor fabrication facilities sent shock waves of availability fears through the tech industry, and resulted in brief but substantial price up-ticks.

Your ability to secure sufficient flash memory supply at an acceptable price is, of course, dependent not only on your demand but also that of your peers in both similar and dissimilar industries. To that point, it would be imprudent to ignore the actions of Apple, a dominant consumer of NAND flash memory. Unfortunately, Apple's endemic and longstanding reticence regarding public disclosure of its upcoming product plans greatly complicates the task of forecasting its demand trends.

The main article mentioned Apple's September 2005 HDD-to-flash memory conversion marked by the launch of the iPod nano and the simultaneous obsolescence of the iPod mini. The recently introduced and flash memory-based iPhone also seems to be a marketplace success. And, with the even more recent introduction of the iPod touch, Apple's solid-state iPod line ranges as high as 16 Mbytes in capacity. As I recently mentioned, I suspect that the 80- and 160-GByte iPod classics will mark the end of the line for Apple's HDD-based portable multimedia players.

Look at the list of companies supplying (or planning to supply) hybrid HDDs and SSDs, and you'll encounter a mix of vendors both with and without 'captive' flash memory capacity. Just as Sandisk's direct access to NAND flash memory supply has aided its efforts in the portable multimedia market, as compared to competitors who must buy their flash memory on the open market (and deal with the resultant profit-sapping price mark-ups), this same access will likely also help the company deliver cost-effective SSDs. Samsung and Toshiba's mass storage divisions are similarly benefited by having in-house NAND supply (that is, as long as demand remains strong). It'll be interesting to see how both companies resolve the contending priorities of their NAND flash component, HDD (potentially moving to hybrid HDD) and SSD product groups going forward, and how the companies' systems divisions will influence the evolving strategy direction.

Your ability to source cost-effective flash memory is further complicated by the pinout, erase block architecture, command set, specification and other device-to-device incompatibilities, both within a given supplier's product line and across multiple suppliers. The IM Flash Technologies (Intel and Micron joint venture) -championed ONFI (Open NAND Flash Interface) project strives to resolve these mismatches, but pragmatically its membership roster (37 companies as of late January) currently includes only second-tier flash memory suppliers. Until ONFI snags Samsung or Toshiba, its influence will be muted.

At May's WinHEC, Dell, Intel and Microsoft unveiled another industry alliance. The NVMHCI (Non-Volatile Memory Host Controller Interface) Working Group plans, as its name implies, to "provide a standard software programming interface for nonvolatile memory subsystems. The interface would be used by operating system drivers to access NAND flash memory storage in applications such as hard drive caching and solid-state drives." As part of this project, working group representatives plan to de-emphasize or even eliminate legacy commands that apply to HDDs but are meaningless for non-rotating SSDs. Note, however, that this working group does not plan to focus on standardizing the flash memory components that reside behind the solid-state storage controller.

Continue reading with 'Solid-State Storage: Levels And Cycles'...