Brian's Brain

Back in the mid '90s while I was still working at Intel, due both to my existing role as a flash memory architect and technical marketing engineer and my prior work as an ASIC designer for the company's Series 2 Memory Cards, I was peripherally involved in a skunk works project to develop a flash memory-based HDD. Intel never formally introduced the product, because it was unable to identify a sufficiently large market to justify the investment required to take the concept to production. And the few-dozen-MByte density (my memory's faded on the exact specs) that was available in the lab prototype pales in comparison to the 100+ GByte capacity delivered by modern 2.5" HDDs based on rotating magnetic storage.

Capacity and cost-per-MByte issues aside, the prototype worked, and worked well. And the solid-state approach does have some merit, specifically in the areas of ruggedness, power consumption and sustained read access performance, which is why companies like M-Systems and SimpleTech have picked up the torch. Microsoft, in partnership with Samsung, last week at WinHEC unveiled further details on an intriguing hybridization of flash memory and rotating storage that attempts to deliver much of the benefit of a 100% solid-state approach while minimizing the full blown implementation's shortcomings. Microsoft first unveiled its hybrid drive concept at last year's WinHEC, and I admit I was pretty skeptical. I'm still skeptical, albeit less so, and my cynicism center on business issues and has little to do with technical feasibility.

As the block diagram shows, Microsoft's approach adds a 64- or 128-MByte flash memory to an otherwise conventional HDD design, although modification of the drive controller will be necessary to support bidirectional bulk data direct transfers between the drive platter(s) and the flash media. Microsoft settled on the recommended flash memory density after analyzing typical O/S and application file size and access frequency patterns under Windows XP, along with forecasts for its upcoming Longhorn O/S. By employing Samsung's OneNAND flash, which offers a wide 16-bit data bus and limited direct-execute capability, it's possible to implement both file cache and already-existing firmware storage functions within the same chip, addressing incremental cost and board space concerns.

Microsoft's exercise has several objectives; speed system boot (and recovery from hibernation), minimize power consumption by maximizing the percentage of time that the drive platters are spun down (balanced against the incremental energy required to spin up a platter versus that required by a platter that's already rotating...that good old static-versus-dynamic coefficient of friction factor that we all vaguely remember from high school physics class) and increase HDD subsystem ruggedness and operating life (again, by maximizing the percentage of total time that the platters are spun down and the drive heads are parked). The company believes that it'll be able to drop average power consumption of the HDD subsystem by 80% and, best-case, increase system operating time while on battery power by 10%. For the nitty-gritty details, see this PowerPoint presentation and this paper in Adobe Acrobat format.

I'm not terribly concerned about the slow write times of flash memory, since system- and drive-based RAM buffering already exists and, if necessary, the drive can spin up its platter(s) and transition the writes to it/them once it's/they're ready. Nor am I concerned about the limited cycling capability of flash memory; intelligent wear leveling algorithms can maximize its useful life, and as it peters out the reads and writes can migrate over to a more traditional direct-to/from-platter scheme. So, where's the skepticism I mentioned earlier coming from?

The hybrid hard drive sounds like a great deal for Samsung's HDD group (who hopes to have products in production by the end of next year, coincident with Longhorn's forecasted release timeframe), and for Toshiba's as well, especially if the company comes out with a OneNAND-like part. But what about all the other HDD manufacturers, who don't happen to have their own captive flash memory supply? Drive profits are razor-slim in today's highly competitive and capacity-rich market, if they even exist at all, and incremental bill of materials costs further bloated by NAND flash suppliers' markups won't be enthusiastically embraced. Samsung's flash memory strategic marketing manager Ivan Greenberg swore to me that his external customers get the exact same pricing as his sibling divisions do....but I'm not 100% confident I believe him (and even if I do, I doubt Fujitsu, Hitachi, Maxtor, Seagate, Western Digital and the rest of the HDD gang will).

Here's what the gang at Slashdot think. What are your thoughts?