SSDs: Flash Memory Advancements And HDD Counterattacks

Things have, as always, remained interesting on the solid-state versus hard disk drive front since I last covered the topic in detail a month-and-a-half ago. Back then, I said:

In-between the "where SSDs clearly win" and "where HDDs clearly win" ends of the spectrum, there’s a large "either/or" grey zone. Here’s where incremental flash memory supply will go, on an impermanent basis based on whether we’re in a constrained-supply or oversupply state at a given point in time.

Thanks to recent reported order scale-backs by the single largest flash memory consumer, Apple, suppliers find themselves once again with more product sitting in warehouses and in-line at fabrication facilities than there are identified homes for it. Lithography shrinks, which increase the amount of bits a given-sized fab can churn out in a given amount of time, don’t help the oversupply situation, either…witness, for example, the Intel/Micron partnership’s quite impressive recent litho jump down to 34 nm.

Intel’s SSD announcements at Computex were fairly modest…4 and 8 GByte drives (with a 16 GByte scheduled for availability by year-end) using archaic PATA interfaces, but sized perfectly for Atom CPU-targeted systems. However, the 34 nm lithography’s first product, a 32 Gbit MLC (i.e. 16 Mcells) NAND flash memory device, hints at more ambitious SSD aspirations. And indeed, recent comments by Intel’s Pete Hazen (who I used to work side-by-side with) suggest that the company’s specifically targeting high-capacity read-mostly enterprise applications where SSD’s performance strengths will be most evident.

Intel’s not the only one aiming for the enterprise, though. Sun’s girding to release SSDs, too, and a recent blog entry from CEO Jonathan Schwartz intimately ties the hardware aspirations to the company’s failure-resistant ZFS file system…support for which, as it happens, will also appear in the Server variant of the upcoming Apple OS 10.6. Can you connect the dots?

For mainstream applications in which HDD capacities and floor costs also enable them to play, suppliers continue pushing down the price-per-GByte at a mind-boggling pace. Take 1 TByte HDDs, for example. Back in early November of last year, I expressed delight that prices had fallen as low as $260. Now they’re $110 lower, fueled by factors such as PMR. That’s even cheaper (on a $/GByte basis) than the $80 Seagate 500 GByte HDD (16 cents per GByte) I bought from TigerDirect the other day as a replacement for a DOA unit. And, at 15 cents/GByte, they’ve dipped well below the magical 20 cent/GByte threshold where demand tends to explode.

Flash suppliers are cranking out their own cost reductions in response; earlier this month, for example, Hynix announced its intention to roll out triple-level-cell technology (I think they mean triple-bit-cell technology, i.e. eight-level-per-cell flash memory, since today’s MLC two-bit-per-cell technology already has four levels per cell) later this year. I’m frankly still skeptical that yield, performance and reliability will be up to snuff.

As far back as 1992, I’ve been predicting (using the 8" to 5.25" to 3.5" floppy disk as one historical metric) that capacity capabilities of magnetic media would outstrip the capacity demands of mainstream applications, and that as a result the form factor of the average shipped HDD would steadily decrease. Take Toshiba’s just-unveiled 160 GByte 1.8" HDD, for example. Or for an even more compelling case study, take a look at Western Digital’s VelociRaptor 10,000 RPM HDDs.

I’ve admiringly written about prior-generation Raptors before…specifically their clever positioning in-between PATA/SATA and SCSI/SAS drives, with price tags comparable to the former but performance akin to the latter. These latest-generation VelociRaptor drives also come in 3.5" form factors…but the drives themselves are actually 2.5" models, with passive cooling surrounding them to make up the volume difference. And, because the platters are narrower in radius than with past ‘true’ 3.5" Raptors, seek times are diminished, leading to higher speeds in random access-rich applications. Don’t just take my word for it; check out ExtremeTech’s review. And what about future 20,000 RPM HDDs? We’ll just have to wait and see…

For the moment, I’ve got one VelociRaptor in hand, and am yearning for a second so that I can configure the pair in RAID 0 striped mode and see what they can really do. Cool? Yeah, I think so, too.