Brian's Brain

One in a series of posts…

Several weeks ago, we found out Micron's initial product plans for the NAND flash memory silicon flowing out of its partnership fab network with Intel. Tuesday was Intel's turn to inform, and the company was in top form during a dedicated briefing to an overflowing crowd of journalists and analysts in a large conference room.

Like Micron, Intel plans both SLC (single-level-cell) and MLC (multi-level-cell) SSD variants, the latter in both 2.5" and 1.8" form factors. For now, like Micron's competitive offerings, they'll be manufactured on 50 nm lithography-based silicon, with a next-generation roadmap to 34 nm-derived products. Pricing was not announced.

X18-M and X25-M (MLC-based 'mainstream' SSDs)

• 80 GByte variant currently sampling, in production within 30 days
• 160 GByte variant sampling in Q4, in production Q1 2009
• Up to 250 MByte/sec read speeds
• Up to 70 MByte/sec write speeds
• 85 µsec read latency

X25-E (SLC-based 'extreme' SSDs)

• 32 GByte variant currently sampling, in production within 90 days
• 64 GByte variant sampling in Q4, in production Q1 2009
• Up to 250 MByte/sec read speeds
• Up to 170 MByte/sec write speeds (which, note, are lower than the sustained speeds Micron claimed for its products a few weeks back; Intel marketing manager Troy Winslow bristled and responded with confidence in his SSDs' competitive capabilities in real-life applications when I pointed this out during Tuesday's briefing)
• 35,000 IOPS (4 KByte read); 3,300 IOPS (4 KByte write), both measured using Iometer
• 75 µsec read latency

Intel's been working on flash file system wear leveling and performance optimization technologies for several decades; yours truly, for example, was in his past professional life a technical advisor to the team that developed (but ultimately decided not to productize) a 32 MByte 2.5" PATA SSD in the early 1990s based on dual-die 32 Mbit NOR flash memory technology. This accumulation of expertise is evident in these SSDs' specifications as well as their underlying designs.

Parallel interface channels from the memory controller go out to each of the up-to-10 flash memory components (each containing up to 4 NAND flash memory die per package) resident within a SSD enclosure. Each channel can therefore issue simultaneous read, write and/or erase operations to multiple die in parallel, balancing incremental performance against resultant incremental power consumption. And, at the system interface side of the equation, befitting Intel's championing of the SATA interface (which also comes in an external variant, of course), the drives support native command queuing and dynamic reprioritization (which arguably delivers less incremental benefit as compared to HDDs, by virtue of SSDs' lack of rotating storage media).

All sounds good, right? What's with the 'uncertainties' in this post's title, then? Part of my pause derives from price…specifically price-per-GByte vs the HDD alternative. As I mentioned the other day, I'm pleasantly surprised at how much application code I'm able to squeeze into a 20 GByte virtual HDD. But I've explicitly avoided storing a lot of bit-intensive data…specifically photos, music tracks and video clips…to that particular partition, something that the average consumer won't know (and won't want) to do. And the longstanding 'bigger numbers are better' selection criteria that drive most consumers' purchase decisions can't be ignored.

As such, although I predict a bright future for SSDs both in enterprise apps (where performance and power consumption are key parameters) and small-form-factor systems (where ruggedness is paramount, and capacity isn't so critical), it'll be a stretch for SSDs to broadly succeed in mainstream desktop and laptop PCs, at least in the near term. As such, my Intel contacts privately tell me, I shouldn't expect to see any corporate defocus on the Turbo Memory concept which Intel's partners have so far largely declined to embrace. A small flash memory-based cache in-between (or depending on your perspective, alongside) the HDD delivers a reasonable facsimile of the full SSD benefit suite, at much lower incremental platform cost increment compared to the full HDD-to-SSD conversion.

The other primary counterbalance to my SSD enthusiasm involved profitability. In Intel's recent earnings briefing, CEO Paul Otellini was adamant that he wouldn't let the NAND division act as a drag on overall corporate fiscal success. Cynically speaking, given the longstanding cyclical nature of any memory business and especially the consumer electronics-dominated NAND flash memory industry, I'd call Intel's management bafflingly naïve if they didn't realize from the beginning what they were getting into. Nonetheless, the specter of a possible spin-off or sale of the group looms over its accomplishments this week.

With that said, though, I'm fairly confident that Intel will hang tight with its NAND flash memory investment, because the technology is (perhaps for the first time…I don't particularly count the past flash BIOS embrace) strongly aligned with the company's microprocessor-dominant focus. Intel intimately realizes that any system imbalance that doesn't enable its CPUs to fully shine from a performance standpoint will be to its long-term detriment. A focus on system balance, for example, explains the company's past advocacy of Rambus' Direct RDRAM, along with its current tendency to embrace latest-and-greatest DDR SDRAM technologies.

Grasping for system balance nirvana explains the pursuit of ever faster internal and external interconnect topologies such as incremental USB, PCI Express and SATA generations. And it also explains why SSDs were everywhere from demo, keynote and technical session standpoints at IDF…including in all of the Nehalem reference systems I perused. Whether the end form is a performance-optimized SSD or cost-modulated Turbo Memory cache in any particular design, NAND still wins. And, although the word 'bundling' is anathema from a FTC reprimand standpoint, the ability for an Intel sales representative to successfully promote a 'bag o'chips' package containing not only the CPU and core logic but also flash memory cannot be underestimated…to the detriment of competitive NAND suppliers.

On that note, one comment made by Winslow on Tuesday particularly sticks in my mind. "We will strongly align with Intel Architecture designs", he stated; "we will not pursue the USB flash stick business."