What you think you know about flash memory may be wrong, according to a provocative panel discussion at the Flash Memory Summit Aug. 23, 2012. Panelists were asked to come up with the "top ten things you need to know about flash memory today," and while not all had ten items, all had some original and controversial insights into the status and future of flash memory and solid state drives (SSDs).
The panel was chaired by Andy Marken of Marken Communications, and it featured the following panelists in order of presentation:
- Andy Tomlin, vice president of Solid State Development, Western Digital
- Jered Floyd, CTO, Permabit
- Knut Grimsrud, fellow, Intel
- Kevin Kilbuck, marketing director, Micron Technology
- Jim Handy, chief analyst, Objective Analysis
A similar panel was held last year, and my report on that is here. For this year's panel, as of this writing, presentation slides for all panelists except Floyd are available here. Following are the major points made by the panelists.
Andy Tomlin - Firmware Takes Longer Than You Think
Everyone agrees that firmware is critical for developing an SSD product, but few know how long it takes, Tomlin said. "If you are starting out with a new firmware code base for a new product, it's a minimum of two years," he said. "If you think it takes less it will probably take longer because you'll make poor decisions along the way."
Tomlin also noted that 10K CPU cycles per IOP (input/output operations per second) is a reasonable budget, and multiple CPUs or extensive hardware automation represent the only ways to change the equation. Finally, he noted that it may take 6 months to get power cycling to work for an SSD. Suggestion: "Constantly explore new tools and methodologies that will help you get an advantage."
Jered Floyd - Flash is Actually Cheaper Than Disk
Floyd presented a "top ten" list in an interesting fashion, as shown below. He said the list was based on things he'd heard at the conference that were wrong. Floyd focused his discussion on the top five items. He didn't get to the tenth item, which is illegible in this photo, but it says that "flash will save the world."
He's serious, though, about his argument that flash is actually cheaper than disk. That's because enterprises often buy large disk drive arrays and utilize small portions of them just to get the IOPs they need. They could get those IOPs with a single flash device, he said. It's really all about the use case. "No matter how expensive the flash is per terabyte, it might still be saving you money." And flash can save even more with data optimization (item 2 on list), he noted.
Speaking to his fourth item, Floyd said that flash will continue to scale for a long time and noted that "people always say the end of the road is coming and it's always three years out."
Knut Grimsrud - SSDs Drive Interface Transition
Grimsrud said he would offer just two thoughts. One is that SSDs are driving a storage interface transition from Serial ATA to PCI Express. "Existing SSDs are hitting the limits of what Serial ATA can do, and it's clear that Serial ATA is not going further than where it is today," he said.
His second thought is that we are viewing SSD performance metrics in an "old fashioned way" and that "metrics of merit" other than average IOPs are needed. We also need to look at variability from the average IOP rate. Also, power efficiency is increasingly becoming a metric of merit.
Kevin Kilbuck - Flash Is Not Just About Price
Kilbuck lent additional support to Floyd's point about flash being cheaper than disk. He noted that NAND flash displaced other media, such as camera film, well before it actually lowered costs. That's because it offered other advantages, such as the ability to take a picture and see it instantly. "NAND flash doesn't have to be lower cost per bit to displace hard drives in many applications, and it's already happening," he said.
What about hybrid drives? "In a few years we'll look back and laugh at people still using hard drives," he said.
Jim Handy - NAND Prices Stagnate, but Flash Applications Grow
Jim Handy, who spoke at last year's "top ten" panel, explained how he updated his list for this year. The new list is as follows.
A few explanations:
- (1) All data centers will use flash but not necessarily SSDs - we're seeing flash appliances that don't use an SSD form factor.
- (2) NAND is in an oversupply situation and prices will not rebound until 2013. There will be another collapse in mid-2015.
- (3) As controllers get more sophisticated, TLC (triple-level cell) technology will come into the enterprise.
- (5) and (8) PC makers will increasingly use SSDs for cache alongside hard drives. This architecture will replace a lot of SSD-only PCs and will result in a "lower density mix" for flash.
- (9) The only people who will really notice when NAND flash goes away are the people who make NAND flash controllers, because "the difficult job of managing NAND flash will be easier with any of the new technologies."
Handy also said that flash will probably scale down to 10nm and that it won't be replaced with another technology for 6-8 years.
In all, it was an interesting and thought-provoking panel session and a fitting conclusion to the 2012 Flash Memory Summit. To view the conference schedule, click here. For a summary of Cadence activities at the conference, click here.
Related Blog Post
Keynote: New Memory Technologies Challenge NAND Flash and DRAM