"Sure the margins are low - but we'll make it up on volume!" At first glance it didn't make much sense to me why Intel announced flash drives last week. Going up against Samsung in flash, a company that has been reducing flash cost per bit at about 70% per year, hardly seems like a smart move. Intel's target market isn't disk drives There are several unusual aspects to Intel's announcement:
- The drives are puny - 4 GB max until year end
- USB interface - every thumb drive is a potential competitor
- Single-level flash - more costly
Let's look at each in turn.
Boy, that new drive is really, um-m, tiny You can buy 16 GB thumb drives for about $140 online. So why come to market with a tiny little 4 GB drive? First, cheap thumb drives are mongrel products. The flash and the controller chips are bought on the spot market so you never know what is in them. Plus they use cheaper multi-level cell (MLC) chips, while Intel uses single-level cells (SLC) in its drives. The reason for the small capacity: Intel's goal is to price its drives below where disk drives can be profitably sold, which is about $65 for a single-platter high-volume drive.
Contract prices for 4 GB of SLC chip today is about $32, add in $3 for everything else (controller, USB connector, PCB, packaging) and you can price at $58 for volume sales with a 40% gross margin. Add in the greater reliability, lower power usage and lighter weight of flash and that is a pretty compelling story if your device can live with 4 GB. There's no technical reason Intel couldn't sell an 8 GB drive today instead of December as planned. They just expect that 8 GB of SLC will cost about what 4 does today.
Compared to SATA that USB I/O really . . . creeps Flash drives are about access time, not bandwidth. So even the slow USB 2.0 and a not-too-impressive 100 IOPS blows the doors off competing 1.8" disks: competitive bandwidth and double the IOPS with much faster accesses. Plus USB is really cheap and universally available so OEM integration time should be about zero.
SLC flash - more costly and more reliable The flash vendors are very conservative on their read/write cycle specs. The rumor mill has it that SLC can handle close to 1 million R/Ws, almost 10x the 100k spec both Samsung and Intel publish. I'd been thinking that flash vendors would go to smaller sector sizes - 128k is the norm today - to give their wear-leveling algorithms more play room. But even without that a 4 GB SLC drive can handle 3.2 billion 128k sector writes in its entire life. Surprised?
Do the math: 4 GB has 32,000 128k sectors x 100,000 R/W cycles. That is over 1.7 million writes a day for five years or 20 IOPS every second, 24 hours per day. Factor in that most I/Os are reads and that is a fine life span. You can see why they use SLC - that spec sounds a lot more impressive once you run the numbers.
Will Intel win in the flash drive market? IMHO they have an uphill battle. They've done some smart things, like focusing on a price point disks can't touch and using USB instead of SATA or PATA. But Samsung is already in the game in huge volumes and it isn't used to the rich margins Intel's CPU business commands. When Intel's VP of NAND products requests capital, he'll have to show the executive committee that he'll get a return that is somewhere in the vicinity of Intel's core CPU business. That won't be easy.
Comments welcome, of course.