Jerome,
The following article can be found at www.wwpi.com (Storage Inc.)
M-Systems Targets Flash At The Low End Of The HDD Market
ÿ
by Hal Glatzer
ÿ
"I'm talking about embedded storage systems," he explained, "like test equipment, point-of-sale terminals, informational kiosks, ticketing machines, and so on. They need maybe only four- to eight megabytes, ten megabytes maximum, onboard. Even a PBX with outgoing messages for voice mail doesn't need a gigabyte: 200 high-quality voice messages take up only about 20MB.
No one argues that solid-state storage is cheaper per megabyte than hard-disk storage. But Dov Moran insists that it's more economical at low capacity-points. He's president of M-Systems, an Israeli company which buys those "flash" memories and packages them in hard-disk emulators that he targets at the low end of the storage market:
"The OEM price of a disk emulator is about $4/MB; we can sell a 10MB device for $40. But have you tried to buy a hard disk today with only 10MB capacity? Everything is a gigabyte or more. If you can find one-say you found a 100MB drive-it'd be a few years old already, and it's still going to cost you $100-150, no matter what the capacity. Compared to a disk-emulator, it will also take up more room, it'll require ventilation, a larger power supply, and so on. With all that baggage, an HDD can't beat our price. I'll go so far as to say the break-even point between an HDD and a flash memory today is about 40MB; and predict that a 40MB flash device will come down to $150 [OEM] soon."
M-Systems was founded nine years ago, had its first public stock offering in 1993, when it reported revenues of $1.3 million, and obtained secondary financing with a smaller offering in 1996. For 1997 it reported $19 million in revenues, "and we went to profitability in the last quarter [97Q4], with a gross margin of almost 40 percent," Moran said.
The company has 75 employees at its Tel Aviv headquarters, and 23 overseas, including 15 in the U.S.; he recently hired Chuck Schouw away from Sony, to preside over U.S. operations.
But Not PDAs Or Cameras
Flash chips are either electronically pro-grammable ROMs (EPROMS) using a technology called NOR; or electronic-ally erasable and programmable ROMS (EEPROMS), called NAND. M-Systems employs both, although "we are mostly going with NAND," he told Storage Inc. "NAND has excellent performance characteristics, capacity and price-per-megabyte that can't be achieved by NOR. Although it's slightly slower at reading than NOR," he conceded, "the write speed of NAND is tremendous: it has a ten-to-one advantage over NOR. And that's important, since we are using them for disk-emulation, not simply as a ROM.
"Our approach," he explained, "is to buy standard flash components, and add our fast `TrueFFS' technology-an algorithm in firmware that does the actual disk-emula-tion. That creates a product that looks to the computer system like a hard disk."
M-Systems buys those components-mainly 8MB NAND chipsets-from Samsung and Toshiba, and he has also started to get Intel's new, denser "multi-level" NAND products.
But two established markets for solid-state storage that Moran is not targeting are mobile computers-handheld PCs and personal digital assistants-and digital photography. "We could compete against SanDisk and Intel and Toshiba, who currently serve those markets," he said, "but we prefer to compete in other areas."
That may be because, on one hand he's their customer; on the other, their competitor. But since he's not in the PDA or digital camera markets, he's apparently unafraid to predict the for-tunes of their offerings there. "Intel's Miniature Card will die without more design-wins," he said. "You can't survive with less than ten percent of the market, which is what they have. You need 50-60 percent. Toshiba's SSFDC is limited in storage capacity, and though it's popular in Japan it's not seen much in U.S. design-wins. I have no doubt that SanDisk will be the survivor."
Data Storage, Not ROM Replacement
The market for EPROMs and EEPROMs is large. Moran estimates it's at least $3.5 billion today: "That's for all the current apps: programming, executable code, and so on. But the data storage market as a whole is about $40 billion. And that's where our goal lies: to be a major player in the data storage market by the year 2000.
"Our newest product is also our smallest: we call it `DiskOnChip 2000'. It has a capacity ranging from 2MB-72MB right now, and we expect to release a 144MB version later this year. It looks like any other component you'd see on a printed-circuit board, and you'll find a socket for it in some of the CPU boards that are coming out of Taiwan now. The Taiwanese," he added, "are very fast at bringing new technologies to market. They design fast, and they move fast."
Cooperation with board-makers enables Moran to target system designers with unique offerings. "Nobody makes a 2.5-inch [form factor] SCSI HDD," he said. "The 2.5-inch HDDs you see out there are intended for notebook computers, and notebook computers aren't SCSI. But we make a 2.5-inch SCSI drive. System designers don't want to change their dimensions.
A major focus for M-Systems's marketing, however, is telecommunications. "If you're a telecom utility," he said, "losing data is very costly. Losing billing data is bad enough, but if something makes you lose connectivity, you can get penalized for it by a regulatory agency. Telecom apps put a lot of stress on hard disks, with all that I/O demand, but companies can't afford to swap hard disks every few months. We give them a five-year warranty-that's a few years longer than the HDD warranties; and we guarantee them that we'll sell them the same product, if they need it, over that five-year time span."
Another potential market the so-called network computer (NC), or thin client, or Internet appliance. The prospects have waned, lately, for what were originally conceived of as "diskless" PCs, as the price of full-featured PCs has continued to fall. But assuming they become viable pro-ducts, Moran insisted, "Every NC user will need a fast boot, and a space for storing private e-mails, and so on. Those things shouldn't reside on the main server. There is a requirement for local storage."
HDDs would seem to make the most sense in that application because of their ubiquity, and the fact that their price/ MB is still dropping-a fact largely attributable to shrinking manufacturing costs.
The competitive potential of flash-based storage devices won't be realized until their prices tumble considerably further down. But Moran is optimistic. "What happened in the HDD business will happen here too," he said. "They're using imperfect components to maximize the yield. Our TreFFS algorithmic technique overcomes the solid-state equivalent of bad sectors: it maps the flash, and eliminates them. And it can continue to do it even after the component is on line [i.e., installed], so if it finds one during the work it's doing, it knows how to skip it.
"This means we can add manufacturing capacity, knowing we'll get some imperfect pieces, but then we can sell those as lower-capacity devices; we don't have to throw them away."
|
