Flash memory seemed like a miracle chip when it made its commercial debut a decade ago. A new type of electrically erasable programmable read-only memory (EEPROM), it did not need a constant supply of electricity to store data, unlike random-access memory (RAM). Packaged in standardized formats such as CompactFlash and SmartMedia, it has become ubiquitous in personal digital assistants, digital cameras and audio devices.
Unfortunately, flash is more expensive to produce than RAM. And that has generally limited capacities of consumer-priced memory cards to 32 or 64 megabytes-confining in a world where 1-MB photographs routinely flit across the Internet. The highest-capacity type, called NAND for the type of logic gate it predominantly employs, also takes far longer to write data than do competing magnetic and optical alternatives.
But cheaper and faster flash memory is on its way.
Cutting Costs, Bit-by-Bit
Flash proponents are aggressively preparing new manufacturing processes to drive the cost per bit-the memory industry’s yardstick-down far enough so that NAND flash can compete with RAM as a high-capacity internal storage medium. It could also challenge optical media, such as DVDs, as a medium for exchanging movies and other large media files.
For example, at lower costs, flash could replace DVDs in video players more compact than today’s, says Rich Wawrzyniak, director of non-volatile memory at Semico Research, a Phoenix-based market-research firm. Instead of volatile RAM, certain startup operations could be handled in flash memory, allowing the long-promised “instant on” PC. Medical patients could store their latest x-rays on a “smart” card they carry around to doctor visits. “There’s no end to the kinds of things you would want to carry around,” Wawrzyniak says.
Reducing costs is largely a matter of squeezing more transistors onto each silicon wafer manufactured. This year, for example, one of the leading flash makers, Toshiba America Electronic Components (TAEC) of Irvine, CA, began making 256- and 512-Mb chips. Toshiba employs a new .16-micron manufacturing process that produces smaller transistors, and thus more transistors per wafer, than the previous .20-micron process.
By yearend, the company says, it will quintuple performance to 10 megabytes per second-faster than top-end DVD players, for instance. And next year Toshiba expects to deploy a .13-micron process to manufacture a new multi-level cell technology that allows two bits to be stored in each memory cell, effectively doubling capacity.
The practical effect of all this manufacturing prowess has been a halving of per-MB costs since last year to a little more than $1 per megabyte, says Kevin Kilbuck, TAEC’s business development director for flash memory products. Sixty-four-MB memory cards, the consumer “sweet spot,” now cost $50 to $100 at retail, he says. Toshiba could drive prices down to a quarter per megabyte or less by 2005, with one-Gb cards perhaps following shortly after.
But only the lower-capacity chips will reach 10 megabytes per second initially, with the first one-Gb chips staying at two megabytes per second, Kilbuck admits. “The trick is to store two bits per cell and not degrade performance,” he says.
Though Kilbuck claims Toshiba is leading the price/performance drive, the semiconductor industry tends to adopt new technologies in lock step, and competitors Samsung and Fujitsu are working on similar products.
Other Non-Volatile Memories
Chipmakers also are working on other types of non-volatile memory.
One, ferroelectric RAM (FRAM), uses ferroelectric materials to store bits. It provides the superior speed of regular RAM but uses less power than flash memory, according to the company holding many of the FRAM patents, Ramtron International of Colorado Springs, CO.
Another technology promising similar advantages, magnetic RAM (MRAM), uses magnetic rather than electronic charges to store data. Among its backers are IBM and Infineon Technologies of Munich, a major memory maker.
FRAM could appear next summer as a cheaper, single-chip alternative for today’s cellular phone-system software. (Today this is typically stored on a combination of static RAM and NOR flash, a second type of flash memory that employs logic gates better suited for sophisticated logical operations than NAND, which excels at sequentially reading large blocks of data.)
“Ferroelectric RAM can basically handle both the functions of the fast [static RAM] and the non-volatile flash,” says Matt Schmidt, a spokesperson for Infineon, a partner with Toshiba on FRAM technology. Both new types of memory could replace flash in many applications, though MRAM is not likely to arrive before 2004, Schmidt says.
Security concerns are another hurdle for flash memory and its rivals, emphasizes Wawrzyniak.
Music companies and other content providers need assurances against piracy before they’ll be willing to put their intellectual property on flash-memory video and audio players. Perhaps more importantly, he says, they’re jockeying to make sure the security standard supports their plans to use memory cards to achieve economies and competitive advantages in new distribution channels.
“You can read [NAND flash] out like a tape,” Wawrzyniak says. “What’s holding them up is the right method of copyright protection.”
Wawrzyniak remains confident that the semiconductor industry will work out the manufacturing and the security issues for non-volatile memories. “It’s the holy grail,” he says. “Semiconductor manufacturers have been looking for this for 30 years.”