Flash Memory for Dull Devices

New trick keeps flash memory from being a power hog in some devices

Photo: Kevin Fu

25 January 2011—A new approach to flash memory [PDF], the solid-state storage technology behind smartphones and USB thumb drives, promises to open the door to a smarter generation of low-power devices—even ones without batteries. The trick, say researchers, is to know when not to read the flippin’ manual.

The problem is that flash memory, which was first developed when the plugged-in desktop computer was king, needs more electricity for writing than tiny batteries may be able to provide. Minimum flash memory requirements range from 2.2 to 4.5 volts today. But low-power microprocessors, such as Texas Instruments’ embedded systems workhorse MSP430, run on as little as 1.8 volts.

If the flash memory sits inside a relatively big gizmo such as a netbook or an iPad, the competing specs are not a problem: Just run the CPU at one voltage and the memory at another. But for tiny and embedded devices, like those inside RFID chips or pacemakers, everything runs off the same power rail.

Designers of such systems basically have two options, says Kevin Fu, assistant professor of computer science at the University of Massachusetts Amherst. They can either boost the CPU’s voltage up to the flash memory’s minimum—and suck the life out of the battery—or forget flash memory altogether and work with a gadget that forgets everything.

But Fu recently began to wonder what would happen if he tried operating flash memory at the low-power CPU’s voltage. In fact, he and four other researchers from UMass and Texas A&M found that a few tricks—and a willful disregard for the flash memory manufacturer’s specs—can give low-power systems designers the best of both worlds.

The most effective way to write to flash memory at less-than-spec voltage levels, says the project’s lead researcher, Mastooreh Salajegheh, boils down to one word: persistence. A single bit of flash memory is like a nanoscale desert island on the computer chip that outside electrons jump to by the quantum mechanical phenomenon of "tunneling." Accumulate enough electrons on the island and it flips from its default "1" state to a "0."

Operating at its standard voltage, a single flash memory cell would be written to at a near 100 percent confidence level. But if the cell is operating below its spec voltage, some electrons can still tunnel in. It takes just a few tries to accumulate enough charge to get the bit to flip.

And when conserving battery power—not minimizing CPU clock cycles—is the goal, this one trick can work wonders. The researchers saved 34 percent of the device’s overall energy budget by running at 1.8 volts and using the "persistence" method instead of at the 2.2-volt minimum specification for flash.

However, Fu says, "this is not a panacea. It’s not applicable to all things." But, he adds, to engineers working to squeeze more life out of shrinking devices with shrinking battery space—such as remote-control key fobs or digital picture frames—the new technique could provide a welcome boost.

Plus, he says, some systems already run on puny batteries or sometimes no batteries at all: RFID tags, electronic passports, and smart nodes in wireless networks. Their electricity comes from harvesting the UHF waves from external reader devices. With the persistence technique, ICs without batteries could soon have their own kind of nonvolatile storage.

Martin Rinard, professor of computer science at MIT, says Fu’s group’s optimization trick is clever and may indeed clear the way for new applications.

"Kevin [Fu] is taking this device in which the manufacturer is saying, ’Hey, Kevin. All bets are off,’ " Rinard says. And yet, with the help of persistence, the flash memory game for low-power systems designers is now on.

Fu's team will be presenting this research next month at the USENIX File and Storage Technologies conference in San Jose.

About the Author

Mark Anderson is an author and science writer based in Northampton, Mass. In the January 2011 issue he described an alternative way to measure the power of supercomputers.

Related Stories

Advertisement
Advertisement