Processor chips are the brains of today's consumer digital devices, but memory is actually at their heart, with flash memory being the favored approach for cards that plug into mobile phones, cameras and PCs. Whereas hard drives store large amounts of long-term data, RAM—also called "solid state" memory—retains information outside the hard drive, where it can be accessed quickly and repeatedly. Flash memory, the cheapest RAM variety at only about $1.50 per gigabyte (after that is dynamic RAM, or DRAM, more than a dozen times more expensive), does not require much power and can retain data after a device is powered down, key to gadget-makers ability to turn out smaller, more powerful devices.

But flash has its limitations and will someday reach a scaling barrier that would leave subsequent generations of digital cameras and cell phones unable to store significantly more information or operate orders of magnitude faster than their predecessors—at least not without costing a lot more. "What it comes down to is that we've gotten spoiled with our gadgets," says Stan Williams, Hewlett–Packard senior fellow and founding director of HP's Information and Quantum Systems Lab. "We've come to expect exponentially faster devices, but technologies like flash are getting toward the end of their ability to scale."

Flash is also less durable than static RAM (SRAM) or DRAM, wearing out and becoming less reliable over time. SRAM, which costs about $450 per gigabyte, is often used for cache memory in microprocessors and can rapidly read and write data.

For these reasons HP, IBM and others are grooming new technologies to be the replacement for flash as soon as these limitations catch up to the demands of manufacturers and consumers.

A smarter type of memory
A recently announced alliance between HP and Hynix Semiconductor aims to fill the imminent gap with a denser and more energy-efficient technology called resistive random access memory (RRAM) that could roll off the assembly line in the next few years.

RRAM technology (sometimes called ReRAM) is in a relatively early stage of development but has managed to garner attention from HP, Sharp, Samsung and several other companies that have received RRAM-related patents in the past decade.

The key component of HP's RRAM, so named because data flows with the help of changes in electrical resistance, is a switch called a memristor (or memory resistor). In addition to holding nonvolatile memory (it can retain information even when off), HP says a memristor can also perform computations, a feature that other types of RAM do not have. This versatility gives the company hope that it will someday be able to replace memory and central processing units (CPUs) with a single chip that performs both functions.

Although scientists have known about memristor dynamics for about 40 years, only recently have they been able to design the technology into integrated circuits, Williams says. RRAM involves a subtle motion of atoms, changing the film of a resistor by a matter of nanometers. "People have been talking about this for 40 years but weren't able to fully understand or control the process," Williams says.

Whereas a one-bit flash memory can be switched on and off hundreds of thousands of times (a switch might involve taking or deleting a picture on a digital camera), a memristor has a much longer lifespan than flash, Williams says. On the memory sticks and digital cameras where flash has flourished, switching is not done in the same volume as in other types of memory, so flash can get away with its lower level of endurance, he says.

In April, HP Labs announced its discovery that a memristor could also perform logic, enabling computation to one day be performed in chips where data is stored, rather than on a specialized CPU. HP researchers also say they have designed an architecture where multiple layers of memristor memory can be stacked on top of one another in a single chip to increase storage capacity without taking up much more space.

Although much of RRAM and memristor's future is speculative, HP has no shortage of plans for the technology. The company's goal is to develop and commercialize a RRAM chip in roughly three years that is faster and more durable than flash yet uses less power and features twice the number of bits for storage, Williams says. "We would redesign products around memristor technology," he adds. HP's long-term play is for RRAM to compete with flash, DRAM and even hard disks.

Competing with DRAM, however, requires much greater endurance. A DRAM used by a supercomputer doing climate modeling might read or erase data one quintillion (one million trillion) times over the course of three or four years, Williams says. "That's daunting, but that's what we're going to have to do to compete head-to-head with DRAM," he adds.

For all of their excitement over memristors HP still faces a number of challenges in bringing its RRAM to market. For one, "HP's memristor is made using a titanium oxide, which is not common to put on semiconductors and appears to be difficult to manage today," says Jim Handy, an analyst with the semiconductor market research firm Objective Analysis in Los Gatos, Calif. HP disputes this characterization of titanium oxide, saying there are established protocols for applying it to semiconductor material and that titanium oxide is compatible with current complementary metal-oxide-semiconductor (CMOS) technology used to build integrated circuits.

Racetrack memory
Rival IBM is taking a different approach to develop its next-generation memory, one that uses the spin of an electron to store data. IBM calls dubs it "racetrack memory," because it uses magnetic domains to store information in columns of magnetic material (nanoscale "racetracks") arranged on the surface of a silicon wafer. Such a chip needs no moving parts larger than an electron to read and write bits, boosting speed and reliability, Stuart Parkin, an IBM fellow and manager of IBM Research–Almaden's Magnetoelectronics Group, wrote in the June 2009 Scientific American article "Data in the Fast Lanes of Racetrack Memory."

Whereas chips with horizontal racetracks alone could out-compete flash memory, creating "forests" of vertical racetracks on a silicon substrate would yield three-dimensional memory chips with data-storage densities surpassing those of hard drives, Parkin says. "With racetrack, we've demonstrated the principle works," he says, adding that he expects to see a prototype of racetrack memory within three years and a device featuring it within five.

Other approaches
Along with RRAM, there are several other types of random access memory under development, although given their relative immaturity, none has made so much as a dent in flash's market dominance. Phase-change random access memory (PRAM), like RRAM and racetrack, is nonvolatile. PRAM is novel because it uses electrical current to store data in a glassy substance called chalcogenide, whose atoms are rearranged when it is heated. Chalcogenide is also used in CDs and DVDs, which store information with the help of a laser that heats and rearranges chalcogenide atoms.

Magnetoresistive RAM (MRAM) stores data using magnetic elements rather than electric charge or current flows, although newer developments in the technology seek to use electrical switching to increase MRAM's density and make it a stronger candidate to replace flash before many other RAM options come to market.

Nano RAM (NRAM), which works by placing billions of carbon nanotubes on a silicon chip to store data, is under development by Woburn, Mass.–based Nantero. When an electrostatic force is applied, the carbon nanotubes move up and down to represent 0s or 1s. The company says it has demonstrated that NRAM has the speed and capacity to surpass other memory types and can be manufactured in existing chip-making facilities (called "fabs").

Nantero and Lockheed Martin last year developed a radiation-resistant version of NRAM that NASA tested on a the Atlantis space shuttle. NASA incorporated NRAM into special autonomous testing configurations installed into a carrier at the aft end of the shuttle's payload bay. The companies say this mission was an important first step in the development of high-density, nonvolatile, carbon nanotube–based memories for spaceflight applications.

Nantero continues to work on NRAM to get it ready for the commercial market as replacement for flash and embedded memory that would sit on a variety of different types of chips, including microprocessors, application-specific integrated circuits (ASICs) and field-programmable gate arrays (FPGAs). One of the company's biggest challenges, however, will be overcoming the difficulty of making uniform carbon nanotubes that can perform predictably on mass-produced NRAM chips.

Will flash flame out?
In any case flash still could enjoy years of dominance as companies continue to develop new related technology. Toshiba, for example, in June announced it had created a 128-gigabyte flash memory module, which the company claims is the highest capacity flash to date. Toshiba says it was able to achieve this level of memory by stacking together 16 eight-gigabyte flash chips.

"It is unclear exactly when flash will hit the wall in terms of its ability to scale," Handy says. "There's no way to really know when flash will reach its limit." Until then, it will be difficult for any of the emerging memory technologies seeking to replace flash to compete with its low cost.