The first successful rewritable nonvolatile memory, flash, is a mainstay in numerous devices that require data to be retained when the equipment is switched off: cellphones, digital cameras, and PDAs. Flash chips represent a big market—US $15 billion to $17 billion.

Flash doesn't do well, however, at writing very large quantities of data quickly. It also begins to leak charge after several hundred thousand cycles, making it unsuitable for devices like a computer's main memory in which data are constantly rewritten. So it's not surprising that a lot of technologies are vying to be the premier next-generation nonvolatile memory, including polymer ferroelectric memories, magnetic RAM, and nanowires. But by a growing expert consensus, the most likely candidate to succeed flash memories is a technology based on phase-change materials.

In these materials, the permanent storage of information depends on phase changes induced by heat. Normally in an amorphous state, such materials become crystalline when heated. After being heated still more, they melt and, when rapidly cooled, revert to the amorphous state. In the crystalline state, these materials conduct electricity better because, essentially, the regular crystal structure offers less opportunity for electrons to scatter.

Such materials were first intensively studied by the inventor Stanford Ovshinsky during the 1950s, and they are now widely used in rewritable CDs, in which the phase changes are prompted by a laser beam. Information is read from the optical variations that accompany these phase changes. But such materials were not developed to the point of being commercially usable in semiconductors, at least partly because of the high currents required to effect the phase changes and unwanted interactions between circuit elements and contacts.

Still, many leading electronics and semiconductor companies, among them AMD, Energy Conversion Devices, Intel, Panasonic, Philips, Sony, and ST Electronics, have maintained R&D programs to develop phase-change memories. Some of those companies have pursued a memory design descended from Ovshinsky's work that is called ovonic unified memory, or OUM. It consists of an amorphous semiconductor film deposited on a large electrode; currents passing through point electrodes on the opposite side of the film from the big electrode heat the film locally, changing it from amorphous to crystalline or resetting it to amorphous. Alternative designs contain separate resistive heaters to make that local phase change.

Because the phase changes occur in areas of the film not thermally insulated from their surroundings, the voltages required for programming the chips have proved to be too high. For new generations of circuitry, which typically operate at only about 1 volt, phase changes must occur at very low voltages.

Until recently, finding a way to facilitate these low-voltage phase changes eluded researchers. Now, however, two teams are reporting greater success with OUM devices. Florian Merget of the Institute of Semiconductor Electronics at RWTH Aachen University in Germany has come up with a different approach. Instead of using a film, he deposited a tiny strip of an amorphous semiconductor compound of germanium, antimony, and tellurium on a layer of silicon dioxide. He then connected the strip to lateral contacts connected to current sources. With this configuration, heat dissipation can be better controlled and smaller voltages can be used to change phase. Merget and his colleagues demonstrated that it was possible to stimulate phase changes with very low currents and voltages of about 1 volt.

Meanwhile, a team at Royal Philips Electronics NV, in Amsterdam, reported success in this April's issue of Nature Materials with a similar device made with a phase-change material recently developed at Philips. It is based on an antimony-tellurium material already used in DVDs that has been doped with one or more of the elements germanium, indium, silver, or gallium. This doped material changes phase significantly faster than the germanium-antimony-tellurium compounds in other experimental OUMs.

In the earlier materials, the phase change occurred through nucleation: the crystalline structure, when heated, starts growing in many spots in the amorphous material simultaneously. In the Philips amorphous material, a crystalline structure emerges in just one area first and then grows quickly and expands over the whole heated region of the memory cell, explains Karen Attenborough, a senior scientist at Philips who leads the research.

"This is the first time that someone has looked at these fast-growth materials and seen the potential these materials have for memories," she says. The programming speed in the Philips material is far better than the 10-microsecond programming speed of flash memories: the Philips team reports that it has obtained a programming speed of 30 nanoseconds, and it expects to reach a switching time of 5 ns [see micrograph, "fast"].

Now, both the Aachen and Philips teams are studying ways to create arrays containing memory cells. Although the strip and its two contacts limit the packing density of these memory cells, they will still fit on top of part of the "selector," the associated transistor used for reading out the individual memory cell, says Attenborough.

Of course, integrating these antimony-tellurium materials in real circuits will be the true test, says electronics engineer Andrea Lacaita of the University of Milan. "To say that there is a real advantage in using this material, we should have at least results on microarrays; it is not always true that the active material that you get on top of a memory array has the same electrical properties as measured in a single sample," he says.

Moreover, Lacaita points out that, contrary to earlier expectations, it probably will be possible within a few years to make flash memories with features as small as 45 nanometers. This development may give flash added life. "In the next decade there is no way that they will be replaced by other memories," Lacaita believes.