Morphing Memory

Superfast atom shuffling inspires data-storage alternatives

Anyone who purchases an electronic camera, cell phone, voice recorder, travel disk, or PDA, typically brings home a stick, card, or some other medium containing a chip ready to store information via a technology known as flash memory. Last year, consumers worldwide bought almost $12 billion worth of flash products, which depend on electrons to store data. The semiconductor industry expects global demand to surpass $18 billion by 2007.

BIT BY BIT. Even low currents heat thin metallic vanes in this array of phase-change memory cells viewed by an electron microscope. Unobstructed vanes abruptly heat adjacent areas (arrows) of the phase-change material called GST and cause rapid atomic rearrangements between amorphous and crystalline states, representing bits 0 and 1, respectively. Pellizzer/STMicroelectronics

HOT SPOT. In an electron micrograph of a crystalline, indium-selenide layer, a heat pulse created a semicircular blob of amorphous material. Gibson et al./Applied Physics Letters

GOING THROUGH A PHASE. Diagrams depict a phase transition within the chalcogenide alloy layer (red) of a phase-change-memory cell. A pulse of electric current through that material and gray metal regions around it changes the material from crystalline (left) to amorphous (right). H. Dieker/RWTH Aachen

Nonvolatile memory systems, in which data remain intact even when the power is off, are widespread as the magnetic-disk drives of computers. More recently, portable consumer products have taken advantage of nonvolatile memory provided by fast, high-capacity microchips. In these products, flash rules.

Although camera buffs, for instance, can today store hundreds of images on a stamp-size chip costing less than $100, they’re demanding more data-dense, cheaper storage components. Engineers working to create the next generation of data-storage devices consider flash to be “the technology to beat,” says Matthias Wuttig of RWTH Aachen University in Germany.

Several technologies have potential to dominate the future of microchip nonvolatile memory. The newest contender relies on a principle already at work in any computer that can burn a rewritable CD. A laser heats spots on an inner layer of the CD to between 300°C and 600°C for a few nanoseconds. That’s all it takes to rearrange the atoms in that layer in a way that imprints one bit of digital data—the proverbial 1 or 0. Over the past decade, phase-change material, a class of silvery semiconductors about as soft as lead, has emerged as a star ingredient of write-your-own optical disks.

Now, researchers are striving to recast electronic memory chips by taking advantage of this material. Rather than accumulating electrons to store data, these upcoming chips instantly toggle patches of atoms between order and disorder.

What makes phase-change material particularly suitable for fast-memory devices is that it “can go from amorphous to crystalline [or back] with minimum motion of the atoms,” notes Gary A. Gibson of Hewlett-Packard Laboratories in Palo Alto, Calif. Consequently, it can switch with lightning speed between arrangements that have dramatically different optical properties or electrical resistances. “This is really magic,” Wuttig says.

Phase-change memory developers are resurrecting a decades-old invention that was eclipsed by the success of such materials in optical disks. In the 1960s, Stanford R. Ovshinsky of Energy Conversion Devices in Rochester Hills, Mich., made the seminal discoveries that revealed the potential for those materials to be a medium for electronic—as well as optical—data storage.

Scientific and commercial interest in the electronic version of the technology has exploded in the past few years, Ovshinsky says. That version is known as phase-change random access memory, or ovonic memory, in reference to Ovshinsky. Those who are most bullish about it, forecast that the technology could end up stealing not only flash-memory markets but also those now dominated by volatile-memory technologies, such as the dynamic random access memory (DRAM) and static random access memory (SRAM) used by computers.

It’s elementary

Although in the kitchen, making ice cubes and softening butter aren’t the speediest operations, freezing and melting serve as the basis for the new form of fast computer memory. On microscopic scales, materials can freeze and melt at blinding speeds. Associating 1 or 0 with each of these states of matter provide the makings of memory.

Compounds known as chalcogenides have opened new vistas of data storage because of the changes they undergo when suddenly heated. At the heart of each of those compounds is one or more of such elements as sulfur, selenium, and tellurium—which appear in oxygen’s column of the periodic table—combined with other semiconducting or metal-like elements such as germanium, indium, and antimony.

In optical disks, a laser’s heat switches a chalcogenide patch between an orderly crystalline form and a more disordered, amorphous one. Because the mirrorlike crystalline patches bounce light in a given direction than the somewhat translucent, amorphous patches do, a detector in a CD or DVD player can almost flawlessly discern which bits are 1s and which are Os.

In the new applications, jolts of electric current, rather than bursts of light, trigger the reversible crystal-to-amorphous structural change. Going from crystal to amorphous is straightforward. The electric jolt instantly melts the patch of chalcogenide, and when the nanoseconds-long zap ends, the patch’s temperature plummets so quickly that the jumbled atoms freeze in place before they can snap back into crystalline order.

Going the other way, from an amorphous to a crystal state, the scientists apply a slightly longer, less-intense dose of current that warms, but doesn’t melt, the amorphous chalcogenide patch. The technique takes advantage of the greater stability and lower energy of the crystalline state. The warmth mobilizes the atoms of the amorphous state just enough so that they can rearrange themselves into an orderly lattice.

For reading back the recorded information, differences in electrical resistance reveal whether bits are amorphous or crystalline, or 0s or 1s.

Ovshinsky became enchanted by the magic of phase-change materials more than 40 years ago. While studying an alloy of tellurium, arsenic, silicon, and germanium, he noted that pulses of electricity caused reversible changes in the alloy’s phase and electrical properties.

Even as rewritable optical disks grew into an approximately $70 million industry, Ovshinsky and his company didn’t forget about the electrical option. “We never stopped working on it,” he says.

And it took a lot of work. The earliest chalcogenide formulations crystallized too unevenly or slowly, or they required large electrical currents to switch between states.

Over time, however, materials scientists found ways around these shortcomings. During the 1990s, Ovshinsky and his colleagues unveiled phase-change-memory prototypes whose bits could be read and written a million times as fast as those of chalcogenide structures of earlier decades. The improved materials and designs required so little current to write data that devices made with them could incorporate reduced-size transistors or diodes, making the units more compact.

Cell mates

Today, phase-change-memory developers are girding to take on flash memory. Most are betting on the same chalcogenide alloy of germanium, antimony, and tellurium already widely used in optical disks. A company known as Ovonyx, which Ovshinsky started, owns patents for a material made of germanium, antimony (chemical abbreviation Sb), and tellurium and known as GST. Ovonyx licenses this technology to various companies.

As other types of electronic memory chips do, each GST-based chip contains a grid of millions of identical structures, called memory cells. Each cell is typically made of a sliver of GST plus an electronic component, such as a transistor, for controlling current through the sliver. The cell is encased in thermally and electrically insulating material to prevent waste of heat or current. This cell serves as a single data bit.

The cell often includes a tiny bar of electrically resistive material to enable rapid heating. “I think of it as almost like a match under the chalcogenide material,” says physicist Gregory Atwood, who heads the phase-change-memory program at Intel.

Teams at electronics companies, most of which are also flash-memory makers, are working hard to optimize the layouts and dimensions of their phase-change-memory cells. The competitors include the U.S. firms Intel and IBM; the European companies BAE Systems, Infineon, and STMicroelectronics; Korea’s Samsung; and Japan’s Elpida Memory.

“We believe the technology has good promise,” Atwood says. Intel, for example, has made a prototype phase-change memory chip of 4 million bits (megabits) and is now developing a 128-megabit model, he notes. Next month, Samsung researchers are scheduled to report on a 256-megabit device at a meeting in Kyoto, Japan on cutting-edge chip technology. Even so, phase-change memory still lags behind current products.

“For any new technology, it’s difficult to challenge this juggernaut of flash,” Atwood acknowledges. Flash chips as large as 8 billion bits (gigabits) are due out this summer. Moreover, flash technology promises to continue speeding up and shrinking down at least until the end of this decade. After that, however, size reduction of flash-memory cells will approach physical limits, so competitors may catch up.

In addition to phase-change memory, there’s magnetic random access memory, or MRAM, based on the electron’s inherent magnetic property of spin. In another approach, called ferroelectric RAM, abbreviated as FRAM or FeRAM, materials store electric fields in orientations that can represent digital bits. All these methods use a data-writing process that’s faster than flash’s.

Different strokes

Although most champions of phase-change memory are betting on GST-based memory cells, some researchers have taken another tack. A group in the Netherlands, for instance, has devised a chalcogenide memory cell made almost entirely of antimony and tellurium.

In recent experiments, reported in the April Nature Materials, Martijn H.R. Lankhorst and his colleagues at Philips Research Laboratories in Eindhoven found that cells made with the antimony-tellurium cells work at lower voltages than GST-based cells do. For portable electronics, that could translate into longer battery life.

Another team, led by Hewlett-Packard’s Gibson, has dispensed with cells. Aiming for maximum storage capacity, the researchers created a tiny array of electron-beam emitters that can be maneuvered above a silicon chip coated with selenium-based chalcogenides. With bursts of electrons, the individually controlled emitters can simultaneously write data onto underlying locations of the chip.

In the Jan. 31 Applied Physics Letters, Gibson and his colleagues reported fabricating spots only 150 nanometers across. Such small bits could potentially store data more than twice as densely as typical cells can. Other developers, including the Fremont, Calif.–based Nanochip, write even smaller spots by pumping current through an array of tiny tips in contact with GST. Such high data densities could be attractive for many portable electronic gadgets, Gibson says.

The array techniques “offer a potential significant advance in data density,” says Stephen J. Hudgens of Ovonyx. “But the technology is presently immature.”

Clean sweep?

When phase-change memory finally hits the market, flash memory won’t be the only domino to fall, predict Ovshinsky and some other phase-change-memory enthusiasts. Indeed, Ovshinsky has dubbed GST-based, phase-change technology as “ovonic universal memory.” Much discussed by data-storage engineers, universal memory would combine the best traits of all the different memory types now in use.

“At a single, nanostructured spot [of chalcogenide], you have the potential to replace flash, SRAM, and DRAM,” Ovshinsky says.

But, as Lankhorst notes, even the best chalcogenides today can’t switch endlessly between states, as can standard silicon-based memories such as DRAM. And researchers working on next-generation magnetic or ferroelectric memories may end up leapfrogging into the lead of the nonvolatile-chip-memory competition.

All these technologies have their potential, and they have their drawbacks. Says Gibson, “I wouldn’t want to bet a lot of money on which one is going to win the race.”

More Stories from Science News on Tech

From the Nature Index

Paid Content