Numonyx Makes Stackable Phase-Change Memory

17 November 2009—A new type of memory based on phase-change materials that can be stacked in layers could lead to much denser memory chips at lower costs, according to the researchers at Intel and Numonyx who developed it.

"We think it's quite a significant breakthrough," says Greg Atwood, a senior technology fellow at Numonyx who is working on the memory with Intel researchers. Numonyx is a Swiss joint venture of Intel and STMicrolectronics.

The memory uses phase-change materials, which alter their structure when exposed to heat or electrical current, ust as heat changes the phase of ice to liquid water. When heated, the material can switch between a conductive crystalline and a resistive amorphous form, providing the binary states necessary for memory. Because the material retains its state until switched, the memory is nonvolatile, requiring no electricity to retain its information. The researchers used a chalcogenide glass, a popular phase-change material that can include elements such as sulfur or tellurium. Numonyx would not say exactly what the chemical makeup of its compound is.

The team created a 64-megabit test chip in a cross-point configuration, with individual memory cells defined by the intersection of perpendicular lines. Typically, such devices are built with each cell containing a phase-change memory element next to a selector, which isolates one memory element from the next. Some of the best selectors to date are silicon diodes, says Al Fazio, an Intel fellow and director of memory technology development who also works on the project. In this case, however, the team used a component called an ovonic threshold switch, or OTS, as the selector. The OTS is also a chalcogenide glass that's laid down as a thin film and is compatible with the phase-change material in the memory elements. Applying a current causes the OTS material to switch on, but unlike the material in the memory component, the OTS's phase does not change.

Silicon selectors must be built on the chip itself and thus can be only one layer thick. The phase-change material and the OTS can be laid down on top of the chip in layers, with each new layer adding more memory capacity. It's this ability to stack the layers that could ultimately lead to a very dense memory. Atwood says it's not yet clear how many layers can be stacked on top of one another. Each additional layer adds both some cost and the possibility of more defects with each processing step.

Several companies have been chasing phase-change memory for years. Today's computer memory often consists of a mix of NAND flash memory and dynamic random access memory, or DRAM. NAND flash is low cost and nonvolatile, but it can take many milliseconds to read data and hundreds of milliseconds to write it. DRAM works much faster, but it's costlier, and it can't store data when the power is turned off. By contrast, phase-change memory is nonvolatile and should be faster than flash and perhaps cheaper, the researchers say. Using more nonvolatile memory could cut a computer's power requirements as well.

Also promising is that engineers should be able to shrink phase-change memory to much smaller dimensions than it uses now, allowing the technology to keep pace with ever-shrinking circuits on the chip. Atwood says it's becoming harder to shrink flash memory, but phase-change materials should be stable even down to memory-cell sizes of 3 nanometers. Numonyx's stackable memory prototype has a cell size of 4 F²—where F is the size of the smallest feature in the device—compared with a range of 6 to 8 F² for DRAM or about 5 F² for single-bit-per-cell NAND flash. However, stacking can make the effective cell size much smaller.

Numonyx is already marketing limited quantities of phase-change memory based on 90-nm cells and hopes to have samples of a 45-nm device available this year. In September, Samsung announced it had begun selling phase-change random access memory, or PRAM, with 512-megabit capacity for mobile devices. Samsung said its device can erase 64 000 words of data in 80 milliseconds, more than 10 times as fast as flash. The company says it expects PRAM to become one of its core memory products. Other companies, including IBM and Philips, are also developing phase-change memory.

Atwood and Fazio refused to project how long it might be before their stacked phase-change technology is ready for commercialization, emphasizing that it is still in the research stage. "NAND flash today is the mainstream technology that's out there, and we're going to try to continue that as long as possible," Fazio says. "We see this as more of a fundamental long-term type of thing."

The researchers plan to release more details when they present a paper on their work at the IEEE International Electron Devices Meeting in Baltimore on 9 December.