Phase-Change Memory Technology Moves toward Mass Production

by George Lawton

Numonyx, a joint venture of chipmakers Intel and ST Microelectronics, says it has built a prototype of an advanced phase-change memory (PCM) chip that it hopes will become a breakthrough product in nonvolatile memory. The company reportedly demonstrated a 45 nm PCM chip with a 1 Gbit capacity, which will start mass production next year. Numonyx joined forces with Samsung Electronics earlier this year to begin mass production of a 512 Mbit chip. In June, the two companies announced the development of a common specification for phase-change chips, giving designers a common interface to work from.

Numonyx also announced R&D advances in the technology. Last year, it began research on a multilevel PCM that would let each chip cell store multiple bits. Last month, it announced two key advances that would support multilayer stacking of PCM material. Both technologies promise to increase PCM capacity once the technology has been refined for commercialization.

"We envisioned PCM as a potential successor to flash, but found that it had some unique new characteristics of its own," said Greg Atwood, a Numonyx senior research fellow who started working on the technology at Intel in 1999. Researchers expect PCM to be faster than flash memory and to consume less power than DRAM.

Understanding PCM

A 1970 Electronics magazine article coauthored by Gordon Moore described an early 128-bit PCM prototype at Intel. However, this tiny chip consumed too much power — 25 V at 200 mA (5 W) — and the technology was never widely used. Later, Stanford Ovshinsky, a research scientist who founded Ovonyx, made several ground-breaking technology advances that have since been licensed to all major PCM players. Today's PCM devices are energy efficient and on par with flash memory for speed.

The PCM chip is made of chalcogenide, a material that consists of at least one of the group-16 chemical elements in the periodic table, such as oxygen, sulfur, selenium, and tellurium. Light or heat can easily switch this material between a polycrystalline and amorphous state, which also changes its optical and electrical resistivity properties.

In a PCM chip, individual bits are heated up, then written to by using electrical current and read by measuring each bit's resistance. Rewritable CDs apply the same principle, using a high-powered laser to change a cell's state and a lower-powered laser to read its state.

Where PCM Fits

PCM is nonvolatile like flash memory, so it requires no power when it's not being read or written to. It can be read and written to nonsequentially, like DRAM and NOR flash, which makes it ideal for storing and running code. In contrast, the more cost effective and widely used NAND flash can be read only in long sequences and written to only a page at a time, limiting its utility to reading and writing whole files. According to Atwood, PCM has a data-seek time on the order of 100 nanoseconds versus microseconds or milliseconds for flash.

PCM can be read at the same fast rate as DRAM. Writes are considerably slower than DRAM but on par with flash memory. Atwood said that today's PCM can write at 10 Mbytes per second compared with several hundred megabytes per second for DRAM.

PCM has an average life cycle of 1 million writes and could go higher, which is much higher than flash at 100,000 writes but lower than DRAM, which is infinite. This limit could pose a problem if flash was used as the primary storage in a memory-intensive application, such as an enterprise server.

However, Atwood believes that PCM might be ideal as a secondary memory in such applications, helping reduce the size and hence the power requirements of DRAM. Unlike PCM and other nonvolatile memories, DRAM must be fully powered at all times.

Increasing PCM Capacity

Numonyx is researching two techniques to increase memory capacity. One would increase the number of bits stored in each cell; the other, the number of layers used to store data. Both technologies are in a preliminary stage and are not expected to be used commercially for several years.

The technique to store multiple bits in each cell actually stores data in analog elements. The element's state is converted into digital data by measuring the chalcogenide material's resistivity. The technique uses an increase in measurement precision to differentiate between more bits in each cell, but it's costly and can raise the error rate. Flash memory has employed a similar technique, but it's a more mature technology.
The other R&D advance uses a chalcogenide material made from germanium, antimony, and tellurium, which can be stacked between specially crafted metal layers. This technique involves a new device called an ovonic threshold switch, which replaces the bipolar transistor traditionally used in PCM and allows the device to address multiple layers. Numonyx has so far demonstrated only the first layer of this new design, but Atwood said the first layer is the hardest. Additional layers can use standard process technologies. Although there's no theoretical limit to the number of layers that can be stacked, Atwood said that the company is likely to create only four-layer chips in practice because each layer can potentially introduce defects and hence reduce the yield of working chips.

Challenges and Applications

The biggest challenge PCM faces is cost, said Jim Handy, a memory-industry analyst with Objective Analysis. The first-generation PCM chips are considerably more expensive owing to reduced economies of scale and substantially less R&D than traditional silicon-based technologies such as DRAM and flash memory. "We have over $200 billion of R&D as an industry in understanding silicon compared with on the order of $200 million on PCM," said Handy.

In the near term, designers will employ conventional PCM designs from Samsung and Numonyx, said Stephen Hudgins, technical consultant to Ovonyx, which licenses PCM technology to both companies. PCM will have a hard time competing with NAND flash for many applications because of the higher cost, but it will become more competitive as memory devices scale. Flash memory is fundamentally limited by the challenges associated with storing electrons in smaller memory elements. PCM doesn’t face these challenges because it doesn't store electrons. It can therefore scale down to smaller sizes and higher densities. 'Ultimately, economics will dictate against NAND flash," Hudgins said.

Hudgins expects PCM to replace NOR flash memory and possibly DRAM for code storage in portable devices. Replacing DRAM could provide substantial power savings and longer battery life.

George Lawton is a freelance technology writer based in Monte Rio, California. Contact him at glawton@glawton.com.