Computing giant IBM has demonstrated the first working sample of multi-bit phase-change memory, which it claims offers a 100-fold improvement in write latency compared to traditional NAND flash.
The company's research arm has been working on phase-change memory, or PCM, for years, but has thus far been unable to transfer its theoretical capabilities onto a real implementation. Today's announcement changes that, with a working sample of multi-bit PCM created for the first time.
Phase-change memory works, as the name suggests, by physically changing the properties of the materials from which it is constructed. As a result, it is inherently non-volatile - meaning that data stored within the memory doesn't disappear when the power is cut, unlike traditional DRAM.
Designed as a universal, non-volatile memory technology to replace NAND flash in SSDs - and, potentially, to eventually replace volatile DRAM - IBM's PCM technology is capable of reading and writing data up to 100 times faster than flash. With a write latency of 10 microseconds, PCM promises some major performance boosts if it makes it to production.
IBM's PCM implementation doesn't just improve the latency of the memory, however: the company claims that a PCM storage device could withstand at least 10 million write cycles, compared to 30,000 cycles for enterprise-grade NAND flash or 3,000 cycles for consumer-grade versions.
"By demonstrating a multi-bit phase-change memory technology which achieves for the first time reliability levels akin to those required for enterprise applications," explained Dr. Haris Pozidis, the manager of Memory and Probe Technologies at IBM Research, "we have made a big step towards enabling practical memory devices based on multi-bit PCM."
The team behind the breakthrough were able to solve the issue of short-term drift - a problem in multi-bit PCM that causes read errors over time - using a novel modulation coding technique, smashing one of the biggest barriers to the commercialisation of PCM as a storage technology.
Based on a 90nm CMOS process, IBM's test chip was able to prove long-term retention of bits stored in a sub-array of 200,000 PCM cells for a period of over five months. That's significantly longer than the company has managed before, and is a hopeful indication that PCM can reach the levels of reliability required for a commercial implementation.
Thus far, however, IBM has shied away from suggesting a potential launch date for products based around the technology.