SSD 2.0: Phase-Change Memory Challenges NAND Flash

Phase-change memory started as a topic of discussion in the 1970s, when it was touted as a possible replacement for magnetic storage because of its high read/write speeds, low volatility, and high storage density. Now, PCM is actively being developed as the next technology after NAND for flash memory, because it has the potential to address NAND's shortcomings as well as the limitations of conventional hard drives.

"There are a number of advantages that PCM has over NAND and hard drives," says Spike Narayan, functional manager of IBM's Science and Technology Group in San Jose, Calif. "First, PCM is a very high-performance device when compared to NAND or hard drives. Second, PCM is non-volatile and energy-conserving. When no information is being read or written, there is no power consumption. This is significant, especially as more data centers look for storage that is energy efficient. Third, PCMs write endurance is far superior to that of NAND. Right now, NAND flash memory has a lifespan of 10,000 to 100,000 write cycles -- but PCM is better by a factor of three to five times."

For writes, phase-change memory uses a medium called chalcogenide, a glassy substance containing sulphur, selenium, or tellurium. These semiconductors can be changed from one phase to another through the application of heat, which automatically erases any data present. Write performance with phase-change memory far surpasses that of NAND, because PCM bits do not have to be erased (as with NAND) before data is written.

"What you have is a two-step write process of NAND being reduced to a single step with PCM," says Cliff Smith, phase-change program manager for Numonyx BV , a developer of memory products that was formed in 2008 by ST Microelectronics, Intel Corp. (Nasdaq: INTC), and Francisco Partners . "We’re envisioning that PCM will enter the storage market, but that it won't replace other existing media. Enterprises will be able to take advantage of PCM's unique capabilities by placing it in front of NAND in an SSD [solid-state disk] array."

Smith says that this architectural approach would allow PCM to perform the quick reads and writes and then push other writes into NANDs, which are better suited for heavy and consolidated block writing. Chunks of data of 512 bytes "are not a good fit for larger blocks of NAND, because it forces you to manage system resources and overhead."PCM will initially be used as cache memory in enterprises, Smith says, because it can rival, and in some cases exceed, the performance of DRAM (dynamic random access memory). NAND will continue for a time as the leading form of innovative storage for enterprises, but as more PCM gets introduced, PCM price points (and enterprise uptake) will improve.

"Materialization of flash memory in the enterprise will occur over the next two years," says IBM's Narayan. "After that, it is likely that we will begin to see PCM appear in niche applications. It is safe to say that the onslaught of flash will pave the way for PCM. Eventually, we are also likely to see a one-to-one replacement of flash by PCM."

While PCM is not yet established as enterprise storage, there already are some thoughts as to how it can best be employed in an enterprise setting.

"Network routers and switches are areas where PCM could be put to good use," says Numonyx's Smith. "These devices control highly complex systems of hardware and software, and do numerous data logs. In an environment that uses PCM for these data logs, you have the optimum memory for working with small chunks of data quickly. If you try to use NAND or NOR for this, you have to manage the media."

Numonyx has produced 90-nanometer PCM that looks like NOR flash memory with write cycles very similar to flash. But as the company moves to product using 45-nanometer PCM in 2009, write cycles are likely to go up to 2 million to 10 million, Smith says."The problem facing NAND and DRAM has been scalability," says IBM's Narayan. "Once we get below 32 nanometers with either, they are not very effective. Scalability is one of the things we really looked at, and the data thus far is telling us that PCM can be scalable to between 3 and 20 nanometers."

PCM is not likely to appear much in enterprises over the next two or three years, but it is a technology that enterprise CTOs are already focusing on since their technology roadmaps extend out five to seven years.

"If you look today at how flash memory is being suggested to the enterprise, it is mostly though flash manufacturers," says Narayan. "But it is the ability to integrate flash into storage systems that becomes an art, and that will develop enterprises' understanding of how flash can best be used. Industry players also need to look at PCM technology and how it can be leveraged from a systems standpoint to achieve a viable value proposition for PCM in the enterprise."