Phase-change memory has been around for a while, but it looks like we could be on the verge of a breakthrough in several key areas that have made it unusable in enterprise settings.
IBM announced recently that it has developed an elusive multi-bit cell technology that enables PCM to store large amounts of data over long periods of time. Most importantly, it provides a stunning 100-fold increase in throughput over today's fastest SSD technology, all the while providing a non-volatile storage environment that can provide upwards of 10 million write cycles. A commercial product is expected to hit the channel within five years.
A key feature of the chip is that it can store two bits of data per cell without the degradation and corruption issues that have hampered PCM designs to date. The device features 90 nm circuitry coupled with a new modulation technique that prevents data loss as the device's CD-like material shifts between crystalline and amorphous phases during the read/write process.
Like NAND Flash, PCM is primarily a cellular/mobile technology, although there is no reason why it could not find its way into higher-order enterprise systems as well. Naturally, then, power consumption is a primary concern. PCM works by applying heat to its base material to affect the phase change, which requires a fair amount of energy. However, research is progressing on a number of fronts to bring the power requirements down. The A*STAR Data Storage Institute, for example, is working with new dielectric materials that provide high degrees of both thermal and electrical insulation.
Ultimately, the use of PCM could produce computing systems that function increasingly like the human brain, according to David Wright of the UK University of Exeter. Phase-change material provides the opportunity to perform simultaneous memory and processing functions - something that has eluded silicon-based technology for some time. But before we begin a conversation about computer overlords, note that the most practical applications would be in medical and scientific research requiring artificial simulation of biological neurons.
Clearly, a lot can happen in five years to either increase the performance of SSDs in the enterprise or diminish PCM's prospects. Odds are good, though, that top-tier enterprises will be deploying PCM as a matter of course by the end of the decade.
But if past is prologue, this isn't likely to devolve into a zero-sum competition with Flash technology. Rather, PCM could easily carve out its own tier in the ever-growing enterprise storage infrastructure.