Arthur Cole spoke with Dr. Haris Pozidis, manager of memory and probe technologies for IBM Research-Zurich.
Phase-change memory (PCM) technology took a major leap forward recently with an announcement from IBM that it had developed a multi-bit version that enables large amounts of long-term, non-volatile storage. Naturally, this had many people talking of Flash's imminent demise, although as IBM's Dr. Haris Pozidis explains, the more likely scenario is an additional tier of PCM storage for improved data handling. The technology also needs a little more time on the drawing board before a commercial product is ready.
"PCM can make it more practical for cloud providers to put massive amounts of data in the cloud so it can be accessed and processed."
Cole: IBM has a new phase-change technology that some say could be suitable for enterprise applications. What exactly have you developed and what were the chief obstacles that had to be overcome?
Pozidis: PCM is challenged with a phenomena called "resistance drift." Because of structural relaxation of the atoms in the amorphous state, the resistance increases over time after the phase change, eventually causing errors in the read-out due to the drift. Resistance drift is particularly detrimental for multilevel-cell (MLC) PCM due to the small gap between the tightly packed resistance levels. To overcome this issue, we developed and applied an advanced modulation coding technique that is inherently drift-tolerant. The modulation coding technique is based on the fact that, on average, the relative order of different resistance levels does not change due to drift in contrast to their absolute values. Using this technique, we were able to mitigate drift and demonstrate long-term retention of bits stored in a sub-array of 200,000 cells of our PCM test chip, fabricated in 90-nanometer CMOS technology.
In this demonstration, each cell stores two bits of information - 00,01,10,11 - as opposed to the conventional one bit per cell - 0 or 1 - storage. The result is that we have been able to withstand the drift for over six months now, maintaining a very low error rate, which had never been done before.
Cole: Comparisons to Flash technology were almost instant. Are we really looking at a wholesale replacement of Flash, or are there specialized applications/environments that PCM will cater to?
Pozidis: Flash is expected to face difficult scaling problems in the next few years. In addition, when you compare write cycles, Flash cannot compete with PCM. Even enterprise-class Flash specially engineered for enterprise applications at the cost of lower speed can sustain 30,000 write cycles, which is by far not suitable for the enterprise. PCM can sustain 10 million write cycles. Read/write speed is another area where PCM shines compared to Flash; not only is PCM 100 times faster, but it also enables very small accesses of data, down to a single byte, whereas Flash always needs to access blocks of 8 kb even for small random reads or writes, which wastes power. So we see a large gap in the memory hierarchy for PCM, particularly for enterprises and the cloud. PCM can make it more practical for cloud providers to put massive amounts of data in the cloud so it can be accessed and processed.
But there are consumer applications for PCM as well. The mobile handset makers are talking about it for better battery efficiency and faster performance. Also, let's not forget DRAM. Today, we have DRAM sitting next to the CPU, and this will not change in the foreseeable future. DRAM will stay and will continue to talk to the CPU because it is fast and durable. But there is hot data, which is accessed often, and then there is some data that may not need to be exchanged all of the time. This kind of less-frequently accessed data can be stored to PCM, which sits right next to a smaller DRAM. The PCM, which is much larger, acts as a repositor. If the data becomes hot again it will move back to the DRAM. This is a much more efficient approach than moving such data all the way to the SSD or the disk, which costs in performance and power consumption.
Cole: The timeline for a commercial product is said to be five years. What further development is needed before the technology hits the channel?
Pozidis: We are looking at different materials now to make the technology faster, and we think we can get the density to match Flash at 3 bits per cell while keeping the speed and write endurance high. We also need to get the cost down, and I believe the consumer market will help us here. Once PCM begins to find its way into mobile phones for code storage - most probably the first application of the technology - we will get scale to also compete with Flash on price. We just opened a $90 million nanotechnology center on the campus of IBM Research in Zurich where we will pursue this work, and it is my opinion that there are no real fundamental limits that will prevent us from achieving our goal.