The bulk of commercialized PCRAM is in Intel Optane SSDs and DIMMs.
TORONTO — Earlier this year, phase change memory (PCRAM) was touted as one of three emerging memories to watch, in large part because of an uptake in Intel’s Optane SSDs and DIMMs. But what about non-3D Xpoint-based PCRAM?
This year’s report by Objective Analysis and Coughlin Associates, Emerging Memories Ramp Up, forecasts the revenue growth of 3D XPoint at $16 billion by 2029 in the form of Optane, driven in large part to its sub-DRAM prices. Although it’s never been explicitly stated to be PCRAM, the consensus is that that is what 3D Xpoint is. But outside of the technology jointly developed by Intel and Micron, there is ongoing research to realize the potential of PCRAM.
As an emerging memory, PCRAM has a long history, but it’s only been in the last decade that research has picked up. Jim Handy, principal with Object Analysis, said the reason the technology continues to hold such promise is its really small crosspoint cell and its ability to be stacked. Intel, for example, is stacking two decks right now, he said, but recently revealed it can go up to four decks. “They intend on reducing costs by increasing the number of decks.”
But outside of the 3D Xpoint technologies, Handy isn’t aware of anything else that’s been commercialized despite a great deal of research and development activity. Micron Technology recently announced the X100 NVMe SSD, the company’s first product in a new family of high-performance memory solutions based on 3D XPoint technology.
Handy said PCRAM is like all emerging memories in that each one has something it excels at, but if you talk to the MRAM crowd, it will say it has everything going for it, and PCRAM doesn’t, and vice versa. “The question is, is there enough of a market for that one particular thing that it will be able to create a market for itself? For the most part, it doesn't happen.” He said one area that holds promise for PCRAM is building neural networks, although other technologies such as ReRAM are also being looked at too.
The only commercialized PCRAM to date is in the form of Intel Optane DIMMs and SSDs, which the company has positioned as another layer in the storage hierarchy above 3D NAND SSDs and below DRAM. (Source: Intel)
IBM Research is looking at the potential of PCRAM. Haris Pozidis, who manages the cloud storage and analytics group at IBM Research in Zurich, said the focus over the past several years has been on enabling multi-bit storage as the only way to reduce costs by storing more bits per cell. The research has included running workloads on Intel Optane SSDs to better understand the implications of having a faster device. With the long-awaited release of DIMMs, he said, there’s interest in the feasibility of storing an entire database in this type of non-volatile memory. “If you designed the file system to take advantage explicitly of the persistent memory capabilities, then the performance benefits that you can get can be very, very significant.”
PCRAM’s potential to scale is one of its most appealing characteristics compared to MRAM, which has seen more commercialization, said Pozidis. “The area of the memory cell in MRAM is about 10 times bigger than the area of the memory cell in PCRAM. This translates into much smaller number of cells in the same byte size, and, therefore, scalability of MRAM is very doubtful.” Similarly, he doesn’t see the maturity of ReRAM to be as far along as PCRAM even though it is scalable. While ReRAM will continue to attract investment, the benefits of PCM are more immediate.
PCRAM still faces the barrier that any emerging memory does — making an economical argument that it can dethrone an incumbent technology, particularly flash, said Pozidis. The next milestone for PCM to compete with flash is up — by putting more cells on top of each other, while cramming more bits into each cell. From a materials perspective, he said, PCRAM is in a good position, as demonstrated by the commercial availability of 3D Xpoint.
The materials involved in PCRAM go back near 50 years, said Etienne Nowak, head of the Advanced Memory Lab at Leti, which means they are well understood even though research has only increased in the past decade. PCRAM is usually comprised of typically glassy materials from the chalcogen family such as sulfur, selenium, and tellurium. Using heat as the programming mechanism, these phase change materials transition from highly amorphous material arrangements to crystalline arrangements. The order of materials determines resistance levels, and proper deposition of materials is necessary as the center of the wafer must have the same composition as the edge as this affects the phase change.
Micron, which jointly developed 3D Xpoint technology with Intel, just released its X100 NVMe SSD, the company’s first product in a new family of high-performance memory solutions based on 3D XPoint technology. (Source: Micron)
Despite familiarity with the materials, there’s still room to improve, said Nowak. “In terms of physics, we are still not at the limit of this material.” From a manufacturing perspective, the small dimensions create challenges as the materials involved all have unique deposition issues. “The challenge now from our point of view is more continuing to scale without losing the capability,” he said. “But theoretically, when we go smaller it gets better.”
Intermolecular, Inc., a part of the Performance Materials Business of Merck KGaA, Darmstadt, Germany, which operates as EMD Performance Materials in the U.S. and Canada, works with semiconductor manufacturers in the world to help them solve some of their materials challenges. Chief technology officer Karl Littau said PCRAM still doesn’t scale as well as the manufacturing technologies for vertical 3D NAND because it must be built up layer by layer, each with its own critical lithography and etching steps. This means a four-layer PCRAM device costs twice as much as a two-layer PCRAM device, so while Optane is out there, it's still not clear what version of non-volatile memory will be the winning one, he said. “You can't make a 128-layer device and have it be cost competitive against existing technologies.”
Even if the materials are well understood, they are fundamental to making a device that performs well and is economical to produce, said Littau. “When Optane was released, people were really excited because clearly Intel and Micron felt that they had solved those problems sufficiently, that they could release a viable product. Time will tell exactly how successful it is in meeting all of that,” he said. It’s hard to argue with the basic performance it’s demonstrated, said Littau, but it’s not clear whether it’s economical enough.
Littau said many manufacturers are keeping an eye on how successful Optane is before introducing their own products. Learning curves are a normal part of the process for new technologies such as Optane and they will always be more expensive to manufacture in the early stages. “It's not clear yet where on the learning curve this is going to flatten out.”