Intel and Micron chose not to reveal the materials they are using or the switching mechanism at play in their 3D XPoint non-volatile memory launched earlier this week. But there is some circumstantial evidence for it being phase-change memory technology implemented in a 2Xnm or 1Xnm manufacturing process. The scaling, novel mixes of chalcogenide glass and other developments may have provided the claimed performance improvements.
The 2Xnm feature size can be derived from calculations based on a typical die size, the assumption of a 4F2 memory cell, and the given information of two planes of 64 billion memory cells in the 128Gbit 3D Xpoint memory. A die area of 1 square centimeter yields a feature size of 19nm or less. A larger die would push the upper limit up to the 27nm region.
The secrecy has given rise to much speculation about the underlying 3D XPoint technology. Rob Crooke, general manager of the non-volatile memory group at Intel, and Mark Durcan, CEO of Micron, described it as a fundamental breakthrough. They also said it was "bulk switching" suggesting a non-filamentary behavior.
Intel did say that 3D XPoint should not be described as a ReRAM but the nomenclature definition of what is or is not a resistive RAM is not formal. Intel also said the 3D XPoint memory cell operates via changes in resistance of the bulk material, which makes it a resistive RAM by the broadest definition, but includes the possibility of it being a phase-change memory.
Also I am not sure I would describe phase-change memory as a bulk phenomenon as it results from a thermal pulse that proceeds from an electric current through the material. But it could be argued that in a sufficiently small cross-point a significant amount of the active material undergoes the phase change.
And then there are similarities between 3D XPoint and physical layout diagrams of previous presentations on PCMS – a stacked arrangement of phase-change memory and a selector diode.
Searches of databases of patents and patent applications yielded at least 20 patents assigned to Intel or Micron in recent years that either reference phase-change, PCM or PCMS directly or if they are written more generally about non-volatile memory, reference PCM/PCMS as a specific embodiment. On top of this there are many similar patent applications not yet granted.
And it is harder to find mentions of other non-volatile memory types in those same searches. A list of the patents found so far is at the foot of the article. The frequency with which PCMS comes up seems to support the view that phase-change is the likely mechanism behind 3D XPoint memory but is not conclusive.
The executives in the press conference described 3D XPoint as fundamentally new technology and yet a single-layer PCMS memory was shown more than five years ago in October 2009 prior to the presentation of PCMS at IEDM in December 2009. (see Intel, Numonyx claim phase-change memory milestone)
I guess it depends on your definition of new.
Durcan was asked in the press conference to compare 3D-XPoint with phase-change memory and say why 3D-XPoint would be more successful He said: "Relative to phase-change, which has been in the marketplace before and Micron has some experience with, this is a very different architecture in terms of where it sits in memory hierarchy because of the dramatic improvements in speed and volatility and performance."
This appears to support the idea that 3D XPoint is not PCM but improves upon it.
Phase-change memory has a long and difficult history and perhaps Intel and Micron choose to define 3D XPoint to be something different to PCM?
Micron had previously tried to sell 90nm and 45nm phase-change memories before removing them from its website and said in January 2014 that it was reviewing whether it would continue with phase-change memory technology.
On its website in recent days those PCM datasheets are present again and Micron states: "Micron continues innovating with PCM. After two generations of PCM process technologies, we’re developing a follow-on process to achieve lower cost per bit, lower power, and higher performance. PCM is one of several emerging memory technologies that Micron is investing in."
A 2Xnm or 1Xnm phase-change memory process allowing construction of the two-layer 128Gbit 3D XPoint memory is consistent with that statement. But, as Micron says, there are other emerging memory technologies.
There could also be technology hybrids. What properties would a memory cell have if made of multiple interleaved chalcogenide and metal-oxide layers?
Related links and articles:
Some granted US patents relevant to PCM/PCMS (not an exclusive list).