Al Fazio, Intel Fellow and technology and manufacturing group director, memory technology development, Intel Corp. (Santa Clara, Calif.), considers phase-change memory (PCM) to be a technology that will ensure the continuation of Moore’s Law in the non-volatile memory (NVM) sector, assuming that its manufacturability hurdles are overcome.
 |
| Al Fazio, memory technology development director, Intel. (Source: Intel) |
“Developments have largely moved out of the R&D stage for PCM techn
ology, and things are entering the heavy-duty engineering phase,” Fazio said. “However, until it becomes manufacturable, nothing is proven and it remains in the category of ‘future technology.’” He added that, in a historical context, few technologies get over that obstacle. “We’re working with memory technologies invented some 30 years ago that have evolved beyond their hurdles to what they are today. PCM has its problems, but getting over them is probably not a matter of fundamental research, but of just straightforward engineering.”
All semiconductor memory technologies in production today trace their roots back to the late 1960s, early 1970s — SRAM, DRAM and flash, the latter having evolved from EPROM, which was developed during that time. Fazio said that, when considering new technologies that appear as if they might change the face of the memory landscape, it is useful to step back and take a look at that history. “All these technologies — SRAM, DRAM and flash — have significant scaling problems associated with them,” he said. “They’re based on MOSFET technology, and their scaling challenges are motivating researchers to consider alternative technologies, such as MRAM, PCM and FeRAM.” Fazio is quick to admit that, from an NVM perspective, for the rest of this decade and the beginning of the next, it seems clear that nothing will supplant flash, specifically NAND flash — it will remain the mainstream flash memory technology over the next several years.
“There are many alternative technologies that can display memory characteristics, keeping in mind that for memory characteristics, all you need is a hysteresis curve,” Fazio said. “However, you’ve got to realize that when you talk about what can be the main memory technology at the end of this decade, the physics of its underlying technology must first be understood; you cannot still be exploring the physics of the thing if you’re going to go into manufacturing. It has to have moved out of university basic rese
arch and be in corporate fabrication facilities, and it has to be at the point where you can be looking at large, multi-megabit arrays.”
 |
| Before achieving full-scale manufacturability with phase-change memory, engineering hurdles must be overcome. (Source: Intel) |
Going by Fazio’s statement, there would seem to be only three memory technologies that more or less meet those criteria: MRAM, FeRAM and PCM, with the latter being the most promising. This is mostly driven by the fact that PCM has the best scaling capability. “There is very good ongoing research with MRAM and FeRAM,” Fazio said, “but they present scaling difficulties. PCM, however, appears to have the capability to scale down to 5 nm and beyond. That would make it a technology with sufficient longevity to last for several generations; nevertheless, there still remains much to be proven.”
The situation is not desperate. If one considers the pace of technology development, particularly in NAND flash, a new iteration of the technology has been delivered on almost a yearly basis. This means that by the end of the decade, we will still have another two or three generations to go; however, that does not mean there will be no opportunity for innovation. As Fazio put it, “There’s a lot going on, and it’s always easiest to make an incremental improvement over the devil you know and take it forward another generation than it is to invent a completely new alternative technology.” This is why he believes NAND will be dominant for the next several years.
PCM is already proven in certain applications; it is the material that is used in rewritable CD-ROMs. There, it is used with a laser — photon energy changes its material from an amorphous to a crystalline state. “We’re trying to use an electrical current — IR heating — instead of light to change the memory,” Fazio said. “While it’s doable, it hasn’t been proven on a large manufacturing base. That’s a major hurdle — moving this feasibility, the basic capab
ility, to the manufacturing floor.” Another PCM concern is that although there are many studies showing that it is scalable down to probably <5 nm dimensions, producing a device around it is a different matter. “How do you build the device structure, what’s the wordline/bit line configuration of that array, how do you get all of this to work in those small dimensions? I have little doubt that it will be solvable, but it will require work,” Fazio said.