LONDON – STMicroelectronics NV plans to introduce phase-change memory as an embedded non-volatile memory option for use with its microcontrollers.
The technology is being prototyped at the 90-nm manufacturing node during 2013, said Jean-Marc Chery, chief technology officer of ST, speaking at a one-day conference for analysts held here Thursday (May 16). Although the embedded PCM technology will be offered to customers in 2014 ST's main motivation is to prepare for the introduction of a 28-nm phase-change memory process that will prototype in 2016 or 2017, Chery said. PCM would replace NOR flash memory for MCUs, which is not expected to scale beyond the 40-nm node, Chery said.
ST is currently ready for production with a 55-nm embedded flash process for MCUs based on a one-transistor NOR memory element. The company is also developing a 40-nm NOR-based embedded flash process for automotive applications, Chery said.
"We continue to invest in PCM because it is fully compatible with the 28-nm HKMG CMOS process," Chery told analysts. Chery told EE Times that while the initial embedded PCM introduction would be subject to thermal limitations the company is making progress on finding ternary mixes of elements that would have improved performance at high temperature.
In a chart Chery showed analysts that the 90-nm PCM MCUs would be offered for general-purpose and secure applications but not automotive applications. The chart showed that the 28-nm PCM embedded non-volatile memory would serve all three application sectors.
Phase-change memory is non-volatile and based on changing the material phase and electrical resistance of a chalcogenide layer by the use of electrical heating. It has been touted as possible replacement for both flash memory and DRAM, but the technology has been in research for decades and now faces numerous competitor technologies.
For phase-change memory the typical germanium, antimony, tellurium element ratio is 2:2:5, but this has the disadvantage of a relatively low melting point. This manifests itself in the problem that preprogrammed memories could be erased during soldering onto a printed circuit board. Although in-system programming can get around this problem the temperature limitation it can also impact the ability to guarantee 10-year retention at elevated temperatures.
ST began work on phase-change memory in 2000 and joined forces on research with Intel. In 2005, ST and Intel and agreed to codevelop a 90-nm PCM technology. In 2008, ST and Intel combined their discrete memory businesses to form the Numonyx NV joint venture which was subsequently bought by Micron Technology Inc. (Boise, Idaho). Micron is offering stand-alone 128-Mbit phase-change memory at 90-nm and a 1-Gbit device at 45-nm.
With the spin dial always set a maximum when presenting to analysts it would appear that some lessons of the past have been forgotten; blinded perhaps it would appear by the rosy glow of PCM futures selling.
The long and more recent history of phase change memory (PCM) has been plagued by claims that Flash memory will not scale, the problem it did and still does. Moving the target from NAND to NOR is again just asking to be proved wrong. I guess the analysts failed to ask the question what is now different in the approach of ST now and why will it not result in the same end point, i.e. the failure of Numonyx and its absorption into Micron. Especially if some of the same cast of characters are involved.
Prototyping PCM at 90nm, for 2013, with a 55nm Flash NOR ready for production is an interesting approach. The long search for amorphous PCM materials (inherently meta-stable) that offer thermal stability is again part of PCM history. There is from ST an implied prediction that for the 28nm lithographic node embedded PCM their “progress” will have yielded a result, with a device suitable for all three application areas by 2016.
It would also be useful to know what the embedded PCM will offer over the flash NOR, that is the USP. While there is no mention of the planned memory bit capacity for the target applications; from a competitive viewpoint it would appear that MRAM and FeRAM are likely to be serious competitors, whereas for 1Gb they still have some way to go vis-a-vis PCM.
It is interesting that for pre-programmed devices it is suggested that melting is the problem. The crystallization temperature for PCM materials is always lower than the melting temperature so I would have thought bits in the pre-programmed amorphous state would be more vulnerable to the effects of soldering.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.