LONDON – Non-volatile memory bits can be made by applying phase-change material to previously created nanometer-scale gaps in carbon nanotube filamentary conductors, according to a paper published by researchers at the University of Illinois, Urbana-Champaign, Illinois.
The configuration achieves programming currents as low as 0.5 microamps for set and 5 microamps for reset operations, two orders of magnitude lower than state-of-the-art devices, the authors claim in their ScienceExpress paper, published March 10.
The researchers, led by Professor Eric Pop of University of Illinois, grew single-wall and multi-wall carbon nanotubes with diameters of between 1 and 6 nanometers and then used electrical breakdown, e-beam writing or an atomic force microscope to form a gap of less than 100 nanometers in each wire. The devices than received a sputtered layer of germanium-antimony-tellurium (GST) so that the gaps in the CNTs are filled with the phase-change material. Although amorphous GST is laid down over the entire device, the switching occurs only in the nanogap of the CNT, which is the location of highest electric field and heating effect.
The authors claim their results address the scaling of both the size and power reduction that is possible with programmable PCM bits. The paper reports on reversible switching with programming currents between 1 and 8 microamps, two orders of magnitude lower than state-of-the-art PCM devices. The PCM material switches between an off resistance of 50-megaohm to an on resistance of 2-Mohm but which can be as low 500-kohm, depending on sample preparation.
The authors report hundreds of switching cycles. Conventional PCM devices have been reported at more than 1 million cycles endurance.
The creation of nano-debris in the formation of the CNT gaps does not appear to be a problem, according to Professor Pop. "The
nanogaps are formed by an oxidation process, so it is possible that the C atoms
combine with O atoms forming gas (CO or CO2), not solid debris. Regardless,
debris does not seem to be a problem since the nanotubes show a “clean cut”
after imaging with atomic force microscopy (AFM)," Professor said in email correspondence with EE Times.
Professor Pop and his team only worked with individual devices. "That being said, I think the integration could be pretty high if the nanotubes could be closely spaced. Perhaps one could fit even up to 100 parallel nanotubes side-by-side (~10 nm spacing) and the memory bits could still work without cross-talk. This is just an educated guess based on some calculations we’ve done, not something that we have achieved in practice, yet, he said.
The authors go on to suggest that 5-nm GST bits in a CNT gap could operate at 0.5-V and less than 1 microamp, such that nanosecond switching times would result in sub femtojoule-per-bit energy consumption. There is also the option to use GeSb which has a lower switching threshold. The authors claim the results are "encouraging for ultra-low power electronics and memory based on programmable PCM with nanoscale carbon interconnects."
Professor Pop agreed that blowing billions of CNTs electrically serially would be too time consuming and that doing it one pass would consume too much power. "Yes,
we’ve considered this, but we don’t think this is necessarily the way to go for
a commercial application," said Professor Pop. "Commercially, nanogaps would probably be made through some large-scale etching
process, in large CNT arrays," he added.
However, the paper does not address
the introduction of additional manufacturing complexity, so the
development is a long way from being proven as a useful commercial
Micron Technologies Inc. and Samsung Electronics Co. Ltd. are the two companies most active in commercial phase-change memory. Both have 1-Gbit designs but neither company is boasting of high volume sales into commercial designs and concerns have been raised over the ability of conventional PCM-on-silicon memories to scale and compete with flash memory which is already at higher than 1-Gbit capacity implemented on processes with finer minimum dimensions.
Volatile Memory; I think you are being a little bit hard on your interpretation of the motives of Prof Pop and his colleagues a “charlatans”. There maybe some associated with PCM but I think it is unfair to describe this particular group in the way you have. Although this has not changed my views on PCM scaling, my view is that Prof Pop et al are doing exactly what a University Department should be doing. In the end the paymasters (including taxpayers) must decide on the value of the work, but the choice of work must be left to the Profs and Department heads.
Real progress is made by trying to prove the established wisdom (or lack of it) wrong. They are teaching new graduates the difficulties and pit-falls of handling carbon nano-tubes and working with structures with dimensions of a few nm. This will surely enhance their ability to contribute at the leading edge in their future careers, while in the near term there are likely to be bi-products of their work that have nothing to do with PCM. I think there is even one associated with PCM and their gap structure they may have missed, but notwithstanding that they already have the basis of a discrete low current fuse.
I am in the process of preparing PCM Progress Report Part 2 and I will discuss my view of the shortcomings of the CNT-PCM work in that-EETimes editors willing.
An interesting feature is the phase-change region is surrounded by the high-resistance amorphous GST. This is similar to ReRAM. Realistically, standard semiconductor processing steps will crystallize this surrounding amorphous, which cannot be tolerated without wrapping the CNT's or any electrodes in dielectric isolation. With crystalline lower-resistance surroundings, the RESET current has to be larger, since you have to address all possible current paths.
If I understood Mr. Neale's series of four articles correctly, Prof. Pop and colleagues are simply a bunch of charlatans as they have done nothing to address the real problems that plague PCM/PRAM.
And the last time I checked, no commercial product on the market uses any phase-change memory, and of course, neither Numonyx nor Samsung is selling any PCM/PRAM.
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.