LONDON – Researchers from University College London (UCL) have developed a silicon oxide based non-volatile resistive memory structure by following a promising line of research similar to that pursued by teams at Rice University and the University of Pennsylvania.
Resisitive RAM (ReRAM), sometimes called a memristor, is now being researched extensively as a potential replacement for flash memory which is not expected not to scale in plan much below 20-nm minimum dimensions. Some further memory capacity scaling may be achieved by stacking flash memory cells vertically but it is thought that resistive memory could displace flash memory if it can offer planar as well as z-directions scaling. Silicon oxide is now being researched as a potential alternative to metal-oxide types of ReRAM.
The UCL team has developed a silicon oxide memory, described in a recent paper in the Journal of Applied Physics, which they claim performs the switch in resistance much more efficiently than has been previously achieved.
The paper, entitled Resistive switching in silicon suboxide films, reports that the UCL team have worked with silcon-rich silicon dioxide. This differs from the work of Professor I-Wei Chen at the University of Pennsylvania who has reported the use of inclusions of atomically dispersed platinum in silicon oxide material. The UCL work appears to be closer to that of a team under James Tour at Rice University (see Making memory out of silicon oxide).
"The resistive switching phenomenon is an intrinsic property of the silicon-rich oxide layer and does not depend on the diffusion of metallic ions to form conductive paths. In contrast to other work in the literature, switching occurs in ambient conditions, and is not limited to the surface of the active material," the researchers state in the abstract to their paper.
They go on to propose a switch that is operated by field-driven formation and current-driven destruction of filamentary conductive pathways and report on/off resistence ratios of 10^4:1 and higher. The conductive pathways are 10-nanometers in diameter or smaller and programming currents can be as low as 2-microamps, with transition times on the order of nanoseconds.
"Our ReRAM memory chips need just a thousandth of the energy and are around a hundred times faster than standard flash memory chips," said Tony Kenyon, one of the researchers at UCL's department of electronic and electrical engineering, in statement. The UCL devices also display a continuously variable resistance that depends on the last voltage that was applied. This is an important property that allows the device to mimic how neurons in the brain function.
UCL ReRAM test circuits on broken wafer. Source: UCL/Adnan Mehonic
The behavior was discovered by accident as researchers at UCL were working on silicon-based light emitting diodes but noticed that devices appeared to be unstable. UCL PhD student, Adnan Mehonic, was asked to look specifically at the material's electrical properties and he discovered that the material was flipping between conducting and non-conducting states.
The technology has potential application beyond memory storage. The team is also exploring the use of silcon-rich silica as a logic switch for use in processors.
UCL team has been backed by UCLB, the technology transfer company of UCL, and has recently filed a patent. UCL said UCLB is in discussions with a number of semiconductor companies.
Here is an important detail, many seem to have overlooked :
["The UCL devices also display a continuously variable resistance that depends on the last voltage that was applied."]
Current Flash already allows Multi-Bit cells, so any alternative is going to have to have the same ability, in order to be far enough ahead of flash, to be worth the bother.
My lasting impression of MLC is someone showing a distribution with tail bits and saying "this shows MLC capability"..ok.
If the resistance depends on the last voltage applied, that may not be so useful for memory, but maybe for a memristor.
ReRAM is the most awaited replacement of Flash RAM if it get successful implementation as compared to FlashRAM. It really need Chemical Engineers and Nano Technology experts working behind the early invention of the possibilities associated with ReRAM.
I think the key to a successful product introduction is the controller in front of the bucket-o-bits. As long as the controller interface is such that the CPU is indifferent to what's behind it then they can quickly get competitive products out into the market in standard storage formats. That'll free them up to sell the advantages of this technology over NAND - faster, lower-power, lower costs etc.
Interesting claims they have on being first. Others besides James Tour have published recently on SiOx RRAM--Jack Lee of UT-Austin, and Hyunsang Hwang of Gwangju University, Korea with Luigi Pantisano of IMEC. There are also a few published reports from the '70's, one by MJ Howes and another by RM Anderson.
Two more examples, which were previously cited by Blaise Mouttet, infamous Memristor denier:
D. R. Lamb and P. C. Rundle, "A non-filamentary switching action in thermally grown silicon dioxide films", Br. J. Appl. Phys. 18, 29-32 (1967)
Dow Corning also did work on SiO2 ReRAM in the early 1990's (US Patent 5283545).
I prefer memristor realist rather than "memristor denier." I initially supported HP's work on memristors until I figured out that it was mostly propaganda to support their business agenda rather than legitimate science.
But why is it important how they call their device...
Isn't only important thing what device can do?
If HP produce device to match flash or to do something useful - call it memristor, or ReRAM or whatever it is irrelevant.
The "memristor" represents a mathematical model formulated by Leon Chua in the 1970's. If the memristor model is a wrong mathematical model for ReRAM then its acceptance will probably hinder scientific development.
If the figures of merit are speed, size, energy and so on, calling something memristor or not wouldn't affect it.
Nobody really ended up with any real disadvantage for calling ReRAM 'a memristor' as nobody is using Chua's model to improve any important parameters anyway..
People are using name memristor (instead ReRAM) as it is more catchy but nobody really backs up anything on Chua's model.
I simply can't see any real problem with that.
If I got it right this is the bulk effect and Rice's study is more related to SiO2 substrates where the formation is on the surface - one of the reasons it doesn't work in the air.
NDR effect is known for many years but the operational device with properties to match flash and be cheap is different thing...
In the overall cost of a RRAM chip, deposition of a few hundred angstroms of most metal oxides (eg. TiO2, Al2O3, HfO2, NiO, etc) is not significantly more expensive than deposition of silicon oxide, so it's not likely a cost driver.
On their website they stated 3k, Tour's group
10k, and Hwang's group 100k.
So silicon dioxide is more or less similar to other metallic oxides in these terms.
HP will come there first anyway - but it is interesting to see if silicon can fight back once again.
Very nice article in IEEE Spectrum about SiOx RRAM.