Resistion; The need for forming brings with it at least at two problems and maybe more. If forming requires conditions different normal operation then that adds a burden to the initial user of the as-born device or as you suggest the test budget. In that it is first necessary to apply a forming cycle to every cell prior to testing. This would not be negligible as you suggest for a 16Gbit device. The second consideration is relaibility, does a device in its just formed state have the same level of reliability as a device in its F+1 write/erase cycle state.
Example: Long ago in the early days of Phase Change Memory (PCM) if the devices were made in the amorphous state then sometimes the natural annealing of the amorphous material in its as deposited state would cause the threshold voltage to drift above the value at which the array could be operated. With very small arrays the techique we used was to thermally crystallize the material during fabrication then at the test stage run multiple reset pulses, to erase all of the active material in each pore, then test. Not something I would recommend for 16Gbits. The later introduction of the edge contact (dome/mushroom) PCM solved that problem to a degree and allowed PCMs to be fabricated in the crystal conducting as-born state; even so the initial pulse is still a dome forming pulse.
My understanding is the initial state for resistive memory determines the initial operation. For an insulator like HfO2, there would be an initial step similar to breakdown, but not as drastic as in antifuse OTP. For an initially conducting state, the initial operation would be the RESET, which should be very high current for something metallic.
Compliance - we spent over 5 years gettting rid of the bugs. So, we are entering now a development stage - as you may know, in industrial research we have first a prove of concept and then R&D to show that it is real and useful. After that there is an engineering benchmarking phase, the develpment phase, in which design and processes cand comply with product specs and reliability. We are at this point. The teams separate - one continues further R&D for other applications and the technology team does the Test Chips at a given node, Typically this is about 2 years. We are entering that curve now. So 2 years is a good number. The nice thing here is that this is fab compatible and we do not need Platinum electrodes as the otherRRAMs usually do.
Great question. The answer is yes if we can start with a p-type oxide. This covers than all the perovskites, like SrTiO3 and most d-block and f-block elements that have an oxide. But, HfO, NbO, TaO are not p-type. So, when they make HfO, the filaments are needed and I cannot really tell you what is the detail physics of how these materisl open and close fialments, But, I can tell you that so far, I see only sandwiches of different stoichiometry or even different materials (like Tio/TaOx) in order to make this work, The interface between two non-stoichiometric oxides is a very rich area for new things to happen, even superconducting phases have been detecte, But, as a practical matter, banking on that ina manufacturing process for memory - remember Gbits means 1 billion resistors have to be perfect - is a bit too optimistic.
Yes. True. But counter doping these GBs - like grain boundary decoration is a way to eliminate that. So, if you read the large post I just made to resistion, you will appreciate that the story is a bit more complex than accidental microstructure mishaps. The conduct/insulate switch is at the heart of this, not whether there is or not a filament. Even with a filament, what makes it become insulating....it is not reasonable to believe that a metal streak can become an insulator, So, as you will read, it is an insulator that can become a metal-like material and never a metal that can become an insulator. So, filament termination is where the locus of an Insulator to metal transition can occur. Better than to lose the need of making a filament all together and prepare the material to naturally go through a MIT as prdicted by quantum theory (Hubbard Hamiltonian). So, polycrystallinity is not an issue here because we do not need electronic states that are "extended" like in semiconductors (and metals), but only a locallised electron-electron strong correlation as described in the other post. I know, it is weird, but physics is better than metaphysics. And, quantum phenomena cannot be described by classical pictures.
Compliance- please read post I just made to resistion and ask yourself if frying a device with higher voltage and high current to form random filament creation is good in say a 10 nm technological Node. A bad start in making so that you have rsistance switching.
Resistion - we should analyse what doping really means: in semiconductors, doping is only about 1% at the most. So, you are right, we are cahnging the material with higher than 5% doping levels. But, the doping in semiconductors is to produce in the lattice, places wher (for example) Phosphorus, which is group five, would covalently bind with Silicon (group IV), and have an extra electron without a bond. Then ta the room temperature, that electron is free to move in what we call conduction band. This thermal activation is key, as phophorus now is ionized (Ndonnor is positive). We then can accelerate that electron by applying a field and thus we have n-type conduction in Si. Here, we have a different scenario for the effect of doping. Nickel has incomplete 3d shell, it is like many other transition metals, a case where the 3d shell has less than 10 electrons, does not get filled and already a 4s orbital is created. When bonded with oxygen or another suitable "ligand", these orbitals do not bond nice and covalently with the cation. It is a complex situation depending on hoe the oxygen wraps around the Ni ion. This wraping around is called the coordination number - and how many and what charge arrangement, determines what charge the Ni will be able to be. In the case of NiO, we have Ni(+2) and O(-2). And the wraping around is a perfect octahedron with 6 vertices, each sharing 1/6 of an oxygen atom. In the middle of the octahedron is the Nickel ion. This is in theory. In practice, the oxygen atoms simply are not perfectly counter balacing the Ni "charge". So, if an oxygen atom is not there, the area is uncompensated and the Ni simply emits an electron to ounterbalance it. This happens with any transition metal, as the electrons can hop in and out of available empty states (like in the case of Ni, we have the shell with only 8 electrons, Cobalt only 7, Titanium only 1...etc). This hopping is a form of local trap if the oxygen vacancy is there. And, it can be undone with raising the temperature. In the other hand, if we could keep all Nickel ions at a fixed and suitable oxidation number, that is a suitable coordination with its surrounding of perfect oxygen atom compensation, we could proceed with the quantum rearrahgement of electrons that renders NiO an insulator - and that is fone in two steps in nature: first, spin is maximized by what is called Hund's rule. This makes the electrons in the d8 orbital to split states and be at the maximum same spin direction. Not to violate the Pauli exclusion principle, these two electrons cannot be in the same state of energy. Then, we have other Ni ions with the opposit spin that sees the opportunity to jump to the ion that has an unpaired electron of opposite spin, and then this does happen. The double occupancy of this hybrid state creates a repulsion between these electrons and an energy barrier, purely electrostatic appears which makes the NiO become an insulator even though it has this 3d8 energy configuration (9 and 10 being unfilled states would be a conduction band), In 1937, two dutch physisist showed that the normal Wilson-Bloch Band theory would be violated by transition metal oxides because they technically should be conductors and they are not, Neville Mott took that to heart and explained it many years latter with the argument that I just used. In 1963, in a brillliant series of papers, John Hubbard showed that the electrostatic repulsion is a real thing and was able to show how the metal to insulator transition is controlled by this double occupancy of the hybrid state that came out of the old 3d8. In this scenario, doping takes a very different meaning than in a semiconductor. What we call doping here, is a purposively introduced ligand to take place where the oxygen is gone, There are only a few of these. In the high school chemistry experiments, Cobalt and Nickel compounds show brilliantly vivid different colors when mixed with different levels of amonia. In that case, in a liquid, the amonia is a ligand that whenbound to the transition metal, will form different energy windows (like a local absorption/reflection energy deifference) such that different light frequencies get absorbed or not. This is a result of manipulating the oxidation number or coordination "sphere" around the TMO. In our case, we used CO as the most common and stable ligand in the solid state reaction that the material goes through, and in this way, the coordination of Ni is at the very start kept as +2 and the other quantum arrangements of electrons as described above can proceed naturally. That is, the double occupancy will make the material an insulator and if the potential energy for some reason would change, it would become a conductor. The question is, what changes the potential energy well (its spread due to the net positive charge of the core electron+nucleous). The answer is free electrons surrounding it. As an insulator, the second electron became bound due to the double occupancy. How to make it leave? Here is the elegance of nature which is exploited in CeRAM - Electron injection at Vset increases the electron density up to the point that the reach of the positive ion-core positive potential well is reduced and the bound second electron is released, thus the electrostatic repulsion is gone and the gap is gone, and thus a conductive state appears. So, as you can see, in a very thin film of thickenesses in the order of 69 nanometers and bounded by two metal electrodes, the Schottky barriers of the insulator/electrode and the lattice termination is a plethora of defffects such as oxygen vacancies. This is always the same because thermodynamic equilibrium forces thes defects to always exist. What scientist did up to now, was to form filaments of the metal and bank on a memory effect due to electrochemical reactions that appear due to these random changes in oxidation number near the surface. But here, we fis the coordination sphere and move the electron-electron double occupancy region to a thin area in the middle, So, we have a controlled and reversible electron in - electron out "reaction" in the hybrid orbital ( now 3d9). This is a quantum level switch completely regulated by the electron injection or electron deficit created by the lower voltage hole injection. Without the oxygen vacncy traps and without the gross metallic streak defects caused by filament formation, we have a pure quantum phase transition as demonstrated by switching at 4K, which we show in our web site. The simplicity of the operation and the implication of ultrafast switching is a feast to a Many Body Theorist in physics, Here, perhaps for the first time we have command of the Metal_Insulator transition isolated as if in an active region and controlled by electron screen electron density variation similar to the effect known as Degeberacy pressure. Only as recent as 2012, scientist could show that applied pressure of 2.4 Million Atmospheres could make NiO become a metal. Here we can use the mesoscale of the device (thickness) and inject enough electrons to have a density-pressure equivalent(more on this in coming papers). In conclusion, when the coordination sphere is fixed, the electrostatic potential of double occupancy of electrons is of such a high energy that to undo this we cannot even show at up to 400 C. In the case of a filament memory, already at 200s, the detraping of electrons in oxygen vacancies is destroying the insulating state. So, back to your comment, it is like a New Material, but not really. The NiO is just being corrected to behave as if it was a bulk sample and the electron density limited by the compliance is just right to allow a beautiful quantum switch to come into being which is nonvolatile and robust in storage temperature. We are making this quantum switch not only for memory but also for logic (if possible) as if the quantum phase transition is known to be in the 10s of femtosecond order. Thank you for your interest as always.
The Other Tesla David Blaza5 comments I find myself going to Kickstarter and Indiegogo on a regular basis these days because they have become real innovation marketplaces. As far as I'm concerned, this is where a lot of cool ...