Contributor Ron Neale joins us again to review a recently-published article in the journal Nature Scientific Reports. While the main focus of the paper is on using a nitrogen environment to generate stable memory selectors from ZnTe, it also provides some new inputs through which he finds further support of his theories of Forming and device behavior.
A recently-published Nature Scientific Reports article by a research team from Hanyang and Kunsan Universities in The Republic of Korea focuses on the solution to a manufacturing problem, but their research produced more than that. (Bidirectional-nonlinear threshold switching behaviors and thermally robust stability of ZnTe selectors by nitrogen annealing, by Gabriel Jang et. al.) This team has produced new results which confirm that a change to threshold switches based on polycrystalline binary Zn0.5Te0.5 compositions may provide the answer to BEOL (Back End of Line) elevated temperature memory array fabrication problems which are not directly associated with the operation of the device itself.
This new work has also provided us with a novel technique for establishing that threshold switching is filamentary, as well as some insights as to the dramatic effects of Forming and the thermal nature of switching.
As you read this post I recommend for you to open the Nature article and review the figures side-by-side with what I am saying here. The Nature article is a little unusual in that the text refers to graphs that don’t appear in the article itself, but are provided as “Supplementary Information” in a file that must be downloaded using an obscure link almost at the bottom of the page, below all of the article’s references. This section consists of no more than a hotlink to a Microsoft Word file. This file contains the most interesting figures in the article.
The key problem is that the amorphous materials most frequently used for selectors in memory arrays are meta-stable and will eventually phase change or crystallize at some temperature. If the operation of a device depends on retaining its as-fabricated state but it suffers a phase change, or is crystallized during some BEOL fabrication step, the device can no longer serve its useful purpose.
The very significant move from the more normal amorphous to polycrystalline material does not remove the elevated temperature BEOL degradation problem. The degradation that ZnTe material suffers is not from an amorphous-to-crystal phase change but from phase separation. Notwithstanding that difference, both types of change result in selector device degradation when subjected to elevated temperatures.
The basic claim of the new work is simple, annealing of ZnTe in nitrogen produces thermally robust material, and the resulting selector devices are similarly robust, whereas the more conventional processing step of annealing in vacuum results in phase separation and an unstable material and devices. The team were able to establish that nitrogen annealing resulted in a material which, when subjected temperatures as high as 400° C, did not undergo any phase separation.
For the structural analysis films in the pristine, or as-fabricated state were annealed at different temperatures between 200° and 400° C in vacuum or nitrogen and then subjected to analysis by X-ray crystal diffraction (XRD) and X-ray photo electron spectroscopy (XPS).
The results of that analysis provided a clear indication that the phase separation changes which occurred in the vacuum-annealed films did not occur in the nitrogen-annealed films. An additional conclusion from the structural analysis was that Zn has little or nothing to do with the threshold switching, it merely provides the scaffold which allows the Te to carry the switching load.
This raises the question: does annealing in nitrogen act to nitrogen-dope ZnTe and provide a stable structure, while vacuum annealing results in the desorption of any absorbed gas destabilizing the structure?
From all of their detailed structural analysis the research team established, with a high degree of certainty, that ZnTe material will be able to deal with any BEOL problems stemming from elevated temperatures.
ZnTe device performance
Introducing a new selector material as a solution to a BEOL fabrication problem only has value if at the same time the electrical characteristics of the devices meet those required for use as a memory array selector. The Korean team have gone a long way towards satisfying both requirements.
A first look at Figure 1(c) of the main article (not the Supplementary Information) suggests that the threshold switching symmetry in the I-V characteristics of ZnTe-based selectors will be able to offer a route to future stacked memory arrays of ReRAMs and PCMs, as the matrix isolation device.
For the present generation of PCM products ZnTe has the potential to remove arsenic from the processing line. Arsenic is currently used for doping GST to turn what is normally a memory device material into a non-crystallizing selector material. The presence of uncontrolled arsenic will have an effect on the silicon and, perhaps more importantly, it could cause the GST memory write characteristics to change.
The device structure used by the researchers is illustrated on the left-hand side of the figure below. It is a sandwich of polycrystalline ZnTe with an FCC (Face-Centered Cubic) structure between tungsten electrodes. The Supplementary Information reveals that the active material thickness for the experimental work ranged from 10nm to 60nm while the main body of the article tells us that the device area, defined by the size of the bottom electrode, ranges from 420nm to 1,414nm. The bottom electrode also acted as a close-coupled resistor, or heater.
On the right hand side, the top illustration shows the likely structure of the device after Forming, with a filament of modified material (purple) consisting of many conducting percolation paths (yellow), while the lower illustration shows a speculative view of the conductive region (pink) expanding radially after threshold switching.
Bidirectional endurance testing of ZnTe selectors involved measuring the current while switching with a 2-volt pulse, followed by reading with a 1-volt pulses. On and off currents of between 10-4 and 10-6 amps respectively were recorded for endurance of over 107 cycles. An interesting feature of those tests was the off state for one direction of current is slightly different from that of a reversed current when reading with 1 volt. This may be evidence of a contact effect for a device with one electrode larger than the other.
Overvoltage switching delay time measurements for ZnTe devices were consistent with what is observed in other competing amorphous selector materials. A 3.5-Volt pulse applied to a device with a threshold voltage of 2 Volts resulted in a delay time of 50nsecs, for an overvoltage of 0.5Volts.
The team confirm that the pre-threshold switching and post switching I-V characteristics of the polycrystalline ZnTe devices are also similar to those reported for other amorphous chalcogenide compositions such as doped-GST and GeSe, from which it is implied that the same or similar switching mechanism is involved.
Modelling the Device
As with the more well-established selector materials pre-threshold switching conduction I-V characteristics were modelled as a low-voltage Ohmic conduction followed by Poole-Frankel conduction leading to the point where switching occurs.
The equation used by the researchers models pre-switching conduction as electrons hopping between traps or defects and, as well as including the distance between traps, it also includes trap concentration, the time to escape traps, and the barrier height associated with the trap, plus device thickness and thermal activation effects.
The authors employed an ingenious technique to establish the accuracy of the detailed equation they use to describe pre-conduction breakdown. They ran sets of simulations with a series of different values for the constants in their equation to obtain sets of characteristic curves, and then found a best-match with the experimental results obtained from device structures. This enabled them to determine the constants of the equations.
Using their best-match technique, they were able to establish values for all of the constants in their equation. One of the constants is the distance between traps, and the best-match was found to be 6.5nm. This would imply that for the thinnest device (10nm) the pre-switching electrical characteristics are determined by a single trap in the inter-electrode space.
Threshold Voltage vs. Thickness
There is one other noteworthy characteristic the Korean team provide. It is threshold voltage as a function of thickness. The figure above is an embellishment of Figure S3 (b) of the Supplementary Information. This is the blue line measured on the left vertical axis. I have added a red line that I will explain shortly. The blue line depicts threshold voltage over the active material’s range of thickness (t) in nanometres) from 0nm to 60nm, and follows a simple equation of the form.
Vth = (0.025) x thickness (t) + 0.5
Oddly enough, the constant and the Y-axis intercept tell us that there is a finite voltage at zero thickness! It is possible that this observation has something to do with a voltage drop across a series resistor or a semiconductor-metal contact effect. Another simplistic explanation would be that the act of Forming reduces the thickness of the active material. This could be caused by the ingress of tungsten, which would offset the values of Vth versus thickness to the left by 20nm. While that might be possible for the thicker devices, for the 10nm thick devices this would be unacceptable.
For illustrative purposes I have added a line representing the instantaneous power at switching as function of threshold voltage (red line, right vertical axis). The data for this line was calculated by digitally extracting it directly from Figure S3 (a) of the Supplementary Information. This power data will be discussed in relation to device modelling in the later paragraphs of this article.
The ZnTe Devices Require Forming
Yes, like other selectors that have gone before, ZnTe selectors require Forming. I happen to believe a clear understanding of what happens during Forming is a key to any future NV memory success. (See my earlier series: NV Memory Selectors: Forming the Known Unknowns.) This new work helps in that direction.
From the main article’s Figure 1 (c) we can see that the ZnTe selectors used for the reported study have Forming voltages in the range 3 to 4 Volts for subsequent operating threshold switching values of from 1.3 to 1.4 Volts, with holding voltages of about 1 Volt, which may include a contribution from the integrated series resistor or metal semiconductor contact effects.
As I have illustrated in this figure, the research team suggested a model of operation for their ZnTe selectors is one based on the theory that Forming creates a modified filament of material which completely links the upper and lower electrodes. The team uses the term “CFs” for conducting filaments, suggesting that they consider that a filaments-within-a-filament model more accurately describes the Formed material, with more highly conducting percolation paths as the secondary filaments.
This modified filament material is where the threshold switching occurs or at least is initiated. It is a model which is generally in accord with the universal model offered by this writer in the post Observations on the “Universal Law” for NV Memory Cells. The team’s experimental results almost inevitably lead to a conclusion in favour of filament formation, not just for a filament that is present while current is flowing, but rather for a mixed picture of a filament of permanently-modified material in which the after-switching current flows. The degree to which the current expands into the pristine material, as illustrated below, is unknown, although included as a suggested possibility.
Figures 1c and 2b of the Nature article show “S” shaped (V-I) curves. The fact that the negative resistance of threshold voltage switching is “S” shaped already provides a rather strong pointer in the direction of a conducting filament. The results on which the Korean team based their filament formation conclusion raise some interesting questions and are worth special attention.
The main article’s Figure 2 (a) shows that prior to Forming, and irrespective of the area of the device, the pre-switching current density J, as a function of voltage (J-V), is the same for all device cell sizes, indicating that the whole of the area of the device is equally conducting.
After Forming, the (I-V) characteristics for the same set of devices also overlapped (shown in Figure 2 (b) of the main article) which leads to the conclusion that the result of Forming is a filament of modified more conducting material in the pristine ZnTe which threshold switches when sufficient voltage is applied.
As well as evidence of a filament, the article’s results appear to point to a remarkable structural change of some sort caused by Forming. This is not addressed by the researchers in their paper, other than to mention percolation paths, which must include the possibility of precipitated zinc or metallic Tellurium.
The data from the overlapping (J-V) curves in Figure 2 (a) of the main article indicates for the pristine material, at voltage value of 0.5 Volts, that the current density is 0.1 Amps/sq-cm for all device areas. For the device with a smallest electrode diameter, 420nm, this calculates to a current of 1.38 x 10-10 Amps. This calculation can be repeated for the rest of the devices with diameters of 420, 618, 922 and 1,414nm. (I chose 0.5V because at this voltage the devices are considered to be in the Ohmic or linear conduction region.) The data from the overlapping (I -V) device characteristics in the main article’s Figure 2 (b) indicates that for the Formed material, also at 0.5 Volts, the current is 6.8 x 10-8 Amps, again for all four devices in the measured set.
As a starting point, assume that, for the smallest device, the whole of the volume of material directly between the electrodes is modified as a result of Forming and is, in effect, the filament.
This would represent a current change and a conductivity change by a factor of 680/1.38 = 490.
But what if only a portion of the material’s volume was changed by Forming? A further simple calculation then shows that if the filament was, say, one tenth of the area of the device, the conductivity of the Formed material must have changed by a factor of greater than 4,900. While a change by a factor of 490 for the material in the larger filament would be large, a non-volatile memory-like change by a factor of nearly 5,000 would appear unlikely, or would certainly require an explanation to account for any other possible implications of that change.
If, as it appears after Forming, the I-V characteristics still follow the trapping model of conduction, then one simple answer might be that the density of shallow traps has increased. This carries with it the implications of significant structural change. As mentioned earlier, an alternative might be the precipitation of Zn metal, as opposed to the ingress of titanium, as was mentioned earlier.
In contrast to careful and controlled annealing, steps are used to stabilize the material prior to forming the active material, that is the material where the threshold switching initiates appears to have undergone some significant structural changes during forming.
The researchers’ description of the results of Forming suggests that while the polycrystalline pre-Formed ZnTe layer has zinc-blended crystal structures, that structure of large crystallites is destroyed by Forming, and when high electric fields are applied to the resulting structure percolation paths allow conduction within the modified area. This might tempt one to suggest that conduction along the more numerous grain boundaries could account for the percolation paths and conductivity change.
Unresolved Issues
The devices used for the reported experimental work are large. When ZnTe based devices are scaled to sub-20nm dimensions, as will be necessary if ZnTe is ever to progress to competitive memory products, then the formed material will be the complete device, and will only a slender relationship to the nitrogen annealed structure, unless the diameter of the filament somehow scales with the area of the device.
While it is acceptable to address and solve BEOL fabrication problems on material in bulk, how valuable is that data when the material is in device structures, and after they have been subjected to a switching event? Is it safe to draw conclusions regarding the performance of selector materials in the face of evidence that some very significant unexplained structural changes appear to be occurring during the forming step?
One could argue that at least the pristine material surrounding the filament will be stable, providing, that is, that Forming has not resulted in radial movement of some constituents of the composition to change the composition of the filament and some of the surrounding pristine material. In the past there has been some evidence that different levels of nitrogen in ZnTe can modulate the electrical conductivity. (See: Nitrogen doping of ZnTe for the preparation of ZnTe/ZnO light-emitting diode, September 2013 Journal of Materials Science 48(18):6386-6392 DOI: 10.1007/s10853-013-7438-y.)
Then there is to me yet another important question: What is the operating temperature of the conducting filament during and after switching? Given that information, one would be able to suggest that the pristine material around the filament would be safe from phase separation. As a guide, it is known that the current that resets a phase change material is able to bring that material to a temperature of 600° C prior to quenching. The selector, as a matrix isolation device, certainly for PCMs, will also carry that current and is the same type of structure and has the same order of device resistance.
Considerations of filament temperature raise a further question: After threshold switching does the conducting filament expand radially into the region of the active material around the initiating filament? There might be a clue if, like chalcogenide selector materials, the post-switching resistance characteristics (Rs) can be described by an equation of the form Rs =Vc/I + Rc (i.e the characteristics of a constant current-density filament expanding and contracting in series with an internal or external resistor). This would result in a constant voltage Vc across the media.
If those constant voltage characteristics are not observed, and the voltage rises as the current increases, then a model, with all the current flowing solely in what must be a highly modified filament, would be more likely.
If heating of a narrow filament initiates switching, then it might be expected that power at switching would be proportional to the thickness of the device. To that end there is one other characteristic which I think it might be possible to extract from the data on ZnTe in the Supplementary Information’s Figures S3 (a) and (b), which shows threshold voltage and current for devices with thickness over the range 10nm to 60nm.
Making a Model
With all of this background can we model the selector’s performance?
If, in Figure S3 (a) of the Supplementary Information, we draw a line through the switching current and voltage at the point of the switching transition, it traces an increase in power by a factor of ~6 for an increase in ZnTe film thickness by a factor of six (60nm/10nm). (My actual best numbers are a factor of 5.55 +/- 0.3). If correct, that result would match a model where the modified link or percolation path resulting from Forming must be heated along its length to the same temperature to initiate switching.
For illustrative and discussion purposes, I have extracted from the Supplementary Information’s Figure S3 (a) an average value of the instantaneous power at the point where threshold switching occurs. This power is 9.6x 10-8 Watts/nm for each value of thickness. These values are shown as the red line in my earlier figure that compares threshold voltage versus thickness. Using that extracted data at the point of switching I have produced the set of (I-V) curves below. My extracted values of power for each device thickness occur at the intersection of a region of what would be the straight line of Poole-Frenkel conduction and the vertical switching transition. As readers will find [Ref1] these curves are almost indistinguishable from the curves in the Supplementary Information’s Figure S3 (a).
This does give rise to the possibility of a different model of device operation, illustrated below, where Forming results in a filament region of more disordered polycrystalline material with more grain boundaries than the annealed ZnTe. Conduction along grain boundaries then provides the percolation paths which lead to threshold switching. The researchers used the term “defected Te and Zn” which might be considered something similar. However, there is the possibility that the more polycrystalline material, when heated, becomes more conducting and more disordered and melts; there is in the phase diagram of ZnTe a solid-liquid phase in the 400° to 500° C range. In this model the liquid state material provides the percolation conduction paths, with liquid Tellurium one possible candidate for these paths. While current is applied that state is maintained. When the current is removed, the conducting filament returns, or is quenched, to its disordered high resistance mixed amorphous-polycrystalline off state, with the amorphous material dominating the pre-switching (I-V) conduction characteristics.
Conclusion and Caution
The Nature article is an exemplary piece of structural analysis by a team in Korea which suggests that ZnTe might be capable of solving the problem of material degradation caused by elevated temperature during the fabrication of NV memory arrays, while at the same establishing that ZnTe, in selector device form, has the potential to provide useful switching characteristics as a memory matrix isolating device. This latest work has also provided some interesting insights on Forming and switching for this new polycrystalline contender as the selector device for the stacked NV memory arrays of the future.
The move from amorphous to polycrystalline material for thin film selectors and the fact that the pre-switching and post- switching electrical characteristics are identical, carries with an interesting implication. It would appear that for any high resistance material, when a sufficient voltage is applied it will threshold switch, as long as a sufficient number of traps are available to provide a percolation path for conduction. The post-switching conducting state may be the result of: melting, some other form of structural change such as electromigration, phase separation, bond flipping or crystallization. The conducting state before threshold switching does not appear to need to have to direct connection to that which occurs after switching.
I have tried to contact the authors with a view to obtaining comment and clarification on some of the points and questions I have raised. To date that effort has been unsuccessful, they will be provided with space for any comments they would like to add. As always, some caution must be observed in drawing conclusions with respect to numerical data extracted from the pages of published work.