Transient Thermometry and High-Resolution Transmission Electron

Subscriber access provided by University of Sussex Library

Article

Transient thermometry and HRTEM analysis of filamentary resistive switches Jonghan Kwon, Abhishek A. Sharma, Chao-Yang Michael Chen, Andrea Fantini, Malgorzata Jurczak, Andrew A. Herzing, James A. Bain, Yoosuf N Picard, and Marek Skowronski ACS Appl. Mater. Interfaces, Just Accepted Manuscript • DOI: 10.1021/acsami.6b05034 • Publication Date (Web): 28 Jun 2016 Downloaded from http://pubs.acs.org on July 11, 2016

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

ACS Applied Materials & Interfaces is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Applied Materials & Interfaces

Transient thermometry and HRTEM analysis of filamentary resistive switches Jonghan Kwon,1,* Abhishek A. Sharma,2,* Chao-Yang Chen,3 Andrea Fantini,3 Malgorzata Jurczak,3 Andrew A. Herzing,4 James A. Bain,2 Yoosuf N. Picard,1 Marek Skowronski1 1

Department of Materials Science and Engineering, Carnegie Mellon University, Pittsburgh, PA 15213, USA 2 Department of Electrical and Computer Engineering, Carnegie Mellon University, Pittsburgh, PA 15213, USA 3 IMEC, Kapeldreef 75, Leuven 3001, Belgium 4 National Institute of Standards and Technology, Materials Measurement Laboratory, 100 Bureau Drive, Gaithersburg, Maryland 20899, USA

Abstract We present data on the filament size and temperature distribution in Hf0.82Al0.18Ox-based Resistive Random Access Memory (RRAM) devices obtained by transient thermometry and high-resolution transmission electron microscopy (HRTEM). The thermometry shows that the temperature of the nonvolatile conducting filament can reach temperatures as high as 1600 K at the onset of RESET at voltage of 0.8 V and power of 40 µW. The size of the filament was estimated at about 1 nm in diameter. Hot filament increases the temperature of the surrounding high resistivity oxide, causing it to conduct and carry a significant fraction of the total current. The current spreading results in slowing down the filament temperature increase at higher power. The results of thermometry have been corroborated by HRTEM analysis of the as-fabricated and switched RRAM devices. The functional HfAlOx layer in as-fabricated devices is amorphous. In devices that were switched, we detected a small crystalline region 10-15 nm in size. The crystallization temperature of the HfAlOx was determined to be 850 K in an independent annealing experiment. The size of the crystalline region agrees with thermal modeling based on the thermometry data. Scanning TEM coordinated with electron energy loss spectroscopy could not detect changes in the chemical make up of the filament. * Both authors have contributed equally. ACS Paragon Plus Environment


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Keywords RRAM, thermometry, filament, HRTEM, STEM-EELS

Introduction Resistive Random Access Memory (RRAM) has emerged as a leading candidate for storage class memory.1 Such devices typically consist of a metal/insulator/metal (MIM) structure and exhibit switching phenomena that result in a non-volatile change of resistance by the application of electrical bias. It is now widely accepted that shunting and rupturing of a local conductive path (filament) is responsible for the two resistance states. Various material systems have been explored for RRAM application, including HfOx which is commonly used as a gate dielectric in CMOS process.2,3 The HfOx-based RRAM devices have shown excellent switching performance in terms of switching speed (on the order of nanoseconds), superior endurance (> 1010 cycles), low-energy operation (< 0.1 pJ), and scalability ((10×10) nm2).4 Recently, alloying of HfOx and AlOx has been explored as means to enhance device performance by increasing both endurance and reliability.5–7

Although there has been a large body of research conducted to quantify the processes involved in resistive switching, many of the aspects of the process remain unclear. The most important is the nature of the structural change that makes the filament conducting. In HfOx, along with other wide band gap oxides such as TaOx and TiOx, the switching is attributed to redistribution of oxygen vacancies by the applied electric field. Accordingly, the filament in the reminder of this report will be defined as the small diameter cylinder composed of reduced conducting oxide.

ACS Paragon Plus Environment

Page 2 of 41

Page 3 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Such a filament is nonvolatile and retains its properties after the bias is terminated. Its three fundamental parameters are the diameter, the composition (vacancy density) and the temperature it reaches during switching.2 As pointed out below, the close-to-stoichiometric oxide surrounding the filament can conduct significant current density when heated. This is not counted as filament.

Since resistive switching behavior is directly tied to the characteristics of the underlying filaments, methods for estimating their size are of fundamental importance and have been the topic of an ongoing research effort. By determining the filament size, a deeper insight into the switching mechanism can be gained which will then enable the engineering of superior devices. There are large differences in filament size estimates in the literature. Some experimental results place the filament size in the 100 nm and over range by using a variety of characterization tools: transmission electron microscopy (TEM),8 X-ray spectroscopy,9 and atomic force microscopy (AFM).10,11 However, device scaling suggests that filaments are below 10 nm in diameter.4 This dimension has also been assumed by most modeling efforts.12–14 While assumptions are not equivalent to evidence, the modeling produced a self-consistent picture of both the SET and RESET processes as being due to the thermally activated motion of oxygen vacancies.12 A change from 10 nm to 100 nm filament diameter would result in the filament temperature lower by hundreds of degrees at the given dissipated power and virtually freeze the motion of vacancies, making these models unfeasible. On the other hand, some of the TEM results15 and AFM-based tomography16 indicate the filament size at about 10 nm.

Although the above results appear contradictory, it is possible that the filament size in different devices formed in different conditions vary widely. Most of the experimental data mentioned



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

above have been obtained from devices that have dissipated significant power during SET (≈ 20 mW).15,17 Notable in this respect is the paper by Kwon et al..15 While the structural features analyzed by HRTEM indicated the Magnéli phase inclusion of about 10 nm diameter, the crater formed by the electroforming process was about 10 µm and in concert with the level of dissipated power. HRTEM cannot analyze features of this size and by necessity the images showed only a small fragment of the entire crater possibly missing some of its parts. Recently fabricated devices formed with a low current compliance achieved by on-chip transistors and dissipated power of (10 - 20 µW) are likely to form a much smaller filament.4 However, these results do require experimental verification.

The important consequences of the filament size are the filament temperature generated during switching and the consequence for the ultimate scaling limit. Clearly, the high temperature expected in small diameter filaments would affect the reliability of switching devices.18 On the other hand, high temperatures are necessary to achieve fast switching.12,14 The estimates of the filament temperature so far have largely relied on simulations that make a number of assumptions about the conductivity, size, and geometry of the filament.14,19,20

In the present work, we attempt to resolve the contradictions among recent studies by performing a detailed electrical and high-resolution microstructural analysis of nanometer scale devices with on-chip ballast. All devices used in this study dissipate orders of magnitude lower power than the above examples and exhibit stable switching. We use high-speed pulsed thermometry to estimate the temperature excursions in nanoscale RRAM devices based on I-V characteristics and validate the physical picture by direct evidence of microstructural changes in the form of local


Page 4 of 41

Page 5 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


crystallization.21,22 This work uniquely juxtaposes HRTEM with thermometry to provide insights into: (1) the interplay between the device size (nanoscale) and the filament heating in devices that operate at powers much more relevant in RRAM applications; (2) identify the peak temperature - limiting mechanisms in the RRAM cell; (3) validate these mechanisms and thermal excursions through the use of direct HRTEM studies; and (4) confirm the role of power compliance on filament size and thermal excursions. The results indicate that the filament diameter was overestimated by more than an order of magnitude while its temperature was greatly underestimated.

Experimental Materials Crossbar devices with stacks consisting of TiN (40 nm) / HfAlOx (5 nm) / Hf (10 nm) / TiN (30 nm) have been fabricated on SiO2/Si substrates (details of the device configuration are provided in supporting information (Figure S1)). The functional HfAlOx layer was deposited via atomic layer deposition and the bottom TiN and the top Hf/TiN electrodes were sputter deposited. The Hf layer acts as an oxygen getter layer that allows formation of the oxygen deficient, substoichiometric HfAlOx functional layer, facilitating creation of the conductive filament under applied electrical bias. The active area of the devices is defined by the region where the bottom electrode and the top electrode overlap. All devices were encapsulated by a 30 nm Si3N4 and a 200 nm SiO2 passivation layer. The device size used for the current study was nominally (85 × 85) nm2 unless specified otherwise.

Electrical Testing



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Thermometry of the devices was carried out by collecting the pulsed I-V characteristics as a function of stage temperature. This was accomplished by launching ultra-short (80 ps) pulses to the device in Ground-Signal-Ground configuration along a 50 Ω transmission line.20 We extracted the voltage and current across the device using a time domain transmissometry setup23 discussed in the supporting information. The pulsed I-V-T characteristics obtained in this way are not affected by the self-heating of the device and can be used as a look-up table for interpretation of the DC characteristics.20 The pulsed I-V characteristics were also used to determine the temperature coefficient of resistance (TCR) of the filament at ≈ 0.16 %. This value is consistent with the TCR of metals with poor electrical conductivity24 and was used in electrothermal simulations to implement self-consistency between electrical and thermal domains.

Electro-Thermal Simulation Finite element simulations using the COMSOL package* were carried out making similar assumptions to those used in our previous publications.20,22 In addition, one of the significant deficiencies of past approaches has been addressed. Specifically, we have included a thermal boundary resistance (TBR) at the metal/oxide interface, which is crucial in understanding the high-temperature excursions that cause current spreading in RRAM device at high-powers. A TBR of 200 MW/m2K typically exists between metals and semiconductors/insulators because the mode of heat transfer in metals (electrons) is fundamentally different from that of semiconductors/insulators (phonons).25 At the interface, heat transfer between the oxide and the metal occurs only via phonon exchange with the temperature increase of the metal electrode lattice resulting in subsequent increase of the electron gas energy. The mismatch at the interface manifests itself in the form of TBR.25 In our system, the TBR diminishes the vertical heat flow *Certain commercial equipment, instruments, or materials are identified in this paper in order to specify the experimental procedure adequately. Such identification is not intended to imply recommendation or endorsement by the National Institute of Standards and Technology, nor is it intended to imply that the materials or equipment identified are necessarily the best available forParagon the purpose. ACS Plus Environment

Page 6 of 41

Page 7 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


from the filament to the electrodes, which increases the local temperature of the filament. This, in turn increases the lateral heat flow, resulting in significant heating of the oxide surrounding the filament and its the electrical conductivity. As the consequence, the current spreads outside of the filament. In order to accurately simulate this effect, we have measured the pulsed I-V characteristics of an as-fabricated unformed device as a function of stage temperature, as shown in Figure S4. The I-V characteristics show a current with a non-linear dependence on both voltage and temperature. We fit this I-V-T surface and use it as the conductivity σ(V, T) of the oxide surrounding the filament in the simulation. We also assume that thermal properties of the amorphous and crystalline HfAlOx are the same, as the reported thermal conductivity values in typical oxide (for example silica) thin film ranges from 1.1-1.4 W/m-K.26,27 Moreover, changing the thermal conductivity of the surrounding region does not significantly affect the thermal resistance (and consequently temperature) seen by the filament, due to the heat dissipation occurring mostly through the electrode to thermal ground.

Microstructure Analysis The TEM assessment of microstructural changes was performed on crossbar type devices shown in Fig. S1b. Electron transparent TEM specimens were fabricated using a FEI Nova Nanolab 600 dual-beam focused ion beam (FIB) system equipped with an Omniprobe in situ lift-out capability. Membrane samples about (4×1.3×2) µm3 in size were milled out using a 30 keV Ga+ ion beam. The membranes were then transferred to Omniprobe copper grids and ion milled down to ≈ 100 nm thickness at 30 keV with final polishing steps performed using a 5 keV beam. The final thickness of the specimens was ≈ 80 nm. More details of TEM specimen preparation can be found in supporting information (Figure S5). This procedure allowed the entire device structure



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

to be captured within the TEM specimen to detect all the possible switching-induced structural changes in the device.

High resolution (HR) TEM imaging was conducted using a Cs-corrected FEI Titan G2 80-300 TEM at 300 keV. The specimens were oriented toward a zone axis referenced by the Si substrate in order to minimize the overlap between the layers. In order to image all existing crystallites through the thickness of the specimens, multiple images at the same location were collected with different objective lens strength (through focus imaging series).

Results and Discussion Local Temperature Extraction Figure 1 shows the DC forming (black) and switching I-V characteristics of the TiN/Hf/HfAlOx/TiN stack obtained with the current compliance set to 50 µA, 100 µA and 200 µA (during forming and switching). In all of these cases, the current compliance was applied by means of an on-chip transistor or resistor ballast. This minimized the capacitance of connections between the ballast and the devices, practically eliminated the energy dissipation due to capacitance discharge and allowed for a much better control of current during electro-formation than the off-chip ballast. As seen in Figure 1, the resistance of the low and high resistance state of the device decreases by nearly a factor of 4, as the compliance current during forming increases from 50 µA (red) to 200 µA (blue). We use the pulsed I-V characteristics of the device (see Experimental section) to extract the temperature excursions in the device.


Page 8 of 41

Page 9 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Figure 1. Quasi-DC forming and switching I-V characteristics of the devices at different compliance current at forming. A linear scale I-V can be found in Figure S1.

The extracted temperature increase (∆T) is plotted as a function of the applied power, P, in the low resistance state (LRS) of the device formed with the current compliance of 50 µA, as shown in Figure 2.

Figure 2. Temperature increase vs. power in the low resistance state of Hf0.82Al0.18Ox-based device.

The slope of the ∆T versus power (P) curve represents the thermal resistance (Rth) that is experienced by the heat-source. In the steady-state:



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

∆T = Rth×P Here, Rth depends on the thermal conductivity of the filament and the surrounding oxide, the distance over which the heat is transported (i.e. the thickness of the functional layer) and the area of the heat source. Out of these parameters, the area of the filament is the only parameter that can change with dissipated power as thermal conductivity has a weak dependence on temperature.20 At low dissipated power (< 1 µW) the temperature increases linearly with power corresponding to a constant cross-section of the conducting non-volatile filament. At higher power, the slope continuously decreases and at power close to the RESET reaches 0.05 K/nW. This implies that the effective cross section of the area where the heat is generated increased significantly. We interpret this change in the following way. At low power, the current is conducted solely by the permanent small diameter metallic filament. As the power increases, the temperature of both the filament and the surrounding oxide increases raising the electrical conductivity of the oxide matrix and decreasing that of the filament. The heat now is generated in a much larger volume with the significantly lower Rth. The effect only persists when the bias is on; the next switching cycle starts again with a small permanent metallic filament. In other words, we can consider the nonvolatile filament as the region detected using pulsed I-V (without Joule heating). This is the region with local oxygen deficiency. The region around the filament gets hot at high biases and thus becomes conducting. This is purely an electro-thermal byproduct of self-heating of the nonvolatile filament. However, as the electrical thermometry gives the size of the heat source, this region is detected in the measurement leading to a reduction in ∆T vs Rth slope.

This model was made quantitative by using finite element simulation to track changes in Rth with power. Figure 3 shows a COMSOL-generated plot of thermal resistance for filaments of various


Page 10 of 41

Page 11 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


sizes, using a purely thermal simulation. To generate this plot, we simulate the steady-state temperature response of the device assuming conducting regions (filaments) of increasing size – from 0.5 nm up to 400 nm radius. This simulation plots the ratio of the temperature rise to the applied power (arbitrary value of 10 µW assumed) to obtain the thermal resistance Rth. The thermal resistance is akin to a bulk equivalent of the resistance seen by the region which dissipates power (filament and hot zone), i.e. thermal resistance between the filament and the thermal ground. To improve upon our previous simulation,20,22 we have included a TBR of 200 MW/m2K as a component of this bulk Rth.

Figure 3. Thermal resistance for varying filament size. The inset presents the schematics of the device used in simulation. Red region represents the non-volatile metallic filament and the surrounding pink region represents the hot but otherwise untransformed oxide.

The plot in Figure 3 is used as a lookup table to extract the radius of the heat source, in this case the conductive filament, for a measured Rth. The diameter corresponding to an Rth of 4 K/nW observed at low power is about 1 nm. This corresponds to the size of the permanent conductive filament serving as the heat source. It must be noted that this thermal resistance (Rth) is approximately 100 times greater than that reported in our previous work.20 This is due to a much



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

smaller size of devices used in this study than the ones in previous work ((85×85) nm2 versus (1×1) µm2), smaller thickness of the electrodes (5 nm versus 20 nm) and smaller thermal conductivity of the TiN bottom electrode compared to Pt (1.3 W/m⋅K versus 17 W/m⋅K). All three result in the electrodes being much less efficient as heat sinks.

Figure 4. Electro-thermal simulation to estimate temperature distribution at 0.8 V and 40 µW dissipated.

With the increase of voltage and the temperature of the filament, the surrounding oxide heats up substantially. This causes the electrical conductivity of the surrounding region to increase,20 and the current to spread beyond the metallic filament. As the volume where the power is dissipated increases, the corresponding thermal resistance is reduced. This is apparent in Figure 3 where at voltages close to RESET, the slope of the ∆T versus power decreased to 0.05 K/nW corresponding to a diameter of the hot zone of about 50 nm. One must note that this diameter is averaged by the conductivity and the temperature is also similarly averaged. The line scan of the temperature as a function of position obtained from the electro-thermal simulation (Figure 4) is shown in Figure 5a. This simulation assumed the filament size obtained


Page 12 of 41

Page 13 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


from low-power filament diameter (1 nm, Figure 3) with its temperature dependent conductivity obtained using pulsed I-V characterization (see Figure S4). The approximate temperature coefficient of resistance (TCR) is 0.16% in the investigated temperature range; however, for precision, a non-linear fit to the pulsed I-V which incorporates TCR reduction at high temperatures. For the surrounding material, the electrical conductivity which is a non-linear function of voltage and temperature is used, as obtained from calibrations of an unformed device. The simulation is an axisymmetric simulation that allows electro-thermal conduction in all directions, allowing us to view the current spreading described in Figure 2. The peak local temperature obtained in the simulation is ≈ 1600 K in the middle of the nonvolatile filament. The temperature of the surrounding oxide reaches ≈ 850 K 7 nm away from the 1 nm permanent filament (Figure 5a). The corresponding current density distribution is shown in Figure 5b. The I-V characteristics of the device generated by this simulation match the experimental results, further corroborating the effect of heat and current spreading.

(a)

(b)

Figure 5. COMSOL simulation of a device electroformed with current compliance of 50 µA (Figure 1) under bias of 0.8 V. (a) Temperature distribution. (b) Current density as a function of distance from the center of the filament.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Microstructure Analysis In order to experimentally verify the magnitude and extent of temperature excursions during switching, we have analyzed the microstructure of the device stacks by HRTEM. Figure 6a is a cross-sectional view of the pristine device showing TiN/Hf/HfAlOx/TiN stack, and Figure 6b and 6c provide magnified views at the left and right edge of the device, respectively.

Figure 6. Cross-sectional TEM image showing the as-fabricated nominally (85×85) nm2 device. The TEM sample was cut parallel to the top electrode of the crossbar device. Low magnification (a) and high magnification views of the left (b) and right (c) edge of the device.

The TEM samples were sectioned parallel to the TiN top electrode of the crossbar device. As a consequence, the top electrode extends across the entire length of the sample (TiN layer in the upper half of the Figure 6a) and the bottom electrode is oriented perpendicular to the crosssectional TEM sample surface. The two TiN electrodes are polycrystalline and exhibit a microstructural morphology consistent with columnar growth. The HfAlOx functional layer is slightly lighter than the Hf oxygen getter layer because of the atomic density difference. The light grey contrast surrounding the TiN bottom electrode is the SiO2. Small recesses in the functional layer seen at the edges of the top electrode are likely due to an etching process during


Page 14 of 41

Page 15 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


device fabrication. The recesses provide landmarks to confirm that the magnified images in Figures 6b and 6c were recorded at the edges of the device. Some small regions of dark contrast that can be seen in the SiO2 layer are due to re-deposition of sputtered materials from the FIB specimen preparation process. Note that these features are not seen in the device area and so do not affect phase contrast imaging of the device layers. The HfAlOx functional layer in asfabricated device is amorphous. No signs of crystallinity were observed in HfAlOx even when performing a through focusing series and FFT analysis (see Figure S6). This conclusion has been further confirmed by examining a much larger volume of HfAlOx in the as-deposited state for a different, larger device ((1×1) µm2).

Similar analysis has been conducted on devices that were electroformed with current compliance set at 10 µA (Figure 7a), 50 µA (7b), and 200 µA (7c). The small device sizes (the actual size of the active area was (65 × 65) nm2) allowed for inspection of the entire active device volume after biasing and, as the consequence, for detecting any microstructural changes. The micrographs are high magnification images of the right edge of the devices (the recess is seen at the lower right corner).



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Figure 7. HRTEM micrographs showing a localized crystallite embedded in an amorphous matrix electroformed with (a) 10 µA, (b) 50 µA, and (c) 200 µA current compliance. The inset in (b) is an inverse FFT of the crystallite region showing clear lattice fringes.

Imaging through-focus series revealed one crystalline region embedded in the amorphous HfAlOx layer for each device examined. The phase contrast regions, or lattice fringes, visible in these crystallites are denoted by dashed lines in Figure 7. An interplanar spacing of ≈ 0.3 nm was measured for the lattice fringes using Fast Fourier Transform (FFT) analysis and likely corresponds to the {101} crystallographic planes for tetragonal HfAlOx. The crystallite size is ≈ 10 nm at 10 µA compliance current and ≈ 15 nm at both the 50 µA and 200 µA compliance current. We interpret the presence of crystalline regions as an indication of bias-induced crystallization of the HfAlOx caused by local heating of the oxide around the conducting filament. One has to note that the crystalline HfAlOx is not necessarily conducting and the crystallites are not the filament itself. These are merely interpreted as microstructural features indicative of appreciable, local Joule heating. As discussed above, the estimated size of the metallic filament (≈ 1 nm diameter) is significantly smaller than the crystallite size. The filament is believed to have the same crystalline structure as the crystallites with the only difference being the oxygen deficiency. While nominal resolution of the instrument used to acquire these images is < 0.1 nm, it is difficult to observe the filament due to the presence of the surrounding material along the electron beam path in cross-sectional view configuration. We have attempted to image


Page 16 of 41

Page 17 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


the filament by using scanning TEM (STEM) electron energy-loss spectroscopy (EELS) with results described in the Chemical Analysis section.

Figure 8. HRTEM micrographs of specimen annealed for 2 seconds at: (a) 820 K, (b) 870 K, (c) 920 K, and (d) 970 K. The inset in (c) is an inverse FFT of the crystallite region showing clear lattice fringes.

This above assertion is supported by the results of an annealing experiment. The devices used in this experiment were from the same wafer as the (65 × 65) nm2 device described above. Separate dies cut from the wafer were annealed at different temperatures for 2 seconds in a rapid thermal annealing furnace and the FIB-TEM samples were prepared from large (3 × 3) µm2 crossbar devices. The annealing time was chosen to set the similar time period of the DC sweep test. Five different locations in each specimen were randomly selected and through focus imaging was



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

used to assess degree of crystallinity. Figure 8 shows cross-sectional views (one of the through focus series) of the devices annealed at 820 K, 870 K, 920 K, and 970 K.

Figure 9. Degree of crystallization as a function of annealing temperature. A clear transition is observed around 850 K.

The top dark layer in all Figure 8 images corresponds to Hf and is followed by slightly brighter HfAlOx and TiN layers. We have found no signs of crystallization in specimens annealed up to 820 K. Above this temperature, all annealed samples contained crystallites with size and density increasing with temperature. The crystallites are outlined by white dashed lines in Figure 8 and the observed d-spacing was again ≈ 0.3 nm in all cases. It appears that the crystallites in HfAlOx nucleate on the Hf grains with the same orientation of lattice fringes (shown in Figure S7). The degree of crystallization (volume percent) as a function of annealing temperature is estimated and plotted in Figure 9. Since the first crystallites appeared in the annealed samples at 870 K, we set 850 K as the approximate crystallization temperature of HfAlOx. This annealing study indicates that the crystallite size (≈ 15 nm) observed in the TEM analysis is consistent with the extent of the hot zone (≈ 14 nm) extracted from the thermometry and simulated by the COMSOL


Page 18 of 41

Page 19 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


model (diameter corresponding to the size of the volume where temperature exceeds 850 K as seen in Figure 5a). I-V measurements for annealed devices show no changes in device resistivity, thus confirming that the crystallization does not correspond to filament formation itself, but rather is due solely to localized heating.

Chemical Analysis In order to identify filament configuration and understand its chemical nature, we conducted STEM-EELS analysis using a Cs-corrected FEI Titan 80-300. The 50 µA CC specimen (Figure 7b) chosen for the chemical analysis was re-polished using 5 keV ion beam to further reduce the thickness of the specimen. The thickness was determined to be ≈ 40 nm by the Log-Ratio Method.28 The STEM-EELS analysis did not detect any local oxygen deficiency both within and beyond the crystallite region. Instead, we observed stochastic oxygen signal fluctuation along the HfAlOx functional layer indicating that oxygen content variation was below the sensitivity limit of the current STEM-EELS experiment, corresponding to < 4 % of oxygen content.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Figure 10. STEM-EELS O-K edge signal collected away from the crystallite in 50 µA CC specimen. (a) Integrated O-K edge map of Hf/HfAlOx/TiN stack. Each pixel corresponds to (1×1) nm2. (b) Integrated O-K edge line scan along 6th row (marked by an arrow in (a)).

To assess the detection limit of our experiment, we have collected an O-K edge map from the same specimen in an area far from the crystallite (Figure 10a) with a spatial resolution of 1 nm/pixel and 2 second/pixel acquisition time. The acquisition time (electron irradiation dose) was limited by the beam damage to the functional layer. The marked region from 4th to 8th row of the image is the HfAlOx functional layer (5 nm thickness) and the slight contrast change at the interfaces is due to the overlap of the Hf/HfAlOx and HfAlOx/TiN layers. Figure 10b is an O-K edge line scan along the 6th row in the area map (marked by an arrow in Figure 10a). The line profile oscillates around the average with a standard deviation of 3.9 %. Since there is no reason to suspect the oxygen content to vary in the HfAlOx at these length scales, this value represents


Page 20 of 41

Page 21 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


the oxygen detection limit in the investigated structures. Even in case of pure Hf filament embedded in the 40 nm thick HfAlOx matrix in current experiment, the oxygen deficiency along the electron beam path would amount to about 2 %, which is below the detection limit. It should be noted that the reported values of the oxygen deficiency in the filament were much lower than 100% (< 13 at.%).12,14,29 Increasing acquisition time, reducing specimen thickness, and changing chemical make up of the functional layer may help to increase the detection limit but all of these approaches are expected to be limited. At present, EELS is unlikely to produce an oxygen deficiency estimate in this material system, principally because in the cross-sectional sample configuration, the beam traverses the entire oxide thickness which is many times the estimated diameter of the filament.

Endurance Cycling The thermometry, finite element modeling, and TEM analysis yielded a self-consistent picture of the temperature rise in the device and the filament size. We use the same thermometry approach to understand the evolution of the filament during endurance cycling. Figure 11 shows the endurance cycling data for typical RRAM devices that were cycled with a current compliance of 50 µA, VSET of 2.5 V and VRESET of -1.75 V, showing an endurance failure after ≈ 107 cycles. The traces in red and blue represent the mean values, whereas the traces in gray represent data from individual devices. The resistance of the OFF state of devices decreased during cycling and after 107 cycles approached the value of the ON state resistance. Failure is defined by devices no longer showing the RESET at the selected bias. At this point, devices are still functional but would require higher VRESET pulse amplitude to switch to the OFF state.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

OFF state

ON state

Figure 11. Endurance cycling of the HfAlOx device at VSET of 2.5 V and VRESET of -1.75 V with current compliance 50 µA using pulse width of 100 ns.

Since the devices fail by "stuck-at ON state", one can interrogate the state of the non-volatile filament. A series of such devices was tested by pulsed I-V versus temperature and the temperature was assessed by the same thermometry approach as for just formed devices. A comparison of the ∆T versus power of as-formed and failed but otherwise identical devices is shown in Figure 12. The resistance of the failed device was lower by a factor of 10.

Figure 12. (a) ΔT vs. power at low-power after endurance failure. (b) Rth vs. radius profile showing filament size evolution upon endurance cycle (following Figure 3).


Page 22 of 41

Page 23 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


The red circles represent data obtained on the as-fabricated device while the black squares correspond to device after the endurance cycling. Since the slope of the plot at high biases underestimates the temperature due to heat and current spreading, we use low-bias slope for the Rth extraction and the filament size estimate. The post cycling Rth of 2.3 K/nW (changed from 7 K/nW) was observed to indicate a filament diameter of approximately 3 nm. This is interpreted from using the slope from Figure 12a and mapping it on to the thermal simulation in Figure 12b (same as Figure 3). These estimates of filament size are smaller than many previous estimates 8– 10

but consistent with AFM tomography16 and modeling performed on similar devices.19 As this

technique combines thermometry with a TEM-based microstructural verification, the filament size estimates can be directly used to estimate scaling prospects of RRAM devices. Moreover, increased filament size results in a larger volume of hot oxide consistent with the higher power levels needed to switch the device.30 The possible causes of RESET failure mechanisms have been previously discussed as being due to temperature activated changes in the number of oxygen vacancies at the oxide / bottom electrode interface.18,31 The increase of the filament diameter may suggest that a gradual closure of the resistance change window (HRS gradually collapsing to LRS) occurs due to high temperature excursions in the device prior to SET and RESET. These high temperatures are sufficient to create more oxygen vacancies by interactions with the electrodes or by incomplete retraction and eventual accumulation of vacancies to form a wider filament.32 Soret effect also can play an important role in switching failure modes.14,33 Once the widening of the filament has started, the Soret effect causes an agglomeration of oxygen vacancies by diffusion towards the filament resulting in the net reduction in the resistance of the device.33 This additionally results in the decrease in the voltage across the device, thus increasing the voltage across the ballast transistor. Similarly to suggestions in the



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

previous work,18 this increases the drive current and leads to a positive feedback that eventually causes a RESET failure. More extensive chemical /structural investigation and mass transport simulations are needed to verify the exact mechanism responsible for the endurance failure with thermometry acting as an efficient diagnostic tool. Another unique insight that this methodology provides is the role of the surrounding material in current spreading and limiting the peak temperature of the filament. Previous works12,15,19,33 have mainly focused on the effect of temperature, temperature gradients and field on vacancies and the filament. This work brings into perspective the role of surrounding stoichiometric oxide which may be central in determining the optimal operating and switching power required for aggressively scaled devices.

Conclusions We present estimates of the maximum temperatures reached during the switching cycles in (85×85) nm2 HfAlOx RRAM devices using transient thermometry. For these devices, the temperature of the metallic non-volatile filament, could reach as high as 1600 K at the onset of RESET. The thermometry estimates the filament size to be as small as 1 nm in diameter. At high biases the surrounding oxide heats up and starts to conduct significant current causing the current density to decrease. The temperature can reach 850 K 7 nm from the filament at RESET voltage. The thermometry and modeling were verified by HRTEM of the as-fabricated and electroformed devices. The images show a single crystallite in each of the formed devices with the size in 10-20 nm diameter range. Independent TEM experiment determined the onset of HfAlOx crystallization at 850 K. The size of the crystallites agrees with the simulated temperature distribution at RESET bias. In a parallel STEM-EELS experiment, no chemical change has been observed due to the inherent detection limit of cross-sectional STEM-EELS. Lastly, the endurance cycling of


Page 24 of 41

Page 25 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


the devices resulted in a resistance drop by a factor of 10 after 107 cycles, which corresponds to a filament diameter increase by a factor of 3. The thermometry of these devices, showed the comparable change of thermal resistance. This methodology and observations could play a pivotal role in understanding the role of the surrounding oxide region in thermal dynamics, peak temperature, switching mechanism and scaling behavior.

Acknowledgements This work was supported by NSF Grant No. DMR1409068 and SRC Contract No. 2012-VJ2247. Also, the authors acknowledge use of Materials Characterization Facility at Carnegie Mellon University supported by grant MCF-677785.

References (1)

Lee, H. Y.; Chen, P. S.; Wu, T. Y.; Chen, Y. S.; Wang, C. C.; Tzeng, P. J.; Lin, C. H.; Chen, F.; Lien, C. H.; Tsai, M. J. Low Power and High Speed Bipolar Switching with a Thin Reactive Ti Buffer Layer in Robust HfO2 Based RRAM. Electron Devices Meet. 2008. IEDM 2008. IEEE Int. 2008, 1–4.

(2)

Wong, H.-S. P.; Lee, H.-Y.; Yu, S.; Chen, Y.-S.; Wu, Y.; Chen, P.-S.; Lee, B.; Chen, F. T.; Tsai, M.-J. Metal–Oxide RRAM. Proc. IEEE 2012, 100, 1951–1970.

(3)

Kim, Y. M.; Lee, J. S. Reproducible Resistance Switching Characteristics of Hafnium Oxide-Based Nonvolatile Memory Devices. J. Appl. Phys. 2008, 104, 114115.

(4)

Govoreanu, B.; Kar, G. S.; Chen, Y.; Paraschiv, V.; Kubicek, S.; Fantini, A.; Radu, I. P.; Goux, L.; Clima, S.; Degraeve, R.; et al. 10x10nm2 Hf/HfOx Crossbar Resistive RAM with Excellent Performance, Reliability and Low-Energy Operation. In International



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Electron Devices Meeting; IEEE, 2011; pp. 729–732. (5)

Huang, C.-Y.; Jieng, J.-H.; Jang, W.-Y.; Lin, C.-H.; Tseng, T.-Y. Improved Resistive Switching Characteristics by Al2O3 Layers Inclusion in HfO2-Based RRAM Devices. ECS Solid State Lett. 2013, 2, P63–P65.

(6)

Yu, S.; Gao, B.; Dai, H.; Sun, B.; Liu, L.; Liu, X.; Han, R.; Kang, J.; Yu, B. Improved Uniformity of Resistive Switching Behaviors in HfO2 Thin Films with Embedded Al Layers. Electrochem. Solid-State Lett. 2010, 13, H36.

(7)

Raghavan, N.; Degraeve, R.; Fantini, a.; Goux, L.; Wouters, D. J.; Groeseneken, G.; Jurczak, M. Stochastic Variability of Vacancy Filament Configuration in Ultra-Thin Dielectric RRAM and Its Impact on OFF-State Reliability. 2013 IEEE Int. Electron Devices Meet. 2013, 21.1.1–21.1.4.

(8)

Strachan, J. P.; Pickett, M. D.; Yang, J. J.; Aloni, S.; David Kilcoyne, A. L.; MedeirosRibeiro, G.; Stanley Williams, R. Direct Identification of the Conducting Channels in a Functioning Memristive Device. Adv. Mater. 2010, 22, 3573–3577.

(9)

Strachan, J. P.; Strukov, D. B.; Borghetti, J.; Yang, J. J.; Medeiros-Ribeiro, G.; Williams, R. S. The Switching Location of a Bipolar Memristor: Chemical, Thermal and Structural Mapping. Nanotechnology 2011, 22, 254015.

(10)

Jeong, D. S.; Schroeder, H.; Breuer, U.; Waser, R. Characteristic Electroforming Behavior in Pt/TiO2/Pt Resistive Switching Cells Depending on Atmosphere. J. Appl. Phys. 2008, 104, 123716.

(11)

Joshua Yang, J.; Miao, F.; Pickett, M. D.; Ohlberg, D. A. A.; Stewart, D. R.; Lau, C. N.; Williams, R. S. The Mechanism of Electroforming of Metal Oxide Memristive Switches. Nanotechnology 2009, 20, 215201.


Page 26 of 41

Page 27 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


(12)

Larentis, S.; Nardi, F.; Balatti, S.; Gilmer, D. C.; Ielmini, D. Resistive Switching by Voltage-Driven Ion Migration in Bipolar RRAM — Part II : Modeling. IEEE Trans. Electron Devices 2012, 59, 2468–2475.

(13)

Chen, Y. Y.; Degraeve, R.; Clima, S.; Govoreanu, B.; Goux, L.; Fantini, A.; Kar, G. S.; Pourtois, G.; Groeseneken, G.; Wouters, D. J.; et al. Understanding of the Endurance Failure in Scaled HfO 2 -Based 1T1R RRAM through Vacancy Mobility Degradation. In IEDM; 2012; pp. 10–13.

(14)

Kim, S.; Choi, S.; Lu, W. Comprehensive Physical Model of Dynamic Resistive Switching in an Oxide Memristor. ACS Nano 2014, 8, 2369–2376.

(15)

Kwon, D.-H.; Kim, K. M.; Jang, J. H.; Jeon, J. M.; Lee, M. H.; Kim, G. H.; Li, X.-S.; Park, G.-S.; Lee, B.; Han, S.; et al. Atomic Structure of Conducting Nanofilaments in TiO2 Resistive Switching Memory. Nat. Nanotechnol. 2010, 5, 148–153.

(16)

Celano, U.; Goux, L.; Degraeve, R.; Fantini, A.; Richard, O.; Bender, H.; Jurczak, M.; Vandervorst, W. Imaging the Three-Dimensional Conductive Channel in FilamentaryBased Oxide Resistive Switching Memory. Nano Lett. 2015, 15, 7970–7975.

(17)

Szot, K.; Rogala, M.; Speier, W.; Klusek, Z.; Besmehn, A.; Waser, R. TiO2-a Prototypical Memristive Material. Nanotechnology 2011, 22, 254001.

(18)

Chen, C. Y.; Goux, L.; Fantini, A.; Redolfi, A.; Clima, S.; Degraeve, R.; Chen, Y. Y.; Groeseneken, G.; Jurczak, M. Understanding the Impact of Programming Pulses and Electrode Materials on the Endurance Properties of Scaled Ta 2 O 5 RRAM Cells. IEDM 2014, 355–358.

(19)

Govoreanu, B.; Clima, S.; Radu, I. P.; Chen, Y.; Wouters, D. J.; Jurczak, M. Complementary Role of Field and Temperature in Triggering ON/OFF Switching



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 28 of 41

Mechanisms in Hf/HfO2 Resistive RAM Cells. IEEE Trans. Electron Devices 2013, 60, 2471–2478. (20)

Yalon, E.; Sharma, A. A.; Skowronski, M.; Bain, J. A.; Ritter, D.; Karpov, I. V. Thermometry of Filamentary RRAM Devices. IEEE Trans. Electron Devices 2015, 62, 2972–2977.

(21)

Borghetti, J.; Strukov, D. B.; Pickett, M. D.; Yang, J. J.; Stewart, D. R.; Williams, R. S. Electrical Transport and Thermometry of Electroformed Titanium Dioxide Memristive Switches. J. Appl. Phys. 2009, 106, 124504.

(22)

Sharma, A. A.; Noman, M.; Abdelmoula, M.; Skowronski, M.; Bain, J. A. Electronic Instabilities Leading to Electroformation of Binary Metal Oxide-Based Resistive Switches. Adv. Funct. Mater. 2014, 24, 5522–5529.

(23)

Strickland, J. A.; Gordon, L. Time-Domain Reflectometry Measurements; Tektronix, 1970.

(24)

Ashcroft, N. W.; Mermin, N. D. Solid State Physics; Rinehart and Winston: New York, 1976.

(25)

Chen, G. Nanoscale Energy Transport and Conversion: A Parallel Treatment of Electrons, Molecules, Phonons, and Photons; Oxford University Press, 2005.

(26)

Burzo, M. G.; Komarov, P. L.; Raad, P. E. Thermal Transport Properties of Gold-Covered Thin-Film Silicon Dioxide. IEEE Trans. Components Packag. Technol. 2003, 26, 80–88.

(27)

Delan, A.; Rennau, M.; Schulz, S. E.; Gessner, T. Thermal Conductivity of Ultra Low-K Dielectrics. Microelectron. Eng. 2003, 70, 280–284.

(28)

Egerton, R. F. Electron Energy-Loss Spectroscopy in the TEM. Reports Prog. Phys. 2009, 72, 016502.

(29)

Degraeve, R.; Fantini, A.; Chen, Y. Y.; Clima, S.; Govoreanu, B.; Goux, L.; Wouters, D.


Page 29 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


J.; Roussel, P.; Kar, G. S.; Pourtois, G.; et al. Reliability of Low Current Filamentary HfO2 RRAM Discussed in the Framework of the Hourglass SET/RESET Model. 2012 IEEE Int. Integr. Reliab. Work. Final Rep. 2012, 3–7. (30)

Yu, S.; Wong, H. S. P. A Phenomenological Model for the Reset Mechanism of Metal Oxide RRAM. IEEE Electron Device Lett. 2010, 31, 1455–1457.

(31)

Degraeve, R.; Fantini, A.; Roussel, P.; Goux, L.; Costantino, A.; Chen, C. Y.; Clima, S.; Govoreanu, B.; Linten, D.; Thean, A.; et al. Quantitative Endurance Failure Model for Filamentary RRAM. 2015 Symp. VLSI Technol. Deigest Tech. Pap. 2015, 188–189.

(32)

Kwon, J.; Sharma, A. a.; Bain, J. a.; Picard, Y. N.; Skowronski, M. Oxygen Vacancy Creation, Drift, and Aggregation in TiO2-Based Resistive Switches at Low Temperature and Voltage. Adv. Funct. Mater. 2015, 25, 2876–2883.

(33)

Strukov, D. B.; Alibart, F.; Stanley Williams, R. Thermophoresis/diffusion as a Plausible Mechanism for Unipolar Resistive Switching in Metal-Oxide-Metal Memristors. Appl. Phys. A Mater. Sci. Process. 2012, 107, 509–518.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31


Page 30 of 41

Page 31 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24




1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22


Page 32 of 41

Page 33 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26




1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

(a)


(b)

Page 34 of 41

Page 35 of 41 (a) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22

(b) TiN

Hf


TiN (Top electrode)

HfAlOx Hf

HfAlOx

TiN

(c) TiN

Hf

TiN (Bottom electrode)

HfAlOx


20 nm

5 nm

TiN

5 nm

(a)


5 nm

Page 36 of 41

Hf

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 d = 20101 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40

HfAlOx

(b)

5 nm

~0.30 nm

(c)

5 nm


(a)

Page 37 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 d = 35101 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55


Hf

HfAlOx

5 nm

(b)

5 nm

(c)

~0.30 nm

5 nm

(d)


5 nm


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28


Page 38 of 41

7×105

Hf

Page 39 of 41


1

2 HfAlO x 3 4(5 nm)

Counts

5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39

3×105

(a)

TiN (bottom electrode)

0 (counts)

700x103 600x103 500x103 400x103 300x103 200x103 100x103 0

(b) 0

5 ACS Paragon 10 Plus Environment 15 20

Distance (nm)

25


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23


Page 40 of 41

(a)

Page 41 of 41

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

(b)



Transient Thermometry and High-Resolution Transmission Electron

Recommend Documents