Solid-State Electronics 206 (2023) 108667 Contents lists available at ScienceDirect Solid State Electronics journal homepage: www.elsevier.com/locate/sse Assessment of the variability of the I-V characteristic of HfO2-based resistive switching devices and its simulation using the quasi-static memdiode model E. Salvador a, *, M.B. Gonzalez b, F. Campabadal b, J. Martin-Martinez a, R. Rodriguez a, E. Miranda a a Departament d’Enginyeria Electrònica, Universitat Autònoma de Barcelona, 08193 Cerdanyola del Valles, Spain b Institut de Microelectrònica de Barcelona, IMB-CNM, CSIC, 08193 Cerdanyola del Valles, Spain A R T I C L E I N F O A B S T R A C T Keywords: Variability of the conduction characteristics of filamentary-type resistive switching devices or resistive RAMs Memristor (RRAMs) is a hot research topic both in academia and industry because it is currently considered one of the major Variability showstoppers for the successful development and application of this technology. In this work, we thoroughly Resistive switching investigate the statistics of the cycle-to-cycle (C2C) variability observed in the experimental current–voltage (I-V) HfO2 curves of HfO2-based memristive structures using the fitdistrplus package for the R language. This exploratory analysis allows us to identify which parametric probability distributions are the most suitable candidates for describing our data. This study involves graphical tools such as the density, skewness-kurtosis (S-K), and quantile–quantile (Q-Q) plots. The analysis is completed with the aid of goodness-of-fit statistics (Kolmogorov- Smirnov, Cramer-von Mises, Anderson-Darling) and criteria (Akaike’s and Bayesian). The selected distributions are incorporated into the SPICE script of the quasi-static memdiode model for resistive switching devices and used for simulating uncorrelated C2C variability. Finally, a one-way sensitivity analysis is carried out in order to test the impact of the model parameters variation in the output characteristics of the device. 1. Introduction compact model for such devices would be of utmost importance. In this work, uncorrelated cycle-to-cycle (C2C) variability is introduced in the Filamentary-type resistive switching devices or resistive random- quasi-static memdiode model (QMM) [6–7] through the random access memories (RRAMs) are electroformed metal-oxide-metal struc­ assignment of the model parameter values. We consider here the QMM tures that can alter their resistance state upon the application of an because we are only interested in simulating the devices when subjected external stimulus (voltage, current, light, etc.). In these structures, a to ramped voltage with fixed ramp rate. For pulsed measurements, localized vacancy or ion conducting bridge spans the oxide layer and where switching times are of utmost importance, a dynamical approach connects or disconnects the opposite electrodes in a non-volatile fashion. is required. Uncorrelated C2C means in this context that the generated This means that when power is turned off, the last resistance state of the cycles are considered independent so that no trends in the low (LRS) or device remains unaltered. Nowadays, this kind of structure is considered high (HRS) resistance state are expected. Trends can be included by a promising candidate for a plethora of applications including infor­ adding the appropriate wearout rules in the definition of the model mation storage, neuromorphic computing, logic circuits, cryptography, parameteres. The analysis reported here mainly focuses on the spread of and many more. [1–3] Nevertheless, one of the major drawbacks that the I-V curves around the median characteristic. It is worth mentioning this technology must face concerns with its intrinsic variability linked to that for specific applications, the model should be recalibrated since the morphologic changes of the conducting filament occurring at the atomic parameter values may be affected by the device state history if it in­ scale [4,5]. This is particularly important for the high resistance state volves any kind of irreversible effect. After the inclusion of variability in where a few atoms participate in the conduction process. Since vari­ the model parameters, the attention concentrates on the general ability is always present in RRAM operation, its inclusion in any behavior of the QMM. We have investigated how sensitive the model is * Corresponding author. E-mail address: emili.salvador@uab.cat (E. Salvador). https://doi.org/10.1016/j.sse.2023.108667 Received 10 January 2022; Received in revised form 15 March 2023; Accepted 2 May 2023 Available online 11 May 2023 0038-1101/© 2023 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). E. Salvador et al. Solid State Electronics 206 (2023) 108667 Fig. 1. A) experimental i-v curves: 450 cycles and median curve. b) voltage correction illustrating the snapback effect in a single cycle and parameter specification. to systematic modifications of the parameter values. The approach fol­ curves is illustrated. The snapback voltage is defined as lowed is also valid for any other model and it is technically referred to as VSB = Vapplied − Ri ⋅Imeasured where Ri is an internal series resistance (it can a one-way sensitivity analysis (SA). To be more precise, the sensitivity also include any external series resistance). The value of Ri is selected so we are discussing relates to how uncertainty in the simulation output as to provide an almost vertical increase of the current after the SET can be apportioned to different sources of uncertainty in the model in­ event. This vertical increase occurs at a constant voltage called the puts [8,9]. This paper is organized as follows: in Section 2, the fabri­ transition voltage VT. [12] Both Ri and VT are obtained individually for cation process of the devices under investigation and the preparation of each cycle in the measurement set. A number of observables are indicated the experimental data are described. In Section 3, the experimental in Fig. 1.b: the high (IHRS) and low (ILRS) resistance state currents are observations are statistically analyzed. In Section 4, the quasi-static obtained by extracting the current values at a fixed voltage, V = 0.2 V, memdiode model is presented. The inclusion of variability in the the RESET voltage VR is obtained from the voltage corresponding to the model script is discussed in Section 5. In Section 6, experimental and maximum current reached for negative bias. The last current data point simulated results are compared and, in Section 7, the one-way SA is before the SET event is the SB current ISB. All these observables are performed. Section 8 reports the conclusions of this work. extracted from the experimental curves using Matlab coding. In the following Section, the statistical distribution of these observables is 2. Devices and preparation of experimental data investigated. The devices studied in this work are HfO2-based metal-oxide-metal 3. Analysis of the experimental results structures [10]. The oxide thickness is 10 nm and the active area of the devices is 5x5 µm2. The bottom electrode consists in a 200 nm-thick In this Section, we investigate the statistical distribution of the ob­ W layer and the top electrode is a 200 nm-thick TiN layer on top of a 10 servables mentioned in Section 2 using the fitdistrplus package for the R nm-thick Ti layer acting as oxygen receiver material [11]. After the language [13]. The obtained results will be used to define the model electroforming process, the devices are cycled 450 times with a bipolar parameters in the QMM SPICE script. First, the Cullen and Frey voltage sweep with limits ± 1.5 V. The electrical measurements were skewness-kurtosis (SK) or Pearson plot is used as an indicator of the performed using the Semiconductor Parameter Analyzer (SPA) Agilent appropriateness of the different candidate distributions (normal, 4156C. The equipment was controlled via GPIB bus and programmed lognormal, gamma, Weibull, logistic, etc.). This plot compares the kur­ with Matlab software. The measured I-V curves are plotted in grey in tosis and the squared skewness corresponding to the available data. For Fig. 1.a. The median curve is plotted in blue. In Fig. 1.b, the snapback some specific distributions (normal, uniform, logistic, exponential), (SB) correction for a single cycle extracted from the set of experimental there is only one possible value for the skewness and kurtosis. Thus, the 2 E. Salvador et al. Solid State Electronics 206 (2023) 108667 Table 1 Goodness-of-fit statistics (Kolmogorov-Smirnov, Cramer-von Mises, Anderson- Darling) and criteria (Akaike’s and Bayesian) for a) VT , b) for VR and c) for RI . the blue dot and the bootstrapped data show a large kurtosis (leptokurtic behavior) which means that the tails of the distribution are heavier than those corresponding to a normal distribution. Since the SK plot must only be used as a first indicator, further analysis is required before concluding which distribution better suits. Next, the different candidate distributions (normal, lognormal, gamma, and Weibull) are fitted and compared with the experimental observables. The goodness-of-fit (GoF) statistics (Kolmogorov-Smirnov, Cramer-von Mises, Anderson-Darling) and criteria (Akaike’s and Bayesian) are shown in Table 1. Fig. 3.a-c show the density plots for VT, VR, and Ri , respectively, corresponding to the distributions investigated. The quantile–quantile (Q-Q) plots, which graphically compare the experimental distributions with the parametric models, are illustrated in Fig. 3.d-f, for VT, VR, and Ri , respectively. The combination of the information obtained from the GoF statistics and criteria, and the Q-Q plots were used to determine the most plausible Fig. 2. A), b) and c) show the cullen and frey plots for VT , VR and Ri distribution for each observable. In summary, the analysis carried out respectively. indicates that VT follows a lognormal distribution while VR and Ri are better described by a normal distribution. The same procedure was distribution is represented by a single point on the plot. For other dis­ followed for the rest of the observables resulting in: normal distribution tributions, areas of possible values are represented, consisting in lines for ISB and lognormal distributions for IHRS and ILRS. This information (as for the gamma and lognormal distributions), or shaded areas (as for will be included in the model parameters of the QMM script and will be the beta distribution). Notice the location of the observation (blue dot) used to generate a set of simulated curves with variability. with respect to the theoretical symbols and lines. The yellow points are bootstrapped data (random sampling with replacement). Fig. 2.a-c show 4. Brief introduction to the QMM the SK plot for VT, VR, and Ri , respectively. For VT and VR, the figures show that the experimental observations are close to the star symbol As mentioned in Section 1, variability simulations and sensitivity (normal distribution). Nevertheless, the VT observation as well as the analysis were performed with the QMM model [7]. This model describes bootstrapping cloud are slightly shifted almost coinciding with the the conduction characteristics of bipolar resistive switching devices dashed line representing the lognormal distribution. In contrast, for Ri , 3 E. Salvador et al. Solid State Electronics 206 (2023) 108667 Fig. 3. A), b) and c) show density plots for VT , VR and Ri , respectively, with normal, gamma, lognormal, and Weibull distributions. d), e) and f) show Q-Q plots after selecting the best fitting distribution. 4 E. Salvador et al. Solid State Electronics 206 (2023) 108667 Table 2 LTSpice QMM script including variability in the most significant parameters taking into account the previously extracted best candidate distributions. In red the Gaussian distributions and in blue the lognormal distributions. using a hysteresis operator which keeps track of the memory state of the 5. Simulation with variability using the QMM device. The origin of the switching is related to the formation of a conducting filament (CF) spanning the dielectric film caused by the Once the model parameter distributions were established (see Sec­ application of an external field (SET process for the transition HRS to tion 3), they were included in the QMM model script (see Table 2). Some LRS). The CF is created as a consequence of the displacement and of the model parameters were assumed to be constant to avoid over- accumulation of metal ions or oxygen vacancies (depending on the de­ randomness in the output curves. The LTspice gauss function and its vice type). The CF can be ruptured by the application of a field with transformation were used to generate the appropriate parameter values opposite sign (RESET process for the transition LRS to HRS). According and variability. Simulations were performed having in mind the to the QMM, the I-V characteristic reads: particular features of the experimental curves, such as the voltage span and the number of cycles. The parameters (including variability) are I(V) = I0 (λ)sinh{α(λ)[V − (RS (λ) + Ri )I ] } (1) calculated at the beginning of each cycle and are kept constant until the where I0 (λ) = I0min +(I0max − I0min )λ is the current amplitude factor, next cycle. The goal was to reproduce as close as possible the median RS (λ) a variable series resistance, and α(λ) a fitting parameter. I0min and curve and the spread of the experimental data set. Since ultimately the I0max are minimum and maximum current values (same for RS and α), simulated I-V curves are the result of a sequence of operations in which a respectively. According to (1), for low currents, I depends exponentially number of random parameters are involved, there is always an under­ on V whereas for high currents, I depends linearly on V. Equation (2) lying connection among the mean and dispersion values that need to be expresses the relationship between the memory state λ and the voltage considered in order to achieve a consistent set of curves. To control this across the filament’s constriction VC = V − Ri Ithrough the recursive interdependence an iterative optimization process was followed to hysteresis operator: adjust the simulation parameters for obtaining close agreement with the experimental curves. The simulated observables and their variability { [ (←) ]} λ(VC ) = min Γ− (VC ), max λ VC , Γ+ (VC ) (2) obtained from a complete simulation run were analyzed first. With this information at hand the parameter values were modified in the appro­ (←) priate direction (with the help of the one-way SA approach) and the where λ VC is the memory value a timestep before (in practice dictated simulation was performed again. This process was repeated until coin­ by the simulator timestep). Γ+ and Γ− are the so-called positive and cidence is achieved, within certain error margins, between the experi­ negative ridge functions (sigmoidals), respectively, and are expressed as: mental and simulation results. Notice we included variability in 7 out of 15 model parameters, this combination was found to be reasonable for 1 reproducing the experimental curves. For other devices, less variable Γ± (VC ) = (3) 1 + exp[ − η± ⋅(VC − V ± ) ] parameters might be enough to reproduce their behavior, in fact a basic variable set of curves can be obtained by including the ‘gauss’ LTSpice which represent the creation (+) and dissolution (-) processes of the CF. function in 4 parameters (imin, imax, isb and vr). η± (etas and etar in the model script) are the set (+) and reset (-) tran­ sition rates and V ± are the threshold voltages for set (+) and reset (-). 6. Comparison between experimental and simulation results The model uses other parameters for the fine-tuning of the simulated curves (see the script in Table 2). The model is implemented in LTspice Simulations using the QMM model were performed taking into ac­ using an equivalent circuit with behavioral current sources and includes count the iterative optimization process discussed in the previous Sec­ the snapback (isb) and snapforward (gam) effects. tion. In Fig. 4, experimental (black) and simulated (blue) curves are 5 E. Salvador et al. Solid State Electronics 206 (2023) 108667 Fig. 4. Experimental and simulated curves comparison. a, d) I-V curves, b, e) I-V curves in logscale and c, f) I-V curves applying the SB voltage correction for experimental and simulated respectively. 6 E. Salvador et al. Solid State Electronics 206 (2023) 108667 Fig. 5. Comparison of experimental and simulated parameter distributions: a) IHRS , b) VT , c) ILRS , and d) VR . compared using three alternative plots for the same set of I-V curves: a) parameters: the high resistance state current a) IHRS and the RESET and d) linear–linear axes, b) and e) log-linear axes, and c) and f) line­ voltage b) VR [14]. Both figures show the presence of a trend in the ar–linear axes (with SB correction). The fitting is reasonably good and experimental observables. Notice that our approach is unable to simulations reproduce the main features of the experimental curves in generate this trend since the LTspice simulations considered in this work all the cases. The most conspicuous difference occurs in LRS, mainly do not include cross-correlation among parameters. Aiming to compare because of a peculiarity of the available experimental data. This will be the fluctuation of the observables without trends, Fig. 6 c) and d) illus­ discussed later. A deeper and fairer comparison between the experi­ trate the difference in the observable between cycle k + 1 and cycle k mental and simulated observables is presented in Fig. 5. This figure both for IHRS and VR, defined as ΔIHRS and ΔVR, respectively. As ex­ shows the statistical distributions corresponding to the four main ob­ pected, the plots show that the experimental and model fluctuations are servables. Aiming to compile information from the different sections of of the same order of magnitude regardless the cycle number. In order to the I-V curves, the following observables were specifically selected: IHRS quantify this fluctuation, different indicators are presented in Table 3: for the HRS curve, VT for the SET transition, ILRS for the LRS curve, and the mean value μ, i.e. the first moment of the distribution, the standard VR for the RESET transition. This compilation was performed using the deviation (σ), i.e. the second moment of the distribution, the inter­ corrected I-V curves for VT and VR , and the raw I-V curves, i.e. without quartile range (IQR), a measures of the data spread, and the mean ab­ applying the SB correction for a direct evaluation of IHRS and ILRS . As solute deviation (MAD), which represents the average distance between shown in Fig. 5, the density plots corresponding to the experimental and each point and the mean value. Table 3 compares the four indicators simulated data, IHRS and VT are reproduced correctly. VR presents a (experimental and simulated), for the four different observables indi­ curious effect: taking a closer view at the raw I-V curves, the dispersion cated in Fig. 6. It can be seen that the indicators are similar for all the in the simulations is clearly higher than in the experimental case, but for observable which points out that the fluctuations are correctly captured. the corrected I-V curves, the dispersion is well reproduced. Instead, ILRS exhibits a big difference. The reason is clear, for the case under study, 7. One-way sensitivity analysis of the output curves the experimental ILRS histogram shows a double peak which likely cor­ responds to two different configurations of the CF or different coexisting As mentioned in Section 1, a one-way sensitivity analysis (SA) was CFs. Since the approach we are reporting here only considers a single set also carried out as part of this study. This analysis was performed of parameters for each distribution, the simulated curves cannot considering a voltage sweep high enough to achieve the full SET and reproduce the two distributions experimentally observed. Our parame­ RESET states of the I-V cuves without introducing irreversible damage to ters are only able to generate a single peak as illustrated in Fig. 5.c. This the devices. To understand why the applied voltage is an important issue inconsistency illustrates that caution should be exercised when unex­ for this analysis, Fig. 7 illustrates the role played by the maximum pected deviations in the experimental data occur and additional features applied voltage (Vapp) on the simulations. The figure includes indicators are required to simulate them appropriately. for the trends of the four different observables as the maximum applied As an indicator of the importance of considering the time series voltage is increased. Notice that if the voltage is set to 1.5 V (black evolution and the correlation among parameters, Fig. 6 shows experi­ curve), the simulated curve does not reach the full HRS, which means mental and simulated data corresponding to the C2C variability of two that the RESET process is not complete. Therefore, if the referred voltage 7 E. Salvador et al. Solid State Electronics 206 (2023) 108667 Table 3 Different indicators for studying the fluctuations of: HRS current, reset voltage, and the difference in the observable between cycle k + 1 and cycle k for IHRS and VR (ΔIHRS and ΔVR). VR IHRS ΔVR ΔIHRS Exp σ 0.0455 1.08e-05 0.0458 7.63e-06 Sim σ 0.0381 7.84e-06 0.0541 1.12e-05 Exp IQR 0.0624 1.29e-05 0.0555 7.53e-06 Sim IQR 0.0523 8.96e-06 0.0784 1.29e-05 Exp MAD 0.0455 1.08e-05 0.0458 7.63e-06 Sim MAD 0.0381 7.84e-06 0.0541 1.12e-05 Fig. 7. Simulated I-V curves for different applied voltages. The change in the observables is indicated. or alpha (I-V slope parameter), c) Imax , and d) Imin (maximum and min­ imun currents). Table 4 summarizes the magnitude and trend associated with the variation of each observable in terms of all the modified model parameters (a change of ± 30% from its reference value was considered for the analysis). The reference parameters are those obtained from the fitting of the median I-V curve (see Fig. 1). Notice that Table 4 uses colors and signs. Red color is associated with almost no dependence between the observable and the model parameter; orange indicates that for a 10% model parameter variation, less than 10% variation is detected in the observable and green corresponds to a variation larger than 10%. (+) or (-) indicate direct or inverse dependence, respectively. Reading the first column of Table 4, we can see the impact of all the inputs over the observable IHRS . It is observed for IHRS a strong and positive dependence on alpha, Imin, and gam. It also exhibits strong and inverse dependence on Vapp and etar, weak and positive effect from Vr and Imax and it is scarcely affected by Ri, etas, Rs, and Isb. This kind of test is of utmost importance for investigating the sensibility of the Fig. 6. Time evolution of experimental and simulated parameters for: a) HRS considered simulation model when subjected to variability. current, b) reset voltage, c) and d) the difference in the observable between cycle k + 1 and cycle k for IHRS and VR (ΔIHRS and ΔVR). 8. Conclusions is not high enough, it will seriously affect the sensitivity studies per­ In this work, we investigated uncorrelated C2C variability in RRAM formed on the rest of observables. This is an important point to consider devices using the QMM model. The analysis consisted in comparing within this analysis. experimental I-V curves, obtained from HfO2-based devices with SPICE In what follows, we will only take into account complete SET and simulations. First, the experimental curves were analyzed and the main RESET processes. This can be achieved with a maximum applied voltage observables extracted. By means of the different tools available in the of ± 2.5 V (see Fig. 7 in green). The parameters were swept one at a time fitdistrplus package from the R language, the best candidate distributions (50 steps each) in a reasonable range (this process does not include C2C for the experimental observables were determined. The obtained infor­ variability) and the obtained I-V curves analyzed. Fig. 8 shows the mation was included in the model script. Simulations indicate that the relative variation of the observables as a function of the relative varia­ QMM model can reproduce reasonably well the main features exhibited tion of selected model parameter: a) ηr or etar (reset transition rate), b) α by the experimental curves in terms of mean value and variability. It was 8 E. Salvador et al. Solid State Electronics 206 (2023) 108667 Table 4 Intensity and sign (direct or opposite) of four observables against all the analyzed parameters. + and – signs for direct and inverse dependence, respec­ tively. Green: greater than10%, orange: less than10% and red 0%. Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Data availability Data will be made available on request. Acknowledgements This work was supported by the Spanish Ministry of Science, Inno­ vation and Universities through project PID2019-103869RB-C32. ‘Departament de Recerca i Universitats de la Generalitat de Catalunya’ is acknowledged for the 2020 FISDU 00261 grant. References [1] Yu S, Chen PY. IEEE Solid State Circuits Magazine 2016;8(2):43–56. [2] J. S. Lee S. Lee, and T. W. Noh, Appl. Phys. Rev., vol. 2, no. 3, 031303 (2015). [3] Jo SH, Chang T, Ebong I, Bhadviya BB, Mazumder P, Lu W. Nano Lett 2010;10(4): 1293–301. [4] Chen A, Lin MR. IEEE IRPS 2011;843–846. [5] A. Saltelli S. Tarantola, F. Campolongo, and M. Ratto, John Wiley & Sons, Ltd, (2004). [6] Salvador E, Gonzalez MB, Campabadal F, Martin-Martinez J, Rodriguez R, Fig. 8. One-way sensitivity plots for the relative change of the observables a) Miranda E. Solid State Electron 2021;185:108040. IHRS, b) ILRS, c) VS and d) VR against the relative change of the model param­ [7] Miranda E. IEEE TNano 2015;14:787–9. eters: etar, alpha, Imax and Imin. [8] A. Saltelli , M. Ratto, T. Andres, F. Campolongo, J. Cariboni, D. Gatelli, M. Saisana and S. Tarantola, John Wiley & Sons, Ltd, (2008). [9] Bargallo M, Martin-Martinez J, Maestro M, Acero MC, Nafría M, Campabadal F. also shown that the study of time series and chained-parameter analysis IEEE Trans Electron Devices 2016;63(8):3116–22. [10] Piccolboni G, Molas G, Garbin D, Vianello E, Cueto O, Cagli C, et al. IEEE Electron for correlated C2C variability simulations is a required action. We also Device Lett 2016;37(6):721–3. investigated the impact that variations in the model inputs (parameters) [11] Poblador S, Gonzalez MB, Campabadal F. Microelectron Eng 2018;187–188: have in the model outputs (observables) using a one-way sensitivity 148–53. [12] Karpov V, Niraula D, Karpov I. Appl Phys Lett 2015;109:1–5. analysis. The role of each model parameter was assessed and presented [13] Delignette-Muller ML, Dutang C. J Stat Softw 2015;64:1–34. graphically summarizing the intensity and detected trend between [14] Alonso FJ, Maldonado D, Aguilera AM, Roldán JB. Chaos. Solitons and Fractals parameter and observable (direct or inverse). 2021;143:110461. 9 E. Salvador et al. Solid State Electronics 206 (2023) 108667 Emili Salvador Aguilera Is a PhD student at the Electronical Javier Martin-Martinez Javier Martin-Martinez received the Engineering department from the Universitat Autònoma de M.S. degree in physics from the Universidad de Zaragoza, Barcelona under the supervision of Enrique Alberto Miranda Zaragoza, Spain, in 2004, and the Ph.D. degree from the Uni­ and Rosana Rodríguez, from the same department. Prior to the versitat Autònoma de Barcelona (UAB), Bellaterra, Spain, in PhD, coursed my Bachelor degree in Nanoscience and Nano­ 2009. He was with the Università degli Studi di Padova, Padua, technology and Master degree in Advanced Nanoscience and Italy, and IMEC, Leuven, Belgium. He is currently an Associate Nanotechnology at the Universitat Autònoma de Barcelona. He Professor with UAB. His main research interests include the carried out my Master thesis at the Institut Català de Nano­ characterization and modeling of failure mechanisms in MOS­ ciència i Nanotechnology (ICN2) working in the graphene FETs and also RRAM characterization and modeling for neu­ transistors field. romorphic applications. Mireia Bargallo Gonzalez received the degree in physics from Rosana Rodriguez received the Ph. D. in Electrical Engineering the University of Barcelona, Barcelona, Spain, and the Ph.D. from Universitat Autonoma de Barcelona (UAB) in 2000. Fun­ degree on the topic of stress analysis and defect characteriza­ ded by the Fulbright program, she worked on devices and cir­ tion techniques of semiconductor materials and devices, from cuits reliability at the IBM Thomas J. Watson Research Center Katholieke Universiteit Leuven, Leuven, Belgium, in 2011. She (USA). Currently, she is associate professor at the UAB. Her pursued her Ph.D thesis with the Interuniversity Microelec­ research is focused on the variability and reliability of advanced tronics Center (imec), Leuven, Belgium. In 2011, she joined the CMOS devices. She is interested in the electrical characteriza­ Institut de Microelectrònica de Barcelona (IMB-CNM, CSIC). tion and modeling of process-related and time–dependent Her current research interests include electrical characteriza­ variability sources as Random Telegraph Noise (RTN) and tion, modeling and applications of resistive switching devices. aging mechanisms as Bias Temperature instability (BTI) and Hot Carrier Injection (HCI). Her research includes the study of the variability impact on the performance of single devices and digital and analogical circuits. She is also interested in the characterization of resistive switching devices (memristors) and their application for non- volatile memories, computing and neuromorphic applications. Francesca Campabadal received the Ph.D. degree in physics from the Universitat Autònoma de Barcelona, Bellaterra, Spain, ENRIQUE MIRANDA is Professor at the Universitat Autonoma in 1986. She joined the Institut de Microelectrònica de Barce­ de Barcelona (UAB), Spain. He has a PhD in Electronics Engi­ lona, Consejo Superior de Investigaciones Científicas, Barce­ neering from the UAB (1999) and a PhD in Physics from the lona, Spain, in 1987, where she is currently a Research Universidad de Buenos Aires, Argentina (2001). He received Professor. Her current research interests include the deposition numerous scholarships and awards including: RAMON y CAJAL of high-k dielectric layers, their electrical characteristics, and (UAB), DAAD (Technical University Hamburg-Harburg), the resistive switching phenomena in RRAM devices. MATSUMAE (Tokyo Institute of Technology, Japan), TAN CHIN TUAN (Nanyang Technological University, Singapore), WALTON award from Science Foundation Ireland (Tyndall National Institute), Distinguished Visitor Award (Royal Acad­ emy of Engineering, UK), CESAR MILSTEIN (CNEA, Argentina), Visiting Professorships from the Abdus Salam International Centre for Theoretical Physics, Slovak Academy of Sciences, Politecnico di Torino, Leverhulme Trust (University College London, UK), and Nokia Foundation (University of Turku, Finland). He serves as member of the Distinguished Lecturer program of the Electron Devices Society (EDS-IEEE) since 2001 and as Associate Editor of Microelectronics Reliability since 2003. He has authored and co-authored around 250 peer-review journal papers. 10