Modulating the Filamentary-Based Resistive Switching Properties of HfO2 Memristive Devices by Adding Al2O3 Layers
Modulating the Filamentary-Based Resistive Switching Properties of HfO2 Memristive Devices by...
Kalishettyhalli Mahadevaiah, Mamathamba ;Perez, Eduardo;Lisker, Marco;Schubert, Markus Andreas;Perez-Bosch Quesada, Emilio ;Wenger, Christian;Mai, Andreas
2022-05-11 00:00:00
Article Modulating the Filamentary-Based Resistive Switching Properties of HfO2 Memristive Devices by Adding Al2O3 Layers 1, 1 1,2 1 Mamathamba Kalishettyhalli Mahadevaiah *, Eduardo Perez , Marco Lisker , Markus Andreas Schubert , 1 1,3 1,2 Emilio Perez-Bosch Quesada , Christian Wenger and Andreas Mai IHP Leibniz-Institut für Innovative Mikroelektronik, 15236 Frankfurt (Oder), Germany; perez@ihp-microelectronics.com (E.P.); lisker@ihp-microelectronics.com (M.L.); schuberta@ihp-microelectronics.com (M.A.S.); quesada@ihp-microelectronics.com (E.P.-B.Q.); wenger@ihp-microelectronics.com (C.W.); mai@ihp-microelectronics.com (A.M.) Technische Hochschule Wildau, 15745 Wildau, Germany BTU Cottbus-Senftenberg, 01968 Cottbus, Germany * Correspondence: kalishettyhalli@ihp-microelectronics.com; Tel.: +49-335-5625-352 Abstract: The resistive switching properties of HfO2 based 1T-1R memristive devices are electrically modified by adding ultra-thin layers of Al2O3 into the memristive device. Three different types of memristive stacks are fabricated in the 130 nm CMOS technology of IHP. The switching properties of the memristive devices are discussed with respect to forming voltages, low resistance state and high resistance state characteristics and their variabilities. The experimental I–V characteristics of set and reset operations are evaluated by using the quantum point contact model. The properties of the conduction filament in the on and off states of the memristive devices are discussed with respect to the model parameters obtained from the QPC fit. Keywords: bi-layers; quantum point contact model; memristive device; embedded applications; Citation: Kalishettyhalli Mahadevaiah, M.; Perez, E.; Lisker, variability; conductive filament; CMOS compatibility M.; Schubert, M.A.; Perez-Bosch Quesada, E.; Wenger, C.; Mai, A. Modulating the Filamentary-Based Resistive Switching Properties of 1. Introduction HfO2 Memristive Devices by Novel applications, such as edge computing [1], big-data processing [2,3], image Adding Al2O3 Layers. Electronics recognition [4] etc., demand efficient computing techniques and advancements in 2022, 11, 1540. https:// memory storage technologies [5]. CMOS compatibility, low power consumption, low cost, doi.org/10.3390/electronics11101540 good endurance, fast switching, etc., are among the other features which are expected Academic Editor: Anna Richelli from the new memory technologies [6]. Besides the fact that the oxide-based memristive Received: 18 March 2022 devices can exhibit all the above-mentioned features, it is also possible to monolithically Accepted: 9 May 2022 integrate them with the CMOS logic on the same process nodes [7]. This adds to an ad- Published: 11 May 2022 vantage of using the memristive devices for embedded storage applications [8]. The memristive devices fabricated on silicon substrates are mainly used for memory Publisher’s Note: MDPI stays neu- storage [9], embedded [8] and neuromorphic computing applications [10]. Apart from sil- tral with regard to jurisdictional icon substrates, the memristive devices are also realized on polymer substrates for appli- claims in published maps and institu- cations in the field of flexible electronics [11]. Oxides such as TiO2 [12], NiO [13], ZnO [14], tional affiliations. etc., among many others, are used as a physically flexible switching material in memris- tive devices. Further, the memristive devices are fabricated on glass substrates, which have low thermal conductivities, compared to silicon substrates [15]. The glass substrates Copyright: © 2022 by the authors. Li- favor the diffusion of oxygen vacancies [16]. An improved resisting switching perfor- censee MDPI, Basel, Switzerland. mance in terms of higher memory window (MW), better endurance characteristics and This article is an open access article lower value of reset voltages are reported in various oxides, such as HfOx [15], SnO2 [16], distributed under the terms and con- TaOx [17] and MoO3 [18]. This work mainly focuses on the memristive devices which are ditions of the Creative Commons At- fabricated on the silicon substrates. tribution (CC BY) license (https://cre- ativecommons.org/licenses/by/4.0/). Electronics 2022, 11, 1540. https://doi.org/10.3390/electronics11101540 www.mdpi.com/journal/electronics Electronics 2022, 11, 1540 2 of 13 Among other memristive device types, the filamentary-based memristive devices have the advantages of good retention, fast switching and CMOS compatibility [19]. Re- sistive switching in the filamentary-based memristive devices is due to the redox reactions taking place at the switching layer, under the influence of electric fields [20]. Hafnium oxide (HfO2), acting as a memristive switching layer, is one of the most extensively stud- ied material in the literature [21,22]. The availability of the deposition processes and its CMOS compatibility are the main reasons for the wide usage of the material [23]. How- ever, HfO2 memristive devices integrated in back-end-of-line (BEOL) CMOS technology exhibit increased intrinsic device variabilities [24]. This challenge of reducing the variabil- ity in HfO2-based memristive devices led to the investigations on further CMOS compat- ible materials [25]. Various oxides, such as Al2O3, TiO2, Ta2O5, SiO2, etc., among many oth- ers, have been used in combination with HfO2 layers [26]. The material combinations are used either as ionic doping or in the form of stacked memristive layers, i.e., bi-layer, tri- layer and multi-layer memristive devices [27–29]. Due to their diverse potential applications, Al2O3|HfO2 bi-layers have been widely investigated in the literature [30,31]. The type of process and the precursors used for the deposition of Al2O3 layers plays a vital role for the performance of the memristive bi-layer devices [30–33]. Further, the aluminum oxide layer is widely used as a tunnel barrier in various kinds of memristive devices [34,35]. Various improvements of the resistive switching properties are reported in terms of uniform switching voltages and reduced dispersions of the high resistance state (HRS) [31], analog switching properties [33], etc. The conduction filament (CF) properties in single layer memristive devices are dis- cussed frequently with respect to the quantum point contact (QPC) model by various re- search groups [36–38]. Memristive switching oxides, such as HfO2 [38] and TaOx [39], are mainly used for the study. However, in the case of memristive bi-layers, in particular, Al2O3|HfO2 based devices, the resistive switching properties of the devices are discussed quite often, but the properties of their CF with respect to QPC are very seldomly dis- cussed. Hence, in this work, the filamentary-based resistive switching properties of memristive HfO2 devices are altered by adding the Al2O3 layers, which are deposited by using atomic layer deposition (ALD). Three different memristive layer stacks are com- pared in terms of their resistive switching behaviors. Further, the modulation in their con- ductive filament properties is analyzed within the frame work of the quantum point con- tact (QPC) model. The experimental I-V curves of the memristive devices from the set and reset operations are fitted using the QPC model for the low-resistance state (LRS) and the high-resistance state (HRS), respectively. Finally, the CF properties are discussed with re- spect to the model parameters obtained by fitting the experimental I–V characteristics to the QPC model. 2. Experimental The integrated 1T–1R memristor devices are fabricated using the standard 130 nm CMOS technology of IHP. The CMOS transistor of gate length 130 nm and gate width 150 nm has its drain terminal connected to the bottom electrode (BE) of the memristor device. This forms a series connection between the memristive module and the CMOS transistor. Figure 1 shows the EDX with TEM cross section of the integrated 1T-1R memristive de- vice. The integration of the memristor module into BEOL CMOS technology reduces the parasitic RC. The memristor module, which is essentially a metal–insulator–metal (MIM) struc- ture, is placed between metals 2 and 3 in the AlCu BEOL interconnects as shown in Figure 1. The BE of the memristor module consists of sputter deposited TiN of 150 nm thickness. The switching layers are deposited on top of the BE, using a CMOS compatible thermal ALD process at 300 °C. Aluminum oxide (Al2O3) layers are deposited by alternate pulsing of trimethylaluminum (Al2(CH3)6) as a precursor and water (H2O) as a reactant. Further, the deposition of HfO2 takes place by the alternate pulsing of hafniumtetrachloride (HfCl4) as a precursor and water (H2O) as a reactant. In order to avoid gas phase reactions, an Electronics 2022, 11, 1540 3 of 13 inert gas purge is performed after every pulse, which removes the unreacted precursor and reactants, and the byproducts of the self-termination reactions from the deposition chamber. The switching layers are deposited in three different types of stacks, namely, V1, V2 and V3, as illustrated in Table 1. The V1 variant is the reference sample, which consists of a single layer HfO2 of 8 nm thickness. The V2 and V3 variants consists of thin Al2O3 layers of 1 and 2 nm thickness, respectively deposited on top of the TiN BE, in addition to the HfO2 layer of 8 nm thickness, which is deposited successively without vacuum break- age. Eventually, V1, V2 and V3 device types comprise total dielectric layer thicknesses of 8, 9 and 10 nm, respectively. The presence of the thin Al2O3 layers is verified by the TEM cross section with EDX analysis as shown in Figure 2. The top electrode (TE) deposition of 7 nm thick Ti and 150 nm thick TiN above the dielectric switching layers prepares the MIM stack for subsequent process steps. The patterning of the MIM stack is one of the crucial steps in the memristor module fabrication and was realized by standard MIM module fabrication of a qualified SiGe–BiCMOS technology. The approach consists of an improved fabrication technique with a spacer and encapsulation process steps. Further details can be found in [40]. Figure 1. TEM cross-section of an integrated 1T–1R memristor device, fabricated in 130 nm CMOS technology. The inset micrograph illustrates the memristor module. Figure 2. Cross-sectional TEM images with EDX elemental mapping of memristor mod- ules in (a) V1 (b) V2 and (c) V3 variants. Electronics 2022, 11, 1540 4 of 13 Table 1. Variants of memristive devices with respective layer thicknesses. Description V1 (nm) V2 (nm) V3 (nm) TiN TE 150 150 150 Ti 7 7 7 HfO2 8 8 8 Al2O3 - 1 2 TiN BE 150 150 150 The TEM and the energy dispersive X-ray (EDX) images were prepared using the Tecnai Osiris tool, which was operated at 200 kV. EDX analysis was performed in the scanning TEM mode using the software Esprit from Brucker. Further, the EDX measure- ments were quantified using the Cliff–Lormier method. The TEM lamella of samples were prepared by using the NVision 40 focused ion beam (FIB) tool from Zeiss. The surface of the samples was protected by using a carbon layer deposited through ion beam deposition technique. The prepared lamellas were lifted out using a micromanipulator. X-ray photo- electron spectroscopy (XPS) depth profile measurements were carried on a PHI5000 Ver- saprobe II tool with an Al Kα X-ray source (1486.6 eV) at 89.7 W. The presence of thin Al2O3 layers was verified by using TEM and EDX analyses as shown in Figure 2. However, the stoichiometry of them was not determined using the EDX technique due to the limitation of their depth resolution. Further, the Al2O3 films were deposited using an industry standard TALD process with negligible nucleation de- lay with respect to their growth cycles [41]. The films were grown layer by layer using a self-terminated surface reaction process; they are reported widely in the literature to be stoichiometric [42], [43]. Additionally, the Al2O3 layers grown on silicon substrates were analyzed using X-ray photoelectron spectroscopy (XPS) depth profile analysis for their stoichiometry as shown in Figure 3. The ratio of O/Al atomic concentrations was deter- mined to be ~1.5, indicating the Al2O3 layers as being stoichiometric. Figure 3. XPS depth profile analysis of Al2O3 layers deposited using TALD process at 300 °C. Electrical measurements of 1T-1R V1, V2 and V3 devices were performed under iden- tical DC conditions at room temperature. The resistive switching performance of the memristive devices was tested with a Keithley 4200-SCS semiconductor parameter ana- lyzer connected to a FormFactor PMV200 manual probe station. The characterization of the memristive devices begins with a crucial and onetime operation step called forming. During forming, the drain voltage (VD) is double swept from 0 to 4 V while grounding the source terminal (S) and biasing the gate terminal (G) to 1.5 V. The forming operation is Electronics 2022, 11, 1540 5 of 13 followed by reset and set operations. The reset operation was performed at a gate bias (VG) of 2.9 V and, the source voltage (VS) was double swept from 0 to 2 V while grounding the drain terminal (D). The set operation was performed similar to forming, except that the VD was double swept from 0 to 2 V while grounding S. Finally, 10 devices of each variant were programmed by 50 subsequent cycles of set and reset operations. 3. Quantum Point Contact (QPC) Modelling The conduction filament (CF) properties of V1, V2 and V3 memristive devices were analyzed using the QPC model. The reset and set I–V characteristics were used to model the conduction properties of the CF in HRS and LRS states of the memristive devices, respectively. The HRS I–V characteristics are modeled as [36] ( ) I = + (1) [ ( ) ] where I is the measured current, V is the applied voltage, β is the potential drop at the cathode and anode interfaces, e is the elementary charge of an electron, h is the plank’s constant, G/Go is the conductance parameter which is also equal to number of CFs at very low voltages, φ is the potential barrier height, and α is the parameter related to the poten- tial barrier thickness (TB). Due to the asymmetry of the potential drop at the two ends of the CF, the β value is estimated to be 1. The presence of a potential barrier disrupting the CF is assumed in the HRS for all the three different types of devices V1, V2 and V3. A value of G/Go equal to 1 is assumed. According to Lian et al., in the case of low voltages and high enough potential barriers, Equation (1) converges as below [37]: ( ) I = + (2) where N is the number of CFs at HRS, which is assumed to be 1. The current limiting transistor is connected in series with the memristor device in a 1T–1R test structure. Con- sidering the real case scenario of testing the memristive devices in the form of arrays, the read operation of the HRS takes place in the linear region of the transistor [44]. The re- sistance of the transistor at this point is negligible compared to the resistance of the memristive device [44]. Hence, the resistance of the select transistor is not taken into ac- count for the HRS simulation using the QPC model. Due to the metallic-like conductivity of the CF in the LRS of the memristive devices, the barrier confinement parameter α collapses to zero, resulting in a linear I–V relation. The resistance of the transistor in this case is comparable to the memristive device and hence cannot be neglected [45]. The value of R is determined from the simulations and electrical characterization of the transistor. Finally, the LRS currents equation within the frame work of QPC model is illustrated as [38] I = (3) 2 −1 where Go = 2e /h = (12.9 kΩ) is the quantum conductance unit, N is the number of CFs and, and R = 3 kΩ is the series resistance extracted from the transistor output characteris- tics. The expression for the width of the potential barrier (TB) in the HRS state of the memristive device is illustrated as [36] T = (4) where m* is the effective mass of the electron within the CF. The radius of constriction (RB) of the CF in the HRS state of the memristive device is expressed as [36] R = (5) Electronics 2022, 11, 1540 6 of 13 where zo is the first zero of the Bessel function Jo [36]. The value of zo is equal to 2.404. 4. Results and Discussion The mean values of the forming voltages with their dispersions versus the total die- lectric thickness of V1, V2 and V3 memristive devices are as shown in Figure 4. Further- more, the corresponding I–V characteristics of the forming operations are illustrated in Figure 5. The forming voltages increase with the addition of the Al2O3 layers in the V2 and V3 devices. According to the literature, the forming voltage of the memristive devices is directly proportional to the thickness of the dielectric and inversely proportional to the square root of the dielectric constant √