Atomic layer deposited nanolaminates of zirconium oxide and manganese oxide from manganese(III)acetylacetonate and ozone

Atomic layer deposition method was used to grow thin films consisting of ZrO2 and MnO x layers. Magnetic and electric properties were studied of films deposited at 300 °C. Some deposition characteristics of the manganese(III)acetylacetonate and ozone process were investigated, such as the dependence of growth rate on the deposition temperature and film crystallinity. All films were partly crystalline in their as-deposited state. Zirconium oxide contained cubic and tetragonal phases of ZrO2, while the manganese oxide was shown to consist of cubic Mn2O3 and tetragonal Mn3O4 phases. All the films exhibited nonlinear saturative magnetization with hysteresis, as well as resistive switching characteristics.

Mn-doped zirconia has also been reported to be a magnetic material [11,12]. Magnetic polarization has been achieved and studied also in Fe, Co and Ni doped zirconia [13], Ca and Mg doped [14], Ag doped [15] and even in undoped zirconia [16,17]. MnO 2 has been used as a catalyst [18] and an electrode [19]. MnO has been used as electrode material [20]. Mn 2 O 3 has been used for water splitting devices [21].
This study is devoted to ALD of laminated thin films consisting of ZrO 2 and manganese oxide. The purpose was to firstly investigate the process of depositing manganese oxide from (Mn(acac) 3 ) and ozone, O 3 , and, thereafter, study the memory effects in nanolaminated zirconia and manganese oxide. Specifically, possible ferromagnetic, ferroelectric and resistive switching (RS) properties of such thin films were probed.

Experimental details
The films were grown in a low-pressure flow-type ALD reactor [38]. Mn(acac) 3 (97%, ACROS Organics) was used as the manganese precursor and evaporated at 183°C. Zirconium precursor, ZrCl 4 (99.9%, Aldrich), was evaporated at 157°C. Both metal precursors were evaporated from a glass boat inside the reactor at a pressure of 220 Pa. Nitrogen, N 2 (99.999%, AS Linde Gas), was applied as the carrier and purging gas. Ozone was produced from O 2 (99.999%, AS Linde Gas) using BMT Messtechnik 802 N generator. The ozone concentration at the reactor inlet, was 245-250 g m −3 in the experiments. The estimated ozone flow rate from the generator was about 68 scc m −1 , while the carrier gas flow rate was kept at about 220 scc m −1 .
One ALD cycle of a binary oxide consists of four sequential pulses. For producing a layer of ZrO 2 , the sequence was ZrCl 4 pulse-N 2 purge-O 3 pulse-N 2 purge. In the case of depositing a manganese oxide layer, the metal precursor was Mn(acac) 3 , while purging with N 2 and oxidizing with O 3 was similar to the ZrO 2 scheme. Pulse times for depositing both metals were 5-2-2-5 s for the sequence given above. Various layered structures were deposited, for example, 100 cycles of manganese oxide followed by a 100 cycles of zirconia, more details are given in table 1.
ZrO 2 -MnO x films were grown on cleansed and etched Si(100) [39] substrate surfaces as well as on surfaces of conductive TiN layers. The TiN bottom electrode layers were pre-deposited on Si(100) wafers with a resistivity of 0.014-0.020 Ω·cm, i.e. on Si boron-doped to the concentrations up to 5×10 18 -1×10 19 cm −3 . TiN was grown by pulsed chemical vapor deposition using a batch TiCl 4 /NH 3 process [40,41] at temperatures of 450°C-500°C in an ASM A412 Large Batch 300 mm reactor at Fraunhofer IPMS-CNT. The films, which were deposited on TiN substrates for electrical measurements, were also supplied with Ti electrodes (area 0.204 mm 2 ) electron-beam evaporated on top of the films. The structure to conduct electrical measurements was, from top to bottom, Ti/ZrO 2 -MnO x /TiN/Si/Al.
Spectroscopic ellipsometer (SE), model GES5-E, was used for the measurements of the films thicknesses. Ellipsometric data was modeled in the range of 1.3-5.0 eV using the Tauc-Lorentz dispersion model. X-ray fluorescence (XRF) spectrometer Rigaku ZSX 400 and program ZSX Version 5.55 were used to measure the elemental composition of the films. Measurements were performed with a semi-quantitative (standardless analysis) program (SQX). SQX is a program to obtain concentrations of elements by theoretical calculation using the fundamental parameter (FP) method and internal sensitivity library.
The x-ray photoelectron spectra (XPS) were collected at normal emission using a Gammadata/Scienta SES100 hemispherical analyzer and a Thermo VG Scientific XR3E2 non-monochromatic dual anode x-ray source (Al-K α /Mg-K α ). The analyzer energy scale calibration was checked against the 4f 7/2 line from cleaned gold foil at 84.0 eV binding energy. Relevant to estimating elemental composition from XPS survey scans, the constant (i.e. independent of photoelectron kinetic energy) analyzer transmission function was checked against accessible core level lines of clean Au, Ag and Cu samples, and additionally asserted by constant magnification in spatial imaging (in the non-energydispersive direction) of a structured test sample through the electron optics over the entire used kinetic energy range). Spectral components were fitted, and elemental content from survey spectra was estimated using CasaXPS software [42]. XPS was exploited, in the first place, to detect the presence of carbon, complementarily to XRF, and study the differences of the carbon bonding in the films grown at two different temperatures, 250 and 300°C, i.e. that providing the most uniform films in terms of thickness, and that probably enabling higher degree of structural ordering, respectively. No preclean or surface etching was conducted before XPS measurements, since deposition of metal electrodes was in order after the deposition of oxide films, and that could not be carried out in a continuous process without exposure to the laboratory air. Nevertheless, since the probing and analysis depth of XPS extends through topmost 4-5 nm of a film, the method allowed one to acquire data adequately, sufficient for the description of bonding between elements constituting the solid layers. Different roles for carbon impurities were expected due to the lower rate of surface exchange reactions at lower temperatures. The crystal structure was evaluated by grazing incidence x-ray diffractometry (GIXRD), using an x-ray diffractometer SmartLab Rigaku with CuKα radiation, which corresponds to an x-ray wavelength of 0.15406 nm.
RS measurements were carried out by means of a semiconductor analyzer Keithley 4200SCS, with samples put in a light-tight probe station. The DC voltage was applied to the top electrode, leaving the bottom one grounded. The current −voltage curves were measured by applying voltage sweeps while incremental voltage pulses were applied in order to obtain the memory maps seen in figure 12. The RS measurements appropriate for recording common current−voltage hystereses as well as those enabling admittance memory mapping have been described in a number of earlier studies, e.g. those devoted to HfO 2 [43] and ZrO 2 -Al 2 O 3 [44] based media. The aforementioned semiconductor analyzer can also be used to perform the Double Swing Quiescent Current (DSQC) technique [45], a novel method which would allow one to detect the polarization curves of thin film samples implying ferroelectric behavior.
Magnetic measurements were performed using Vibrating Sample Magnetometer (VSM) option of the Physical Property Measurement System 14T Quantum Design by scanning the magnetic field from −0.5 to 0.5 T parallel to the film surface at 300 K and −3.5 to 3.5 T at 5 K.

Film growth and composition
Data regarding the composition of films is given in table 1. Since the zirconium precursor contains chloride and the manganese precursor contains carbon, amounts of these elements were measured as well.
Certain issues became related to the uniformity of the film thickness along the gas flow direction in the flow-type reactor used in the experiments. For ZrCl 4 as the Zr precursor in ALD, it is known, that this particular precursor does not decompose thermally even at elevated temperatures. Even at the substrate temperature of 500°C, the thickness of ZrO 2 films grown from ZrCl 4 and H 2 O may not deviate more than 12%-15%, measured along the substrate length [46]. Formation of certain thickness profile in chloride based ALD processes is probably inevitable in such flow-type reactors, possibly caused by the effect of secondary surface reactions, i.e. 'poisoning' of the surface by reaction products, that is HCl [47]. However, in the case of organic and oxygen-containing ligands to metals in the precursors like acetylacetonates, the thermal decomposition of the precursor can start in the gas phase during its transport in the reactor towards substrates, and, in this case at already slightly above 300°C, the oxide layer may, thus, deposit with higher rate, and preferentially, to the regions closer to the leading edge of the substrate. This has, evidently, became a considerable issue in the case of the manganese oxide process, as can be seen in the figure 1.
The XPS studies on elemental composition of the surface region of the films deposited at two different deposition temperatures 300 and 250°C from XPS survey spectra revealed O:Mn content ratio approximately 1.4 for the film grown at 300°C, similarly to the value obtained by XRF. For the film grown at 250°C, this content ratio increased up to 2.5 and, considering the probability of formation of surface carbonate instead of oxide, increasing with decreasing substrate temperature due to incomplete reactions, the results obtained from the film grown at 250°C could be interpreted via 3-4 nm thick surface layers dominantly consisting of MnCO 3 (figure 2).
The Mn 2p spectrum (figure 3), although at least partially demonstrating formation of solid manganese oxide in the present study, is generally not convenient to use for the identification of the Mn charge state due to the underlying (2p 1 -3d final state) multiplet and nontrivial spectral structure, but the main peak maxima positions suggest 2 + and/or 3 + Mn charge states, and the satellite peak (figure 3) at binding energies ∼5 eV higher than the main lines (in the 250°C deposited sample spectrum) can be regarded as that typical of Mn 2+ [48]. The films grown at 250°C remained highly disordered. Multiphase structure started to form and become detected by x-ray diffraction studies or films deposited at 300°C . The main diffraction maxima were attributed to tetragonal Mn 3 O 4 , while Mn 2 O 3 was identified as an additional phase (figure 6), as will be described below. Both GIXRD and Mn 2p XPS implied that the phase composition of the MnO x films grown in the present study, especially at lowered temperatures, are complicated to define unambiguously (some further  insight is provided from Mn 3s XPS, below). Besides, although manganese can take on all charge states from +7 to −3, the most abundant (and stable) oxides have charge states +2, +3 and +4, and the formula Mn 3 O 4 can sometimes be written as MnO·Mn 2 O 3 , although such physical mixture picture would imply exchange interactions such that the magnetic properties would be that of an antiferromagnet (even at low temperatures), different from what we observe (below). Figure 4 depicts oxygen 1s spectra obtained from the same reference films grown at temperatures of 250 and 300°C. The XPS spectrum for O 1s acquired from the film grown at 300°C demonstrated a dominant sharp peak at 529.9 eV, which is a value typical of bulk manganese oxides [49][50][51]. In addition, a shoulder structure was resolved and centered around 531.5 eV, which has been related to the presence of (surface) hydroxyl groups [52], accompanied by a smaller amount of C−O bonded impurity with the component peak above 532 eV and a trace peak above 533 eV. The latter could be associated to physisorbed water vapor. However, the signals detected in the spectral region in the range of 531-534 eV would also accommodate and reveal double-bonded carbon of carbonates [53,54] and/or ketonate ligands [55], likely arising from the residues of the metal precursor also exploited in the present study, amply present in the film grown at 250°C.
The multiplet (exchange) splitting in the Mn 3s [56] due to the remaining single 3s electron in the final state having spin either parallel or antiparallel to the aligned spins of the 3d valence electrons has allowed to distinguish less ambiguously between the different Mn charge states. The size of this splitting depends also on the type of bonding (hybridization) and typically stays at less than half of the theoretically estimated value of ∼13 eV [57]. It is experimentally observed to range from 4.3 eV for Mn 4+ to ∼6 eV for Mn 2+ [50,58]. The Mn 3s splitting measured in the present study (figure 5) exhibited a peak separation of 5.9 eV for the film deposited at 300°C. The spectral shape occurred virtually identical to that of bulk Mn 3 O 4 measured earlier [58]. In the present study, the observation is also supported by the GIXRD results, allowing one to consider the Mn 3 O 4 as dominant crystalline phase in such films. One can note that separation of the peaks in the measured photoelectron spectrum appears to differ significantly from those in the dominantly Mn 3+ containing model compounds [58,59]. At the same time, the spectrum measured from the film grown at 250°C possessed quite a similar shape (figure 5), but the peak separation is perceptibly, even if by as little as 0.1 eV, larger, and the lower binding energy component at 83 eV narrower, which is consistent with the features characteristic of a material containing significant amounts of MnCO 3 , as deduced from estimates from C 1s and O 1s XPS above.   In further experiments aiming at tailoring oxides of manganese and zirconium, the MnOx growth process was carried out at 300°C, providing films with acceptable thickness uniformity and with much less carbon residues. In reference growth experiments, 200, 100, and 50 ALD cycles of manganese oxide at 300°C resulted in films thicknesses of 50, 20, and 7 nm, respectively. This implies that the growth of the film was not linear against the amount of the deposition cycles, but increased from 0.14 to 0.25 nm/cycle. Most likely this is caused by the inhibited nucleation in the early stage of the film growth, whilst after certain film thickness e.g. number of cycles, the linear growth begins. The initial nucleation determines the structure formation and hence the film properties, which in turn is dependent on the type of substrate and its surface functionalization, precursor chemistry and deposition temperature used in ALD process. The composition measurements allowed one to state that the value of x for MnO x was around 1.4, irrespectively of the thickness.

Film structure
As mentioned above, in MnO x films x=1.4, measured by XRF, implying the formation of a mixture of crystalline compounds with different stoichiometry. The GIXRD studies of films deposited at 300°C revealed, that both the cubic phase of Mn 2 O 3 (PDF Card 01-078-0390) and tetragonal phase of Mn 3 O 4 (PDF Card 00-018-0803) were present in the manganese oxide films not combined with ZrO 2 ( figure 6). Moreover, one diffraction maximum attributable to tetragonal MnO 2 (PDF Card 01-072-1982) could be recognized, although the appearance of a solitary reflection is not enough to confidently claim the corresponding phase actually formed in the films. ZrO 2 was found to be formed as a mixture of cubic (PDF Card 01-077-3168) and tetragonal (PDF Card 01-075-9649) phases. These phases can be difficult to distinguish, since several diffraction maxima tend to overlap. However, the 211 reflection of the tetragonal ZrO 2 polymorph was unambiguously recognized (figure 6), neighboring the closely located 311 reflection of the cubic ZrO 2 . The intensities of diffraction maxima are somewhat correlated to the thickness of films, whereby multi-layered structures demonstrating less intense maxima than those observed in non-laminated reference films, since their thicknesses are lower as well.
For a comparison with literature data, Mattelaer et al [21] have demonstrated the possibility to control phase composition of ALD-grown manganese oxides by applying either Mn(thd) 3 and O 3 , or Mn(thd) 3 and NH 3 plasma as precursors, followed by annealing the as-deposited MnO or MnO 2 films either in reducing or oxidative ambient in the temperature range of 450°C-900°C. Nilsen et al [36] have grown MnO 2 films by ALD also from Mn(thd) 3 and O 3 at 186°C in a flow type ALD reactor. In the latter study, additional test experiments revealed that Mn 3 O 4 could be grown at 200°C using O 3 under atmospheric pressure, otherwise the Mn 3 O 4 was achieved after increasing the substrate temperature above 230°C . In addition, Ghods et al [37] have obtained MnO thin films by ALD at substrate temperature of 200°C using Mn(acac) 3 and H 2 O as precursors. In the latter study, the stoichiometry of MnO was determined by x-ray photoelectron spectroscopy.

Electrical and magnetic properties
The films demonstrated nonlinear saturative magnetization with hystereses and moderate but measurable coercive fields similar to those common for ferromagnetic materials. Notably, figure 7 shows room temperature ferromagnetic-like behavior, most markedly apparent in the thinnest nanolaminate, with a saturation value of 1.5.10 −6 emu. Similar magnetic moment values in ZrO 2 films and its laminated structures with other metal oxides have been obtained by various authors [39,60,61].  Low temperature magnetic isotherm or hysteresis loops measured at 5 K only confirmed the existence of Mn 3 O 4 in all of the samples. The coercive field (H c ) measured was 10.1 kOe for the non-laminated MnO x film as well as for most laminates (figures 8 and 9). It was slightly higher for the 50+50+50+50+50 film at about 11.5 kOe ( figure 9). Such high values of H c have previously been reported for Mn 3 O 4 nanoparticles with small crystallite sizes. [65] figure 9 is the same as figure 8, but with the manganese oxide film excluded, since the saturation magnetization value for manganese oxide is so high that other samples are barely distinguishable on the graph in figure 8.
The most prominent feature from our temperature dependent zero field cooled (ZFC) and field cooled (FC) magnetization measurements is the bifurcation (irreversibility) of ZFC-FC loops and a large increase in magnetization below 46 K (figure 10), which coincides with the Curie temperature of Mn 3 O 4 and clearly indicates the presence of the ferrimagnetic oxide in all four of the samples. No cusp or maxima in ZFC-FC curves indicating any of the antiferromagnetic (AFM) transitions could be detected. If the AFM phases exist in the samples, then their magnetic moment is below the sensitivity level of the PPMS VSM used in the measurements. Temperature dependent susceptibility of the pure ZrO 2 was flat and featureless, appropriate for a diamagnetic material.
All films exhibited RS characteristics ( figure 11). This means that, under electrical stimulus, these samples change their resistance state and retain its value even when the power is turned off (non-volatile effect). It has earlier been established that the physics associated with RS involves the movement of both ions and electrons [66]. Furthermore, these devices require an electroforming step [67], an initial electrical stress necessary to activate the switching property. After this process, the characteristic RS current−voltage (I−V ) curves can be obtained. Bivalued curves which present two distinctive conduction states referred to as the high-resistance (HRS) and the low-resistance (LRS).
The pure (non-laminated) MnO x and ZrO 2 samples presented the widest functional windows (figures 11(a) and (b)), along with the 50xMnO x +50xZrO 2 (50+50 sample/film), which was the thinnest film as grown. The window narrowed as the number of cycles increased, as we can see when comparing the 50+50 and the 100xMnO x +100xZrO 2 samples (100+100 sample/film) ( figure 11(c)). This is also shown in figure 12, in which the memory maps [68] drawn from the same films are depicted. This fact can be explained taking into account that the number of ALD cycles determines the thickness of the film (14 nm versus 36 nm, in this case). The 100+100 film also demonstrated the narrowest window together with the multilayer film, which consisted of five sequential layers in stack. In addition to the thickness values (36 nm and 35 nm respectively), this might be due to the   existence of several interfaces between the constituent layers, which could hinder the formation of the conductive filament.
The non-laminated ZrO 2 film had the lowest SET and RESET voltages ( figure 11(a)). However, this film was also the one with the highest current values for both LRS and HRS states (see also figure 12(a)), showing, at the same time, the steepest SET and RESET transitions. At first glance, this could be seen as an advantage, as these transitions are also the fastest of all the samples. Nevertheless, this steep transition implies that ZrO 2 samples alone may not perform well in multilevel applications, and the high current values may simply make it the least suitable for non-volatile-memory applications, as the power consumption would remain high. The results for non-laminated ZrO 2 results are in a good accordance with the ones presented earlier by Ossorio et al [69].
The non-laminated MnO x sample the lowest current values for both high and low resistance states (figures 11(b), (c)), which provide similar resistance values to those determined by Zhang et al [70] when applying Ohm's law to the current values obtained for both HRS and LRS in the memory map that can be seen in figure 12. The memory maps are an image of the state of the RRAM cell after a previous stress stimulus, and are obtained by measuring some representative magnitude of the state at voltages low enough so that the measurement process itself does not alter the state of the device [43,68].
The low current values, along with a notably wide functional window ( figure 12(b)), make the MnO x film the most adequate for memory applications. Additionally, this sample presented smooth SET and RESET transitions, thus showing multilevel capabilities, which have been demonstrated to be useful for neuromorphic applications, i.e. artificial neural networks, as they might behave as electronic synapses [71].
The latter is also true for the MnO x -ZrO 2 mixed samples, which all presented smooth SET and RESET transitions, as well as lower HRS and LRS currents when comparing them to the non-laminated ZrO 2 film. Thus, MnO x -ZrO 2 films might become further investigated and suited to multilevel applications, as implied especially by the 100+100 cycle sample film, which demonstrated the smoothest transitions. On the other hand, these stacks possessed rather narrow functional windows and, consequently, may not regarded as the most appropriate for memory applications, with the exception of the 50+50 cycle sample, which really exhibited high ratio between HRS and LRS as well as appreciably low current values.

Summary
ZrO 2 and manganese oxide films, as well as their layered structures, were deposited by ALD, using ZrCl 4 , Mn(acac) 3 and ozone. XPS results confirmed significant contribution from carbon containing species in the manganese oxide film deposited at 250°C, allowing one to consider even the formation and presence of MnCO 3 . At the same time, in the film deposited at 300°C, essentially weaker role for residual carbon, and formation of Mn 3 O 4 was detected and determined by both XPS and GIXRD. Therefore, 300°C was chosen for the deposition temperature for nanolaminates. Non-laminated ZrO 2 was found to form in its cubic and tetragonal polymorphs, whereas non-laminated manganese oxide possessed cubic Mn 2 O 3 and tetragonal Mn 3 O 4 phases. All the films were found to behave as ferromagnetic-like materials. At the same time, hysteretic charge polarization-electric field behavior was not registered, and also the DSQC method measurements did not imply any ferroelectric behavior. Manganese oxide films demonstrated both saturation magnetization and coercivity values by two orders of magnitude higher at 5 K, compared to the values measured at 300 K. In ZrO 2 -MnO x films and nanolaminates, the polarization values at the aforementioned temperatures reached differed by about ten times. Manganese oxide lost its high magnetization value at 40-45 K.
All films exhibited excellent resistive switching characteristics with almost no difference between their commutation values. However, there appeared remarkable differences in the current values linked to both high and low resistance states, with the non-laminated ZrO 2 sample film showing the highest ratios between these states. The nonlaminated MnO x film possessed the lowest ratio, while in the different ZrO 2 -MnO x samples the ratio between high and low resistance states varied. Both non-laminated samples present great functional windows between each resistance state. Nevertheless, the mixed ZrO 2 -MnO x films have very different windows between these states, with the 50 x MnO x +50 x ZrO 2 showing a great functional window, while the 100xMnO x +100xZrO 2 and the multilayer film show the narrowest windows. This is explained by the variation in thickness and the fact that the multiple interfaces between the layer could hinder the formation of the filament. These results prove the potential interest of the MnO x and ZrO 2 /MnO x -based MIM structures in both non-volatile memories and neuromorphic applications.