Interplay of Precursor and Plasma for The Deposition of HfO2 via PEALD: Film Growth and Dielectric Properties

HfO2 thin films are appealing for microelectronic applications such as high‐κ dielectric layers, memristors, and ferroelectric memory devices. To fulfill the different requirements of each application, the properties of the deposited material need to be tuned accordingly. In this context, plasma‐enhanced atomic layer deposition (PEALD) is a powerful processing route to tailor the properties of HfO2 thin films, especially at low temperatures. Herein, a comprehensive bottom‐up approach is presented, ranging from the synthesis of molecularly engineered Hf precursors to the development of a HfO2 PEALD process and a detailed evaluation where plasma can be exploited to tune the dielectric properties. With the example of the newly synthesized bis‐(dialkylamido)‐bis‐(formamidinato) Hf(IV) precursor, [Hf{η2‐(iPrN)2CH}2(NMe2)2] which is reactive, thermally robust and volatile, successful implementation in a PEALD process for HfO2 at low temperatures is demonstrated. The typical atomic layer deposition (ALD) characteristics of precursor saturation, linearity, and ALD temperature window are demonstrated with constant growth of 0.7 Å per cycle from 125 to 200 °C, yielding high‐purity layers. The effect of plasma pulse duration on the chemical composition alongside structural, topographical, as well as dielectric properties of the films is investigated. For the latter, the films are incorporated in metal‐insulator semiconductor (MIS) structures.


Introduction
The ongoing trend of miniaturization in electronic components has led to new challenges for device manufacturers and brought SiO 2 to the limit of its thickness. [1][6] In terms of the application of hafnia layers as transistor dielectrics should preferably have an amorphous structure because in crystalline structures, grain boundaries can cause current paths. [7]An indication of this phenomenon could be an undesirably high leakage current.On the other hand, the permittivity of crystalline hafnia layers is usually higher than that of amorphous ones, resulting in a trade-off between the permittivity and leakage current that must be considered for the desired application.Thus, the dielectric properties, defect chemistry and the leakage current must be considered as the selection criteria for a high- film.In addition to the transistor application, oxide layers are of considerable interest for resistive switching applications. [8,9]The control of conductivity by an applied electric field is achieved in the oxide layer by charged oxygen vacancies forming and rupturing localized conductive filament paths for the preferential or undesirable transport of charge carriers. [10,11]Conductive filaments were identified to be the directionally aligned crystalline regions in amorphous HfO 2 consisting of monoclinic and orthorhombic oxygen-deficient phases. [12][15] Moreover, due to its high refractive index ranging from 1.8 to 2.2 and low absorption, [16] hafnium oxide is widely used for optical multilayer coatings for ultraviolet (UV) and infrared (IR) applications.The hardness of HfO 2 also makes it useful for the preparation of optical and protective coatings in the production of special glass types for fiber-optic products. [17,18]For laser applications films with low impurities are desired to guarantee low optical losses. [19]hese very different fields of application for HfO 2 require different layer properties with their respective desirable morphologies, crystallinities, and defect densities.For hafnia layers prepared by chemical methods such as chemical vapor deposition (CVD) or atomic layer deposition (ALD), these properties can be specifically tuned by selecting suitable precursors and process parameters, demonstrating the importance of understanding the deposition process and tailoring it to the specific application.22] In addition, plasma enables film-forming reactions of the precursor molecules at temperatures that would not allow film formation in a pure thermal process or when the precursor is not reactive towards standard co-reactants.Low process temperatures enabled via PEALD allow direct growth on flexible or nanostructured components or strongly curved substrates and great efforts have been made to extend the range of HfO 2 based applications to flexible electronic devices. [23,24]On the other hand, in PEALD processes, interaction with plasma radicals can damage the interface or lead to the formation of a thicker interface and thus relativize the advantages of the low process temperatures.For high-quality HfO 2 films, in addition to the precise adjustment of the plasma parameters such as the duration of the plasma pulses, the temperature, gas flow rates, pulse and purge durations the power, etc., the use of suitable precursors is crucial in any ALD-type process.The precursor needs to fulfill the prerequisites of high volatility, thermal stability, and reactivity towards the plasma species as well as the substrate.Thus, intensive research has been conducted to develop optimized precursors that can be divided into several classes.Early reports on ALD of HfO 2 focused on halide-based precursors, especially hafnium(IV) chloride (HfCl 4 ).This compound, however requires high deposition temperatures of more than 300 °C with water, [25] ozone [26] or oxygen [27] as co-reactants due to the strong metal -halide bond.Furthermore, chlorine containing precursors are prone to formation of etching by-products such as HCl that can interfere with the film growth and potentially damage the reactor. [28]s a halide-free alternative, the class of alkylamide containing Hf precursors namely [Hf(NMe 2 ) 4 ], [29][30][31][32] [Hf(NEt 2 ) 4 ], [30,33] and [Hf(NEtMe) 4 ] [30,34] have been commonly been used in thermal ALD processes.These compounds are known to possess limited long term thermal stability under the conditions that are adopted for ALD processes.[37][38] While they were also used in PEALD processes, temperatures of 200 °C and above were still required. [4,39,40]Another issue of monodentate alkylamide ligands are the vacant coordination sites on the Hf center that can facilitate oligomerization reactions, limiting the shelf-life of these compounds. [41]Thus, efforts have been made to enhance the thermal stability without compromising much on the volatility and reactivity of the parent complexes.In this context, heteroleptic precursors have been demonstrated to be a better alternative.It has been found that cyclopentadienyl (Cp) containing precursors show a high thermal stability accompanied by sufficient volatility and have thus been successfully implemented in ALD processes.[44][45] These findings indicate a limited reactivity of Cp based Hf precursors, that can be explained by a high activation energy for the chemisorption and by steric hindrance of the bulky Cp ligand, [46] suggesting that the class of Cp containing precursors is not well suited for lowtemperature deposition of HfO 2 .
Despite increasing interest for the low temperature fabrication of HfO 2 , the library of available Hf precursors for processes operating below 200 °C is still very limited.Among them is a study by Kessels et al.where the authors were able to counteract the limited reactivity of Cp precursors by employing an O 2 plasma as the co-reactant to develop a PEALD process using [Hf(CpMe)(NMe 2 ) 3 ] with a GPC of 1.1 Å from 150 to 400 °C. [22]This highlights the potential of PEALD to enable lowtemperature processes.Long plasma pulses of 8 s were necessary to fully oxidize adsorbed precursor molecules.To implement HfO 2 thin films as a dielectric, shorter plasma pulses would be beneficial as the oxygen plasma can lead to an SiO 2-x interface layer that gives rise to interface trap states which diminish the performance of the dielectric. [47]urthermore, a recent study demonstrated growth of HfO 2 at temperatures as low as 30 °C using [Hf(NMe 2 ) 4 ] and water, but long purging times of 150 s were required to remove residual water from the reaction chamber. [48]Furthermore, the films contain considerable amounts of hydroxyl contributions, an issue that is often observed with water as the co-reactant and that could negatively influence the properties of the HfO 2 layers.Nevertheless, this process demonstrates the high reactivity of precursors containing Hf-N bonds.To exploit this feature while avoiding the previously discussed drawbacks of the monodentate alkylamido ligands, it is a promising approach to introduce bidentate chelating nitrogen containing ligands.This was previously shown by our group with the partial replacement of amido ligands in tetrakis(alkylamido)hafnium by chelating guanidine ligands yielding heteroleptic precursors with enhanced thermal stability compared to their respective parent amide complexes.These heteroleptic complexes are still volatile and reactive towards oxygen sources. [47,49]For example, [Hf{ 2 -(( i PrN) 2 CNMe 2 )} 2 (NMe 2 ) 2 ] resulted in GPCs of 1.0-1.2Å from 100-225 °C in a thermal process using water as the co-reactant. [49]Similarly, [Hf{ 2 -( i PrN) 2 CNEtMe}(NEtMe) 3 ] (Hf(DPGUAN)(NEtMe) 3 ]) was employed in a low-temperature PEALD process with an oxygen plasma, resulting in a GPC of about 1.1 Å from 60 to 240 °C.The obtained HfO 2 films were of high purity and topological quality and successfully implemented as dielectric layers in MIS capacitors. [47]ased on our previous work that yielded promising results using heteroleptic complexes featuring chelating guanidine ligands, [47,49] this study aims to evaluate the related but smaller formamidines as possible ligands for HfO 2 precursors.These ligands have recently found increased interest for other metals such as Y and In from Gordon et al. [50] and our group. [51,52]We transferred this concept to Hf and hereby present a bottom-up approach starting from rational design of the two heteroleptic bis- , their detailed characterization including thermal analysis, as well as scale up of the reactions to the multigram scale.The thermally most promising compound 1 was evaluated for PEALD of HfO 2 , particularly in view of low temperature depositions and the effect of the plasma on the properties of the resulting thin films was studied.This process was developed in the same reactor and under comparable conditions as used for [Hf(DPGUAN)(NEtMe) 3 ], [47] therefore a direct comparison of the two processes is possible and any differences in the resulting HfO 2 films can be correlated to the precursor design.A complete process optimization followed by detailed film analysis, the evaluation of the dielectric behavior was carried out to develop process optimization strategies that allow tuning of the materials properties, based on the desired application and the results are discussed herein.

Precursor Synthesis and Characterization
With the goal of obtaining Hf precursors with enhanced thermal stability compared to the dialkylamido compounds, while maintaining reactivity to enable film growth at low temperatures, two heteroleptic bis-(dialkylamido)-bis-(formamidinato) Hf(IV) compounds were synthesized in an easy to conduct and scalable reaction.Starting from the parent dialkylamides [Hf(NR 2 ) 4 ] (R = Me or Et), complexes 1 and 2 were obtained by a ligand exchange reaction with two equivalents of N,N'-diisopropylformamidine as illustrated in Scheme 1. Formation of the desired compounds by ligand exchange prevents the formation of precipitating salts.Instead, only the respective dialkylamine is liberated during the reaction mitigating the purification procedure especially for larger batches.By introducing two of the bidentate formamidine ligands, the Hf center is coordinatively saturated, helping to enhance the thermal stability and shelf-life of the resulting com-plexes with respect to the unsaturated parent aklylamides. [41,53]hereas the retained all-nitrogen coordination ensures a sufficient reactivity towards oxygen containing co-reactants. [41,54]The use of formamidine instead of guanidine [47,55] as the chelating ligand has several benefits.First, the formamidine features a lower molecular mass which is beneficial for the volatility of the resulting metal organic complexes.Second, due to the different synthesis route, the use of highly toxic diisopropylcarbodiimide is avoided as the formamidine ligand can be accessed using isopropylamine which is easier to handle.Most importantly the reactivity of the compounds is higher which falls in line with the feature recently reported by Gordon and our group for metalorganic complexes featuring N,N'-diisopropylformamidine ligands for the ALD growth of metal oxides at low temperatures. [50,51,56]he purity of compounds 1 and 2 as well as their structure in solution was investigated by 1 H and 13 C nuclear magnetic resonance (NMR) spectroscopy, revealing the successful formation of the desired bis-substituted products.The obtained 1 H NMR spectra are shown in Figure 1 with peaks assigned to the respective protons, while the 13 C ones are summarized in Figure S1 (Supporting Information).The 1 H NMR spectra of 1 just as for 2, show the expected signals whereby the formamidine ligand splits up into three signals as follows: The proton of the CH in the backbone is highly de-shielded due to the electron pulling effect of the two neighboring nitrogen atoms and appears at ppm values of about 8.1 and 8.2, respectively.Noticeably, the isopropyl groups split into a heptet and a doublet that is shielded from the electron withdrawing effect of the nitrogen atoms.Additionally, the dimethylamido group protons of 1 appear as a singlet, while the diethylamido protons of 2 split into a triplet for the shielded CH 3 protons and a quartet for the CH 2 protons.
The purity of the obtained compounds was further verified by elemental analysis (EA), revealing C, H and N values closely matching the expected ones, with values provided in Table S1 (Supporting Information).
The compounds were investigated by electron impact mass spectrometry (EI-MS) (spectra shown in Figure S2, Supporting Information).Selected mass-to-charge (m/z) ratios with intensities and proposed fragments are summarized in (Tables S2 and  S3, Supporting Information).For both compounds neither the molecular peak nor any peaks at higher m/z values with notable intensity were detected.In both cases, the peak with the highest mass to charge ratio appears at m/z = 562 and could correspond to a compound consisting of a Hf center coordinated by three formamidine ligands.As the 1 H NMR spectra show no sign for such a moiety, it is likely that it is formed under the harsh ionization conditions during EI-MS.This indicates the stabilization potential of the chelating formamidine with respect to the simple dialkylamido ligands.This observation is further strengthened by the fragmentation pattern that is dominated by cleavage of the first (m/z = 478 and 506, 100 % rel.intensity each) and second dialkylamido ligand (m/z = 435 each, 26 % and 34 %) respectively for 1 and 2. In addition to the discussed peaks, the spectrum consists of the cleavage of ligand fragments, whereas two peaks with m/z ratios of 128 and 43 appear in both spectra.These can respectively be attributed to the free formamidine ligand and a [N(CH 2 ) 2 + ] moiety.

Thermal Analysis
Compounds 1 and 2 were investigated by thermogravimetric analysis (TGA) as depicted in Figure 2a.This analysis method was employed to investigate the volatilization behavior as well as the thermal stability and, in the process, probe possible influences of the substitution pattern on these properties.Contrary to the expectations from the lower molecular mass of compound 1 with the dimethylamido groups, compound 2 has a slightly lower onset of evaporation temperature than 1 (142 versus 147 °C) which was determined by the temperature at which a weight loss of 1 % occurs.This observation can likely be explained by the higher steric hindrance of the diethylamido ligands, compared to the dimethylamido ones, that decreases the possibility of intermolecular interactions by shielding the Hf center. [57]However, at elevated temperatures the volatilization behavior changes, as can be seen from the step temperatures.These were estimated graphically using the method of tangents and were found to be 213 and 220 °C for 1 and 2, respectively.This indicates that, once the evaporation process starts, the molecular mass becomes the dominating factor for the volatility, explaining the lower step temperature of 1. Compound 1 evaporates in a single step that is completed at around 250 °C with a residual weight of 4 % at the end of the step.Until the end of the measurement, no significant further mass loss can be observed.In contrast, 2 evaporates in two steps, leaving a residual mass of 25 % at a temperature of 270 °C, followed by a further decrease to 19 % at 370 °C.This indicates that compound 2 thermally decomposes at elevated temperatures that are reached during the evaporation process.From this observation it seems that 1 is more thermally stable than 2, but these results need to be interpreted carefully, as the temperature reached during the evaporation process in the thermogravimetry (TG) measurement of 1 is not as high as for 2. To investigate the thermal stability of  1 at higher temperatures, TG was recorded with a higher heating rate of 20 K min −1 .This way, higher temperatures can be attained before the compound in the crucible fully evaporates.The obtained data is shown in Figure S3 (Supporting Information), revealing a single step evaporation with a rest mass of 5 % at 310 °C.This demonstrates that the precursor is still not affected by thermal decomposition and indicates a thermal stability on the TGA timescale of up to 310 °C.
As [Hf(DPfAMD) 2 (NMe 2 ) 2 ] (1) showed a more promising thermal behavior, its vapor pressure p was estimated by stepped isothermal-TG (iso-TG) measurements.Thereby, evaporation rates r v between 140 and 180 °C were obtained.With the Langmuir-equation, [58] p can be calculated for a given T. For ALD precursors, the 1 Torr vapor pressure temperature (T 1Torr ) is established for the comparison of different metalorganic compounds.The Langmuir equation for 1 is shown in Figure 2b, giving a T 1Torr of 149 °C with a respective r v of 382 μg min −1 cm −2 .The related monoguanidinato compound [Hf(DPGUAN)(mNEtMe) 3 ] has a slightly higher T 1Torr of 152 °C, despite its lower molecular mass and higher degree of asymmetry. [47]This observation indicates that the two formamidine ligands that result in a six-fold coordination of the Hf center reduce the intermolecular interactions with respect to the monoguanidinato compound.A similar behavior is already known for homoleptic yttrium formamidinates and the respective guanidinates. [59]

PEALD Process Development for HfO 2
Based on the promising thermal properties, compound 1 was evaluated as a precursor for the PEALD of HfO 2 thin films using a direct O 2 plasma.For this, the precursor was maintained at 135 °C in a stainless-steel vapor draw bubbler and introduced into the reaction chamber in 4 pulses, each separated by a vacuum purge of 100 ms with a total of 500 cycles.A schematic representation of the sequence is shown in Figure S4 (Supporting Information).Investigating the duration of the precursor pulse versus the film growth, depicted as GPC in Figure 3a, showed a clearly saturating growth behavior with a GPC of 0.74 Å, starting at a total precursor dose of 3.2 s.For longer pulse times up to a total of 6.0 s the thickness of the films did not increase any further, validating the typical self-limiting growth characteristic of ALD.To prevent any fluctuations in the growth by operating at the edge of saturation, the precursor dose was set to 4.0 s for the following depositions.
The linearity of the process was investigated by varying the number of cycles and plotting it versus the thickness in Figure 3b.A linear fit of the obtained data points with R 2 > 0.98 proved the linearity of the HfO 2 growth with a GPC of 0.73 Å as obtained from the slope of the linear fit, closely matching the 0.74 Å from the saturation.
Finally, the temperature (T) dependence was examined.Many ALD processes, predominantly thermal ones, show a so-called temperature window in which the reactivity of the precursor is sufficient to achieve self-limiting growth but is not affected by thermal decomposition, resulting in a constant GPC over a defined temperature range.For the presented HfO 2 PEALD process such a constant GPC is observed from 125 to 200 °C as can be seen in Figure 3c.For temperatures below 125 °C, the GPC increases steeply.This observation can be explained by condensation of the precursor on the substrate, possibly due to the substrate temperature being lower than the evaporation temperature.At 225 °C, however the GPC decreases slightly.This could be a sign for an onset of precursor desorption at the elevated temperature.

Thin Film Composition
To investigate the stoichiometry of the HfO 2 films, Rutherford backscattering spectrometry (RBS) in combination with nuclear reaction analysis (NRA) was conducted on films deposited at different deposition temperatures and the results are summarized in Table S4 (Supporting Information).Within the ALD window from 125 to 200 °C, the O/Hf ratio is slightly above the ideal value of 2.0 and the impurity levels are low (C ≤ 3 at.%,N ≤ 2 at.%), giving evidence for true ALD growth within this temperature range.The film deposited at 75 °C shows an increasing O/Hf ratio as well as impurity level, indicating insufficient reactions of the precursor molecules.
Complementarily to RBS/NRA analyses, X-ray photoelectron spectroscopy (XPS) was used to examine the surface chemical composition for a thin film grown at 150 °C on Si substrates with 150 ms oxygen plasma pulse time.
The survey recording showed all expected signals associated with Hf and O including the O KLL Auger line.While the presence of adventitious carbon was indicated by a small signal in the C 1s core level region, no contamination-related signals, e.g., originating from nitrogen were found (Figure 4a) which matches well to the RBS/NRA results described earlier.Ar + sputter treatment (2 × 1 min, 2 × 2) reduced the intensity of the C 1s signal below the detection limit arguing for the absence of C in the film bulk but also induced partial reduction in the Hf 4f core level (see Figures S7 and S8, Supporting Information) which is why chemical species analysis is restricted to the as-introduced surface.
The Hf 4f 7/2 and Hf 4f 5/2 core level signals were found at binding energies of 17.1 eV and 18.8 eV and are shown in Figure 4b.The spin component separation of 1.7 eV matches reports on stoichiometric HfO 2 and so do the binding energy values that are typically reported in regions of 16.6-17.6and 18.3-19.3eV, respectively. [47,49,60,61]Analysis of the O 1s core level region revealed a notable difference in comparison to HfO 2 thin films previously grown in the same PEALD reactor with the related [Hf(DPGUAN)(NEtMe) 3 ] at a nearly identical temperature of 160 °C but slightly longer plasma pulses of 500 ms.Contrasting to HfO 2 films grown from our prior PEALD process, [47] films obtained from [Hf(DPfAMD) 2 (NMe 2 ) 2 ] had only a very minor contribution of adsorbed oxygen species while the major contribution in the O 1s signal arouse from the O 2− lattice oxygen component (see Figure 4c).More precisely, the O 2− lattice oxygen component was found to have an integral share of 96.5 % on the overall signal at a binding energy of 530.1 eV [61][62][63] while adsorbed oxygen species (O (ads) : OH − , H 2 O) [64,65] accounted only for 3.5 % of the overall integral at 532.0 eV.After only one minute of sputtering, the higher energy shoulder of the O 1s sig-nal was already found to vanish completely (see Figure S9, Supporting Information).It is noteworthy that the integral shares were 70.0 % O 2− lattice oxygen versus 30.0 % adsorbed oxygen species in the prior case and that the sample storage prior to the XPS measurements did not significantly differ.Preconditioned that environmental effects on the different HfO 2 samples are negligible, it appears consequential that the difference in chemical composition arises primarily from the choice of precursor and the different plasma pulse lengths in the respective PEALD process.In terms of surface stoichiometry, the O lattice / Hf ratio of 1.95 is close to the ideal case, but slightly oxygen deficient.This contrasts with the bulk stoichiometry, that was found by RBS/NRA to be slightly oxygen rich at the same deposition temperature.
As the plasma pulse strongly affects the composition, topography, and structure of the grown material and thus its functional properties, we investigated films deposited at 150 °C with plasma pulse times of 50, 100, 150, 200, 300, and 500 ms.All of these films were deposited on HF etched n + Si(100) using 143 cycles and resulted in thicknesses of 4 to 15 nm (Table S5, Supporting Information).
From RBS/NRA analysis, summarized in Table S6 (Supporting Information), it was found that the films deposited with plasma pulses of 150 ms and longer were slightly oxygen rich with low impurity levels.The films deposited at 50 and 100 ms on the other hand showed O/Hf ratios of 2.3 and 2.4 with N contents of 13 % and 11 %, respectively.This indicates that a plasma pulse of at least 150 ms is necessary to remove the ligands from the precursor and form HfO 2 .

Thin Film Topology
For the application of thin-films in semiconductor applications smooth films with a root mean square roughness (R RMS ) below 1 nm are beneficial [66,67] while it has been shown that rougher dielectric layers induce grain boundaries and increased trap densities. [68]Thus, the roughness of the deposited HfO 2 films deposited with varying plasma pulse times, was assessed by atomic force microscopy (AFM).The films were measured in a scan range of 1 μm, as depicted in Figure S10 (Supporting Information) and the results are summarized in Table S6 (Supporting Information), alongside the thickness of the respective layers as obtained from TEM.With R RMS values between 0.12 nm (300 ms plasma) and 0.30 nm (50 ms plasma) the films were found to be smooth, with a roughness that is in the same order as that of the underlying Si substrate.Comparison of the films shows an abrupt change in the morphology from the shorter plasma pulses (50 to 150 ms) to the longer ones (200 to 500 ms).The films deposited with shorter plasma pulses show rather large features, while those deposited with larger pulses have smaller, grainy features.This indicates that there is a change in the structure of the films, possibly from amorphous to crystalline with increasing plasma pulse times.

HRTEM Analysis and Electrical Characterization
The quality of the dielectric is highly influenced by the Si/HfO 2 interface and therefore, this interface was analyzed by cross-section high-resolution transition electron microscopy (HRTEM).These depositions were performed on HF etched n + Si(100) substrates to remove the native SiO 2 , allowing us to analyze whether the plasma influences the Si surface during the first cycles of the PEALD process.
In our previous study, using the related precursor [Hf(DPGUAN)(NEtMe) 3 ], HRTEM images revealed the formation of an SiO 2-x interlayer.This was attributed to the plasma pulse of 500 ms oxidizing the Si surface at the beginning of the film growth and was associated with increased interface trapping. [47]Using precursor 1 in this study we subjected MIS capacitors with HfO 2 as the dielectric layer and sputtered aluminum as the gate electrode to HRTEM measurements.
To investigate the effect of different plasma exposures on the films thickness, its interface to the Si substrate and to possibly correlate this to its dielectric properties, HfO 2 films deposited with plasma pulse times between 50 and 500 ms were analyzed.Thereby, the thickness of the HfO 2 layers was found to remain constant with plasma pulses of 50 and 100 ms but then increases with longer pulse times until it remains mostly constant with pulses of 200 ms and longer, as depicted in Figure S11 (Supporting Information).
In Figure 5a-c, representative HRTEM images of the depositions with 200, 300, and 500 ms of plasma pulse show homogeneously smooth, closed films with a sharp interface to the Si substrate and ordered, crystalline regions.The crystalline nature of the films is validated by fast Fourier transformation (FFT) of selected areas of the HfO 2 layers.In these FFT images, several spacings could be assigned to planes of monoclinic hafnia as reported in literature. [69]For all films, the (−111) plane, was detected, while the (112) plane was found in the films deposited with 200 and 300 ms of plasma.Furthermore, the (202) plane could be assigned in the films deposited with 300 and 500 ms of plasma, while the (200) plane shows up in the 200 and 500 ms ones.HfO 2 deposited with 300 ms of plasma is furthermore showing the (112) plane and the one with 500 ms of plasma the (110) plane.The crystalline regions can be seen in more detail from scanning TEM (STEM) images where the HfO 2 layer is visible in an atomic resolution (Figures S13-S15, Supporting Information).Using the average from ten layers, this allowed a direct determination of the d-spacing.In particular, the film deposited with 200 ms of plasma showed three different regions with line spacings corresponding to the (021), ( 200) and ( 111) planes of monoclinic hafnia.From both STEM images of the films deposited with 300 and 500 ms of plasma, on the other hand, only one line spacing corresponding to the (−111) plane could be clearly identified.This plane was also prominent in the FFT images and according to the literature, [69] it shows the most intense reflexes, validating the formation of crystalline monoclinic hafnia in our process when plasma pulse times of 200 ms and longer are employed.
As can be seen from the HRTEM recordings in Figure S16 (Supporting Information), the HfO 2 layers deposited with 50, 100, and 150 ms were likewise homogeneous and closed, despite their lower thickness.However, these layers did not appear to be crystalline as no diffraction spacings were detected in the respective FFT pictures.This observation is well in line with the change in the morphology when going from the 150 ms to the 200 ms plasma pulse time that was observed by AFM.These results indicate that the structure changes from amorphous to crystalline as the plasma pulse time increases and the plasma provides more energy to induce crystallization of the films, even at a temperature as low as 150 °C and is in line with previous studies. [70]To exclude that this change in structure is caused by the thickness, which also increases with longer plasma pulses (compare Figure S11, Supporting Information), a thicker HfO 2 layer of around 48 nm, deposited with a plasma pulse of 150 ms, was subjected to XRD measurements (Figure S17, Supporting Information).No diffraction peaks were observed, thus it can be ruled out that the change in morphology and crystallinity is an effect of the films thickness.Furthermore, none of the investigated films showed evidence for the formation of an SiO 2-x interlayer.This could be explained by the comparably short plasma pulse times combined with efficient reactions between the precursor and the plasma, preventing oxidation of the underlying Si substrate.This observation is in contrast to our previous process with the related monoguanidinato precursor where the formation of an SiO 2-x interlayer was indeed visible. [47]Due to the similar process conditions this difference can be attributed to the difference in the precursor design where the higher reactivity of the formamidinato-based Hf compound could be an important factor for the absence of an interfacial layer.Further detailed and systematic studies on the influence of the precursor reactivity and interface formation, also at different temperatures are needed to prove this claim which is beyond the scope of this work.
To get an insight into how the dielectric properties are influenced by the plasma, we analyzed HfO 2 thin films deposited onto HF etched n + Si substrates with plasma times between 50 ms and 500 ms and vapor deposited Al contacts on top, using capacitance-voltage (CV) and current density electric field (JE) profiling.The electrical characterization of these, proof of principle MIS capacitors revealed that the plasma can be employed to tune the dielectric behavior of the HfO 2 layers.Representative capacitive measurements of the MIS structures based on HfO 2 layers deposited with plasma times from 50 ms (a) and (f) 500 ms are displayed in Figure 6.The GPC of the HfO 2 layers is influenced by the plasma pulse time (compare Figure S11, Supporting Information).Thus, for better comparability, the CV curves were plotted against the electric field in Figure 6 and the capacitance was multiplied by the layer thickness.Since the breakdown field strength also varies with changing layer properties, not all curves could be included in the entire electric field range.The voltage was varied from the state of inversion to accumulation and back again to also capture the electrical hysteresis.The shape of the Cd/E characteristics differs fundamentally between the short plasma times (50-150 ms) and the longer plasma times (200-500 ms).While the MIS structures based on amorphous HfO 2 films are, as expected, very steep and show flat band voltages close to zero (Figure 6a-c), there is a clear stretch-out of the Cd/E curves for crystalline films and a shift of the flat band voltage towards negative voltages (Figure 6d-f).At the same time, the hysteresis is reduced significantly.The hysteresis for the HfO 2 layers is due to charge accumulation by de-fects forming shallow, rechargeable trapping centers.There are two possible causes for this: a) the trapped charges indicate the presence of oxygen defects formed during film deposition [71] and b) the trapped charges are caused by the large lattice mismatch between the silicon substrate and HfO 2 . [72]Since the films become polycrystalline and the hysteresis decreases significantly with longer oxygen plasma time (t ≥ 200 ms), the first reason is more likely in the process presented here.The shift of the flat band voltage for thin films prepared with longer plasma times, indicate a significant higher amount of positive fixed oxide charges.Due to fast surface states, these curves are also stretched out and indicate a distortion by the capacitive coupled plasma.The stretch of the capacitance curves was also observed for HfO 2 films from our previous process with the related monoguanidinato Hf precursor deposited in the same reactor for plasma pulse times of 500 ms [47] and is a consequence of a less suitable direct electron cyclotron resonance configuration of the plasma reactor.The generated fast surface states can be "passivated" by means of heat treatment in a hydrogen containing ambient, e.g., forming gas.Presumably, the hydrogen diffuses to the Si/HfO 2 interface and satisfies dangling bonds, but such an analysis is not in the focus of this work.In the Cd/E characteristics the "shoulder" feature near the flat band voltage originates from Si dangling bonds and may be also minimized by annealing. [73]The current density J as a function of the applied electric field E of typical MIS structures based on HfO 2 films deposited by different plasma times is shown in Figure 7. Differently coloured curves in a diagram represents a JE characteristic of an individual device with identical geometries.The JE characteristics of devices based on amorphous HfO 2 layers (Figure 7a-c) again differ significantly from those of crystalline films (Figure 7d-f).The breakdown strength is for nearly all devices in excess of 6 MV cm −1 .A leakage current density of 10-100 nA cm −2 is observed for amorphous films and an elec-tric field below 3 MV cm −1 .Instead the leakage current of the crystalline films rises exponentially by applying an electrical field leading to one to two orders of magnitude higher leakage current density at 3 MV cm −2 .The clearly higher leakage currents of crystalline films can be explained by the presence of grain boundaries.Additionally, the barrier height generally does not depend on the layer thickness and thus any influence of direct tunneling is unlikely even in the thinnest films here.However, experimental results showed that the band gap of monoclinic HfO 2 is smaller (E g = 5.65 eV) when compared to amorphous HfO 2 (E g = 6.04 eV) [74,75] and thus the reduced barrier height results in enhanced leakage currents.The leakage current through the device increases exponentially with a lower barrier height and results in a larger power consumption of the devices.
The dielectric parameter extracted from the capacitance and JE curves of the MIS capacitances are summarised in Figure 8 as a function of plasma duration.Thereby, the permittivity is obtained from the maximum capacitance in the accu-mulation region, at which the series capacitance of the depletion region is negligible.The interface trap density D it was extracted using the conductance method proposed by Nicollian and Goetzberger. [76]The interface trap density D it is obtained by using the relation: in terms of the maximum conductance G p,max .The interface trapped charges, are charges at the HfO 2 -Si interface within the band-gap and can exchange charges with the semiconductor in a short time.For films grown by plasma pulses of 50 and 150 ms, the interface trap density is one order of magnitude higher when compared to the crystalline films grown by the use of longer plasma pulses.These trap states are typically produced by excess Hf, excess oxygen and impurities.In our case, RBS measurements reveal that oxygen excess and nitrogen impurities are responsible for the enhanced interface trap states in the amorphous HfO 2 films (Table S5, Supporting Information).The dielectric constant of the as deposited monoclinic HfO 2 films reaches up to 12 which is in line with monoclinic films prepared by ALD [47,75] as well as by pulsed laser deposition [77] without annealing.The results show that as the permittivty increases, the leakage current density increases as well.
The refractive index of crystalline HfO 2 films was found to correlate with the mass density of HfO 2 films. [77,78]Thus, for the crystalline films we derived the mass density of the films from X-ray reflectivity (XRR) measurements and the refractive index by use of a spectroscopic ellipsometer.The mass density amounts to 8.84, 9.55, and 9.70 gm −3 for HfO 2 films grown using 200, 300, and 500 ms plasma pulses, respectively.The ellipsometry measurements failed for HfO 2 films grown with 200 ms plasma pulses.Probably these films should be modeled as a mixture of the monoclinic phase and compact amorphous grains.The refractive index for the films deposited with 300 and 500 ms amounts 1.05 and 1.95 respectively.These values are below the bulk value of 2.1 and also slightly below the refractive indices reported for similar HfO 2 films grown via ALD [79][80][81][82] and would suggest even longer pulse and purge times for enhanced optical properties.On the other hand, the electrical measurements have shown, that in the capacitive coupled direct plasma this is associated with a massive degradation of the interfacial properties.Therefore, we refrained from doing so within this study and favor to do advanced process development with this precursor in a commercial ALD reactor with a true remote plasma within the scope of subsequent studies.

Conclusion
Rational precursor design and thermal evaluation resulted in the identification of a promising Hf precursor, namely [Hf(DPfAMD) 2 (NMe 2 ) 2 ] (1), for the deposition of HfO 2 thin films by ALD.The applicability of this precursor to be used for versatile low temperature processes was demonstrated by its successful implementation in a PEALD process.Self-limiting and linear growth could be shown at 150 °C and an ALD window of 125 to 200 °C was identified.By RBS/NRA and XPS measurements it was shown that the films deposited within the ALD window are stochiometric HfO 2 .Further investigations on the influence of the plasma pulse on film characteristics, revealed that the properties of the films can be modulated.This allows the process with precursor 1 to be adapted to the specific requirements of the processing as well as to the desired properties, such as the stoichiometry, structure, electrical and optical properties, of the HfO 2 for different applications.While these results are already promising, post-deposition annealing could be applied to reduce the interface trap density and further enhance the dielectric quality of the films.With the opposing trends of permittivity increasing while the breakdown field decreases with longer plasma pulses, the HfO 2 process presented here is certainly far from industrial maturity, making further optimization necessary.Nevertheless, the results show that the precursor is a promising alternative to the most used alkylamide precursors and therefore we will focus on a transfer of the process to an industrial scale ALD reactor with a remote plasma source.Furthermore, low temperature depositions on flexible substrates as well as an investigation of the memristic properties such as the switching performance of the films will be evaluated.

Experimental Section
Precursor Synthesis: Handling of air-and moisture-sensitive compounds was carried out in gloveboxes (MBraun).All synthesis experiments were conducted in oven-dried and degassed glassware under a dried argon atmosphere (AirLiquide, 99.995 %) using conventional Schlenk techniques.Solvents were purified and dried using an MBraun-SPS-800 purification system.The synthesis of the starting reagents [Hf(NMe 2 ) 4 ] and [Hf(NEt 2 ) 4 ], based on a procedure of Bradley and co-workers, [83] is described in the Supporting Information.
Synthesis of (  (40 mL) and a solution of 5.78 g (45.10 mmol) ( i PrNH)CH( i PrN) in toluene (30 mL) was added.The mixture was refluxed at 120 °C and the solvent and volatile byproducts were removed under reduced pressure, yielding the crude product as an orange wax.The crude product was purified by vacuum sublimation at 90 °C resulting in 11.24 g (21.57mmol, 95 %) of compound 1 as an off-white, sticky solid. 1   ] were dissolved in toluene (40 mL), a solution of 1.28 g (10.02 mmol) ( i PrNH)CH( i PrN) in toluene (30 mL) was added and the mixture was refluxed at 120 °C.After removal of the solvent and purification by vacuum sublimation at 120 °C, 1.62 g (2.81 mmol, 56 %) of compound 2 were obtained as a colorless solid. 1 H-NMR (400 MHz, benzene-d 6 ,  (ppm)): 8.17 (s, 2H, ( i PrN) 2 CH), 3.69 (q., J = 6.87 Hz 8H, N-CH 2 -CH 3 ), 3.45 (sept., J = 6.52 Hz 4H, N-CH-(CH 3 ) 2 ), 1.19 (d, J = 6.56 Hz, 24H, N-CH-(CH 3 ) 2 ), 1.14 (t, J = 6.90 Hz 12H, N-CH 2 -CH 3 ); 13  Precursor Characterization: Storage and sample preparation of compounds 1 and 2 were carried out in an argon filled glove box (MBraun LM 100).Deuterated benzene-d 6 for NMR experiments was degassed before use and stored over 4 Å molecular sieve.All 1 H and 13 C NMR spectra were recorded on a Bruker Avance III 400 HD instrument and referenced to the internal tetramethyl silane (TMS) standard (TMS,  = 0.00 ppm).EA was conducted with a vario micro cube tool (Elementar Analysesysteme) in the CHNS analysis mode.EI-MS measurements were carried out at 70 eV using a Varian MAT spectrometer in the electron ionization mode.The samples were filled into steel cartridges, sealed with lids and individually fed to the spectrometer via a load-lock chamber, which was pumped to ultrahigh vacuum prior the sample transfer to the main chamber.TG and stepped iso-TG measurements were conducted on a Netzsch STA-409 device in a temperature range of 30 to 550 °C.The TG experiments were run under an argon atmosphere with sample sizes of ≈10-20 mg, heating rates of 5 or 20 K min −1 and a nitrogen flow rate (AirLiquide, 99.999 %) of 300 mL min −1 .For the stepped isothermal experiment the temperature was increased in 10 °C steps, whereby the heating rate was set to 40 °C min −1 and held constant for 10 min at each step.The evaporation rates were determined by a linear fit of the mass loss under consideration of the surface area of the crucible.
Thin Film Deposition: PEALD depositions were performed in a custom-built PEALD reactor.A direct electron cyclotron wave resonance (ECWR) O 2 plasma was generated by a radio frequency generator (13.56 MHz) and an active magnetic flux of 2.8 mT using a matching network with a plasma power of 100-300 W in the pressure range of 10 -2 mbar -10 −3 mbar.Precursor 1, [Hf(DPfAMD) 2 (NMe 2 ) 2 ], was heated to 135 °C.The plasma power was adjusted to 200 W, whereas oxygen (AirLiquide, 99.995 %) and argon (AirLiquide, 99.995 %) gas flows were adjusted to 15 sccm for all depositions.Two inch, p-type Si(100) wafers (MicroChemicals) were used for ALD process optimization.The deposition temperature was varied from 75 to 225 °C.Saturation of the precursor and the linearity of the thickness versus the number of cycles were investigated at a deposition temperature of 150 °C.The optimized pulse/purge sequence for the deposition process was found to be (4 × 1000 ms precursor pulse / 1000 ms argon purge / 1000 ms oxygen pulse with 150 ms plasma ignition / 500 ms argon purge) and is depicted Figure S4 (Supporting Information).
Thin Film Characterization: Thin film thicknesses were determined by XRR measurements performed on a Bruker AXS-D8 Discover diffractometer (Bruker Corporation) using Bruker LEPTOS software.Thickness determination was further verified by fitting the recorded XRR curves to simulated ones.A two-layer model (HfO 2 /SiO 2 on Si(100)) was considered as the basis for these measurements.AFM measurements were performed on selected samples using a Nanoscope Multimode V microscope from Digital Instruments operating in the tapping mode.Recorded AFM images were further evaluated with Nanoscope software to assess R RMS values of the HfO 2 films.The bulk composition of thin films deposited on silicon substrates was determined by RBS/NRA.RBS was executed with a 4 He + ion beam of 2.0 MeV.The samples were tilted at an angle of 7°, while the backscattered particles were detected at an angle of 160°with respect to the beam axis.NRA was performed using a 2H + ion beam of 1.0 MeV.Here, emitted protons were detected at an angle of 135°relative to the incoming beam axis.The SIMNRA program [84] was employed for raw data processing and analysis.XPS studies were carried out on a PHI 5000 Ver-saProbe II instrument utilizing Al K photon radiation (1486.6 eV).Representative HfO 2 thin films were analyzed by a combination of survey scans and core level scans for peaks of interest.Step widths were adjusted to 0.5 eV for each survey scan and 0.05 eV for the core level scans.Hereby, the pass energies were adjusted to 187.5 and 23.5 eV respectively.All binding energies of Hf 4f, O 1s, N 1s, and C 1s were referenced to the Fermi edge position.The analysis chamber pressure was maintained at <10 −7 mbar.Recordings were taken and analyzed in terms of chemical species and composition for the as-introduced surface as well as for the film bulk after a succession of Ar + sputter steps (2 × 1 min, 2 kV, 2 × 2).The deconvolution analysis was completed with a Shirley background processing and Gaussian functions using UniFit 2017 software.
MIS Capacitor Fabrication and Characterization: Electrical characterization of selected HfO 2 thin films was carried out on MIS capacitors.For this, HfO 2 was deposited onto n + -type Si at 150 °C with 143 cycles of the optimized pulse/purge sequence and plasma ignition times of 50, 100, 150, 200, 300, and 500 ms.Prior to the HfO 2 deposition, the native SiO 2 was removed by an HF dip.Al gate electrodes with a diameter of 72 μm were e-beam evaporated onto the respective HfO 2 thin film surface through a shadow mask.To extract the permittivity of the dielectric layers, C−V measurements were performed using an Agilent E4282-A LCR meter.For assessment of the I−V characteristics of the MIS structures, a semiconductor parameter analyzer, Agilent 4156-B, was used.The fabricated capacitors were further investigated using an analytical FEI Tecnai Supertwin F20 TEM operated at 200 kV and equipped with a high angle annular dark field (HAADF) detector.TEM cross-section specimens were prepared using a dual-beam FEI Helios G4 CX focused ion beam (FIB) instrument operated at 30 kV.To protect the samples from damage caused by the ion beam, carbon was sputtered on the layers in advance.In the final specimen preparation step, low-voltage (5 kV) ion beam cleaning was applied for 2 min to each side of the TEM sample to reduce damage by the FIB process.

Figure 2 .
Figure 2. a) Thermogravimetric analysis of 1 and 2. b) Vapor pressure -temperature correlation derived from the Langmuir equation.

Figure 3 .
Figure 3. PEALD process development for HfO 2 on Si(100) using 1 and an oxygen plasma.a) Precursor saturation at 150 °C with a precursor temperature of 135 °C b) Thickness as a function of ALD cycles at 150 °C.c) Growth rate variation versus deposition temperature.

Figure 4 .
Figure 4. a) XPS Survey spectrum of the as-deposited surface of a 28 nm HfO 2 thin film grown on Si(100) at 150 °C.Observed core level and Auger features are denoted in red.b) High resolution spectrum of the Hf 4f core level.c) High resolution recording of the O 1s core level.

Figure 5 .
Figure 5. a-c) HRTEM images of MIS capacitors with HfO 2 dielectric layers deposited on HF etched n + Si substrates at 150 °C with plasma pulse times of 200, 300, and 500 ms, respectively.The thickness of the HfO 2 layers is denoted in green.d-f) FFT images from selected areas of the respective films (marked with red squares) with the identified spacings and diffraction planes.

Figure 6 .
Figure 6.Capacitance multiplied with the thickness of the film plotted versus the electric field of MIS capacitors based on HfO 2 films deposited at 150 °C on a HF etched Si substrate with plasma pulse times of: a) 50 ms, b) 100 ms, c) 150 ms, d) 200 ms e) 300 ms and f) 500 ms.The measurements were performed at f = 1 MHz.

Figure 7 .
Figure 7.Typical JE characteristics of MIS capacitors based on HfO 2 films deposited at 150 °C on a HF etched substrate with plasma pulse times of: a) 50 ms, b) 100 ms, c) 150 ms, d) 200 ms e) 300 ms and f) 500 ms.

Figure 8 .
Figure 8. Overview of electrical parameters obtained from capacitance and JE curves of MIS capacitors fabricated with PEALD HfO 2 layers as the dielectric.The films were deposited at 150 °C on HF etched n + Si substrates with plasma pulse times between 50 and 500 ms.