Ferroelectric, Pyroelectric and Piezoelectric Effects of Hafnia and Zirconia Based Thin Films

Von der Fakultät für Elektrotechnik und Informationstechnik der Rheinisch-Westfälischen Technischen Hochschule Aachen zur Erlangung des akademischen Grades eines Doktors der Ingenieurwissenschaften, genehmigte Dissertation

vorgelegt von

Master of Science Physik Master of Science Elektrotechnik

Sergej Starschich

aus Duschanbe (Tadschikistan)

Berichter: Univ.-Prof. Dr.-Ing R. Waser Apl.-Prof. Dr.-Ing M. Heuken

Tag der mündlichen Prüfung: 22.11.2017

Diese Dissertation ist auf den Internetseiten der Hochschulbibliothek online verfügbar.

Abstract

Ferroelectric materials are of great interest for several applications. On the one hand, the ferroelectric field effect transistor (FeFET) is a promising candidate for future high density, nonvolatile memory devices. On the other hand, in the recent years the energy related applications such as pyroelectric and piezoelectric energy harvesting as well as electrocaloric cooling and electrostatic energy storage attracted wide interest. The conventional ferroelectric materials such as lead zirconatetitanate (PZT) are not completely CMOS compatible and therefore a high-density integration for memory application could not be realized up to date. Furthermore, PZT has environmental issues due to the contained lead. Ferroelectric hafnium oxide, which was first reported in 2011, can overcome the mentioned drawbacks of the conventional ferroelectrics, since it is fully CMOS compatible. The ferroelectric phase is stabilized by doping with various dopants. Furthermore, a mixture of hafnium and zirconium oxide (Hf1-xZrxO2) does also stabilize the ferroelectric phase. In this thesis, hafnia and zirconia based ferroelectrics are deposited by a novel CSD (chemicals solution deposition) process and are characterized in respect to their ferroelectric, piezoelectric and pyroelectric properties. The ferroelectric nature of hafnium oxide is shown for several dopants as well as for Hf1-xZrxO2 with different compositions and for pure ZrO2. Especially in the case of ZrO2 this is very surprising since ZrO2 was studied for many years and for several applications without revealing ferroelectric properties. In contrast to atomic layer deposition (ALD), which is most commonly used for the deposition of hafnia and zirconia based ferroelectric film, the CSD technique is appropriate for deposition of thicker films without a strong reduction of the ferroelectric response. This makes hafnia and zirconia based ferroelectrics suitable for applications, where larger film thicknesses are unavoidable such as piezoelectric and electrocaloric cooling devices.

Kurzfassung

Ferroelektrische Materialien sind für viele Anwendungsbereiche von großem Interesse. Zum einen ist der ferroelektrische Feldeffekttransistor (FeFET) ein aussichtsreicher Kandidat für zukünftige hochintegrierte nicht flüchtige Speicher. Zum anderen haben energiebezogene Anwendungen wie pyro- und piezoelektrisches Energy Harvesting, elektrokalorisches Kühlen und elektrostatische Energiespeicherung in den letzten Jahren an Bedeutung gewonnen. Die konventionellen Ferroelektrika, wie beispielsweise Blei-Zirkonat-Titanat (PZT), sind nicht CMOS-kompatibel, wodurch eine hohe Integrationsdichte für Speicheranwendungen bis heute nicht erreicht werden konnte. Des Weiteren verursacht PZT Umweltprobleme aufgrund seines Bleigehalts. Mithilfe von ferroelektrischem Hafniumoxid, von dem im Jahre 2011 erstmals berichtet wurde, können die genannten Probleme aufgrund der CMOS-Kompatibilität überwunden werden. Die ferroelektrische Phase kann sowohl durch Dotierung mit verschiedenen Elementen als auch durch eine Zusammensetzung von Hafniumoxid und Zirkoniumoxid (Hf1-xZrxO2) stabilisiert werden. Im Rahmen dieser Arbeit werden auf Hafnium- und Zirconiumoxid basierte Ferroelektrika mittels einer neu entwickelten Routine zur nasschemischen Abscheidung (CSD) hergestellt und hinsichtlich ihrer ferroelektrischen, piezoelektrischen und pyroelektrischen Eigenschaften untersucht. Dabei wird gezeigt, dass sowohl eine Vielzahl von unterschiedlichen Dotierstoffen, als auch für unterschiedliche Zusammensetzungen von Hf1-xZrxO2 sowie für reines ZrO2 die ferroelektrische Phase stabilisiert werden kann. Dies ist besonders überraschend im Fall von

ZrO2, welches über Jahre hinweg für unterschiedlichste Anwendungen hin untersucht wurde und dabei keine Anzeichen für Ferroelektrizität gefunden wurden. Im Gegensatz zur Atomlagenabscheidung (ALD), welche am häufigsten zur Deposition von Hafnium- und Zirkonoxid basierten Ferroelektrika genutzt wird, ist die Abscheidung mittels CSD zur Herstellung dicker Schichten geeignet. Dadurch eigenen sich auf Hafnium- und Zirkonoxid basierte Ferroelektrika für Anwendungen, bei denen größere Schichtdicken unumgänglich sind, wie beispielsweise bei piezoelektrischen Sensoren und Aktuatoren sowie beim elektrokalorischen Kühlen.

Acknowledgements

This thesis was written during my doctoral research at the Institut für Werkstoffe der Elektrotechnik II (IWE 2) at the RWTH Aachen University. First, I would like to thank Prof. Dr. Rainer Waser for the opportunity to work in his research group in the field of novel ferroelectrics. Furthermore, I would like to thank Prof. Dr. Michael Heuken for being the co-examiner of my thesis. I am deeply grateful to Dr. Ulrich Böttger for supervising my work and for countless advices and discussions. I would like to thank my external collaboration partners for the successful cooperation. I appreciated the joint work with the NaMLab group of Dr. Uwe Schröder and Dr. Tony Schenk and the group of Prof. Dr. Alfred Kersch, Robin Materlik and Christopher Künneth from the Munich University of Applied Sciences. I express my gratitude to Dr. Theodor Schneller, and David Griesche for their support especially in the field of solution and sample preparation by use of CSD. Furthermore, I would like to thank Dr. Stephan Menzel for his support as an expert in the field of resistive switching. A big thank you goes to Petra Grewe and Daliborka Erdoglija for spending so much time for the sample preparation and characterization. I would also like to thank Jochen Heiss, Hartmut Pütz and Gisela Wasse for the support in electronics and electron microscopy. For the help and support concerning the images, I am thankful to Thomas Pössinger and Dagmar Leisten. Additionally, I appreciate the administrative support of Martina Heins and Udo Evertz. I also wish to thank my office mates Andreas Burkert, Astrid Marchewka, Inka Nielen, Camilla La Torre, Andreas Kindsmüller and Petra Grewe and all co-workers at the IWE 2 for providing a great working atmosphere. Special thanks go to Sebastian Schmelzer for supervising my Bachelor and Master thesis and for the support during the first month of my thesis. I furthermore acknowledge the helpful support of my student research assistants Bingjie Chen, Jan Lübben, Lucia Lauxmann, Maximilian Geppert, Maximilian Kühn, Nan Zhang, Parisa Jaberi, You-Ron Lin and Charlotte Böttger.

Contents

1 Introduction 1

2 Fundamentals 3

2.1 Crystal Structure ...... 3 2.2 Ferroelectricity ...... 4 2.3 Pyroelectricity ...... 6 2.4 Piezoelectricity ...... 7 2.5 Resistive Switching ...... 8 2.6 Ferroelectric Field Effect Transistor ...... 9 2.7 Physical Basics of Sputtering ...... 11 2.8 Experimental Methods ...... 12

3 Sample Preparation 23

3.1 Electrodes and Oxide Sputtering ...... 23 3.2 Chemical Solution Deposition ...... 25

4 Ferroelectric and Piezoelectric Properties of HZO and ZrO₂ 39

4.1 Composition dependence ...... 41

4.2 Ferroelectric ZrO₂ ...... 46 4.3 Doped ZrO2 ...... 51

5 Ferroelectric Properties of Doped HfO₂ 59

5.1 Sputtered yttrium doped HfO₂ ...... 60 5.2 CSD prepared yttrium doped HfO2 ...... 64 5.3 Further Dopants ...... 68

6 Wake-up and Degradation 81

6.1 Wake-up ...... 81 6.2 Degradation and fatigue ...... 96

7 Pyroelectric Properties 101

7.1 Yttrium Doped Hafnium Oxide ...... 102 7.2 Pure Zirconium Oxide ...... 104 7.3 Figures of merit ...... 107

8 Conclusions 109

8.1 Summary ...... 109 8.2 Outlook ...... 111

Bibliography 113

1 Introduction

The first ferroelectric hysteresis was measured in 1920 by Valasek in Rochelle salt

(C4H4KNaO6) [1]. After the first evidence of ferroelectricity, further ferroelectric materials were discovered such as KH2PO4 in 1935 [2] and the first oxide based ferroelectric BaTiO3 in 1946 [3]. Ferroelectric properties have been intensively studied over the years for various applications such as microwave tunable applications [4], ferroelectric memories [5], electrocaloric coolers and several further applications [6]. Ferroelectric materials also exhibit pyroelectric and piezoelectric properties and are therefore suitable for microelectromechanical systems (MEMS) [7-9] as well as for pyroelectric sensors [10]. In recent years, ferroelectric materials became attractive for energy related applications such as electrocaloric cooling [11, 12] and also piezoelectric and pyroelectric energy harvesting [13-15]. Up to now, lead zirconate titanate (PZT) is widely used due to its outstanding ferroelectric and piezoelectric properties [16]. Nevertheless, PZT has some serious drawbacks like the containing of lead, which is prohibited in several countries because of environmental issues. Furthermore, PZT as well as other conventional ferroelectrics is not fully compatibly to CMOS technology [17]. In addition, the scaling node of 130 nm for integrated FeRAM is significantly larger compared to e.g. flash memory [18].

Ferroelectric silicon doped HfO2, which was discovered in 2007 and first reported by Boescke et al. in 2011, can overcome the mentioned drawbacks of the conventional ferroelectrics [19, 20]. The ferroelectricity is attributed to the non-centrosymmetric orthorhombic phase with a

Pca21 space group, which typically coexists with the monoclinic and the higher symmetrical tetragonal/cubic phase [19, 21-23]. It was shown, that ferroelectricity in hafnium oxide can be induced by several different dopants such as Y, Gd and Al [24-26]. Furthermore, a mixture of 2 1 Introduction

hafnium and zirconium oxide (Hf1-xZrxO2) does also stabilize the ferroelectric phase [27]. Hafnium and zirconium oxide are completely compatible with known semiconductor fabrication processes and therefore are promising candidates for future high density, nonvolatile memory devices [28-30]. Apart from the memory application, also energy related topics such as electrocaloric cooling, electrostatic energy storage and pyroelectric energy harvesting have been discussed for hafnia and zirconia based ferroelectrics [31-34]. The objective of the present work is to get a better understanding of the origin of the ferroelectric properties in hafnia and zirconia based materials. This includes the influence of different dopants on the stabilization of the ferroelectric phase as well as to understand the wake-up effect, which means the increase of the remanent polarization during initial cycling that is observed in all HfO2 and ZrO2 based ferroelectrics. The most common deposition technique for this new class of ferroelectric materials is atomic layer deposition (ALD). A drawback of ALD is the strong limitation for deposition of thicker films. This work focuses on the deposition by chemical solution deposition (CSD), which offers the possibility of depositing thicker films, which are suitable for energy related applications and MEMS. Therefore, a novel CSD process for hafnia and zirconia based ferroelectrics is developed for the first time within the scope of this work. A dopant screening is performed to investigate the influence of the dopant size and valance state on the ferroelectric properties of HfO2. Besides doped HfO2, also the Hf1-xZrxO2 system is investigated for different compositions and thicknesses, whereas a surface energy model is used to understand the origin of the stabilization of the ferroelectric phase. In addition to the ferroelectric properties, also the piezoelectric and pyroelectric properties are evaluated for the doped HfO2 as well as for the Hf1-xZrxO2 system.

The wake-up effect is exemplarily investigated for yttrium doped HfO2. Thereby, a strong correlation to resistive VCM (valence change mechanism) switching is pointed out for the first time, which is a significant step to understand the origin of the wake-up effect.

2 Fundamentals

2.1 Crystal Structure

All crystallized material can be classified into 32 crystal classes. These crystal classes can be subdivided in groups according to their electrostrictive, piezoelectric, pyroelectric and ferroelectric properties as shown in Figure 2.1.

Figure 2.1: Subdividing of the 32 crystal classes according to their piezoelectric, pyroelectric and ferroelectric properties. 4 2 Fundamentals

All pyroelectric materials exhibit piezoelectric properties and all ferroelectric systems show pyroelectric and piezoelectric behavior, but not vice versa. In the following subchapter, the properties of ferroelectric, pyroelectric and piezoelectric materials are described.

2.2 Ferroelectricity

Ferroelectric materials exhibit a spontaneous polarization in a non-centrosymmetric crystal structure with a polar axis. The direction of the polarization can be switched by applying an external electrical field. A model system for ferroelectricity is lead zirconate titanate (PZT) with a perovskite structure as shown in Figure 2.2.

Figure 2.2: Perovskite structure of PZT for temperatures below and above the phase

transition temperature TC.

Above the phase transition temperature TC a cubic structure is exhibited with the Ti/Zr atom in the center, resulting in a paraelectric phase. Below TC a tetragonal distortion occurs and as a consequence no stable state is maintained in the center of the cell. The center atom has thus two stable states, which is oriented along the polar axis. By switching the central atom, the polarization switches likewise. 2.2 Ferroelectricity 5

The orthorhombic crystal structure of the ferroelectric hafnium/zirconium oxide is shown in Figure 2.3. In contrast to PZT, in the orthorhombic crystal of hafnium/zirconium oxide not only one ion changes its position during polarization switching but four oxygen ions.

Hafnium/Zirconium Oxygen

P P

Figure 2.3: Orthorhombic crystal structure of ferroelectric hafnium/zirconium oxide with the two possible polarization states.

Figure 2.4: shows an idealized (black) and an exemplary (red) ferroelectric hysteresis curve with its characteristic values. The coercitive voltages VC+ and VC- are characterized by the polarity change of the polarization. For the idealized curve, the polarization completely switches at the coercitive voltages. For the exemplary curve the polarization switching is widened because of the different grains within the film, whereby every grain shows a slightly different coercitive voltage. The further increase of the polarization at higher voltages is caused by the inherent dielectric polarization that every material exhibits. From the slope of the linear part of the hysteresis the relative permittivity of the material can be extracted. The remanent polarizations Pr- and Pr-, indicate the remaining polarization if a zero electrical field is applied. 6 2 Fundamentals

Figure 2.4: A ferroelectric hysteresis curve with its characteristic values for an idealized (black) and exemplary (red) curve.

2.3 Pyroelectricity

As well as ferroelectric materials, also pyroelectric materials exhibit a spontaneous polarization in a non-centrosymmetric crystal structure with a polar axis. The difference is that the polarization cannot be switched by applying an electrical field. The pyroelectric effect is described by the change of the spontaneous polarization during a change of the temperature. A characteristic dependence of the spontaneous polarization on the temperature is illustrated in Figure 2.5. This change of the polarization with changing temperature is found for all ferroelectric materials.

Figure 2.5: A schematic first order transition of a pyroelectric material. At the temperature

TC the polarization vanishes due to a phase transition into a paraelectric phase. 2.4 Piezoelectricity 7

The polarization is reduced with rising temperature till the polarization vanishes completely by reaching the phase transition temperature TC. A change in polarization always correlates with a generation of electrical charge. Therefore, during a temperature change ΔT a charge ΔQ is generated as

∆Q =p ⋅ A ⋅∆ T , where A is the area of the device and p is the pyroelectric coefficient dP p = dT . The voltage change across a pyroelectric capacitor device is given by: d ∆V = p ⋅ ∆ T ε ε 0 r , with the thickness d, the vacuum permittivity ε0 and the relative permittivity εr.

2.4 Piezoelectricity

Piezoelectricity is present in materials that have a non-centrosymmetric crystal structure. By applying mechanical stress, the ions within the crystal are shifted against each other leading to a generation of electrical charges and therefore to a generation of an electrical field. This is called the direct piezoelectric effect. For the reverse piezoelectric effect, an electrical field is applied to the material, which leads to a mechanical deformation. As mentioned before, all ferroelectric materials also exhibit piezoelectric properties. Figure 2.6 shows an exemplarily measured (red) and an idealized (black) displacement curve. Further information about piezoelectric characteristics can be found in [35]. 8 2 Fundamentals

Figure 2.6: Exemplarily measured (red) and idealized (black) displacement curve. The red curve is measured with a double beam laser interferometer.

2.5 Resistive Switching

Resistive Random Access Memories (ReRAM) are a promising candidate for future low power, high density and cost efficient memories [36]. The information is stored in the resistance, which can be changed by applying a positive and a negative voltage, respectively. The hafnium- and zirconium oxide based systems used in this work are well known for resistive VCM (valence change mechanism) switching [37-46]. An exemplary switching curve and the corresponding model are shown in Figure 2.7. In general, an electroforming step is required, before stable resistive switching can be realized. During the electroforming step, oxygen is extracted via one of the electrodes and an oxygen deficient filament is formed [47-49]. This is realized by applying a sufficiently high voltage leading to the formation of a plug, which is shown in Figure 2.7(A). This plug mainly consists of oxygen vacancies leading to a valence change of the hafnium cations and an enhancement in the local conductivity [50-52]. Therefore, this type of resistive switching is called valence change mechanisms (VCM) [49]. In the OFF state (A) the gap between the plug and the platinum electrode leads to a high resistance. By applying a 2.6 Ferroelectric Field Effect Transistor 9 negative voltage to the platinum electrode the oxygen vacancies start to migrate to the platinum electrode (B) leading to a low resistive state (ON state) of the device (C). By applying a positive voltage this process can be reversed and the oxygen vacancies move back to the plug (D). There are several other mechanisms to switch the resistivity in different materials. Further information can be found in [53]. However, in this work the described VCM mechanism is used to get a further insight into the origin of the stabilization of the ferroelectric phase.

Figure 2.7: A characteristic resistive switching curve and the corresponding assumed model for a VCM based System. Redrawn from [54].

2.6 Ferroelectric Field Effect Transistor

The ferroelectric field effect transistor (FeFET) is a promising candidate for future nonvolatile memory devices. The ferroelectric material is embedded in the transistor and replaces the gate dielectric as shown in Figure 2.8. 10 2 Fundamentals

Gate

Ferroelectric + + + + + + + + + + − − − − − − − − − −

P

Source + + + + + + + + + + Drain n+ − − − − − − − − − − n+

p-Ge

Figure 2.8: In a ferroelectric field effect transistor the dielectric gate oxide is replaced by a ferroelectric material. The polarization of the ferroelectric material influences the threshold voltage of the transistor.

The ferroelectric polarization induces compensation charges in the gate electrode and in the channel between source and drain. This induced charge in the channel shifts the threshold voltage of the transistor as schematically shown in Figure 2.9. The memory window (MW) is given by the offset of the two resulting characteristic curves. To read out the stored binary information the drain current is measured at a certain gate voltage. FeFET´s were already realized with conventional ferroelectrics but they strongly suffer from the poor interface of the ferroelectric material with silicon. Therefore dielectric buffer layer are needed realize stable FeFET´s but leads to a large voltage drop over the buffer layer due to the large permittivity and remanent polarization of the ferroelectric. These issues can be overcome by doped HfO2, which shows a low permittivity and a smooth interface with silicon [30].

Gate

+ + + + + + + + + + − − − − − − − − − −

P

Source + + + + + + + + + + Drain n+ − − − − − − − − − − n+

p-Ge

Gate

− − − − − − − − −

+ + + + + + + + +

− +

P

ri urn (a.u.) Current Drain M W

− − − − − − − −

Source − Drain + + + + + + + + +

n+ − + n+

p-Ge

Gate Voltage (V)

Figure 2.9: Schematic curves for the drain current in dependence of the gate voltage for the two different polarization states. 2.7 Physical Basics of Sputtering 11

2.7 Physical Basics of Sputtering

In the scope of this work, sputtering is applied for the deposition of all electrodes and furthermore for yttrium doped hafnium oxide. Sputtering is based on the physical bombardment of the target material with ions leading to an ablation of the target material and deposition on the substrate. A schematic of the sputtering process is shown in Figure 2.10. The target and the substrate are positioned opposite of each other. Due to the relative low pressure of 10-3 to 10 -1 mbar needed for the ignition of the plasma, a vacuum system is required usually consisting of a two-stage vacuum pumping system. The process gas, mostly argon, is used to ignite a non- thermal plasma by applying a high voltage between the target and the substrate. The electric field accelerates free electrons, which ionize the gas atoms trough collisions with sufficient kinetic energy. The electrons and ions, which arise from the collision, are accelerated, leading to a cascade of further ionizations. To obtain a higher ionization rate, magnets are placed behind the target. The magnetic field extends the trajectory of the electrons leading to even more ionization for each electron and a higher electron density near the target. Sputtering can be realized by either a DC voltage or a RF voltage of usually 13.56 MHz. For the DC voltage, the ionized argon atoms are accelerated towards the cathode, where the target material is positioned. One big disadvantage of DC sputtering is that only conductive materials can be sputtered. By using an insulating material, the target is charged positively leading to a compensation of the applied voltage and therefore to a stop of the deposition process. By applying an RF voltage, also insulating materials can be sputtered. Therefore, an oscillating voltage is applied to the target where the substrate is grounded. During the negative half of the alternating voltage the target is ablated and charged positively as described for the DC voltage. During the positive half of the voltage the electrons compensate the positive charge. Due to the higher mobility of the electrons compared to ions the negative charge at the target dominates, leading to a so called self-bias. This DC self-bias voltage causes primarily the ablation of the target during RF sputtering. By adding a reactive gas like oxygen or nitrogen to the process gas, reactive sputtering can be realized. Thereby, the reactive gas is ionized which leads to a chemical reaction with the target atoms during deposition. With this technique, metallic targets can be used to deposit the metal oxides or nitrides. The titanium nitride electrodes used in this work are reactive sputtered from a titanium target. The doped hafnium oxide is sputtered from hafnium and yttrium oxide targets as well as reactively sputtered from metallic hafnium and 12 2 Fundamentals yttrium targets. For further information concerning the sputter deposition technique, the reader is advised to literature [55][56][57].

Figure 2.10: Schematic sketch of a sputter deposition process.

2.8 Experimental Methods

2.8.1 Electrical Characterization

For the characterization of ferroelectric films, electrical measurements play an important role. The remanent polarization and the coercitive electrical field can be extracted from the hysteresis as described in the fundamentals section. For all electrical measurements, a voltage is applied at the device and the current response is measured. The interpretation of the measured current is crucial, because not only the charge from the ferroelectric switching influences the resulting hysteresis but also parasitic effects such as leakage currents. For an ideal capacitor, the switched charge is defined as:

= 2. Pr is the remanent polarization and A the area of the capacitor. For a real capacitor the charge results in:

= 2 + . 2.8 Experimental Methods 13

σ is the conductivity of the material, E the electrical field and t the time. The second part of the sum is a parasitic charge, which can exert a strong influence on the measured current, especially if materials with high conductivity are used or the measurements are performed at high electrical fields or at elevated temperatures. A detailed review of the influence of leakage currents on the measured ferroelectric hysteresis can be found in [58]. The mentioned parameters can lead to an incorrect interpretation of the measured hysteresis curves. To confirm the ferroelectric properties of the investigated system, further measurements like capacitance or displacement measurements can be performed.

Polarization Measurements

Most of the polarization measurements in this work are performed with an aixACCT TF Analyzer 2000. A typical cycle sequence applied to the sample and the resulting polarization data are shown in Figure 2.11 and Figure 2.12, respectively.

Figure 2.11: Standard voltage sequence of a hysteresis measurement with an AixACCT TF Analyzer 2000

The 1st and the 3rd cycles set the sample in a defined polarization state and are not shown in the results. The 2nd cycle is applied after a relaxation time of one second. This facilitates the opportunity to measure the negative relaxed remanent polarization Prrel- as shown in Figure 2.12 (see green dashed line). The 4th cycle depicted by the dotted blue line indicates the positive relaxed remanent polarization Prrel+. Commonly the relaxed remanent polarization is not mentioned despite being a more meaningful value than the remanent polarization Pr because it provides a much more reliable statement about the permanent polarization of the 14 2 Fundamentals material. In most cases only the dynamic hysteresis is shown (red line) which is composed of the second parts of the 2nd and 4th cycle.

Figure 2.12: Exemplary hysteresis curves resulting from the cycling order shown in

Figure 2.11 with the positive (Pr+) and negative (Pr-) remanent polarization and the positive

(Prrel+) and negative (Prrel-) relaxed remanent polarization.

In the example shown in Figure 2.12 there is only a slight difference between the remanent and the relaxed remanent polarization. For this reason this work will mostly use the simplification of only showing and comparing the dynamic hysteresis curves. If not mentioned otherwise a frequency of 1 kHz is used for all measurements. In terms of a higher current resolution and more flexibility with the measurements, also a Keithley 4200 with Pulse Measurements Units (PMU) is used for polarization measurements. The cycle sequence is shown in Figure 2.13. In contrast to the TF Analyzer no relaxed remanent polarization is recorded. The dynamic hysteresis is directly measured after the prepolarization. 2.8 Experimental Methods 15

Figure 2.13: Cycle sequence used for the polarization measurements with the Keithley 4200. The dynamic hysteresis is measured directly after the prepolarization.

Wake-up Measurements

Wake-up means the initial increase of the remanent polarization during cycling, where mostly a constricted hysteresis is observed in the pristine state, which opens up after cycling. This effect is observed for almost all hafnia and zirconia based ferroelectrics and was also found for conventional ferroelectrics such as PZT [59, 60]. All of the investigated layers in this work need a certain amount of so called wake-up cycles to exhibit the ferroelectric properties. These wake- up cycles are realized by rectangle pulses as shown in Figure 2.14. Thereby the frequency of the rectangle pulses can be varied between 1Hz and 100 kHz.

Figure 2.14: Cycle sequence for the wake-up cycling. The wake-up is realized by bipolar rectangle pulses.

16 2 Fundamentals

Capacitance Measurements

Additionally to the hysteresis measurements, capacitance over voltage (C(V)) measurements can be performed to confirm the ferroelectric nature of the investigated material. A HP4284A LCR meter is used for the C(V) measurements shown in this work. Although the capacitance measurements are also sensitive to parasitic current, the resulting influence on the measured curves is completely different compared to the hysteresis curves. Figure 2.15(a) shows the basic circuit called ‘auto-balancing bridge method’ that is used to measure the complex impedance.

(a) (b)

Figure 2.15: (a): ‘Auto-balancing bridge method’ used for measuring the complex impedance of the device. (b): A characteristic C(V) curve from a 5.2% yttrium doped hafnium oxide layer.

To get the complex impedance it is necessary to measure the voltage drop over the device under test (DUT) and the current flows through it. The low potential at the device is automatically driven to zero volts by the operational amplifier with a negative feedback loop. The voltage drop over the DUT can be therefore measured directly by Vx. The current Ix trough the device is equal to the current trough the resistor Rr. The voltage drop over the resistor is given by the measured voltage Vr. By determining the two voltages Vx and Vr the complex impedance can be calculated:

= 2.8 Experimental Methods 17

The resistor Rr is called the range transistor and determines the range of the measured impedance. Figure 2.15(b) shows a characteristic C(V) curve for a ferroelectric layer with the two peaks located at the ferroelectric switching points.

2.8.2 Double-Beam Laser Interferometer (DBLI)

The electrical measurements are performed by measuring the current trough the device. High leakage currents or other effects causing parasitic currents lead to a possible distortion of the results. By measuring the mechanical piezoelectric displacement with a Double-Beam Laser Interferometer the piezoelectric properties can be quantified without measuring an electrical current. Due to the characteristic displacement curve of a ferroelectric material, the ferroelectric phase can be indicated by measuring the voltage dependent displacement. Applying an electrical field to a sample causes a bending of the film due to the induced strain. The bending can be measured with a Double-Beam Laser Interferometer as shown schematically in Figure 2.16. A laser beam (He-Ne laser λ=632.8 nm) is divided and directed to the upper and the lower side of the sample. The reflected beams are brought to interference and hence small deviations of the film thickness can be detected. For this purpose, the backside of the sample is polished for good optical reflectivity. The fast measurement method is used for to record the piezoelectric response as described by Gerber et al. [61]. The measurements are carried out at a frequency of 1 kHz on devices with an area of 4x10-4 cm² An ideal piezoelectric response for a ferroelectric material was shown in Figure 2.6. For further information concerning the DBLI measurement technique for ferroelectric thin films, the interested reader is referred to literature [62, 63]. 18 2 Fundamentals

Piezo actuator + reference mirror HeNe Laser Lambda/4 pl. Lambda/2 pl.

Mirror

Lambda/4 pl.

Sample

Lambda/4 pl.

Photo detector

Figure 2.16: Schematic of the used Double-Beam Laser Interferometer (Figure redrawn from [35]).

2.8.3 X-Ray Analysis

X-Ray Diffraction

Besides electrical and mechanical characterizations, the X-Ray Diffraction (XRD) provides a great approach as a non-destructive crystallographic investigation. The basis for this analysis was given by Bragg with his law

= 2 . The law describes the constructive interference for diffracted waves with a wavelength λ. When the phase shift is a multiple of 2π a constructive interference occurs. To be able to measure the lattice constant d of a crystal, the wavelength needs to be in the same order of magnitude as the lattice constant. Since the lattice constant is in the magnitude of less than one nanometer, cupper x-rays with a wavelength of λ=1.5406 Å are used to perform the measurements. In Figure 2.17 2.8 Experimental Methods 19 an example for constructive interference is illustrated. The two waves appear at the two lattice plains at an angle of incidence θ. Due to the distance d between the two plains a path difference between the two waves occurs. The path difference is shown in the green square and equals exactly two periods fulfilling Bragg's law for constructive interference. There are two basic techniques to perform XRD measurements, which are shown in Figure 2.18 and described in the following section.

Figure 2.17: Illustration of Bragg's law for constructive interference. The two incident waves are diffracted at the two crystal plains. Due to the path difference (marked as green square) which is exactly two periods, constructive interference appears.

Bragg Brentano

In the Bragg Brentano geometry Figure 2.18(a) the incident optics and the detector move on the half circle keeping the same incident and diffracted angle. One drawback for measuring thin films is the small penetration depth of the film. For example, if a 10 nm film is investigated and considering the incident angle of 30°, a penetration length of only 20 nanometer results. Additionally the x-ray penetrates the substrate leading to a very strong diffraction signal of the substrate superimposing the diffraction of the thin film.

20 2 Fundamentals

Grazing Incidence

To overcome the drawbacks for thin films of the Bragg Brentano method, the grazing incidence geometry can be used as shown in Figure 2.18(b). The incident angle is kept constant slightly higher than the critical angle for total reflection. The parallel collimator used as detector is moving on the half circle to measure the diffracted beam at different angles. Capturing the example from the Bragg Brentano section, by using an incident angle of 0.5 °, a 10 nanometer thick layer would be penetrated in a length of more than 1000 nanometers resulting in a much stronger pattern signal. Another benefit is that by using such small incident angles, single crystal substrates like silicon do not contribute to the diffraction pattern and thus no superimposition occurs.

Figure 2.18: Bragg Brentano (a) and grazing incidence (b) setups to measure XRD patterns. For the Bragg Brentano geometry the incident and the reflected beam preserve the same angle compared to the perpendicular. For the grazing incidence measurement the incident beam is parallelized and the angle is kept constant while the detector is scanning the different diffraction angles.

X-Ray Reflectometry

Layer thickness, mechanical density and surface roughness can be determined by using X-Ray Reflectometry (XRR). No crystalline films are necessary for the measurement of the mentioned parameters. The measurement setup is similar to the Bragg Brentano geometry in Figure 2.18, whereby the x-ray mirror for a parallel beam is used instead of the incident beam optic. The beam angle varies from 0.2° to 3° depending on the parameters of the investigated material. In the present work the XRR measurements are mainly used to determine the layer thickness of 2.8 Experimental Methods 21 the thin films. The thickness of the film can be measured when two interfaces are present where a change in mechanical density appears. An illustration of the XRR measurement is shown in Figure 2.19.

a1a2

d a3

Figure 2.19: Illustration of the physical basics for thickness measurement. The incident beam is reflected at both interfaces leading to a path difference resulting in a constructive or destructive interference.

The incident x-ray beam reaches the surface of the thin layer at an angle a1. One part of the beam is reflected at the same angle, the other part penetrates the layer with the thickness d. Due to the different densities of the layer and the surrounding air, the x-ray is redirected at the angle a2. After passing through the layer, the beam is reflected at the interface between the layer and the substrate. Finally, the beam is redirected again at the interface between the air and the layer. Because of the transition through the layer the two outgoing beams exhibit a path difference leading to constructive and destructive interferences mostly depending on the angle of incident and the layer thickness. Therefore, by varying the incident angle the layer thickness can be determined. Figure 2.20 shows a generic XRR measurement. For low incident angles a total reflection occurs. At the critical angle θc the beam penetrates the layer and fringes are measured depending on the layer thickness. The density of the top layer ρ can be calculated from the critical angle θc as first proposed by Kiessig [64]

≈ wherein λ is the wavelength and re the electronic radius. The thickness of the layer can be calculated by using the so called Kiessig fringes [65], which are caused by the above mentioned reflection at the different interfaces. 22 2 Fundamentals

Figure 2.20: Exemplary XRR measurement of a 27 nm thick hafnium oxide layer deposited on oxidized silicon substrate. The 450 nm thick oxide does not contribute to the fringe pattern, because the layer is too thick and therefore the x-rays do not penetrate the whole layer.

1 = 2 − For accurate XRR measurements the interfaces need to be very smooth, otherwise the diffuse reflection leads to an attenuation of the interference pattern.

3 Sample Preparation

In this chapter the sample preparation and especially the preparation of the functional oxides is discussed in more detail. The platinum and titanium nitride (TiN) electrodes used in this work as well as yttrium doped hafnium oxide is deposited by sputter deposition. Chemical solution deposition is used for the deposition of doped HfO2, Hf1-xZrxO2 and pure ZrO2.

3.1 Electrodes and Oxide Sputtering

3.1.1 Electrodes

The 100 nm thick platinum bottom electrodes are DC sputtered from a six-inch target on oxidized silicon wafers with a 10 nm TiO2 adhesion layers at 150 °C. Whereas for the top electrode 50 nm platinum are sputtered at room temperature. Titanium nitride bottom electrodes with a thickness of 30 nm are RF sputtered from a six inch titanium target on silicon wafers at room temperature with 10 % nitrogen in the process gas to realize reactive sputtering.

3.1.2 Oxide sputtering

The ferroelectric layers are deposited by reactive sputtering as well as by sputtering from oxide targets. Both processes are realized by magnetron RF-sputtering from one-inch targets. The sputter chamber used for the deposition is in detail described in Ref. [66]. The great advantage 24 3 Sample Preparation of the chamber is the possibility of co-sputtering, which means the simultaneous sputtering from two targets. This provides the possibility of controlling the yttrium doping concentration without using several targets with different compositions. The doping with yttrium is realized by adjusting the sputtering power at the yttrium or yttrium oxide target, respectively. For the sputtered yttrium doped HfO2, titanium nitride is used as the bottom and top electrode.

Sputtering from oxide targets

Controlling the deposition from the hafnium oxide target turns out to be highly challenging. The deposition rate showed a strong dependence on the evacuation time, which was not observed for any other targets. The deposition rate doubled when the evacuation was carried out for several days compared to an evacuation of 20 minutes. This effect could not be explained so far. To obtain comparable results all steps during the layer preparation are kept the same for all shown results, i.e. the duration of the substrate loading, the evacuation time and the duration of the presputtering are the same for all samples. For the different doping concentrations, the sputtering power of the hafnium oxide is held at a constant value of 12 W, whereas the sputtering power of the yttrium oxide is varied between 3 W and 14 W resulting in a functional layer thickness of approximately 30 nm. At higher sputtering power a higher amount of yttrium is incorporated. The exact yttrium concentration in the hafnium oxide is unknown. For this information, further investigations would be necessary.

Reactive sputtering

For the reactive sputtering of the yttrium doped hafnium oxide layers metallic hafnium and yttrium targets are used. During the deposition, an argon/oxygen process gas is used with an oxygen amount of 2 %. A sputtering power of 30 W is used for the hafnium and 5 W for the yttrium with a resulting layer thickness of 22 nm. As well as for the sputtering from the oxide targets, the exact yttrium doping concentration is unknown.

3.2 Chemical Solution Deposition 25

To avoid oxidation of the electrodes during the annealing step the crystallization is carried out in nitrogen atmosphere at 800 °C for 30 s. In contrast to the CSD prepared samples, the crystallization step is performed after deposition of the top electrode.

3.2 Chemical Solution Deposition

Chemical Solution Deposition (CSD) is a well-known and widely used deposition technique for various material systems including functional oxides [67]. It has many advantages compared to physical vapour deposition (PVD) like sputtering. For example, the doping concentration can be adjusted very accurately during the solution preparation and therefore no elaborated measurements after deposition are needed to figure out the composition. Other advantages are the time efficient sample preparation and the cost efficiency. The solution used for manufacturing the functional layers is prepared by the hybrid and the MOD method [68, 69].

Hybrid Method

The hybrid method consists of a Sol-Gel and a MOD part. For the Sol-Gel educt, hafnium ethoxide with a purity of 99.9 % is used. All preparations with hafnium ethoxide are carried out under inert gas atmosphere by the use of standard Schlenk technique and a glove box. The desired amount of Hafnium ethoxide is weighted into a Schlenk flask and dry ethanol is added. After heating at 60 °C for thirty minutes in an oil bath a transparent solution is formed which is stabilized by adding one equivalent of 2,4-pentanedionate per hafnium ion. For the MOD- educt the desired dopant 2,4-pentanedionate is used. The dopant 2,4-pentanedionate is dissolved in a mixture of propionic acid and propionic anhydride (5:1) at 100 °C. Finally, the two solutions are combined and the concentration is adjusted with propionic acid.

MOD Method

The second method used for the solution preparation is the MOD process for both educts. The dopant 2,4-pentanedionate and Hafnium 2,4-pentanedionate are weighted together in a Schlenk 26 3 Sample Preparation flask and afterwards are dissolved in a mixture of propionic acid and propionic anhydride (5:3) at 140 °C. The adjustment of the concentration is carried out by adding propionic acid and propionic anhydride (5:3).

The two methods are compared for 5% neodymium doped hafnium oxide. Except for the solution preparation, all treatment steps during the layer deposition are identically. The hybrid method is much more difficult and time consuming, since all steps with the used hafnium ethoxide need to be carried out in inert gas atmosphere. The advantage of the hafnium ethoxide compared to the hafnium 2,4-pentandionate, which is used for the MOD method, is the higher purity (99.9 % compared to 98.1 %). The ferroelectric hysteresis for both methods are shown in Figure 3.1. No noteworthy difference for the remanent polarization can be found. The only considerable difference is the coercitive field shift for the hybrid method hysteresis. Therefore, all further solutions are prepared by the MOD method.

Figure 3.1: Comparison between the hybrid and the MOD preparation techniques exemplary shown for 5% neodymium doped hafnium oxide.

3.2.1 Workflow

After preparing the solution the layer is deposited by spin coating on the substrate. A sufficient amount of the solution is brought on the substrate and spin coated at 3000 rotations per minute for 30 seconds. Figure 3.2 shows a sketch of the spin coating process. The rotation speed and 3.2 Chemical Solution Deposition 27 time are constant for all samples prepared within the scope of this work. Therefore, the resulting layer thickness for each spin coating is given by the concentration of the solution.

Figure 3.2: A schematic diagram of the spin coating process. The solution is brought on a substrate, which is placed on the probe holder. Subsequently the probe holder rotates to achieve the target thickness of the wet layer.

After each coating, a heating step is carried out at 215 °c for five minutes. For the deposition of thicker layers the coating and the pyrolysis step are repeated until the desired layer thickness is achieved as shown in Figure 3.3. Finally, a crystallization step is performed at 800°C in an

Ar/O2 atmosphere for 90 seconds in a rapid thermal annealing oven (RTA). If not mentioned otherwise, 100 nm thick bottom and 50 nm thick top platinum electrodes are used for the CSD prepared layers.

Figure 3.3: Workflow for the sample preparation: The solution is deposited on the substrate and afterwards a heating step is performed for five minutes on a hot plate. These two steps are repeated several times to achieve the desired thickness. Afterwards a crystallization step is performed in a RTA oven. Finally, the top electrodes are deposited by a negative lift off process 28 3 Sample Preparation

3.2.2 Yttrium doped HfO2

Yttrium doped hafnium oxide with a concentration of 5.2 % is used for the development of the deposition process. For all shown results here, 1000 wake-up cycles at 1 kHz are performed before measuring the P(V) and C(V) curves. A detailed investigation of the wake-up effect is given in chapter 6.

Solution Investigation

For the investigation of the solution the thermogravimetric analysis (TG) is applied. The thermogravimetric analysis can be used to investigate crystallization, vaporization, sublimation and mass loss due to decomposition of a solution. Therefore, with TG the pyrolysis and crystallization temperatures of the prepared solution can be evaluated. The partially dried solution is placed in a furnace with a thermocouple for accurate temperature measurement. Furthermore, a precise balance is used to measure the mass loss. For better accuracy an inert reference sample (in this case aluminum oxide) is used. For further information, the reader is advised to literature [70]. For this purpose a solution is prepared by the MOD method with hafnium and yttrium 2,4-pentandionate at 5.2 % yttrium doping. For the analysis, the solution is dried in a round bottom flask by means of a heating mantle until all volatile constituents are removed. An amount of 23.3 mg precursor powder with Al2O3 as reference is used for the analysis. The results of the analysis are shown in Figure 3.4. The vaporization of small amounts of residual solvent (~ 1 %) occurs at lower temperatures, i.e. below ~ 160 °C, which is indicated by the endothermic peaks in the differential thermal analysis (DTA) curve. This is followed by a further continuous weight loss which merges into the range of the exothermic decomposition of the metallo-organic precursor molecules. This pyrolysis starts at approx. 250 °C and reaches its maximum at 415 °C indicated by the exothermic peak of the DTA signal. At ~490 °C the pyrolysis is completed and no further weight loss is observed for higher temperatures. In this phase of the thermal transformation process, the largest weight loss of approx. 40 % is observed. Assuming that the yttrium and hafnium 2,4-pentanedionate compounds are solely dissolved during the solution preparation, the overall weight loss during pyrolysis should be 63.6 %, which is significantly higher than the 40 % observed in the DTA analysis. 3.2 Chemical Solution Deposition 29

Figure 3.4: Results of the thermogravimetric analysis. The red curve shows the differential thermal analysis (DTA) of the temperature. For values above 0 µV the process is exothermic, below it is endothermic. The blue curve shows the percentage mass loss of the investigated material (redrawn from [71]).

In order to explain this discrepancy a chemical reaction of the initial metal acetyl acetonato complexes and the propionic acid, which is used as a solvent during solution synthesis have to be taken into account. Further evidence that chemical bonds are formed during the solution preparation is that pure yttrium 2,4-pentandionate does not solve permanently in propionic acid/propionic anhydride, but in combination with the hafnium 2,4-pentandionate it dissolves well. Similar observations have been reported in the literature for zirconium acetylacetone based precursor solutions [72-74]. Since hafnium and zirconium show nearly identical chemical behavior, it is concluded that a reaction with propionic acid takes place here as well. However, a simple complete ligand substitution by propionate groups can be ruled out, because the expected mass loss of 55.4 % is significantly higher as the observed loss (40%). Most probably a more complex Hf cluster structure containing bridging oxo groups and carboxylate groups is formed, similar to what has been found in case of the reaction of Zr(acac)4 with propionic acid [72]. Moreover it is assumed that the yttrium compound is linked to this hafnium cluster, e.g. via a carboxylate group. The fact that yttrium acetyl acetonate only dissolves well in propionic acid in the presence of the Hf-cluster points to this opportunity, but the exact structure of the precursor molecule is not known yet and its determination is beyond the scope of this 30 3 Sample Preparation work. The small exothermic hump above 550 °C with its maximum at ~586 °C probably indicates the crystallization of the powder.

Influence of the solution concentration

The precursor solution can be prepared with different concentrations, which have an influence on the deposition. With a higher concentration more Zirconium/Hafnium 2,4-pentandionate molecules are present in the wet layer after spin coating. This results in a higher layer thickness for each spin coating. A limitation for the maximum concentration is the solubility of the molecules in the propionic acid. Figure 3.5 shows the influence of the concentration on the hysteresis curve. A solution with a concentration of 0.5 M was prepared and diluted to 0.25 M and 0.1 M. For the hysteresis curves, a lower concentration seems to improve the remanent polarization. Considering the current curves it can be seen that there is almost no difference in the peak height. Overall, it can be concluded that the solution concentration has no crucial influence on the electrical results.

Electrical Field [MV/cm] Electrical Field [MV/cm]

Figure 3.5: Dependence of the concentration of the precursor solution on the electrical

hysteresis for 5.2 % doped HfO2.

3.2 Chemical Solution Deposition 31

Process temperature dependency

Figure 3.6 shows the evolution of the film thickness of a single coating step in dependence of the applied process temperatures and models of the assumed corresponding layer structure during each step with the corresponding XRR measurements. Figure 3.6(a) shows schematically the wet, as-deposited layer after spin coating with the corresponding thickness of 114 nm measured by XRR. This film contains the metallo-organic hafnium precursor clusters distributed in a considerable amount of propionic acid. By applying the relative low temperature of 215°C during the first heating step the film starts to densify accompanied with the reduction of the film thickness to 38 nm (Figure 3.6(b)). The higher density after the heating step is indicated by the larger critical angle θc1 in the XRR measurements. Based on the thermal analysis most likely, the vaporization of the solvent occurs predominantly and almost no pyrolytic decomposition takes place. When higher temperatures are applied at this stage, lower film thicknesses are obtained (inset of Figure 3.6) accompanied with a higher degree of pyrolytic decomposition. In order to induce crystallization the second heat treatment step is performed by an RTA furnace at 800 °C. During this process, the main part of the pyrolysis and finally the crystallization inextricably take place leading to an organic free yttrium doped hafnium oxide layer with a layer thickness of 15 nm (Figure 3.6 (c)). In order to investigate the influence of different temperatures of the first heat treatment step on the ferroelectric properties, a series of temperatures attained by a hot plate between 180 °C and 295 °C is applied to thin film samples. It turns out that the lowest applicable temperature which is necessary for the multi-coating process is 215 °C. If the first layer is heated below this temperature, it dissolves in the subsequent spin-on process and no increase of the film thickness can be obtained. The change of the electrical properties at the different temperatures of the first heating step is shown in Figure 3.7 for the permittivity and the ferroelectric hysteresis.

32 3 Sample Preparation

Figure 3.6: Influence of the two different heating steps on the deposited layer. The three sketches in the upper part of the figure show schematically the assumed layer structure for the different conditions. Below each sketch the corresponding XRR measurements are

displayed. θc1 is the critical angle of the deposited layer indicating the surface density and

θc2 is the critical angle of the platinum substrate which is constant for all conditions. The inset shows the evolution of the layer thickness after different temperature treatments at the first heating step (redrawn from [71]).

The best ferroelectric properties result from films at 215 °C where the layer is rather dried than pyrolyzed. By applying a temperature of 295 °C the remanent polarization is reduced by 40 % compared to the heat treatment at 215 °C. A possible explanation is that the presence of the organic part prevents the growth of larger inhomogeneously composed cluster units. This in turn might help to maintain a solution like homogeneity of the metallo-organic precursors in the dried or only partially pyrolyzed films up to higher temperatures, where sufficient energy is available for the crystallization process. Thus, more homogenously composed ferroelectric films with superior properties result. 3.2 Chemical Solution Deposition 33

(a) (b)

Figure 3.7: Influence of the temperature of the first heating step on the ferroelectric hysteresis (a) and the permittivity (b) (redrawn from [71]).

Figure 3.8 shows the influence of the crystallization temperature on the GI-XRD pattern. From the pattern of the different crystal classes, the fundamental problem in identifying the ferroelectric orthorhombic phase can be seen. As the monoclinic and a higher symmetric cubic/tetragonal phase commonly coexist in one layer, it is challenging to distinguish the orthorhombic phase from a combination of the monoclinic and higher symmetric phase [19, 75]. However, in our case for the crystallization temperatures of 600 °C to 800 °C a strong cubic phase is present. Only the crystallization at 900 °C shows some reflections, which could be attributed to the orthorhombic phase, coexisting with the still dominant cubic phase. Figure 3.9 shows the influence of the crystallization temperature on the ferroelectric hysteresis. The hysteresis for the 600 °C crystallization shows high leakage currents indicated by the strong split up of the hysteresis and the increasing current at higher voltages, which could be explained by remains of organic parts in the layer. For 700 °C the leakage current is strongly reduced and a hysteresis with a remanent polarization of approximately 10 µC/cm² is given. For a crystallization at 800 °C the best electrical results are observed especially indicated by the high current peak. The XRD results show that the highest orthorhombic fraction is given at 900 °C, but no electrical results are available for this temperature due to break down of the devices during wake-up. 34 3 Sample Preparation

Figure 3.8: Influence of the annealing temperature on the crystal structure measured with

Gi-XRD for 5.2 % doped HfO2.

Figure 3.9: (a): Ferroelectric hysteresis in dependence of the crystallization temperature

and the corresponding current measurements (b) for 5.2 % doped HfO2. 3.2 Chemical Solution Deposition 35

3.2.3 HZO and ZrO2

Solution Preparation

The solutions with the different hafnia and zirconia ratios are prepared by the MOD method at

140 °C. Thereby, the duration of the solution heating does play a crucial role for ZrO2 as depicted in Figure 3.10. For shorter durations, a strong reduction of the remanent polarization can be observed. The highest remanent polarization is found for a duration of eight hours. The color of the solution for three different time spans is shown in Figure 3.11. Obviously chemical reactions are taking place during the complete heating time leading to a strong change in the color of the solution, which has a strong influence on the ferroelectric properties. The color change of the solution is also observed for the yttrium doped hafnium oxide, but no influence on the ferroelectric properties is found. Although this fact is very interesting from the chemical point of view, no further investigations are made in this direction. For further results shown in this work, the solution was heated for eight hours. Figure 3.12 shows the ferroelectric hysteresis and the corresponding currents for different solution concentrations. No influence of the solution concentration on the ferroelectric properties can be found.

Figure 3.10: Influence of the duration of the solution heating on the ferroelectric properties

for pure ZrO2. A temperature of 140 °C is used. 36 3 Sample Preparation

Figure 3.11: Color of the solution after heating at 140 °C for different time spans. The pictures of the solution in a glass bottle were recorded with a standard digital camera.

Figure 3.12: Influence of the solution concentration on the ferroelectric properties for pure

ZrO2. No significant influence can be found for the different concentrations.

3.2 Chemical Solution Deposition 37

3.2.4 Photolithography

For the electrical characterization of the manufactured layer, patterned top electrodes need to be deposited to achive a MIM (Metal-Insulator-Metal) structure. The working principle of the pattern transfer from a mask on the deposited photo resist is shown schematically in Figure 3.13 for a negative process.

Figure 3.13: Flow chart of the optical lithography process. At the beginning, the spin coated photo resist is illuminated with UV-light trough a structured mask (a). After removing the non-illuminated resist with a developer (b), platinum is sputtered on the sample (c). In the final step the resist is removed during a lift of process (d).

In the first step a photosensitive resist is spin coated on the processed film. Subsequently the resist is illuminated with UV-light through a shadow mask, usually a silica glass plate with chromium thin structures. In the following step, a developer removes those parts of the resist, which are not illuminated and platinum is sputtered planar on the sample. During the lift off step with acetone the remaining resist is removed leading to a patterned structure of top electrodes with different areas. If not mentioned otherwise capacitors with an area of 2.5·10 -5 cm²

4 Ferroelectric and Piezoelectric Properties of

HZO and ZrO₂

In this chapter, the class of ferroelectrics based on Hf1-xZrxO2 (HZO) and doped ZrO2 is investigated in more detail. Hf1-xZrxO2 is of great interest because of its CMOS compatibility, high remanent polarization, large process window and the possibility to obtain constricted hysteresis by changing the composition as reported by several groups for ALD deposited layers

[23, 27, 32, 76-88]. The stabilization of a higher symmetric phase in ZrO2 due to doping is well known [89, 90]. Especially yttrium stabilized ZrO2 is a widely used material for catalytic [91] and fuel cells [92] applications. As shown in the upcoming sections, it is possible to obtain ferroelectricity in pure ZrO2, which means a large reduction of the preparation costs compared to hafnia based ferroelectrics and an easier solution preparation process, because no doping adjustment is necessary. It is demonstrated that a free energy model, which calculates the preferred crystal phase for a parameter set of surface energies and a grain radius distribution, can explain the stabilization of the ferroelectric orthorhombic phase in the Hf1-xZrxO2 (HZO) and pure ZrO2 system. It reveals that the main parameter to control the ferroelectricity in HZO compounds is the grain size. Up to now for the ALD prepared HZO samples a maximum layer thickness of 50 nm was shown by Riedel and coworkers [93]. Alumina interlayers artificially confined the grainsize of HZO to retain the ferroelectricity at this, compared to previous publications, relatively large thickness. With the CSD process, it is possible to deposit significantly thicker films without using interlayers. The calculation results shown in this chapter are carried out by Materlik and coworkers based on their model published recently [23]. Some statements in this chapter are partially rewritten from a previous publication [94]. 40 4 Ferroelectric and Piezoelectric Properties of HZO and ZrO₂

Wake-up

A wake-up is necessary to exhibit the full ferroelectric properties of the prepared layers. Figure 4.1 (a) shows the progress of the remanent polarization over the applied wake-up cycles. The wake up is completed after 1000 cycles, whereas further cycling leads to a reduction of the remanent polarization. The corresponding hysteresis and current curves are shown in Figure 4.1 (b) and (c). After the first cycle, the polarization measurement shows a constricted hysteresis with its characteristic double current peaks. These two current peaks merge to one peak during cycling. For the following shown results, a wake-up with 1000 cycles at 1 kHz is performed before measuring the hysteresis if not mentioned otherwise.

(a)

(b) (c)

Figure 4.1: (a): Remanent polarization over the number of wake-up cycles. (b)+(c): Exemplary polarization curves and the corresponding current measurements. All data are measured with the Keithley 4200. 4.1 Composition dependence 41

4.1 Composition dependence

In this subchapter, the influence of the film thickness and the hafnia/zirconia ratio is investigated and compared to films prepared by ALD. Figure 4.2 shows the ferroelectric hysteresis of CSD prepared layers with the corresponding current curves for different hafnium contents and two different thicknesses of 44 nm and 100 nm, respectively. A thickness dependence can be observed for the Hf1-xZrxO2 system. Most significant is that for the thicker layer the maximum remanent polarization of 12.5 µC/cm² is present for pure ZrO2 and almost no remanent polarization is found for 20 % hafnium content. However, for the 44 nm thick film a maximum remanent polarization of 15 µC/cm² is found for 5 % hafnium content and a remanent polarization of 6 µC/cm² is still present for 20 %. Figure 4.3 shows the remanent polarization over the hafnium content for the 44 nm and 100 nm CSD prepared films and for 9.2 nm thick ALD films redrawn from Müller et al. [27]. For the 9.2 nm thin film a maximum of the remanent polarization is found to be at 50 % hafnia content, whereas for the CSD films the maximum is strongly shifted towards lower hafnia contents. The free energy model of Materlik et al. consisting of ab initio results and surface energies can explain this apparently strong thickness dependence [23].

44 nm

100 nm

Figure 4.2: Influence of the hafnium content in Hf1-xZrxO2 on the ferroelectric properties shown for two different thicknesses. 42 4 Ferroelectric and Piezoelectric Properties of HZO and ZrO₂

Figure 4.3: Remanent polarization over the hafnium content for the 44 nm and 100 nm CSD films and for comparison a 9.2 nm thin film prepared by ALD (Data taken from [27]).

The free energy predicts the preferred crystal phase of a grain with volume V and surface area

A as the phase with minimal free energy expressed as a function of total energy Ux from density functional calculations (DFT), temperature T from the calculated phonon contribution, surface energy γ from grain size, Zr-content x from DFT, entropy S and applied electric field E from the interaction with the polarization P [23, 23]: γ = – + − . Since each crystal phase changes its free energy F differently, when one of the former condition is changed, a new phase can be energetically preferred and a phase transformation can occur.

The ferroelectric Pca21 crystal phase is preferred under distinct circumstances. An important property is the grain radius since the surface to volume ratio A/V determines the individual surface energy effect. The monoclinic phase P21/c has the highest surface energy, therefore all grains below a critical A/V ratio will become monoclinic. In the presence of a grain size distribution, several phases may exist within the films because grains with different size may have a different low energy phase. Volume fractions of phases, effective relative permittivity and remanent polarization can be calculated as averages of pure phase properties over the size distribution. The strain is also taken into account for the calculations, but no significant influence on the ferroelectric phase stability is found for tensile strain up to approximately 5 GPa [23]. 4.1 Composition dependence 43

Figure 4.4(a) shows a sketch of the film structure with columnar grown grains. For the calculations, cylindrical grains are assumed with a radius r and the height t, which is given by the film thickness.

(b)

(a)

2r Pt

t

HZO

Pt Radius [nm]

200 nm

Figure 4.4: (a):Schematic sketch of the columnar growth of the HZO with platinum electrodes. For the free energy model, the grains are modeled as cylinders with the radius r and the height t. (b): The distribution of the grain radius exemplarily shown for the 44 nm

ZrO2 film. The grain radius is extracted with Gwyddion and fitted to a Poisson distribution with shape parameter s = 8.5, which is used for further calculations.

The distribution of the grain radius is fitted to a Poisson distribution (Figure 4.4 (b)):

. = ! The columnar growth of the CSD prepared layers is exemplary shown for the 44 nm thick layer by scanning electron microscopy measurements (SEM) in Figure 4.5. 44 4 Ferroelectric and Piezoelectric Properties of HZO and ZrO₂

Figure 4.5: SEM figure verifying the columnar growth for the 44 nm thick film.

Since the grain radius distribution does not change with film thickness, the shown distribution is used also for the 100 nm thick film. Using the extracted data for the calculation, the strong difference for the layers in Figure 4.3 can be explained by the different surface energies due to the different grain sizes. Figure 4.6 shows again the electrical results from Figure 4.3 including the calculations.

Figure 4.6: Electrical results as shown in Figure 4.3 including the resulting calculations. The solid lines show the calculated results (since no data of the grain radius distribution is given for the ALD films the shape factor is set to s = 5.3 for the calculations in agreement with [34]). 4.1 Composition dependence 45

The free energy model predicts for the 100 nm thick films a maximum remanent polarization of 10 µC/cm² for pure ZrO2 and a maximum of 14 µC/cm² at approximately 4.5 % hafnium content for the 44 nm thick films, which is confirmed by the electrical results. The strong deviation of the CSD prepared films compared to the ALD films can be explained by the lower film thickness and a different grain radius distribution, which does lead to smaller grain sizes. A further reduction of the thickness and therefore a further reduction of the grainsize, would lead to a shift of the maximum polarization to a higher hafnium content. This was shown by

Polakowski and Müller for ALD prepared pure HfO2 [95]. They found a remanent polarization -2 of >10 µC cm for a 6 nm thick HfO2. An increase of the thickness led to a reduction of the remanent polarization. Kim et al. showed that with increasing layer thickness for Hf0.5Zr0.5O2 the remanent polarization disappears [82]. They were able to prevent the reduction of the remanent polarization by using very thin Al2O3 interlayers and therefore interrupting the grain growth. Furthermore, Kim et al. confirmed ferroelectricity for pure hafnium oxide for smaller grain sizes induced by a lower deposition temperature [96]. Therefore, the grain size seems to play a crucial role for the stabilization of the ferroelectric phase in HZO. It should be mentioned that in the publication of Materlik et al. no stabilization of the ferroelectric phase was found for pure ZrO2 [23]. This is because a fixed relation between grain radius and height (fixed surface/volume ratio) was used, which does not favor the ferroelectric phase for pure ZrO2. This ratio was chosen to successfully explain the phases in ALD prepared thin HZO films. For this work, the measured grain radius distribution of the CSD prepared layers is used for the calculations as mentioned before. The calculated dependence of the remanent polarization on the hafnium content and film thickness is shown in Figure 4.7. A high remanent polarization can be found for thin films with a hafnium content of 30 to 70 %. Unfortunately, the HZO films prepared by CSD show high leakage currents below 30 nm film thickness and are therefore not suitable for polarization measurements. 46 4 Ferroelectric and Piezoelectric Properties of HZO and ZrO₂

100 28 [µC/cm²] Polarization Remanent

24 80 20

60 16

40 12

Hf-content [%] 8 20 4

0 0 510 20 30 40 50 60 70 Film Thickness [nm]

Figure 4.7: Calculated dependence of the remanent polarization on the hafnium content and film thickness. The calculations are based on the grain radius distribution of the CSD films shown in Figure 4.4

These results show that the free energy model predicts the stabilization of the ferroelectric orthorhombic phase for a certain surface area to volume ratio. For a larger height of the grains, that means a larger film thickness, the dependence of the remanent polarization on the hafnium content is shifted towards pure ZrO2, which is confirmed for CSD films. On the other hand, by reducing the grain height the distribution is shifted towards pure HfO2, which was shown for ALD films. Therefore, the stabilization of the ferroelectric phase mainly depends on the hafnium content and surface to volume ratio of the grain in HZO films.

4.2 Ferroelectric ZrO₂

A very interesting result from the data shown before is that pure ZrO2 is ferroelectric. Figure 4.8(a) shows the ferroelectric hysteresis curves for different thicknesses varying from

33 nm to 390 nm for pure ZrO2. Figure 4.8(b) shows the remanent polarization for the different thicknesses including the calculated values from the free energy model. 4.2 Ferroelectric ZrO₂ 47

(a) (b)

Figure 4.8: (a) Thickness dependence of the polarization for pure ZrO2. (b) Remanent polarization of thickness including the calculations.

For film thicknesses above 200 nm a saturation of the remanent polarization sets in. This approximation is explained in the model by the calculation of the surface to volume ratio of a cylinder of radius r and height t as A/V = (2πr2+2πrt)/(πr2t) = 2/r+2/t approaching the thickness independent value 2/r for t>>r. Therefore, no thickness limit is given for the ferroelectric properties of ZrO2. ALD prepared films were found to change their grain radius distribution with the thickness. The surface to volume ratio becomes smaller for increasing film thickness, whereby the ferroelectricity disappears [82]. This might be the reason why CSD films can be very thick and are still ferroelectric, whereas ALD films are limited by thickness. A prediction of the model is therefore, that overcoming of the grain radius confinements of the ALD technique will enable thick ferroelectric ALD films.

4.2.1 Piezoelectric Measurements

The piezoelectric measurements are performed for 100 nm and 390 nm thick pure zirconium oxide films. Figure 4.9 shows the results of the piezoelectric measurements of the 100 nm film for a different amount of triangle cycles. No wake-up was performed before the measurements. Figure 4.10 shows the corresponding hysteresis and current curves. 48 4 Ferroelectric and Piezoelectric Properties of HZO and ZrO₂

pm 25 V

Figure 4.9: Piezoelectric displacement curves of the 100 nm film for a different amount of performed triangle cycles. For the initial curve an average of 50 cycles is needed to get a clear result.

For the initial cycle, a constricted hysteresis is found, which is indicated by the two distinguished current peaks for each voltage polarity. The displacement does not show any difference compared to a pure ferroelectric curve. The piezoelectric coefficient shows the highest value for the initial cycle with a maximum slope of 25 pm/V for lower voltages and 10 pm/V for the full range measurement. These values are comparable to ZnO (~13 pm/V [97]) and AlN (<10 pm/V [98]), which are widely used for piezoelectric applications [99]. With further cycling, the piezoelectric response is reduced. The current curves reveal an increase of the leakage current after 500 cycles, which gets dominant after 43k cycles. Almost no ferroelectric switching peak can be found after this amount of cycles. Interestingly, for the piezoelectric displacement the characteristic butterfly curve is still visible after 43k cycles although the piezoelectric coefficient is strongly reduced. 4.2 Ferroelectric ZrO₂ 49

Figure 4.10: Corresponding ferroelectric hysteresis and current curves for the piezoelectric displacement measurements of the 100 nm film.

The piezoelectric displacement results and the corresponding polarization and current curves for the initial hysteresis of the 390 nm film are shown in Figure 4.11 and Figure 4.12. The values for the piezoelectric properties are slightly smaller compared to the 100 nm film. A maximum value for smaller values of 15 pm/V is measured. For the whole voltage range a value of 7 pm/V is given.

pm 15 V

Figure 4.11: Initials piezoelectric displacement curve of the 390 nm film. An average of 100 cycles is used to get a clear result. 50 4 Ferroelectric and Piezoelectric Properties of HZO and ZrO₂

The hysteresis in Figure 4.12 shows also the constricted behavior with the characteristics double current peaks as observed for the 100 nm film. From the previously shown results, it is expected that the piezoelectric properties for even thicker films will be comparable to the results shown for the 390 nm film. As for the thinner films, the piezoelectric response is reduced after cycling. Therefore, the objective should be to get a stable constricted hysteresis without a wake-up or degradation. This can be achieved by doped ZrO2 as shown in the following chapter.

Figure 4.12: Corresponding P(V) and I(V) curve for the piezoelectric displacement measurement of the 390 nm film.

4.3 Doped ZrO2 51

4.3 Doped ZrO2

In this subchapter, the influence of doping on the ferroelectric and piezoelectric properties of

ZrO2 is investigated. A detailed investigation is done for magnesium, whereas also the doping influence of yttrium, strontium, lanthanum and indium are shown. If not mentioned otherwise a film thickness of approximately 60 nm is used. Except for the DBLI data and the corresponding hysteresis, all shown current and therefore the hysteresis curves were measured with the Keithley 4200.

Magnesium Doping

For the Hf1-xZrxO2 a constricted hysteresis is found during the initial cycling (Figure 4.1) which splits up during the wake-up cycling. This is different for doped ZrO2 as shown in Figure 4.13 for 5 % magnesium doping. After a regular wake-up with 1000 cycles at 1 kHz still a constriction is present, indicated by the two additional current peaks. The current peaks approach each other slightly but do not merge together. Figure 4.14 shows the dependence of the ferroelectric hysteresis after wake-up for different doping concentrations varying between 2.2 % and 15 %.

Figure 4.13: P(V) and the corresponding current curves for 5 % magnesium doped ZrO2

before and after a regular wake-up. In contrast to the doped HfO2 and the Hf1-xZrxO2 system, the constriction is stable after wake-up. 52 4 Ferroelectric and Piezoelectric Properties of HZO and ZrO₂

For the lowest concentration of 2.2 % the current peaks almost merge together after the wake- up. With increasing magnesium concentration, the current peaks separate more and more from each other. In addition, the peak height is reduced indicating a reduction of the switched ferroelectric polarization. At 9 % the switching peaks almost vanish whereby at 15 % only a dielectric response is left. Another point is that the leakage current strongly increases with reduced doping concentration, whereby the hysteresis is distorted. Therefore, it seems that with increasing magnesium content a higher symmetrical phase (tetragonal or cubic) is stabilized and therefore no full wake-up is possible. The constricted hysteresis can be explained by a field induced phase transition, where a phase transition occurs during an applied electrical field [23, 27, 79]. This is different from the cycling induced phase transition as described for the wake- up in chapter 0, because after the wake-up a stable ferroelectric phase is given, whereby the field induced phase transition is a reversible process.

2.2% 3% 5%

7% 9% 15%

Figure 4.14: P(V) curves and the corresponding current curves for different magnesium concentrations after a regular wake-up.

Figure 4.15 shows the wake-up in dependence of the doping concentration. For the 2.2 % and 3 % layer a clear wake-up is visible. The current peaks merge during cycling, whereby no further change is found for a higher amount of cycles. For doping concentrations above 5 % the 4.3 Doped ZrO2 53 wake-up cycling has almost no influence on the resulting current curves. Therefore, it can be concluded that no cycling induced phase transition can be realized for magnesium doping concentrations larger than 5 %, whereby a field induced phase transition is present up to 9 %.

2.2% 3% 5%

7% 9% 15%

Figure 4.15: The influence of the wake-up cycling on the P(V) curves for different magnesium doping concentrations.

The XRD pattern for the different doping concentrations are shown in Figure 4.16. A high symmetric cubic phase is present initially. No significant difference can be found for the different concentrations. Therefore, the magnesium doping does not have an influence on the initial crystal phase, but has a strong impact on the field and cycling induced phase transition of the layer. Silicon doped HfO2 shows similar constricted hysteresis with a very strong temperature dependence, whereby for increasing temperatures the constriction of the hysteresis is getting stronger [34, 100].

54 4 Ferroelectric and Piezoelectric Properties of HZO and ZrO₂

Figure 4.17 shows the temperature dependence for three different magnesium concentrations. There is apparently a relatively strong change of the hysteresis with temperature, but a closer look at the current curves reveals that this is mostly due to increasing leakage currents for increasing temperatures. Only slight differences can be found for the switching peak positions and heights at varying temperatures. Such a weak temperature dependence was also found for constricted HZO curves prepared by ALD [32, 101].

9% 7% 5% 3% 2.2%

Figure 4.16: GI-XRD pattern for different magnesium concentrations. 4.3 Doped ZrO2 55

3% 5% 7%

Figure 4.17: Temperature dependence of the P(V) curves for different magnesium concentrations with the corresponding current curves. All curves are measured without performing wake-up cycles beforehand.

Further dopants

The constricted hysteresis observed for magnesium doping can be also found for other dopants. Nevertheless, different dopants have a different influence on the ferroelectric properties. Figure 4.18 shows the polarization measurements for strontium and yttrium doping with varying concentrations. For a concentration of 2 %, small current switching peaks can be found for strontium doping, whereas stronger peaks are present for yttrium. With increased doping concentration, no switching peaks are present for strontium doping, whereby for yttrium at 3 % still a constricted hysteresis is visible. At a concentration of 5 % also for yttrium a pure dielectric behavior is found. Therefore, it seems that for strontium and yttrium the cubic phase is fully stabilized for lower doping concentrations compared to magnesium.

56 4 Ferroelectric and Piezoelectric Properties of HZO and ZrO₂

Strontium 2% 3% 5%

Yttrium 2% 3% 5%

Figure 4.18: Polarization measurements for strontium and yttrium doping with varying doping concentrations. A regular wake-up was performed before measuring the curves.

Lanthanum Indium

Figure 4.19: Polarization measurements for 5 % lanthanum and indium doping. A regular wake-up was performed before measuring the curves. 4.3 Doped ZrO2 57

Piezoelectric properties

The piezoelectric properties with the corresponding current and P(V) curves are shown for different magnesium concentrations in Figure 4.20. The piezo response for all concentrations looks similar to the one for pure ZrO2 shown in Figure 4.9. A strong difference is found for the endurance of the magnesium doped layers compared to pure ZrO2. After 10k cycles especially for the 7 % and 9 % Mg doping, almost no difference for the piezo response can be found. For the lowest concentration of 2.2 % the displacement is significantly reduced after 10k cycles. As already assumed for the pure ZrO2, the high piezoelectric response is caused by the field induced phase transition. For the lowest concentration a wake-up takes place and hence the field induced phase transition is reduced. At higher concentrations the wake-up is reduced as shown in Figure 4.15. Therefore, the field induced phase transition is more stable at higher doping concentrations without a strong reduction of the piezoelectric coefficient. A high leakage current is present for 9 % doping especially after 10k cycles. This strongly influences the resulting P(V) curve but does not affect the measured displacement. Figure 4.21 shows the displacement results with the corresponding P(V) and I(V) curves. The displacement is not reduced during cycling, because almost no wake-up is present. The displacement for the negative voltage even increases during cycling leading to a more symmetrical curve after 10k cycles. 58 4 Ferroelectric and Piezoelectric Properties of HZO and ZrO₂

2.2% 5% 7% 9%

Figure 4.20: Displacement curves with the corresponding P(V) and I(V) curves for 100 nm thick layer at different magnesium doping concentrations.

Figure 4.21: Displacement curves measured with the DBLI with the corresponding P(V) and I(V) curves for 100 nm thick layers at 3 % yttrium doping.

5 Ferroelectric Properties of Doped HfO₂

Bulk hafnium oxide is known to crystallize in the low symmetrical monoclinic phase. Nevertheless, is was shown that there are several ways to induce a higher symmetric phase like the tetragonal or cubic phase in thin films by thermal treatment [102], growth temperature [103], surface energy effects [104] and mechanical encapsulation realized by the electrodes [105]. Furthermore, doping with various elements leads to stabilization of a higher symmetric phase as summarized by Mueller et al. [30]. After the first discovery of ferroelectricity in silicon doped hafnium oxide in 2011 by Boescke et al. [19], it was shown that ferroelectricity can be induced by several dopants, which are also known to stabilize a higher symmetric phase in hafnium oxide (Y [24], Gd [25], Al [26], Sr [106], La [107]). Ferroelectricity was not only observed for layers prepared by atomic layer deposition, but also observed for yttrium doped samples prepared by sputter deposition [108], pulsed layer deposition [109] and for Hf1-xZrxO2 prepared by chemical vapor deposition (CVD) [78] indicating that the stabilization of the ferroelectric phase is independent of the layer preparation technique. The influence of different dopants on the ferroelectric switching properties of hafnium oxide was reported earlier by Schroeder et al. [110]. They showed that dopants with a smaller ionic radius compared to hafnium exhibit a pinched hysteresis, which is stable even after wake-up cycling and was thus, attributed to an antiferroelectric-like behavior. Later this behavior has been explained by a field-induced phase transition from tetragonal to orthorhombic phase during the application of an electrical field [23]. In this chapter, the results for the sputtered and CSD prepared doped hafnium oxide are presented. The influence of the layer thickness, doping concentration, frequency, and temperature on the ferroelectric properties are discussed exemplary for yttrium doped HfO2. Besides yttrium doping, a wealth of further dopants is used to stabilize the ferroelectric phase in HfO2. Thereby, the influence of different ion radii as well as for different valence states is 60 5 Ferroelectric Properties of Doped HfO₂ investigated. Some of the figures and the corresponding explanations in this chapter are partially redrawn and rewritten from previous publications [71] [111].

5.1 Sputtered yttrium doped HfO₂

5.1.1 Sputtering from oxide targets

For the sputtered samples from the oxide targets, the wake-up as carried out for the CSD samples at 1 kHz with 1k cycles does not lead to an increase of the remanent polarization. Instead, a first wake-up with a very small frequency of 1 Hz and one cycle at approximately 2 MV/cm is needed, before the normal wake-up of 1k cycles at 1 kHz leads to an increase in the remanent polarization. The origin of this necessary extra wake-up is not completely understood yet. Most probably, the oxygen vacancies, which lead to the wake-up, are not sufficiently mobile in the initials state and therefore a frequency of 1 kHz is too high to induce a movement of the vacancies. Therefore, a much lower frequency is initially needed to increase the mobility of the vacancies. The hysteresis and current curves for different yttrium contents are shown in Figure 5.1. The maximum remanent polarization is found for an yttrium sputtering power of 10 W. For a power of 14 W no ferroelectric switching can be found in the current curves. The dependence of the remanent polarization on the yttrium sputter power is given in Figure 5.2. A maximum of the remanent polarization is found for 10 W yttrium power. For larger and smaller yttrium concentrations, a reduction of the remanent polarization is observed.

Such a behavior was also found for yttrium doped HfO2 prepared by sputtering [108] and ALD [24]. A very interesting effect is observed for undoped hafnium oxide. A high remanent polarization of >10 µC/cm² is found as can be seen in 5.1 Sputtered yttrium doped HfO₂ 61

Figure 5.1: Hysteresis and current curves for different yttrium sputtering power.

Figure 5.2: Dependence of the remanent polarization on the yttrium sputter power.

Figure 5.3. This was also observed by Olsen et al. who also prepared ferroelectric hafnium oxide layers by sputtering [108]. This observation can be easily understood, when the surface energy model, which was introduced in chapter 0, is taken into account. By sputtering a polycrystalline layer is formed with small grains in contrast to the columnar grains of ALD and CSD prepared samples. These small grains shift the stabilization of the ferroelectric phase to a higher hafnium content and therefore pure hafnium oxide is ferroelectric, as was also shown by Polakowski and Müller [95] for thin ALD layers. Therefore, the sputtered layers confirm the surface energy model. 62 5 Ferroelectric Properties of Doped HfO₂

Figure 5.3: Ferroelectric hysteresis for pure hafnium oxide. The unexpected behavior can be explained by the surface energy model and the small grains for sputtered layers.

5.1.2 Reactive sputtering

For the reactive sputtered layers a fixed concentration is used. At first view, the results look like normal ferroelectric hysteresis curves (Figure 5.4). Despite the high leakage currents, the ferroelectric switching current peaks are clearly visible. Figure 5.5 shows not only the dynamic hysteresis with the corresponding current curves, but also the curves measured to extract the relaxed remanent polarization as described in chapter 2.8.1. Taking the curves after a relaxation time of one second into account, no relaxed remanent polarization is given. After the relaxation time of one second (blue and green curves at 0 V) the remanent polarization vanishes completely. This is also indicated by the current curves. After the relaxation, no current peak can be found in the subsequent voltage sweep indicating ferroelectric switching. That means that the remanent polarization is only present while performing a dynamic measurement and vanishes completely, when the applied field is switched off for one second. Therefore, the layers prepared by reactive sputtering are actually not ferroelectric, because no remanent polarization is present without an applied electrical field. This could be explained by diffusion of the oxygen vacancies. During the applied electrical field, the vacancies are redistributed within the layer leading to a stabilization of the ferroelectric phase. When the field is turned off, diffusion of oxygen vacancies leads to a destabilization of the ferroelectric phase and therefore the loss of polarization. The influence of oxygen vacancies on the stabilization of the orthorhombic phase is in more detail described in chapter 0. 5.1 Sputtered yttrium doped HfO₂ 63

Figure 5.4: Hysteresis curves with the corresponding current loops with and without leakage compensation.

Figure 5.5: Hysteresis curves with the corresponding current loops with and without leakage compensation. Additionally to Figure 5.4 the curves to extract the relaxed remanent polarization are shown. 64 5 Ferroelectric Properties of Doped HfO₂

5.2 CSD prepared yttrium doped HfO2

5.2.1 Yttrium content dependency

The dependence of the ferroelectric hysteresis on the yttrium content is shown in Figure 5.6.

For pure HfO2 no ferroelectric hysteresis is found. With increasing yttrium content, the remanent polarization increases until a maximum is reached for an yttrium concentration of 5.2 %. Further increase of the yttrium content leads to a reduction of the remanent polarization until at 11 % the hysteresis vanishes. The reason for this dependence of the doping concentration on the hysteresis is not understood yet. A similar behavior is observed also for other dopants and will be discussed in chapter 5.3. Figure 5.7 shows the comparison between three different deposition techniques for yttrium doped HfO2. The CSD layers show a very similar concentration dependence compared to the ALD samples prepared with a post deposition anneal [24]. The layers prepared by sputtering (PVD) show a similar behavior with a distinct maximum of the remanent polarization. However, the maximum is shifted to lower yttrium contents [108].

0% 3.75% 5%

6% 7.5% 11%

Figure 5.6: Dependence of the ferroelectric Hysteresis on the yttrium content. 5.2 CSD prepared yttrium doped HfO2 65

Figure 5.7: Remanent polarization depending on the yttrium concentration for different deposition techniques (data taken from [24] (ALD) and [108] (PVD)) (redrawn from [71]).

A comparison of CSD and ALD prepared samples with different thicknesses is done for

Hf1-xZrxO2 in the previous chapter. A strong influence of the film thickness on the ferroelectric properties is found which can be explained by the strongly different grain sizes. Therefore, it can be assumed that the stabilization of the ferroelectric phase in yttrium doped HfO2 is not dominated by the grain radius effect, but by a significant energy contribution from the dopant. Further parameters like chemical energy from the dopant and oxygen vacancies need to be taken into account to understand the stabilization of the ferroelectric phase in yttrium doped hafnium oxide.

5.2.2 Thickness dependence

Figure 5.8. shows the thickness dependence in a range from 14 nm to 66 nm. Except for the 14 nm thin layer, where a smaller remanent polarization is present, no thickness dependence is found. This furthermore confirms the statement that for doped HfO2 the grain size plays a minor role for the stabilization of the ferroelectric phase. 66 5 Ferroelectric Properties of Doped HfO₂

Figure 5.8: Hysteresis curves for different thicknesses of the 5.2 % yttrium doped HfO2 layers.

5.2.3 Frequency dependence

The influence of the measurement frequency on the hysteresis is shown in Figure 5.9. For the frequency range of 500 Hz to 100 kHz no significant influence on the frequency can be found. Only the remanent polarization reduces slightly, whereas the coercitive field increases.

Figure 5.9: Frequency dependence of the ferroelectric hysteresis measured with the

Keithley 4200 measuring unit for 5.2 % doped HfO2. 5.2 CSD prepared yttrium doped HfO2 67

5.2.4 Temperature dependence

The ferroelectric hysteresis and the relative permittivity (k-value) in dependence of the temperature are shown in Figure 5.10. On the left side a relative strong temperature dependence of the relative permittivity is observed. With rising temperature, also the relative permittivity increases as expected by the Curie-Weiss law. The ferroelectric hysteresis is almost unaffected by the temperature change of 255 °C.

38

34

30

26

Figure 5.10: Temperature dependence of the relative permittivity and the ferroelectric

hysteresis for 5.2 % doped HfO2.

5.2.5 Piezoelectric Measurements

The piezoelectric properties are measured with a Double Beam Laser Interferometer as described in section 2.8.2. The results for a 105 nm thick 5.2% yttrium doped hafnium oxide film are shown in Figure 5.11. The piezoelectric displacement shows the characteristic butterfly loop with its minima at the ferroelectric switching points. Outside the switching range a linear behavior is observed for the displacement unequivocally identifying the piezoelectric nature of the layers. A piezoelectric coefficient of ~1 pm/V can be extracted from the displacement curve which is significantly smaller compared to die piezoelectric coefficient of up to 25 pm/V reached for ZrO2 shown in chapter 4.2.1. 68 5 Ferroelectric Properties of Doped HfO₂

Figure 5.11: Piezoelectric measurement with the corresponding polarization curve for a 105 nm thick hafnium oxide layer with 5.2 % yttrium.

5.3 Further Dopants

A huge advantage of CSD is the easy and fast solution preparation. Many further dopants can be incorporated in HfO2 using the same routine as for yttrium. Of special interest is the influence of the size and the valence state of the dopants on the ferroelectric properties. For a good comparability of all dopants, the preparation conditions are the same for all dopants. The

Hf1-xZrxO2 system shows a strongly different behavior compared to the other dopants and was therefore considered separately in chapter 0. The dopants investigated in the chapter are marked in the periodic table in Figure 5.12. All radii shown here are taken from the work of Shannon [112]. 5.3 Further Dopants 69

hydrogen helium 1 2 H He 1.0079 4.0026 lithium beryllium boron carbon nitrogen oxygen uorine neon 3 4 5 6 7 8 9 10 Li Be B C N O F Ne 6.941 9.0122 10.811 12.011 14.007 15.999 18.998 20.180 sodium magnesium aluminium silicon phosphorus sulfur chlorine argon 11 12 13 14 15 16 17 18 Na Mg Al Si P S Cl Ar 22.990 24.305 26.982 28.086 30.974 32.065 35.453 39.948 potassium calcium scandium titanium vanadium chromium manganese iron cobalt nickel copper zinc gallium germanium arsenic selenium bromine krypton 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 K Ca Sc Ti V Cr Mn Fe Co Ni Cu Zn Ga Ge As Se Br Kr 39.098 40.078 44.956 47.867 50.942 51.996 54.938 55.845 58.933 58.693 63.546 65.38 69.723 72.64 74.922 78.96 79.904 83.798 rubidium strontium yttrium zirconium niobium molybdenum technetium ruthenium rhodium palladium silver cadmium indium tin antimony tellurium iodine xenon 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 Rb Sr Y Zr Nb Mo Tc Ru Rh Pd Ag Cd In Sn Sb Te I Xe 85.468 87.62 88.906 91.224 92.906 95.96 [98] 101.07 102.91 106.42 107.87 112.41 114.82 118.71 121.76 127.60 126.90 131.29 caesium barium hafnium tantalum tungsten rhenium osmium iridium platinum gold mercury thallium lead bismuth polonium astatine radon 55 56 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 Cs Ba Hf Ta W Re Os Ir Pt Au Hg Tl Pb Bi Po At Rn 132.91 137.33 178.49 180.95 183.84 186.21 190.23 192.22 195.08 196.97 200.59 204.38 207.2 208.98 [209] [210] [222] francium radium rutherfordium dubnium seaborgium bohrium hassium meitnerium darmstadtium roentgenium 87 88 104 105 106 107 108 109 110 111 Fr Ra Rf Db Sg Bh Hs Mt Ds Rg [223] [226] [261] [262] [266] [264] [277] [268] [271] [272]

lanthanum cerium praseodymium neodymium promethium samarium europium gadolinium terbium dysprosium holmium erbium thulium ytterbium lutetium 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 La Ce Pr Nd Pm Sm Eu Gd Tb Dy Ho Er Tm Yb Lu 138.91 140.12 140.91 144.24 [145] 150.36 151.96 157.25 158.93 162.50 164.93 167.26 168.93 173.05 174.97 actinium thorium protactinium uranium neptunium plutonium americium curium berkelium californium einsteinium fermium mendelevium nobelium lawrencium 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 Ac Th Pa U Np Pu Am Cm Bk Cf Es Fm Md No Lr [227] 232.04 231.04 238.03 [237] [244] [243] [247] [247] [251] [252] [257] [258] [259] [262]

Figure 5.12: Periodic table of elements with the elements used in this work marked in green .

Rare earth metals

The elements of the lanthanide group are trivalent and therefore generate oxygen vacancies if incorporated on the hafnium position in the lattice. The elements within the lanthanide group show very similar chemical properties due to their same valance electron configuration. The main difference is the ion radius which differs between 103 pm (lanthanum) and 89 pm (erbium). The hysteresis and the current curves for doping with different elements from the rare earth metals are shown Figure 5.13. No considerable difference can be found for the hysteresis except a small deviation in the negative coercitive field. In addition, the current curves show no trend for the different elements. Therefore, it seems that the different ionic radii of the elements of the lanthanoid group do not influence the ferroelectric properties significantly. 70 5 Ferroelectric Properties of Doped HfO₂

Figure 5.13: Ferroelectric hysteresis and the corresponding current curves for hafnium oxide doped with various rare earth elements at 5 % (redrawn from [71]).

Boron group

The ferroelectric hysteresis for aluminum, gallium and indium (boron group) doping with ionic radii varying from 54 pm (Al) to 80 pm (In) are shown in Figure 5.14. Like the rare earth metals the elements from the boron group are trivalent, but they have smaller ionic radii compared to the rare earth metals (aluminum: 54 pm, gallium: 62 pm, indium: 80pm). A small hysteresis can be measured for all elements, whereby the remanent polarization slightly increases with increasing ionic radius from 3 µC/cm² (Al) to 5.2 µC/cm² (In), which is also indicated by the increased switching current peak.

Figure 5.14: Ferroelectric hysteresis and the corresponding current curves for hafnium oxide doped with elements from the boron group at 5 % (redrawn from [111]). 5.3 Further Dopants 71

The influence of the doping concentration on the ferroelectric properties is exemplarily shown for aluminum in Figure 5.15. As for the yttrium doping the maximum ferroelectric response is found for a doping concentration of 5.2 %.

Figure 5.15: Influence of the dopant concentration on the ferroelectric properties exemplarily shown for aluminum doping.

Cobalt and nickel

Figure 5.16 shows the ferroelectric hysteresis and the corresponding current curves for cobalt and nickel doping. Both have an almost identical ionic radius of 69 pm and show very similar hysteresis curves with a remanent polarization of 3.8 µC/cm².

Figure 5.16: Ferroelectric hysteresis and the corresponding current curves for doping with cobalt and nickel at 5 %. 72 5 Ferroelectric Properties of Doped HfO₂

Alkaline earth metals

The alkaline earth metals are of special interest, because these elements show similar chemical properties, but the ionic radii strongly vary between 72 pm (Mg) and 135 pm (Ba). The results for the alkaline earth metals are shown in Figure 5.17. Doping with magnesium shows a very small ferroelectric response with a remenant polarization of 3 µC/cm². Strontium and barium exhibit a significantly larger polarization with a comparable maximum remanent polarization of 12.8 µC/cm² and 12 µC/cm², respectively. The doping dependence of the remanent polarization for the alkaline earth metals is summarized in Figure 5.18. As for yttrium doping a window for the doping concentration is given, whereas the maximum of the remanent polarization is found for all three elements at 7.5 mol% instead of 5.2 mol% as found for yttrium. Another difference is the larger doping window where ferroelectricity appears. At 11 mol% for strontium and barium a relative high remanent polarization of 11 µC/cm² and 7.2 µC/cm² can be still observed. For strontium even at 15 mol% a remanent polarization of 8.3 µC/cm² remains, whereas for barium the remanent polarization almost completely vanishes. The larger process window of strontium is not understood yet. The most significant result is, that strontium and barium show a much higher remanent polarization compared to magnesium. The influence of the ionic radius on the remanent polarization will be discussed later. The crystal phases for the varying doping concentrations are exemplarily shown for barium in Figure 5.19. For pure hafnium oxide a dominating monoclinic phase is found indicated by the double peak around 30°. For an increased barium content the cubic phase is stabilized. The phase evaluation monoclinic to a high-k cubic phase is also visible for the relative permittivity as shown in the inset in Figure 5.19. The relative permittivity increases with increasing barium content from 18 to 27 and saturates for concentrations above 7.5 mol%. It is worth mentioning that the highest remanent polarization is found for 7.5 mol%, where the cubic phase is dominant. As mentioned before, a certain amount of wake-up cycles is necessary to exhibit ferroelectric properties. The reason for this necessary cycling is in detail discussed in chapter 0.

5.3 Further Dopants 73

Magnesium

Strontium

Barium

Electrical Field [MV/cm] Electrical Field [MV/cm]

Figure 5.17: Ferroelectric hysteresis and the corresponding current curves for hafnium oxide doped with elements from alkaline earth metals (redrawn from [111]). 74 5 Ferroelectric Properties of Doped HfO₂

Figure 5.18: Remanent polarization for the different alkaline earth metals in dependence of the doping concentration (redrawn from [111]).

(111) (200)

(311) (222)

Figure 5.19: GI-XRD pattern for varying Barium content. The inset shows the relative permittivity in dependence of the barium concentration. The values for the relative permittivity are taken at 0 V before the wake-up (redrawn from [111]). 5.3 Further Dopants 75

Lead

Lead doping shows a significantly different behavior compared to the dopants shown before. As can be seen in Figure 5.20 no ferroelectric hysteresis can be measured, only the paraelectric polarization that every material exhibits. The ionic radius of lead is 77.5 pm and therefore as large as the ionic radius of indium. Unfortunately no 2,4-pentandionate is available for the other elements from the same group. Especially silicon is an interesting candidate, because it is widely used as a dopant in ALD prepared ferroelectric layers.

Figure 5.20: Ferroelectric hysteresis and the corresponding current curve for hafnium oxide doped with lead.

The GI-XRD measurement is shown in Figure 5.21. A strong monoclinic phase is present as can be seen by the two peaks around 30°. It is obviously not possible to achieve a phase transition from the monoclinic into the ferroelectric orthorhombic phase during wake-up for lead doping. 76 5 Ferroelectric Properties of Doped HfO₂

Figure 5.21: GI-XRD pattern for lead doped hafnium oxide.

Discussion

Figure 5.22 illustrates the remanent polarization in dependence of the ionic radius for several dopants. The remanent polarization is extracted from the hysteresis measurements, whereby an electrical field of maximal 3.25 MV/cm is applied for all dopants. A strong dependence of the remanent polarization on the ionic radius is found. Larger dopants induce an almost four times larger remanent polarization compared to smaller ones. The edge, where the remanent polarization increases, is not as perhaps expectable at the hafnium ion radius, but rather at ion radii larger than approximately 85 pm. A further increase of the ionic radius seems to result in a slight reduction of the remanent polarization. A relatively low remanent polarization of 6 µC/cm² for Aluminium doping was also found for ALD prepared films [26]. In another work by Polakowski et al. an optimized aluminum doping process was presented with a remanent polarization of 20 µC/cm² [113]. Also for silicon doping, whereby silicon is the smallest dopant used so far, a higher remanent polarization was found for ALD prepared layers [19, 114]. Doping with silicon shows a strong dependence on the layer thickness and crystallization 5.3 Further Dopants 77 temperature as found by Yurchuck et al.[114]. The best results were found for a relatively high crystallization temperature of 1000 °C and a layer thickness of 9 nm, whereby for thicker films, the ferroelectricity vanishes. Furthermore, an encapsulation with the top electrode during crystallization was necessary for silicon doping to obtain ferroelectricity as shown by Boescke et al.[100]. Possibly, for higher annealing temperatures and/or capsuling also for the smaller dopants used with CSD a higher remanent polarization could be achieved. As mentioned in chapter 3.2.2, for a higher crystallization temperature no electrical characterization was possible. Crystallization after deposition of the top electrode is also not possible, because during crystallization, the layer is densified and therefore the top electrode is not stable. In contrast to the ALD prepared layers, no thickness dependence is found for doped hafnium oxide prepared by CSD. There are some differences between ALD and CSD deposition, which may have a strong influence on the resulting ferroelectric properties. On the one hand, during CSD deposition, except the crystallization step, a maximum temperature of 215 °C is applied after spin coating, whereas for ALD deposition a temperature of up to 300 °C is used during deposition [114]. On the other hand, for the CSD process most of the organic parts are still present in the layers before crystallization and vaporize during the heating step leading to a strong compression of the layer [71]. Whereby for the ALD process after deposition an amorphous layer is present which is crystallized without such a strong change in density. Therefore, strong differences are present between ALD and CSD prepared samples, although for yttrium doping a very similar behavior was shown in chapter 5.2.1.

78 5 Ferroelectric Properties of Doped HfO₂

Hf

Y La Sm

Nd Er Sr Ba

Ga In Co/Ni Mg Al

Figure 5.22: Remanent polarization in dependence of the ionic radius for different dopants (coordination number of VI is used for all elements for the ionic radius). For the shown data the maximum remanent polarization is used. For the alkaline earth metals the maximum remanent polarization is found at a concentration of 7.5 mol%, whereas for all other dopants a maximum remanent polarization is depicted at 5.2 mol% (redrawn from [111]).

Figure 5.23 shows the exemplary GI-XRD results for two larger (Ba, Y) and smaller (Ni, Mg) dopants, respectively. A very similar pattern is found for all dopants. This is unexpected since it is known that dopants smaller than hafnium preferentially stabilize the tetragonal phase, whereas larger dopants tend to stabilize the cubic phase in hafnium oxide [115-119]. Although a cubic phase is present for all films, still a difference between smaller and larger dopants can be found. For the smaller dopants the peaks are shifted towards larger 2Theta values, which indicates smaller lattice constants. This is also expected if a smaller ion is incorporated into the lattice. Furthermore, a small shift of the platinum peaks at 81° and 46° can be observed towards smaller values, which indicates a change of the lattice constants of the platinum electrode. Therefore, it seems that the oxide layer slightly influences the crystal structure of the platinum during crystallization. These small differences in the lattice constants could hinder the transformation from the cubic into the ferroelectric orthorhombic phase during wake-up cycling for the smaller dopants. 5.3 Further Dopants 79

Although the chemical properties within one group (e.g. Mg, Sr, Ba) are similar, the different elements still show some differences e.g. in their electronegativity, which could influence the local bonding and therefore have an influence on the electrical properties. Further experiments like X-ray absorption spectroscopy and X-ray photoelectron spectroscopy as well as calculations on the atomic level could give a deeper insight into the influence of the dopant on the ferroelectric properties. Comparing the ALD and CSD prepared samples, it seems that the dopant size is not the only parameter which needs to be considered to understand the stabilization of the ferroelectric phase in hafnium oxide.

Figure 5.23: GI-XRD results for two smaller (Mg, Ni) and two larger (Y, Ba) doping elements. For Barium the pattern for a doping concentration of 7.5 mol% is used, whereas for the other elements the pattern for a doping concentration of 5.2 mol% are drawn showing the highest remanent polarization from the electrical results (redrawn from [111]).

In summary, it is shown that many divalent and trivalent dopants with varying ionic radii between 54 pm and 135 pm induce ferroelectricity in hafnium oxide. Nevertheless, it is also shown that lead is not suitable for stabilization of a higher symmetrical phase. Most interesting 80 5 Ferroelectric Properties of Doped HfO₂ is the strong dopant size dependence. For doping elements with an ionic radius larger than 85 pm a significantly higher remanent polarization is measured, which is not fully understood yet. Furthermore, it is still unclear why for the alkaline earth metals a maximum of the remanent polarization is found for a doping concentration of 7.5 mol%, whereas for all other dopants shown here the maximum is given at 5.2 mol%. If the amount of induced oxygen vacancies due to the doping plays a role, it would be expected that for the divalent alkaline earth metals the maximum remanent polarization should be shifted towards lower concentration compared to for example the trivalent yttrium. In addition, the larger doping window, where ferroelectricity appears, is not understood yet. No pentavalent dopants are used in this study because no 2,4-pentandionate is available for these dopants.

6 Wake-up and Degradation

In the following two subchapters the wake-up effect and the degradation (fatigue) in yttrium doped hafnium oxide is investigated in detail. Some of the figures and the corresponding explanations in this chapter are partially redrawn and rewritten from previous publications [120, 121].

6.1 Wake-up

As mentioned before, each measured device needs a certain amount of wake-up cycles to exhibit the observed ferroelectric properties. Wake-up means the initial increase of the remanent polarization during cycling, where mostly a constricted hysteresis is observed in the pristine state, which opens up after cycling. The effect was observed for several ALD prepared samples doped with silicon [122], gadolinium [25] and strontium [106] as well as for the

Hf1-xZrxO2 system [84, 86]. Furthermore, for different deposition techniques (sputtering [108], pulsed layer deposition [123]) also initial cycling was necessary to exhibit the maximum remanent polarization. Therefore, it can be concluded that the wake-up is an intrinsic property of the hafnia and zirconia based ferroelectrics. It is worth mentioning that also for PZT a wake-up was found recently, which however is much weaker compared to hafnia and zirconia [59, 60]. Understanding the underlying effect of the wake-up is essential for the application of hafnium oxide in electronic devices. As a possible explanation Zhou and Lomenzo proposed that a depletion layer is formed between the bottom electrode and the hafnium oxide due to oxidation of the TiN during the deposition of the hafnium oxide [122, 124]. This effect can be 82 6 Wake-up and Degradation excluded for our samples due to use of platinum electrodes. Schenk et al. showed two distinct maxima in the I(V) curves in the pristine state, which are biased against each other and merge together during cycling [125]. They furthermore assumed that this internal bias could be attributed to mobile ions that are rearranged after wake-up cycling. Park et al. assumed a phase transition during wake-up from a tetragonal to the ferroelectric orthorhombic phase as the origin of the observed wake-up for 8 nm Hf0.5Zr0.5O2 [84]. For the results shown here yttrium doped hafnium oxide is used with a concentration of 5 mol% and a layer thickness of 42 nm if not mentioned otherwise. Figure 6.1 (a) shows the initial hysteresis measurement on a pristine sample. It can be seen that initially a constricted hysteresis is present, which is also characterized by two current peaks on each voltage side. After 1000 wake-up cycles performed at 1 kHz clear current peaks can be measured and a ferroelectric hysteresis with a remanent polarization of almost 19 µC/cm² is visible (Figure 6.1 (b)).

(a) (b)

Figure 6.1: (a): Initial polarization measurement of a pristine layer for a layer thickness of 42 nm. (b): Polarization measurement after 1000 wake-up cycles at 1 kHz.

The GI-XRD pattern of the layer is shown in Figure 6.2. All observed reflexes can be assigned either to the platinum substrate or the non-ferroelectric cubic phase. The only reasonable explanation is that during the wake-up a phase transition into the ferroelectric orthorhombic phase occurs. Unfortunately, it was not possible to confirm this assumption by GI-XRD measurements after wake-up. 6.1 Wake-up 83

Figure 6.2: GI-XRD pattern of the pristine sample shown in Figure 6.1.

Figure 6.3 shows the ferroelectric hysteresis and the corresponding current curves in dependence of the number of wake-up cycles. During cycling, the two initial current peaks merge together and after 100 cycles almost the maximum current and therefore the maximum remanent polarization is reached.

Figure 6.3: Hysteresis measurements and the corresponding measured current curves in dependence of the number of wake-up cycles. 84 6 Wake-up and Degradation

6.1.1 Frequency dependence of the wake-up

The wake-up is performed at different frequencies as shown in Figure 6.4(a), where a strong frequency dependence of the remanent polarization is found. At higher wake-up frequencies, a higher amount of cycles is necessary to reach the same remanent polarization. By plotting the remanent polarization over the duration of the applied field (#cycles/frequency) the curves at different frequencies show the same dependence (Figure 6.4(b)). From this it can be concluded that not the amount of cycles but the duration of the applied electrical field is essential for the wake-up. That means that 50 cycles at 50 Hz show the same results as 5k cycles at 5 kHz.

(a) (b)

f

...... #cycles #cycles/frequency [s]

Figure 6.4: Dependence of the remanent polarization on the wake-up frequency. (a): Remanent polarization shown over the number of cycles for different frequencies. (b): Remanent polarization shown over the duration of the applied electrical field (#cycles/frequency) (redrawn from [120]).

6.1.2 Temperature dependence

If the wake-up is affected by movement of ions a strong temperature dependence should be present. The current curves and the corresponding hysteresis are exemplarily shown in Figure 6.5 for a temperature range of -160 °C to 100 °C. It can be seen that a higher remanent polarization is reached for higher temperatures. In addition, the wake-up proceeds much faster at the elevated temperature. Already the initial cycle at 100 °C shows a higher remanent polarization, than after 1000 cycles at -160 °C. Further cycling at the lower temperatures does not lead to a further increase of the remanent polarization. Figure 6.6 shows the dependence of 6.1 Wake-up 85 the remanent polarization for different temperatures. The initial remanent polarization as well as the remanent polarization after the wake-up is sufficiently higher for higher temperatures. This strong temperature dependence indicates the presence of ions that are mobile during the applied electrical fields. At elevated temperatures a higher amount of ions are mobile resulting in a higher remanent polarization. Due to the trivalent yttrium ion replacing the tetravalent hafnium ion in the crystal, a high amount of oxygen vacancies is present in the oxide layer [126]. Therefore, most likely mobile oxygen vacancies in hafnium oxide are related to the observed wake-up effect. Current [A] Current Current [A] Current

Figure 6.5: Temperature dependence of the wake-up exemplary shown for -160 °C and 100 °C (redrawn from [120]). 86 6 Wake-up and Degradation

T

T

Figure 6.6: Remanent polarization in dependence of the number of cycles for different temperatures (redrawn from [120]).

6.1.3 Resistive VCM switching

To prove our assumption that movement of oxygen vacancies is the origin of the wake-up effect, the layer stack is modified by a 10 nm thick TiN layer between the hafnium oxide and the top platinum electrode. Furthermore, the thickness of the hafnium oxide is reduced to 15 nm. An asymmetrical stack is used, because it is known to be preferable for VCM resistive switching [40, 42, 44, 127]. Figure 6.7 shows on the left side the initial P(V) cycle and the layer stack is depicted the inset. For the wake-up 100 k cycles are applied at a frequency of 10 kHz at 3.6 MV/cm (Figure 6.7(a)). In contrast to the thicker films without titanium nitride shown before, more cycles are necessary to achieve the maximum polarization, which was also observed by Hoffmann et al [75]. By applying a negative quasi-static triangle signal with the same maximum electrical field of 3.6 MV/cm instead of the wake-up, an electroforming sets in as known from resistive VCM switching of hafnium oxide [40-46] (Figure 6.7(b)). During forming into the high resistive state, a current compliance of 10 µA is applied to protect the device from permanent breakdown. After forming, the characteristic I(V) curves with stable resistive switching are observed. As well as during the forming step, a current compliance is required to protect the device from breakdown during switching. VCM resistive switching is attributed to the movement of oxygen vacancies during the forming step as well as during the resistive switching after a stable plug has been formed [50-52]. 6.1 Wake-up 87

As depicted in Figure 6.7 the electrical field during wake-up is as high as the forming field. The difference is that during the forming step the voltage is applied for a longer time. This could lead to an incorporation of additional vacancies via the electrodes facilitating the filament formation. Alternatively, this long time is required to agglomerate the existing oxygen vacancies in a filamentary region. After the filament has formed, the resistive switching effect will dominate the device behavior. Due to the low resistance of the filament, any ferroelectric response would be masked. An important further observation is that the forming conditions do not change after wake-up. Therefore, after the wake-up the oxygen vacancies are still sufficiently mobile and are redistributed during each performed cycle. This can lead to a device breakdown during cycling as observed by several groups [75, 124, 128]. Mobile vacancies during cycling are crucial for the investigation of the endurance behavior of the device.

10µ

5µ 0 w -5µ ( ) -10µ

-4 -3-2-1 0 1 432

4µ ( ) 0.8m 0 0.4m 0 -4µ -0.4m -8µ -0.8m -4 -3-2 -1 0 1 -1.0 -0.50 0.50.5

Figure 6.7: The left side shows the initial P(V) cycle and the modified layer stack. The wakeup is depicted in the upper path (a). By applying a triangle voltage to an uncycled device at a much lower frequency of 0.1 Hz, an electroforming step takes place with stable resistive VCM switching after the forming (lower path (b)) (redrawn from [120]).

To substantiate the hypothesis that mobile oxygen vacancies explain the wake-up effect, the ion velocity under the experimental conditions is calculated. As a first approximation, the velocity of the oxygen vacancies can be calculated using the Mott-Gurney equation:

∗ =∗∗exp− ∙ sinh , 2 ∗ 88 6 Wake-up and Degradation where is the hopping distance, the attempt frequency, the activation energy, the charge of the ion and the electric field. For a first approximation, the values measured by Zafar et al. are used ( ) [129]. Furthermore, the = 0.25 nm; = 1 ∙ 10 s ; = 0.6 eV following values are used: = 2 (double positively charged oxygen vacancies); = 295 K; = 3.25 MV/cm (the applied electrical field). Calculating with the given paramters a velocity of 1.3 ∙ 10m/s results. By using a wake-up frequency of 1 kHz the voltage is applied for 0.5 ms for each polarity during each cycle. This results in a migration distance of 6.5 nm during each cycle. Therefore, it seems plausible that oxygen vacancies are sufficiently mobile during the duration of the applied electrical field. A dependence of the wake-up on the layer thickness or on the device area is not found. From this observation, it can be assumed that during the wake-up the oxygen vacancies are redistributed within the hafnium oxide layer leading to a phase transition into the ferroelectric phase.

6.1.4 Pulse wake-up

In the following, the wake-up of two different samples is investigated in more detail. For both samples 5 % yttrium doped HfO2 with a thickness of 15 nm is used. For the first stack platinum top and bottom electrodes are used (symmetrical), whereas for the second device (asymmetrical) a 20 nm thick TiN layer is sandwiched between the hafnium oxide and the platinum top electrode. The two layer stacks used for the following measurements are depicted in Figure 6.8

Figure 6.8: The two layer stacks used for the following measurements, which differ by the additional TiN layer (redrawn from [121]). 6.1 Wake-up 89

The regular wake-up for the device with TiN at a frequency of 1 kHz is shown in Figure 6.9.

Figure 6.9: Hysteresis and current curves for a regular wake-up at a frequency of 1 kHz (redrawn from [121]).

During the initial cycles, no current peaks can be found indicating the ferroelectric switching. With ongoing cycling the current peaks become more distinct and a remanent polarization of 10 µC/cm² is reached after 10k cycles. By applying a quasi-static voltage sweep with a series resistor of 10 kΩ to prevent the device from breakdown, an electroforming step is carried out as depicted in Figure 6.10(a). After the forming step, a reproducible resistive VCM switching can be observed as shown in Figure 6.10(b).

800µ (a) (b)

600µ

400µ

200µ

-200µ

Figure 6.10: (a) Exemplary forming step for the unsymmetrical device with a 10 kΩ series resistor. (b) Resulting resistive switching curves after a forming step (redrawn from [121]). 90 6 Wake-up and Degradation

In the following, the wake-up for both layer configurations induced by single and double pulses is investigated. Therefore, four different pulse patterns are applied: (a): a positive pulse (b): a negative pulse (c): a negative pulse followed by a positive pulse (d): a positive pulse followed by a negative pulse. An electrical field of 3 MV/cm is applied for all pulse patterns, whereby the pulse length for every single pulse is varied between 0.5 s and 10 s. The resulting I(V) curves measured after the pulse wake-up for the symmetrical and asymmetrical device are shown in Figure 6.11. In the middle of each figure for the two different configurations, the assumed distribution of oxygen vacancies is shown. Due to the symmetrical electrodes for the symmetrical device (Figure 6.11(1)) a symmetrical distribution of oxygen vacancies is expected, whereby the most vacancies are accumulated at both interfaces between the hafnium oxide and the platinum electrodes as proposed by Pešić et al. [130]. For the two single pulses ((a) and (b)) a small wake-up can be measured indicated by the growing current peak. Thereby the oxygen vacancies drift to the bottom and top electrode, respectively. This redistribution leads to a phase transition to the ferroelectric orthorhombic phase. Both polarities of the single pulse wake-up show almost identical results, whereby the current peaks increase with increasing pulse times. For the two double pulses ((c) and (d)) a more distinct current peak can be found compared to the single pulses. With increasing pulse times not only the current peak increases but also the curves get more symmetrical. Due to the two pulses, a more homogenous distribution of the oxygen vacancies is achieved leading to a larger ferroelectric response. For the asymmetrical device with the TiN top electrode, the distribution of the oxygen vacancies is expected to be strongly different (Figure 6.11(2)). Most of the oxygen vacancies are accumulated at the interface in the initial state due to oxidation of the TiN at as also reported elsewhere [31-32, 130-131]. The same pulse patterns as well as the same pulse lengths are applied to the asymmetrical device as shown before for the symmetrical layer stack. For the positive single pulse (a) a wake-up is observed for all applied pulse lengths. A strong imprint, which means the shift of the hysteresis along the field axis, is given. This is indicated by the asymmetrical current peaks, which are shifted towards negative voltages. The imprint can be explained by the strong accumulation of the positively charged oxygen vacancies at the top interface, whereby a positive bias voltage results. During the applied pulse some oxygen vacancies drift from the top to the bottom leading to a partially formation of the ferroelectric phase. For longer pulse lengths, more vacancies are activated and therefore a stronger wake-up is observed. For the negative single pulse (b) almost no wake-up can be observed. The negative 6.1 Wake-up 91 voltage applied to the top electrode is partially compensated by the voltage bias induced by the vacancies distribution. Furthermore, most of the positively charged vacancies are accumulated at the top interface and therefore a negative pulse does only lead to a small redistribution of oxygen vacancies within the hafnium oxide layer. The applied double pulse with a negative pulse followed by a positive one (c), shows similar results compared to the single positive pulse. The resulting current peaks are indeed slightly higher, but for both pulse patterns, a strong imprint is found for all pulse durations. The first negative pulse has only a small influence on the vacancies distribution, therefore the wake-up is dominated by the following positive pulse. This leads to a final vacancy distribution as for the single positive pulse (a). For the double pulses where first a positive pulse followed by a negative pulse is applied (d), a lower imprint can be observed compared to the pattern (c). This can be understood by the order of the two pulses. The first positive pulse causes a drift of the oxygen vacancies from the top TiN interface towards the bottom interface. Due to the changed distribution the following negative pulse induces a more homogenous distribution of oxygen vacancies within the layer and therefore a lower imprint compared to the (c) pattern is achieved. A further significant difference can be found for the 10 s pulses. On the one hand the current peaks are symmetrical, therefore no internal bias is left e.g. a symmetrical oxygen vacancy distribution is established. On the other hand, the leakage currents strongly increase, which can be attributed to generation of new oxygen vacancies. The generation of oxygen vacancies as the origin of the increasing leakage current was also assumed by Pešić et al. for increasing cycling numbers [130].

92 6 Wake-up and Degradation

Figure 6.11: Resulting IV curves for the symmetrical (1) and asymmetrical (2) device at varying pulse lengths and different pulse patterns, which are depicted in the inset of each graph. Each pulse pattern and time was applied on a virgin device, respectively. An electrical field of 3 MV/cm is used for all pulses (redrawn from [121]). 6.1 Wake-up 93

The difference of the amount of oxygen vacancies between the top and bottom interface can be estimated by the internal bias field as shown by Park et al. [128]. Therefore, the shift of the current peaks for the 3 s pulse duration of the pulse pattern (c) (negative pulse followed by a positive pulse) is exemplary used to examine the amount of oxygen vacancies. For the symmetrical stack, an internal bias of -0.67 MV/cm is observed whereas for the asymmetrical stack an internal bias of -1.25 MV/cm is measured. This indicates that for both stacks a larger amount of oxygen vacancies is accumulated at the top interface compared to the bottom interface. The concentration of charged defects (n) can be simply estimated by using the following equation:

= . Where q is the charge of the defects, the vacuum permittivity and the relative permittivity of the dielectric film. This results in a larger amount of oxygen vacancies at the top interface of

0.56 ∙ 10 and 1.04 ∙ 10 for the symmetrical and asymmetrical stack, respectively. These values are slightly larger as those calculated by Park et al. ( 1.66 ∙ 10), which is due to the larger bias field [128]. This amount of vacancies is small compared to the overall amount of vacancies expected due to the 5 % doping with the trivalent yttrium, which should result in an overall vacancy concentration of 2.5 %.

Another factor that can influence the built in bias is the difference in the work functions of TiN and Pt. Assuming a linear internal potential over the hafnium oxide and work functions of 5.6 eV and 4.8 eV for platinum and titanium nitride respectively, a negative bias of -0.53 MV/cm results. Nevertheless, this is a very simplified assumption, because due to the high doping concentration within the hafnia film no linear behavior of the potential is expected. Furthermore, for the 10s pulse in Figure 6.11(2)(d) symmetric coercitive fields are reached which also disagrees with the assumption of an built in bias due to the different electrodes. Repeating the double pulse pattern does lead to a stronger wake up as shown exemplarily for the (d) pulse pattern in Figure 6.12(a). Furthermore, the leakage current increases which can be indicated by the increasing current at negative and positive maximum voltages. During the 9th cycle a breakdown occurs, whereupon a reproducible resistive VCM switching is observed as depicted in Figure 6.12(b). That means that during the wake-up oxygen vacancies are generated and at a certain point the electroforming step sets in and the vacancies are accumulated to form a conductive path through the oxide layer, which is illustrated Figure 6.13. 94 6 Wake-up and Degradation

The formation of a conductive path consisting of oxygen vacancies was shown by Yalon et al. [132]. Once the conductive path is formed, the device is shortened and therefore no further ferroelectric switching can be observed. Pešić et al. showed that the leakage current is dominated by the oxygen vacancies at the grain boundaries and therefore it is assumed that the generated vacancies are accumulated at a grain boundary, where also the formation of the filament takes place [46, 130, 133].

(a) (b) V ... t 5s

Figure 6.12: (a): Resulting current curves for repeated double wake-up pulses with a length of 5 s for the asymmetrical layer stack. During the 9th cycle an electrical breakdown occurs. A series resistance of 10 kΩ is used to prevent the device from a full break down. (b): Resistive switching VCM curves after the breakdown (redrawn from [121]).

Figure 6.13: The conductive path formed by oxygen vacancies after the electroforming step (redrawn from [121]). 6.1 Wake-up 95

Pulse length and temperature dependence

In the previous subchapter, the pulse wake-up for different patterns at a relatively large time scale was shown. In the present subchapter the influence of the pulse length from 100 µs to 1 s at different temperatures and field strength is investigated. Therefore, a symmetrical layers setup is used as shown in Figure 6.8, whereby a slightly larger film thickness of 20 nm for the oxide is used. Furthermore, only the pulse pattern with a positive pulse followed by a negative pulse (pattern d) from the previous subchapter is used, whereby the shown pulse length means the duration of each single pulse. Figure 6.14 shows exemplarily the P(V) and the corresponding I(V) curves for different pulse length as well as for a regular wake-up at 70 °C. For the shortest pulse of 100 µs only a small change is given as can be depicted from the current curves. For increasing pulse length, the initially two current peaks merge together and only a single current peak is present after 10 ms. Further increasing does lead to an increase of the current peaks and therefore to an increase of the remanent polarization. The regular wake-up additionally shown in Figure 6.8 is realized by 1000 rectangle cycles at a frequency of 1 kHz. Figure 6.15 shows the dependence of the remanent polarization on the pulse length for different temperatures (a) and varying electrical fields (b). For increasing temperatures also the remanent polarization increases as it was also shown in Figure 6.6. Also for increasing applied electrical field, the remanent polarization increases as shown before. Most significant is the logarithmic dependence of the remanent polarization on the pulse length.

V

t

Figure 6.14: P(V) and I(V) curves for different pulse lengths at 70 °C. The wake-up is realized by a positive pulse followed by a negative. Additionally a regular wake-up with 1000 rectangle cycles at a frequency of 1 kHz is shown. 96 6 Wake-up and Degradation

(a) (b)

2.75 MV/cm 3 MV/cm 3.5 MV/cm . .

Figure 6.15: Remanent polarization in dependence of the pulse length for varying temperatures (a) and field strengths (b).

6.2 Degradation and fatigue

In this subchapter, the degradation and fatigue are investigated. Figure 6.16(a) shows exemplary curves of the remanent polarization in dependence of the number of cycles for different frequencies for the asymmetrical device. For lower frequencies, the remanent polarization increases with cycling, which is not expected for fatigue measurements, whereas for larger frequencies a reduction of the remanent polarization is observed. The hysteresis and current curves after wake-up and after cycling are shown exemplarily for 30 kHz and 100 kHz in Figure 6.16(b). For both frequencies, the current peaks are reduced after cycling indicating a reduction of the ferroelectric response. For the 30 kHz the leakage current strongly increases after cycling, which is the reason why the measured remanent polarization increases (the influence of the leakage current on the resulting hysteresis is in detail described in [58]). Since it is assumed that the leakage current is dominated by the amount of oxygen vacancies within the oxide layer as also shown by Pešić et al. [130], it seems that for higher frequencies the generation of oxygen vacancies is strongly suppressed, whereby the endurance of the devices increases. 6.2 Degradation and fatigue 97

(a) (b)

30 kHz 100 kHz

Figure 6.16: (a) Exemplary fatigue results for the asymmetrical stack for different frequencies showing the evaluation of the remanent polarization over the number of cycles. (b) IV and PV curves for a fatigue frequency of 30 kHz and 100 kHz after wake-up and after 106 cycles, respectively (redrawn from [121]).

Figure 6.17 shows the number of cycles that can be performed until the device breaks down in dependence of the cycling frequency for the symmetrical and asymmetrical stack in the range between 10 kHz and 100 kHz. For each frequency and layer stack, ten devices are measured to obtain the device-to-device variation. Before the measurements, a wake-up was performed at all devices with 1 k cycles (symmetrical) and 10 k cycles (asymmetrical) at 1 kHz, respectively. The red circles indicate the asymmetrical devices where no breakdown is observed after the performed cycling. For the symmetrical devices with inert platinum electrodes a much larger number of cycles can be performed until breakdown especially at the lower frequencies. Furthermore, for both layer stacks a logarithmic dependence of the performed cycles in dependence of cycling frequency is observed. An increase of the cycling frequency by a factor of 10 for example, leads to an increase of the cycles-until-breakdown by a factor of 1000 for the asymmetrical devices. For the higher frequencies, also a much larger scattering is observed compared to lower frequencies for the asymmetrical devices. By using a series resistance during the breakdown a reproducible resistive VCM switching is found as exemplarily shown for a breakdown at 100 kHz in Figure 6.18(a). Comparing the strongly different forming conditions, firstly using a quasi static sweep at one polarity (Figure 6.18(b)), then a pulse sequence at 0.1 Hz (Figure 6.18(c)) and then forming at a high frequency of 100 kHz (Figure 6.18(a)), the 98 6 Wake-up and Degradation resulting switching curves look very similar. Therefore, it seems that the applied forming conditions do not significantly influence the resulting filament. The frequency dependent results show that the endurance of the ferroelectric devices is strongly related to the movement of oxygen vacancies. The ferroelectric switching is superimposed by the movement and generation of oxygen vacancies. Controlling these processes is crucial to achieve a high endurance for the ferroelectric switching.

Figure 6.17: Number of cycles that can be performed on the symmetrical and asymmetrical devices till a breakdown occurs in dependence of the cycling frequency. The red circles indicate the devices, where no breakdown was reached after the amount of cycles (redrawn from [121]). 6.2 Degradation and fatigue 99

(a) (b) (c)

Figure 6.18: Resulting resisting VCM switching curves after a breakdown for (a): frequency forming at 100 kHz (b): long pulse forming (Figure 6.12(b)) (c): sweep forming (Figure 6.10(b)).

To summarize, it is shown that the wake-up is strongly related to redistribution of oxygen vacancies, leading to a stabilization of the ferroelectric phase and can be realized by single voltage pulses. Furthermore, the initial distribution of oxygen vacancies can be extracted, showing that for an asymmetrical device oxygen vacancies are accumulated at the TiN interface leading to an internal field bias. The movement of oxygen vacancies is also present during cycling after the performed wake-up and is therefore superimposed on the ferroelectric switching. A degradation of the devices is observed at further cycling for lower frequencies, whereby an increase of the leakage current is measured. In accordance with the model from Pešić et al. [130] it is assumed that oxygen vacancies are generated during cycling leading to the formation of a conductive filament, which was shown by resistive VCM switching. The generation of oxygen vacancies shows a strong frequency dependence, whereby at larger frequencies the generation of oxygen vacancies is suppressed leading to a strongly increased lifetime of the devices.

7 Pyroelectric Properties

Measuring the pyroelectric coefficient is challenging since the change of the = / polarization for a defined temperature variation has to be measured. A review about different measurements techniques is given by Lubomirsky and Stafsudd [134]. The main technique is to measure the current flow, which is proportional to the polarization change, while changing the temperature linearly. A linear change in temperature leads to a constant current flow and the pyroelectric coefficient can be calculated directly from the measured current: ∆Q =p ⋅ A ⋅∆ T ∆Q ∆T =I = p ⋅ A ⋅ . ∆t ∆ t The measurement is carried out by measuring the current without an applied voltage while changing the temperature between two defined temperatures. Even though the exact temperature ramp is not known, the difference of the two temperatures is given. For the calculation of the charge the measured current is integrated over time. In the following subchapters the results for yttrium doped HfO2 and for pure ZrO2 are presented. In the last section some figures of merit for pyroelectric devices are introduced and compared to established pyroelectric materials. Detecting pyroelectric currents is a further hint that a “real” ferroelectric polarization is present in hafnia and zirconia based films.

102 7 Pyroelectric Properties

7.1 Yttrium Doped Hafnium Oxide

Figure 7.1 shows the results for a 5.2 % yttrium doped hafnium oxide film with a thickness of 45 nm. The temperature is raised from -100 °C to 0 °C and then lowered again to -100 °C.

p=1.37*10-5C/(m²K) p=1.43*10-5C/(m²K)

Figure 7.1: Pyroelectric measurement for a temperature ramp from -100 °C‰0 °C ‰-100°C. For the calculation of the charge the current offset is subtracted.

From the non-constant current it can be concluded that the temperature change is not linear due to the nonlinear current flow. After 630 s, when 0 °C is reached, the temperature is lowered again and an opposite current direction can be observed. The pyroelectric coefficient is calculated separately for the increasing and decreasing temperatures as described before by integrating the current over time. The resulting pyroelectric coefficients p are 1.37*10-5 C/(m²K) and 1.43*10-5 C/(m²K). To be sure that no parasitic current is measured the same sample is measured before and after wake-up. The result for a temperature change from 0 °C to 60 °C is shown in Figure 7.2. Before the wake-up no current flow is measured during the temperature change, whereas after the wake-up a negative current is observed. The calculated pyroelectric coefficient is p=1.9*10-5 C/(m²K) and is therefore slightly higher compared to a temperature change between -100 °C and 0 °C. Further pyroelectric measurements for yttrium doping concentrations of 3.75 % and 7.5 % are shown in Figure 7.3 7.1 Yttrium Doped Hafnium Oxide 103 for a temperature change from -100 °C to 0 °C. For higher and lower doping concentrations than 5.2 % a reduction of the pyroelectric coefficient can be observed.

p=1.9*10-5C/(m²K)

Figure 7.2: Pyroelectric measurement for a sample before and after wake-up for a raising temperature between 0 °C and 60 °C. The measured current can be clearly attributed to a pyroelectric current.

p=1.1*10-5C/(m²K) p=0.7*10-5C/(m²K)

Figure 7.3: Pyroelectric measurements for 3.75 % and 7.5 % yttrium doped samples for a temperature range from -100 °C to 0 °C. 104 7 Pyroelectric Properties

7.2 Pure Zirconium Oxide

Figure 7.4 shows the pyroelectric current for a temperature range from 0 °C to 60 °C after one initial measuring cycle is performed as depicted on the left side. Already after one performed cycle a pyroelectric current can be measured with a resulting pyroelectric coefficient of 1.85*10-5 C/(m²K) for the cooling process. For the yttrium doped hafnium oxide samples no pyroelectric current can be measured after one initial cycle (data not shown here).

Figure 7.4: Pyroelectric currents after one hysteresis measuring cycle, which is on the shown left side. The used area is 0.09 mm².

The pyroelectric currents and the corresponding hysteresis after a full wake-up are depicted in Figure 7.5. Compared to the constricted hysteresis in Figure 7.4 a significantly larger current can be measured leading to a higher pyroelectric coefficient of 2.48*10-5 C/(m²K). This coefficient is also slightly larger than for yttrium doped hafnium oxide. To ensure that the measured current are real pyroelectric currents, a measurement is performed on a device, at which no voltage was applied before. The results are shown in Figure 7.6. As expected no current can be measured during the temperature change. This also means that no remanent polarization is present in the layer before an initial measurement cycle is performed. 7.2 Pure Zirconium Oxide 105

Figure 7.5: Pyroelectric currents after a full wake-up. The used area is 0.09 mm².

To bring more evidence that a real pyroelectric current is measured, the device area is changed. As the pyroelectric current is proportional to the device area, by enlarging the area, also the current is expected to rise by the same factor. This is shown in Figure 7.7. A much larger current can be measured, but the resulting pyroelectric coefficient of 2.81*10-5 C/(m²K) is comparable to the one measured at a smaller area. Thus, it is ensured that a real pyroelectric current is measured with the used setup. Smith et al. showed the pyroelectric nature of HZO for different compositions [135]. A maximum pyroelectric coefficient of 4.8*10-5 C/(m²K) was shown for a zirconium content of 64 % which is comparable to the values shown in this work.

Figure 7.6: Current measured for a virgin device where no voltage was applied before the measurement. No pyroelectric current can be measured as expected. The used area is 0.09 mm² 106 7 Pyroelectric Properties

0

-400f

-800f

-1200f

0 200 400 600 800

Figure 7.7: Pyroelectric current after a full wake-up measured at a device with a larger area of 0.25 mm² compared to a device of 0.09 mm².

7.3 Figures of merit 107

7.3 Figures of merit

As shown in the fundamentals the voltage response for a pyroelectric material which is used between two electrodes is given by: d ∆V = p ⋅ ∆ T ε ε 0 r Therefore, not only the pyroelectric coefficient p is important, but also the thickness of the film and the relative permittivity. The advantage of hafnium and zirconium oxide based ferroelectrics is the comparatively low permittivity of approximately 30 as shown in the chapters before. Furthermore, it is proven in chapter 4.2 that with the used CSD technique no thickness limitations exist and therefore thicker layers can be realized resulting in a high voltage response. For the pyroelectric sensor application and the pyroelectric energy harvesting some figures of merit (FOM) are discussed by Bowen et al. [14]. Here in this work only two figures of merit are considered in more detail. For the pyroelectric sensor application the voltage responsivity is used: p p FV = = cEε0 ε r ρ c p ε 0 ε r

Where cp is the specific heat capacity, ρ the density and cE the volume specific heat. For the energy harvesting application the FOM FE´ including the heat capacity is used: p² FE´= (cE )²ε0 ε r A comparison between hafnium and zirconium oxide based ferroelectrics and other conventionally used materials is shown in Table 1. The data for the specific heat capacity for hafnium dioxide is taken from [136]. Compared to the other materials the hafnium and zirconium based ferroelectrics show lower values concerning the FOM´s. Due to its high pyroelectric coefficient and low relative permittivity triglycine sulphide (TGS) has an order of magnitude higher voltage responsivity

FV and even more than two orders of magnitude larger FE´ for the energy harvesting application. Despite the high values for the FOM´s, TGS has the disadvantage of a low Curie temperature of 49 °C. Furthermore, these high values are only give for single crystalline TGS [14]. 108 7 Pyroelectric Properties

-2 -1 -3 .1 -1 3 -1 -11 Materials p[µC m K ] εr cE[MJ m K ] FV[m² C ] FE´[m J ] *10

HfO2/ZrO2 <-28 30 2.77 0.0266 0.0177 ferroelectrics

Triglycine -280 38 2.3 0.362 4.41 sulphide (TGS)

Lead magnesium -1790 2100 2.5 0.039 2.76 niobate – lead titanate

Lead zirconate -380 290 2.5 0.059 0.9 titanate (PZT)

Barium titanate -260 1200 2.5 0.008 0.06

Table 1: Comparison between hafnium and zirconium oxide based ferroelectrics and conventionally used pyroelectric materials. The data for the other materials are taken from [14].

The other materials shown in the table have a large relative permittivity and therefore show comparable values to hafnium and zirconium oxide at least for the voltage responsivity. It should be kept in mind that the large advantage of pure zirconium oxide based devices is the fast, easy and cost efficient layer preparation. Therefore, the hafnium oxide and especially the zirconium oxide based ferroelectrics have a high potential for application in pyroelectric sensor application.

8 Conclusions

8.1 Summary

In this thesis the ferroelectric, pyroelectric and piezoelectric properties of hafnia and zirconia based ferroelectrics are studied. For the first time it is shown that chemical solution deposition (CSD) can be used for the deposition of hafnia and zirconia based ferroelectric layers in a wide thickness range of 14 nm to 390 nm. A dopant screening reveals that a wealth of dopants does induce ferroelectricity in hafnium oxide. Thereby, a strong dependence on the ionic radius is found. For dopants with an ionic radius significantly larger than hafnium, a much stronger ferroelectric response is given. For all dopants, a certain doping concentration window is found where the ferroelectric phase can be stabilized. The origin of this concentration window is not understood up to now.

For the stabilization of the ferroelectric phase, a strong difference is found between doped HfO2 and the Hf1-xZrxO2 (HZO) system. It is shown, that the grain size plays a crucial role for the stabilization of the ferroelectric phase in the HZO system. With increasing grain size, which is realized by an increase of the layer thickness, the maximum remanent polarization is found for zirconium rich compounds. No such dependence on the grain size is found for doped HfO2. Therefore, it is concluded that the grain size plays a dominant role for the stabilization of the

HZO system, whereas it plays a minor part for the doped HfO2.

Ferroelectric ZrO2 is deposited up to a layer thickness of 390 nm, which is significantly larger compared to a layer thickness of 50 nm reached by ALD deposition. Doping of ZrO2 with various dopants shows a stable constricted hysteresis where no wake-up is observed during cycling. This makes doped ZrO2 useable for energy related applications such as electrocaloric 110 8 Conclusions cooling. The thick layers and a measured piezo coefficient up to 25 pm/V make the CSD prepared pure and doped ZrO2 layers suitable for future piezoelectric sensors and actuators. The wake-up effect, which means the initial increase of the remanent polarization during cycling, is studied in detail for yttrium doped HfO2. In the scope of this work, a correlation between the wake-up effect and the movement of oxygen vacancies is proven for the first time. It is shown that during the wake-up cycling oxygen vacancies are redistributed within the layer, leading to a cycling induced phase transition from a cubic into a ferroelectric phase. Furthermore, the mobile oxygen vacancies can lead to a dielectric breakdown during ferroelectric switching whereby a filament of oxygen vacancies is formed, which is proven by resistive VCM switching curves. A further aspect, which has a strong impact on the endurance of the ferroelectric devices, is the generation of oxygen vacancies during cycling by ionic exchange, which leads to an acceleration of the dielectric breakdown. It is shown that the generation of vacancies and the associated dielectric breakdown is strongly frequency dependent, whereby the endurance of the devices strongly increases with increasing cycling frequencies. Beside ferroelectric and piezoelectric measurements, also pyroelectric measurements are performed for yttrium doped hafnium oxide as well as for pure ZrO2. The measurements reveal the pyroelectric nature of the materials with a maximum pyroelectric coefficient of 28 C m-2 K-1. This value is small compared to state of the art pyroelectric materials, but due to the relatively small dielectric constant of the hafnia and zirconia based ferroelectrics, an application in pyroelectric devices is still conceivable.

Apart from the layer preparation by CSD, yttrium doped HfO2 is also prepared by sputter deposition. As for the CSD prepared layers an yttrium concentration window is found for the sputtered layers where the ferroelectric phase is present. The most interesting result from the sputtered layers is the relatively large remanent polarization found for pure HfO2. This can be explained by the grain size model, whereby for smaller grains, which result from the sputter deposition, also pure HfO2 does become ferroelectric. Therefore, the sputtered samples confirm the strong grain size dependence of undoped Hf1-xZrxO2.

8.2 Outlook 111

8.2 Outlook

In this work, fundamental results concerning preparation as well as understanding the stabilization of the ferroelectric phase of hafnia and zirconia based ferroelectrics are presented. Furthermore, the promising potential of this new class of ferroelectrics for various applications is shown. Nevertheless, some essential questions stay unanswered. It is still unclear why ferroelectric properties in doped hafnium oxide can only be observed for a certain doping concentration window. In addition, the dependence of the ferroelectric properties on the ionic radius and the valence state of the dopant is still not understood. Therefore, further fundamental work is necessary to fully understand the stabilization of the ferroelectric phase in these material systems. Beside the memory application, where already state of the art devices were fabricated based on hafnia and zirconia ferroelectrics, no proof of concept was shown for pyroelectric or piezoelectric sensors and actuators up to now. For the latter applications, CSD is extremely interesting: On the one hand, due to its easy, fast and cost efficient layer preparation, on the other hand, due to the possibility to deposit thicker layers, which is essential for almost all piezoelectric applications. Therefore, especially the zirconia based ferroelectric layers should be intergraded into sensors and actuators to get a comparison to state of the art devices based on conventional ferroelectric, pyroelectric and piezoelectric materials.

Bibliography

1. J. Valasek. Piezo-electric and Allied Phenomena in Rochelle Salt. Physical Review, 17, 475 (1920). 2. G. Busch and P. Scherrer. Eine neue Seignetteelektrische Substanz. Naturwissenschaften, 23, 737 (1935). 3. A. von Hippel, R. Breckenridge, F. Chesley and L. Tisza. High Dielectric Constant Ceramics. Industrial and Engineering Chemistry, 38, 1097-1109 (1946). 4. A. K. Tagantsev, V. O. Sherman, K. F. Astafiev, J. Venkatesh and N. Setter. Ferroelectric materials for microwave tunable applications. Journal of Electroceramics, 11, 5-66 (2003). 5. N. Setter, D. Damjanovic, L. Eng, G. Fox, S. Gevorgian, S. Hong, A. Kingon, H. Kohlstedt, N. Y. Park, G. B. Stephenson, I. Stolitchnov, A. K. Taganstev, D. V. Taylor, T. Yamada and S. Streiffer. Ferroelectric thin films: Review of materials, properties, and applications. Journal of Applied Physics, 100, 51606/1- (2006). 6. J. F. Scott. Applications of modern ferroelectrics. Science, 315, 954-959 (2007). 7. H. Funakubo, M. Dekkers, A. Sambri, S. Gariglio, I. Shklyarevskiy and G. Rijnders. Epitaxial PZT films for MEMS printing applications. MRS Bulletin, 37, 1030–1038 (2012). 8. C.-B. Eom and S. Trolier-McKinstry. Thin-film piezoelectric MEMS. MRS Bulletin, 37, 1007–1017 (2012). 9. S.-G. Kim, A. Priya and I. Kanno. Piezoelectric MEMS for energy harvesting. MRS Bulletin, 37, 1039–1050 (2012). 114 Bibliography

10. M. Lee, R. Guo and A. Bhalla. Pyroelectric sensors. Journal of Electroceramics, 2, 229-242 (1998). 11. A. Mischenko, Q. Zhang, J. Scott, R. Whatmore and N. Mathur. Giant electrocaloric

effect in thin-film PbZr0.95Ti0.05O3. Science, 311, 1270-1271 (2006). 12. B. Neese, B. Chu, S. Lu, Y. Wang, E. Furman and Q. M. Zhang. Large electrocaloric effect in ferroelectric polymers near room temperature. Science, 321, 821-823 (2008). 13. C. R. Bowen, H. A. Kim, P. M. Weaver and S. Dunn. Piezoelectric and ferroelectric materials and structures for energy harvesting applications. Energy & Environmental Science, 7, 25-44 (2014). 14. C. R. Bowen, J. Taylor, E. LeBoulbar, D. Zabek, A. Chauhanc and R. Vaishc. Pyroelectric materials and devices for energy harvesting applications. Energy & Environmenal Science (2014). 15. Y. Yang, W. Guo, K. C. Pradel, G. Zhu, Y. Zhou, Y. Zhang, Y. Hu, L. Lin and Z. L. Wang. Pyroelectric Nanogenerators for Harvesting Thermoelectric Energy. Nano Letters, 12, 2833-2838 (2012). 16. N. Izyumskaya, Ya. Alivov, S. Cho, H. Morkoc, H. Lee and Y. Kang. Processing, structure, properties, and applications of PZT thin films. Critical Reviews in Solid State and Materials Sciences, 32, 111-202 (2007). 17. T. P. Juan, Chung-yuan Chang and J. Y. Lee. A new metal-ferroelectric

(PbZr0.53Ti0.47O3)-insulator (Dy2O3)-semiconductor (MFIS) FET for nonvolatile memory applications. IEEE Electron Device Letters, USA, 27, 217-20 (2006). 18. M. Qazi, M. Clinton, S. Bartling and A. P. Chandrakasan. A Low-Voltage 1 Mb FRAM in 0.13 mu m CMOS Featuring Time-to-Digital Sensing for Expanded Operating Margin. IEEE Journal of Solid-State Circuits, 47, 141-150 (2012). 19. T. S. Boescke, J. Mueller, D. Braeuhaus, U. Schroeder and U. Boettger. Ferroelectricity in Hafnium Oxide thin films. Applied Physics Letters, 99, 102903 (2011). 20. T. S. Boescke, J. Mueller, D. Braeuhaus, U. Schroeder and U. Boettger. Ferroelectricity in Hafnium Oxide: CMOS compatible Ferroelectric Field Effect Transistors. 2011 Ieee International Electron Devices Meeting (iedm), 547-550 (2011). Bibliography 115

21. X. Sang, E. D. Grimley, T. Schenk, U. Schroeder and J. M. LeBeau. On the structural

origins of ferroelectricity in HfO2 thin films. Applied Physics Letters, 106, 162905 (2015). 22. D. Martin, J. Mueller, T. Schenk, T. M. Arruda, A. Kumar, E. Strelcov, E. Yurchuk, S. Mueller, D. Pohl, U. Schroeder, S. V. Kalinin and T. Mikolajick. Ferroelectricity

in Si-Doped HfO2 Revealed: A Binary Lead-Free Ferroelectric. Advanced Materials, 26, 8198-8202 (2014).

23. R. Materlik, C. Künneth and A. Kersch. The origin of ferroelectricity in Hf1-xZrxO2: A computational investigation and a surface energy model. Journal of Applied Physics, 117, 134109 (2015). 24. J. Mueller, U. Schroeder, T. S. Boescke, I. Mueller, U. Boettger, L. Wilde, J. Sundqvist, M. Lemberger, P. Kuecher, T. Mikolajick and L. Frey. Ferroelectricity in yttrium- doped hafnium oxide. Journal of Applied Physics, 110 (2011). 25. S. Mueller, C. Adelmann, A. Singh, S. Van Elshocht, U. Schroeder and T. Mikolajick.

Ferroelectricity in Gd-Doped HfO2 Thin Films. Ecs Journal of Solid State Science and Technology, 1, N123-N126 (2012). 26. S. Mueller, J. Mueller, A. Singh, S. Riedel, J. Sundqvist, U. Schroeder and T.

Mikolajick. Incipient Ferroelectricity in Al-Doped HfO2 Thin Films. Advanced Functional Materials, 22, 2412-2417 (2012). 27. J. Mueller, T. S. Boescke, U. Schroeder, S. Mueller, D. Braeuhaus, U. Boettger, L.

Frey and T. Mikolajick. Ferroelectricity in Simple Binary ZrO2 and HfO2. Nano Letters, 12, 4318-4323 (2012). 28. S. Mueller, J. Mueller, U. Schroeder and T. Mikolajick. Reliability Characteristics of

Ferroelectric Si: HfO2 Thin Films for Memory Applications. IEEE Transactions on Device and Materials Reliability, 13, 93-97 (2013). 29. T. Mikolajick, S. Müller, T. Schenk, E. Yurchuk, S. Slesazeck, U. Schröder, S. Flachowsky, R. van Bentum, S. Kolodinski, P. Polakowski and J. Müller. Doped Hafnium Oxide – An Enabler for Ferroelectric Field Effect Transistors. Advances in Science and Technology, 95, 136-145 (2014). 30. J. Mueller, P. Polakowski, S. Mueller and T. Mikolajick. Ferroelectric Hafnium Oxide Based Materials and Devices: Assessment of Current Status and Future Prospects. Ecs Journal of Solid State Science and Technology, 4, N30-N35 (2015). 116 Bibliography

31. Min Hyuk Park, Han Joon Kim, Yu Jin Kim, Woojin Jeon, Taehwan Moon and Cheol

Seong Hwang. Ferroelectric properties and switching endurance of Hf0.5Zr0.5O2 films

on TiN bottom and TiN or RuO2 top electrodes. Phys. Status Solidi RRL, 8, 532-535 (2014).

32. M. H. Park, H. J. Kim, Y. J. Kim, T. Moon, K. D. Kim and C. S. Hwang. Thin HfxZr1-

xO2 Films: A New Lead-Free System for Electrostatic Supercapacitors with Large Energy Storage Density and Robust Thermal Stability. Advanced Energy Materials, 4, 1400610/1- (2014). 33. Min Hyuk Park, Han Joon Kim, Yu Jin Kim, Taehwan Moon, Keum Do Kim and Cheol Seong Hwang. Toward a multifunctional monolithic device based on

pyroelectricity and the electrocaloric effect of thin antiferroelectric HfxZr1-xO2 films. Nano Energy, 12, 131–140 (2015). 34. M. Hoffmann, U. Schroeder, C. Künneth, A. Kersch, S. Starschich, U. Böttger and T.

Mikolajick. Ferroelectric Phase Transitions in Nanoscale HfO2 Films Enable Giant Pyroelectric Energy Conversion and Highly Efficient Supercapacitors. Nano Energy, 154–164 (2015). 35. D. Bolten. Reversible and Irreversible Polarization Processes in Ferroelectric Ceramics and Thin Films., PhD Thesis RWTH Aachen (2002). 36. R. Waser, R. Bruchhaus and S. Menzel. Redox-based Resistive Switching Memories. Wiley-VCH, 683-710 (2012). 37. P. Parreira, G. W. Paterson, S. McVitie and D. A. MacLaren. Stability, bistability and instability of amorphous ZrO2 resistive memory devices. Journal of Physics D: Applied Physics, 49, 95111/1- (2016). 38. M. Wu, Y. Lin, W. Jang, C. Lin and T. Tseng. Low-power and highly reliable multilevel operation in ZrO2 1T1R RRAM. IEEE Electron Device Letters, 32, 1026- 1028 (2011). 39. I. Kaerkkaenen, A. Shkabko, M. Heikkila, M. Vehkamaki, J. Niinisto, N. Aslam, P. Meuffels, M. Ritala, M. Leskela, R. Waser and S. Hoffmann-Eifert. Impedance spectroscopy study of the unipolar and bipolar resistive switching states of atomic layer deposited polycrystalline ZrO2 thin films. Physica Status Solidi a-Applications and Materials Science, 212, 751-766 (2015). Bibliography 117

40. S. Brivio, J. Frascaroli and S. Spiga. Role of metal-oxide interfaces in the multiple resistance switching regimes of Pt/HfO2/TiN devices. Applied Physics Letters, 107, 23504/1-5 (2015). 41. H.Y. Lee, P.S. Chen, T.Y. Wu, Y.S. Chen, C.C. Wang, P.J. Tzeng, C.H. Lin, F. Chen, C.H. Lien and M.-J. Tsai. Low Power and High Speed Bipolar Switching with A Thin

Reactive Ti Buffer Layer in Robust HfO2 Based RRAM. IEEE Transactions on Dielectrics and Electrical Insulation, 1-4 (2008). 42. M. Sowinska, T. Bertaud, D. Walczyk, S. Thiess, P. Calka, L. Alff, C. Walczyk and T. Schroeder. In-operando hard X-ray photoelectron spectroscopy study on the impact of current compliance and switching cycles on oxygen and carbon defects in resistive switching Ti/HfO2/TiN cells. Journal of Applied Physics, 115, 204509 (2014). 43. L. Zhao, H. Chen, S. Wu, Z. Jiang, S. Yu, T. Hou, H. P. Wong and Y. Nishi. Multi- level control of conductive nano-filament evolution in HfO2 ReRAM by pulse-train operations. Nanoscale, 6, 5698-5702 (2014). 44. L. Goux, X. P. Wang, Y. Y. Chen, L. Pantisano, N. Jossart, B. Govoreanu, J. A. Kittl, M. Jurczak, L. Altimime and D. J. Wouters. Roles and Effects of TiN and Pt Electrodes in Resistive-Switching HfO2 Systems. Electrochemical and Solid-State Letters, 14, H244 (2011). 45. P. Gonon, M. Mougenot, C. Vallee, C. Jorel, V. Jousseaume, H. Grampeix and F. El Kamel. Resistance switching in HfO2 metal-insulator-metal devices. Journal of Applied Physics, 107, 74507/1- (2010). 46. M. Lanza, G. Bersuker, M. Porti, E. Miranda, M. Nafria and X. Aymerich. Resistive switching in hafnium dioxide layers: Local phenomenon at grain boundaries. Applied Physics Letters, 101, 193502 (2012). 47. J. J. Yang, F. Miao, M. D. Pickett, D. A. A. Ohlberg, D.R. Stewart, C. N. Lau and R. S. Williams. The mechanism of electroforming of metal oxide memristive switches. Nanotechnology, 20, 215201 (2009). 48. E. Abbaspour, S. Menzel and C. Jungemann. The Role of the Interface Reactions in the Electroforming of Redox-based Resistive Switching Devices Using KMC Simulations. 2015 International Conference On Simulation of Semiconductor Processes and Devices (SISPAD), Washington D.C, USA, 2015 International 118 Bibliography

Conference On Simulation of Semiconductor Processes and Devices (SISPAD), Washington D.C, USA, 293-296 (2015). 49. R. Waser, R. Dittmann, G. Staikov and K. Szot. Redox-Based Resistive Switching Memories - Nanoionic Mechanisms, Prospects, and Challenges. Advanced Materials, 21, 2632-2663 (2009). 50. J. J. Yang, M. D. Pickett, X. Li, D. A. A. Ohlberg, D. R. Stewart and R. S. Williams. Memristive switching mechanism for metal/oxide/metal nanodevices. Nature Nanotechnology, 3, 429-433 (2008). 51. S. Larentis, F. Nardi, S. Balatti, D. C. Gilmer and D. Ielmini. Resistive Switching by Voltage-Driven Ion Migration in Bipolar RRAM-Part II: Modeling. IEEE Transactions on Electron Devices, 59, 2468-2475 (2012). 52. A. Marchewka, B. Roesgen, K. Skaja, H. Du, C. L. Jia, J. Mayer, V. Rana, R. Waser and S. Menzel. Nanoionic Resistive Switching Memories: On the Physical Nature of the Dynamic Reset Process. Advanced Electronic Materials, 2, 1500233/1-13 (2016). 53. D. Ielmini and R. Waser. Resistive Switching - From Fundamentals of Nanoionic Redox Processes to Memristive Device Applications. Wiley-VCH (2016). 54. R. Waser. Redox-Based Resistive Switching Memories. Journal of Nanoscience and Nanotechnology, 12, 7628-7640 (2012). 55. Donald M. Mattox. Handbook of Physical Vapor Deposition (PVD) Processing. William Andrew (1998). 56. P. J. Kelly and R. D. Arnell. Magnetron sputtering: a review of recent developments and applications. Vacuum, 56, 159–172 (2000). 57. G. Franz. Niederdruck-Plasmen und Mikrostrukturtechnik. Springer (2003). 58. T. Schenk, E. Yurchuk, S. Mueller, U. Schroeder, S. Starschich, U. Bottger and T. Mikolajick. About the deformation of ferroelectric hystereses. Applied Physics Reviews, 1, 041103 (14 pp.)-0411 (2014). 59. F. P. G. Fengler, M. Pešic, S. Starschich, T. Schneller, U. Böttger, T. Schenk, M. H. Park, T. Mikolajick and U. Schroeder. Comparison of hafnia and PZT based ferroelectrics for future non-volatile FRAM applications. 2016 46th European Solid- State Device Research Conference (ESSDERC), 369-372 (2016). 60. F. P. G. Fengler, M. Pešic´, S. Starschich, T. Schneller, C. Künneth, U. Böttger, H. Mulaosmanovic, T. Schenk, M. H. Park, R. Nigon, P. Muralt, T. Mikolajick and U. Bibliography 119

Schroeder. Domain Pinning: Comparison of Hafnia and PZT Based Ferroelectrics. Advanced Elekectronic Materials, 3 (2017). 61. P. Gerber, A. Roelofs, O. Lohse, C. Kugeler, S. Tiedke, U. Boettger and R. Waser. Short-time piezoelectric measurements in ferroelectric thin films using a double-beam laser interferometer. Review of Scientific Instruments, 74, 2613-15 (2003). 62. A. Kholkin, C. Wutchrich, D. Taylor and N. Setter. Interferometric measurements of electric field-induced displacements in piezoelectric thin films. Review of Scientific Instruments, 67, 1935-1941 (1996). 63. A. Kholkin, A. Seifert and N. Setter. Electromechanical properties of sol-gel derived

Ca-modified PbTiO3 films. Applied Physics Letters, USA, 72, 3374-6 (1998). 64. H. Kiessig. Total reflection of X-rays. Annalen der Physik, 10, 715-768 (1931). 65. H. Kiessig. Interference of X-rays in thin films. Annalen der Physik, 10, 769-788 (1931). 66. S. Schmelzer. Ultra Thin Oxide Films for Dielectric and Resistive Memory Applications., PhD Thesis RWTH Aachen (Germany) (2013). 67. Theodor Schneller, Rainer Waser, Marija Kosec and David Payne. Chemical Solution Deposition of Functional Oxide Thin Films. Springer (2013). 68. M. S. Bhuiyan, M. Paranthaman and K. Salama. Solution-derived textured oxide thin films - a review. Superconductor Science & Technology, 19, R1-R21 (2006). 69. T. Schneller and R. Waser. Chemical solution deposition of ferroelectric thin films - state of the art and recent trends. Ferroelectrics, Switzerland, 267, 293-301 (2002). 70. A.W. Coats and J.P. Redfern. Thermogravimetric analysis. A review. Analyst (1963). 71. S. Starschich, D. Griesche, T. Schneller and U. Böttger. Chemical Solution Deposition of Ferroelectric Hafnium Oxide for Future Lead Free Ferroelectric Devices. ECS Journal of Solid State Science and Technology, 4, 419-423 (2015). 72. S. Petit, S. Morlens, Z. Yu, D. Luneau, G. Pilet, J. Soubeyroux and P. Odier. Synthesis and thermal decomposition of a novel zirconium acetato-propionate cluster: [Zr12]. Solid State Sciences, 13, 665-670 (2011). 73. K. Knoth, R. Hühne, S. Oswald, L. Schultz and B. Holzapfel. Detailed investigations on La2Zr2O7 buffer layers for YBCO-coated conductors prepared by chemical solution deposition. Acta Materialia, 55, 517-529 (2007). 120 Bibliography

74. M. Yu, N. Bovet, C. J. Satterley, S. Bengio, K. R. J. Lovelock, P. K. Milligan, R. G. Jones, D. P. Woodruff and V. Dhanak. True Nature of an Archetypal Self-Assembly System: Mobile Au-Thiolate Species on Au(111). Physical Review Letters, 97, 166102- 1-166102-4 (2006). 75. M. Hoffmann, U. Schroeder, T. Schenk, T. Shimizu, H. Funakubo, O. Sakata, D. Pohl, M. Drescher, C. Adelmann, R. Materlik, A. Kersch and T. Mikolajick. Stabilizing the ferroelectric phase in doped hafnium oxide. Journal of Applied Physics, 118, 72006/1- (2015). 76. J. Mueller, T. S. Boescke, D. Braeuhaus, U. Schroeder, U. Boettger, J. Sundqvist, P.

Kuecher, T. Mikolajick and L. Frey. Ferroelectric Zr0.5Hf0.5O2 thin films for nonvolatile memory applications. Applied Physics Letters, 99, 112901/1- (2011). 77. M. H. Park, H. J. Kim, Y. J. Kim, W. Lee, T. Moon and C. S. Hwang. Evolution of

phases and ferroelectric properties of thin Hf0.5Zr0.5O2 films according to the thickness and annealing temperature. Applied Physics Letters, 102, 242905/1- (2013). 78. T. Shimizu, T. Yokouchi, T. Shiraishi, T. Oikawa, P. S. S. R. Krishnan and H. Funakubo. Study on the effect of heat treatment conditions on metalorganic-chemical-

vapor-deposited ferroelectric Hf0.5Zr0.5O2 thin film on Ir electrode. Japanese Journal of Applied Physics, 53, 9PA04/1- (2014).

79. S. E. Reyes-Lillo, K. F. Garrity and K. M. Rabe. Antiferroelectricity in thin-film ZrO2 from first principles. Physical Review B: Condensed Matter, 90, 140103/1- (2014). 80. Min Hyuk Park, Han Joon Kim, Yu Jin Kim, Taehwan Moon and Cheol Seong Hwang.

The effects of crystallographic orientation and strain of thin Hf0.5Zr0.5O2 film on its ferroelectricity. Applied Physics Letters, 104, 072901 (5 pp.)-07290 (2014). 81. M. H. Park, H. J. Kim, Y. J. Kim, W. Lee, T. Moon, K. D. Kim and C. S. Hwang.

Study on the degradation mechanism of the ferroelectric properties of thin Hf0.5Zr0.5O2 films on TiN and Ir electrodes. Applied Physics Letters, 105, 72902/1- (2014). 82. H. J. Kim, M. H. Park, Y. J. Kim, Y. H. Lee, W. Jeon, T. Gwon, T. Moon, K. Do Kim

and C. S. Hwang. Grain size engineering for ferroelectric Hf0.5Zr0.5O2 films by an

insertion of Al2O3 interlayer. Applied Physics Letters, 105, 192903/1- (2014). 83. T. Shimizu, T. Yokouchi, T. Oikawa, T. Shiraishi, T. Kiguchi, A. Akama, T. J. Konno, A. Gruverman and H. Funakubo. Contribution of oxygen vacancies to the ferroelectric behavior of Hf0.5Zr0.5O2 thin films. Applied Physics Letters, 106, 112904/1- (2015). Bibliography 121

84. Min Hyuk Park, Han Joon Kim, Yu Jin Kim, Young Hwan Lee, Taehwan Moon, Keum Do Kim, Seung Dam Hyun and Cheol Seong Hwang. Study on the size effect in

Hf0.5Zr0.5O2 films thinner than 8 nm before and after wake-up field cycling. Applied Physics Letters, 107, 192907 (2015). 85. Han Joon Kim, Min Hyuk Park, Yu Jin Kim, Young Hwan Lee, Taehwan Moon, Keum Do Kim, Seung Dam Hyuna and Cheol Seong Hwang. A study on the wake-up effect

of ferroelectric Hf0.5Zr0.5O2 films by pulse-switching measurement. Nanoscale, 8, 1383 (2016). 86. Min Hyuk Park, Han Joon Kim, Yu Jin Kim, Young Hwan Lee, Taehwan Moon, Keum Do Kim, Seung Dam Hyun, Franz Fengler, Uwe Schroeder and Cheol Seong Hwang.

Effect of Zr Content on the Wake-Up Effect in Hf1–xZrxO2 Films. ACS Applied Materials and Interfaces, 8 (24), 15466-15475 (2016). 87. M. H. Park, Y. H. Lee, H. J. Kim, Y. J. Kim, T. Moon, K. Do Kim, J. Mueller, A. Kersch, U. Schroeder, T. Mikolajick and C. S. Hwang. Ferroelectricity and

Antiferroelectricity of Doped Thin HfO2-Based Films. Advanced Materials, 27, 1811- 1831 (2015). 88. M. H. Park, H. J. Kim, Y. J. Kim, W. Lee, H. K. Kim and C. S. Hwang. Effect of

forming gas annealing on the ferroelectric properties of Hf0.5Zr0.5O2 thin films with and without Pt electrodes. Applied Physics Letters, 102, 112914/1- (2013). 89. P. LI, I. CHEN and J. PENNERHAHN. EFFECT OF DOPANTS ON ZIRCONIA STABILIZATION - AN X-RAY-ABSORPTION STUDY.1. TRIVALENT DOPANTS. Journal of the American Ceramic Society, 77, 118-128 (1994). 90. B. Scherrer, S. Heiroth, R. Hafner, J. Martynczuk, A. Bieberle-Hütter, J. L. M. Rupp and L. J. Gauckler. Crystallization and Microstructure of Yttria-Stabilized-Zirconia Thin Films Deposited by Spray Pyrolysis. Advanced Functional Materials, 21, 3967- 3975 (2011). 91. M. N. Tsampas, F. M. Sapountzi and P. Vernoux. Applications of yttria stabilized zirconia (YSZ) in catalysis. Catalysis Science & Technology, 5, 4884-4900 (2015). 92. B. S. Prakash, S. S. Kumar and S. T. Aruna. Properties and development of Ni/YSZ as an anode material in solid oxide fuel cell: A review. Renewable & Sustainable Energy Reviews, 36, 149-179 (2014). 122 Bibliography

93. S. Riedel, P. Polakowski and J. Müller. A thermally robust and thickness independent ferroelectric phase in laminated hafnium zirconium oxide. AIP Advances, 6, 095123 (2016). 94. S. Starschich, T. Schenk, U. Schroeder and U. Boettger. Ferroelectric and

piezoelectric properties of Hf1-xZrxO2 and pure ZrO2 films. Applied Physics Letters, 110, 182905 (2017). 95. P. Polakowski and J. Muller. Ferroelectricity in undoped hafnium oxide. Applied Physics Letters, 106, 232905 (5 pp.)-23290 (2015). 96. K. D. Kim, M. H. Park, H. J. Kim, Y. J. Kim, T. Moon, Y. H. Lee and S. D. Hyun.

Ferroelectricity in undoped-HfO2 thin films induced by deposition temperature control during atomic layer deposition. Journal of Materials Chemistry C, 4, 6864-6872 (2016). 97. J. Briscoe and S. Dunn. Piezoelectric nanogenerators - a review of nanostructured piezoelectric energy harvesters. Nano Energy, 14, 15-29 (2015). 98. M. A. Ahmad and R. Plana. Piezoelectric Coefficients of Thin Film Aluminum Nitride Characterizations Using Capacitance Measurements. IEEE microwave and wireless components Letters (2009). 99. A. Khan, Z. Abas, H. Sm Kim and K. Oh. Piezoelectric thin films: an integrated review of transducers and energy harvesting. Smart Materials and Structures (2016). 100. T. S. Boescke, S. Teichert, D. Braeuhaus, J. Mueller, U. Schroeder, U. Boettger and T. Mikolajick. Phase transitions in ferroelectric silicon doped hafnium oxide. Applied Physics Letters, 99, 112904 (2011). 101. M. H. Park, H. J. Kim, Y. H. Lee, Y. J. Kim, T. Moon, K. Do Kim, S. D. Hyun and C. S. Hwang. Two-step polarization switching mediated by a nonpolar intermediate

phase in Hf0.4Zr0.6O2 thin films. Nanoscale, 8, 13898-13907 (2016). 102. M. Ho, H. Gong, G. Wilk, B. Busch, M. Green, P. Voyles, D. Muller, M. Bude, W. Lin, A. See, M. Loomans, S. Lahiri and P. Raisanen. Morphology and crystallization kinetics in HfO(2) thin films grown by atomic layer deposition. Journal of Applied Physics, 93, 1477-1481 (2003). 103. K. Kukli, J. Aarik, M. Ritala, T. Uustare, T. Sajavaara, J. Lu, J. Sundqvist, A. Aidla, L. Pung, A. Harsta and M. Leskela. Effect of selected atomic layer deposition Bibliography 123

parameters on the structure and dielectric properties of hafnium oxide films. Journal of Applied Physics, 96, 5298-5307 (2004). 104. S. Ushakov, A. Navrotsky, Y. Yang, S. Stemmer, K. Kukli, M. Ritala, M. Leskela, P. Fejes, A. Demkov, C. Wang, B. Nguyen, D. Triyoso and P. Tobin. Crystallization in hafnia- and zirconia-based systems. Physica Status Solidi B-Basic Research, 241, 2268-2278 (2004). 105. D. H. Triyoso, P. J. Tobin, B. E. White, R. Gregory and X. D. Wang. Impact of film

properties of atomic layer deposited HfO2 resulting from annealing with a TiN capping layer. Applied Physics Letters, 89, 132903/1- (2006). 106. T. Schenk, S. Mueller, U. Schroeder, R. Materlik, A. Kersch, M. Popovici, C. Adelmann, S. Van Elshocht and T. Mikolajick. Strontium Doped Hafnium Oxide Thin Films: Wide Process Window for Ferroelectric Memories. ESSDERC (2013). 107. J. Muller, T. Boscke, S. Muller, E. Yurchuk, P. Polakowski, J. Paul, D. Martin, T. Schenk, K. Khullar, A. Kersch, W. Weinreich, S. Riedel, K. Seidel, A. Kumar, T. Arruda, S. Kalinin, T. Schlosser, R. Boschke, R. van Bentum, U. Schroder and T. Mikolajick. Ferroelectric hafnium oxide: A CMOS-compatible and highly scalable approach to future ferroelectric memories. Electron Devices Meeting (IEDM), 2013 IEEE International, Electron Devices Meeting (IEDM), 2013 IEEE International, 10.8.1-10.8.4 (2013). 108. T. Olsen, U. Schroeder, S. Mueller, A. Krause, D. Martin, A. Singh, J. Mueller, M. Geidel and T. Mikolajick. Co-sputtering yttrium into hafnium oxide thin films to produce ferroelectric properties. Applied Physics Letters, 101, 82905/1-4 (2012). 109. T. Shimizu, K. Katayama, T. Kiguchi, A. Akama, T. J. Konno and H. Funakubo.

Growth of epitaxial orthorhombic YO1.5-substituted HfO2 thin film. Applied Physics Letters, 107, 32910/1- (2015). 110. U. Schroeder, E. Yurchuk, J. Muller, D. Martin, T. Schenk, P. Polakowski, C. Adelmann, M. I. Popovici, S. V. Kalinin and T. Mikolajick. Impact of different dopants on the switching properties of ferroelectric hafniumoxide. Japanese Journal of Applied Physics, 53, 08LE02 (5 PP.)-08LE0 (2014). 111. S. Starschich and U. Boettger. An extensive study of the influence of dopants on the

ferroelectric properties of HfO2. Journal of Materials Chemistry C, 5, 333-338 (2017). 124 Bibliography

112. R. D. Shannon. Revised effective ionic radii and systematic studies of interatomic distances in halides and chalcogenides. Acta Crystallographica, Section A (Crystal Physics, Diffraction, Theoretical and General Crystallography), Denmark, A32, 751- 67 (1976). 113. P. Polakowski, S. Riedel, W. Weinreich, M. Rudolf, J. Sundqvist, K. Seidel and J. Mueller. Ferroelectric deep trench capacitors based on Al:HfO2 for 3D nonvolatile memory Applications. 2014 Ieee 6th International Memory Workshop (imw) (2014). 114. E. Yurchuk, J. Mueller, S. Knebel, J. Sundqvist, A. P. Graham, T. Melde, U. Schroeder and T. Mikolajick. Impact of layer thickness on the ferroelectric behaviour of silicon doped hafnium oxide thin films. Thin Solid Films, 533, 88-92 (2013). 115. Y. Yang, W. Zhu, T. Ma and S. Stemmer. High-temperature phase stability of hafnium aluminate films for alternative gate dielectrics. Journal of Applied Physics, 95, 3772- 3777 (2004). 116. T. S. Boescke, S. Govindarajan, P. D. Kirsch, P. Y. Hung, C. Krug, B. H. Lee, J. Heitmann, U. Schroeder, G. Pant, B. E. Gnade and W. H. Krautschneider. Stabilization of higher-kappa tetragonal HfO2 by SiO2 admixture enabling thermally stable metal- insulator-metal capacitors. Applied Physics Letters, 91, 72902/1-3 (2007). 117. S. A. Barolin, M. C. Caracoche, J. A. Martinez, P. C. Rivas, M. A. Taylor, A. F.

Pasquevich and O. A. de Sanctis. Thermal evolution of CaO-doped HfO2 films and powders. Xix Latin American Symposium On Solid State Physics (slafes), 167, 12052/1- (2009). 118. S. Ushakov, C. Brown and A. Navrotsky. Effect of La and Y on crystallization temperatures of hafnia and zirconia. Journal of Materials Research, 19, 693-696 (2004). 119. C. Lee, E. Cho, H. Lee, C. S. Hwang and S. Han. First-principles study on doping and

phase stability of HfO2. Physical Review B: Condensed Matter, 78, 012102 (2008). 120. S. Starschich, S. Menzel and U. Böttger. Evidence for oxygen vacancies movement during wake-up in ferroelectric hafnium oxide. Applied Physics Letters, 108, 032903/1- (2016). 121. S. Starschich, S. Menzel and U. Böttger. Pulse wake-up and breakdown investigation

of ferroelectric yttrium doped HfO2. Journal of Applied Physics, 121, 154102 (2017). Bibliography 125

122. D. Zhou, J. Xu, Q. Li, Y. Guan, F. Cao, X. Dong, J. Mueller, T. Schenk and U. Schroeder. Wake-up effects in Si-doped hafnium oxide ferroelectric thin films. Applied Physics Letters, 103, 192904/1- (2013). 123. T. Shimizu, K. Katayama, T. Kiguchi, A. Akama, T. J. Konno, O. Sakata and H.

Funakubo. The demonstration of significant ferroelectricity in epitaxial Y-doped HfO2 film. Scientific Reports, 6, 32931/1-8 (2016). 124. P. D. Lomenzo, Q. Takmeel, C. Zhou, C. M. Fancher, E. Lambers, N. G. Rudawski, J. L. Jones, S. Moghaddam and T. Nishida. TaN interface properties and electric field

cycling effects on ferroelectric Si-doped HfO2 thin films. Journal of Applied Physics, 117, 134105/1- (2015). 125. T. Schenk, M. Hoffmann, J. Ocker, M. Pešic, T. Mikolajick and U. Schroeder. Complex Internal Bias Fields in Ferroelectric Hafnium Oxide.. Acs Applied Materials & Interfaces, 7, 20224-33 (2015). 126. L. Zhao, S.-W. Ryu, A. Hazeghi, D. Duncan, B. Magyari-Köpe and Y. Nishi. Dopant Selection Rules for Extrinsic Tunability of HfOx RRAM Chracteristics: A Systematic Study. Proceedings of the 2013 Symposium on VLSI Technology (2013). 127. A. Marchewka, R. Waser and S. Menzel. Physical Simulation of Dynamic Resistive Switching in Metal Oxides Using a Schottky Contact Barrier Model. 2015 International Conference On Simulation of Semiconductor Processes and Devices (SISPAD), 9-11 September, Washington D.C, USA, 2015 International Conference On Simulation of Semiconductor Processes and Devices (SISPAD), 9-11 September, Washington D.C, USA, 297-300 (2015). 128. M. H. Park, H. J. Kim, Y. J. Kim, T. Moon, K. D. Kim, Y. H. Lee, S. D. Hyun and C. S. Hwang. Study on the internal field and conduction mechanism of atomic layer

deposited ferroelectric Hf0.5Zr0.5O2 thin films. Journal of Materials Chemistry C, 3, 6291-6300 (2015). 129. S. Zafar, H. Jagannathan, L. F. Edge and D. Gupta. Measurement of oxygen diffusion

in nanometer scale HfO2 gate dielectric films. Applied Physics Letters, 98, - (2011). 130. M. Pešic, F. P. G. Fengler, L. Larcher, A. Padovani, T. Schenk, E. D. Grimley, X. Sang, J. M. LeBeau, S. Slesazeck, U. Schroeder and T. Mikolajick. Physical Mechanisms behind the Field-Cycling Behavior of HfO2-Based Ferroelectric Capacitors. Advanced Functional Materials, 26, 4601-4612 (2016). 126 Bibliography

131. S. Yu, H.-Y. Chen, B. Gao, J. Kang and H.-S. P. Wong. HfOx-Based Vertical Resistive Switching Random Access Memory Suitable for Bit-Cost-Effective Three-Dimensional Cross-Point Architecture.. Acs Nano, 7, 2320-2325 (2013). 132. E. Yalon, I. Karpov, V. Karpov, I. Riess, D. Kalaev and D. Ritter. Detection of the insulating gap and conductive filament growth direction in resistive memories. Nanoscale, 7, 15434-15441 (2015). 133. G. Bersuker, J.Yum, L. Vandelli, A. Padovani, L. Larcher, V. Iglesias, M. Porti, M. Nafria, K. McKenna, A. Shluger, P. Kirsch and R. Jammy. Grain boundary-driven leakage path formation in HfO2 dielectrics. Solid-State Electronics, 65-66, 146-150 (2011). 134. I. Lubomirsky and O. Stafsudd. Invited Review Article: Practical guide for pyroelectric measurements. Review of Scientific Instruments, 83, 51101/1- (2012). 135. S. W. Smith, A. R. Kitahara, M. A. Rodriguez, M. D. Henry, M. T. Brumbach and J.

F. Ihlefeld. Pyroelectric response in crystalline hafnium zirconium oxide (Hf1-xZrxO2) thin films. Applied Physics Letters, 072901 (2017). 136. S. S. Todd. Heat Capacities at Low Temperatures and Entropies at 298.16°K. of Hafnium Dioxide and Hafnium Tetrachloride. Journal of the American Chemical Society, 75, 3035-3036 (1953).

Publication List

S. Starschich, D. Griesche, T. Schneller, R. Waser and U. Boettger. Chemical solution deposition of ferroelectric yttrium-doped hafnium oxide films on platinum electrodes. Applied Physics Letters, 104, 202903/1- (2014).

S. Starschich, D. Griesche, T. Schneller and U. Böttger. Chemical Solution Deposition of Ferroelectric Hafnium Oxide for Future Lead Free Ferroelectric Devices. ECS Journal of Solid State Science and Technology, 4, 419-423 (2015).

S. Starschich, S. Menzel and U. Böttger. Evidence for oxygen vacancies movement during wake-up in ferroelectric hafnium oxide. Applied Physics Letters, 108, 032903/1- (2016).

S. Starschich and U. Boettger. An extensive study of the influence of dopants on the ferroelectric properties of HfO2. Journal of Materials Chemistry C, 5, 333-338 (2017).

S. Starschich, S. Menzel and U. Böttger. Pulse wake-up and breakdown investigation of ferroelectric yttrium doped HfO2. Journal of Applied Physics, 121, 154102 (2017).

S. Starschich, T. Schenk, U. Schroeder and U. Boettger. Ferroelectric and piezoelectric properties of

Hf1-xZrxO2 and pure ZrO2 films. Applied Physics Letters, 110, 182905 (2017).

T. Schenk, E. Yurchuk, S. Mueller, U. Schroeder, S. Starschich, U. Bottger and T. Mikolajick. About the deformation of ferroelectric hystereses. Applied Physics Reviews, 1, 041103 (14 pp.)-0411 (2014).

M. Hoffmann, U. Schroeder, C. Künneth, A. Kersch, S. Starschich, U. Böttger and T. Mikolajick.

Ferroelectric Phase Transitions in Nanoscale HfO2 Films Enable Giant Pyroelectric Energy Conversion and Highly Efficient Supercapacitors. Nano Energy, 154–164 (2015).

S. Nordmann, B. Berghoff, A. Hessel, N. Wilck, B. Osullivan, M. Debucquoy, J. John b, S. Starschich and J. Knoch. A monolithic all-silicon multi-junction solar device for direct water splitting. Renewable Energy (2016).

C. La Torre, K. Fleck, S. Starschich, E. Linn, R. Waser and S. Menzel. Dependence of the SET switching variability on the initial state in HfOx-based ReRAM. Physica Status Solidi A, 213, 316-319 (2016).

F. P. G. Fengler, M. Pešic´, S. Starschich, T. Schneller, C. Künneth, U. Böttger, H. Mulaosmanovic, T. Schenk, M. H. Park, R. Nigon, P. Muralt, T. Mikolajick and U. Schroeder. Domain Pinning: Comparison of Hafnia and PZT Based Ferroelectrics. Advanced Elekectronic Materials, 3 (2017).