There is a growing desire for high-temperature (HT) electronics operating over 300 C for applications such as automotive, aeronautics, geothermal exploration, and space exploration. 1 Non-volatile memory (NVM) provides energy-efficient and extended-duration data storage of media and program instructions and remains elusive at these elevated temperatures. NVM technologies such as complementary metal-oxide-semiconductor field-effect transistors (MOSFETs) and memories such as flash and electrically erasable programable read-only memory (EEPROM) are hindered above 250 C due to degraded data retention and cycle endurance. 2,3 Memories based on ferroelectric oxide, magnetoresistive, phase-change, and resistive mechanisms are limited by phase transitions, increased leakage currents, and reduced signal margin.
SiC-based transistors have shown robust operation at 500 C for >1000 h in air with engineered electrode layers, 4,5 which breakthrough the operation temperature limitation of conventional Si-based devices (<150 C). This is due to the wide bandgap (3.26 eV) of SiC leading to low intrinsic carrier concentrations at elevated temperatures as well as the engineering of stable device layers to avoid diffusion and delamination. 4H-SiC provides a relevant platform on which to integrate ferroelectric nitrides due to the temperature stability, compatible processing techniques, and higher voltage operation than Si devices. 5,6 Ferroelectric materials feature an electrically switchable spontaneous dipole. This characteristic is utilized in NVM applications since the direction of the dipole (i.e., spontaneous polarization) can be controlled for read and write operations. Current commercial ferroelectric random access memory (FRAM) architectures are based on Si CMOS and fundamentally require a transistor, a ferroelectric capacitor, and electrical interconnects for each storage cell. 7 FRAM traditionally uses Pb(Zr,Ti)O 3 and is gradually adopting (Hf,Zr)O 2 as well, but the polarization retention loss is accelerated as temperature increases, thereby limiting operating temperatures to <200 C for extended retention requirements. 8 In wurtzite ferroelectrics, the polar direction is parallel to the c-axis, meaning a film with a prominent c-axis orientation is beneficial for leveraging ferroelectric behavior. AlN-based films preferentially grow c-axis textured because of the strongly anisotropic surface energies, but improved texture quality typically translates to improved ferroelectric and piezoelectric performance, so enormous efforts have gone into fine-tuning the c-axis texture of AlN and Al 1Àx Sc x N films. 9,10 Substrate characteristics (e.g., structure, orientation, and surface roughness) and deposition conditions (e.g., temperature, gas ratio, throw distance, and pressure) are the primary methods for tailoring the c-axis orientation. With regard to the temperature stability of the polar wurtzite structure, AlN is predicted to retain a polar structure of . 2000 C, which provides an enormous margin between the intended operating temperature and phase transition. 11 The addition of Sc or B alloying elements into wurtzite AlN decreases the coercive field (E c ) below the breakdown strength (E BD ), which enables observation of ferroelectric behavior. 12,13 Another way to tune the E c =E BD is by increasing the temperature as shown in previous reports. [14][15][16] Ferroelectric switching in 100 nm thick Al 0:72 Sc 0:28 N films has even been reported up to 600 C. 17 Pradhan et al. recently reported using Al 1Àx Sc x N as a ferroelectric diode for extreme temperatures, which is promising for developing a functional non-volatile memory based on wurtzite nitride ferroelectrics. 18 Since there is a range of compositions that can be used for HT applications, we focus on both Al 0:7 Sc 0:3 N and Al 0:94 B 0:06 N to verify that both systems are compatible.
There are reports on integrating epitaxial nitride ferroelectrics with materials such as W/Al 2 O 3 , Mo/Sc 2 O 3 /Si, GaN, and Pt/GaN; however, the compatibility and complexity for these configurations with current processes focused on HT electronics complicates the fabrication. [19][20][21][22] Designing for Venus, which is an extremely harsh environment, would benefit from Mo to mitigate potential deleterious reactions from occurring. 23 Additionally, Mo can be wet etched with aqua regia, which shows a high selectivity of Mo over AlN. 24 This simplifies the processing and enables a quick route to further integration of ferroelectric nitrides for HT electronics.
To address this lack of a process-compatible NVM capable of operating in these HT environments, here, we demonstrate an integration of ferroelectric nitrides (Al 1Àx Sc x N and Al 1Ày B y N) with a known, high-temperature capable platform: SiC substrate with Mo electrodes.
A single Al 0:7 Sc 0:3 alloy target was reactively sputtered to grow the 160 nm Al 0:7 Sc 0:3 N film, and the 400 nm Al 0:94 B 0:06 N film was grown from a reactive co-sputter process using Al and BN targets as detailed in Sec. IV. Several x-ray diffraction (XRD) techniques were implemented to develop an understanding of the crystallographic relationships for the Al 0:7 Sc 0:3 N/Mo/SiC and Al 0:94 B 0:06 N/Mo/SiC samples (Fig. 1). The XRD θ À 2θ scans reveal the growth of a purely c-axis out-of-plane orientation depicted by the (0002) reflection of the nitride film on the body center cubic (BCC) Mo (110) [Fig. 1(a)]. Respectively, the Al 0:7 Sc 0:3 N and Al 0:94 B 0:06 N (0002) peaks at 35.759 and 36.178 show a distinct difference in the out-of-plane lattice parameters, and the lattice constant values for each layer are summarized in Table I, where measured 4H-SiC values closely match the reported values. 25 The c-axis and a-axis constants for the Al 0:7 Sc 0:3 N unit cell are larger than the Al 0:94 B 0:06 N, likely due to the larger Sc atoms alloyed into the structure than B. The calculated u parameter, which represents the ratio of the length of the metal-nitrogen bond parallel to the c-axis and the c-axis lattice parameter, provides quantification of how the sample structure compares to an ideal wurtzite (u ¼ 0:375) or layered hexagonal (u ¼ 0:5) structure. Since the Al 0:7 Sc 0:3 N sample has a higher u parameter than the Al 0:94 B 0:06 N sample, the polarization is expected to be higher for Al 0:94 B 0:06 N. Faint diffraction fringes on the Mo (110) peak indicate abrupt interfaces of the Mo layer.
The ω-rocking curves in Fig. 1(b) of the Al 0:7 Sc 0:3 N (0002) and Al 0:94 B 0:06 N (0002) indices are similar at 1.1 and 1.3 , respectively, and we do not expect this slight difference to significantly impact the electrical properties. The measurement quantifies how well-aligned the c-axis direction is oriented out of plane (OOP) from the substrate, which influences the ferroelectric and piezoelectric properties for AlN-based materials. The Mo(110) peak exhibits an extremely sharp omega rocking curve of 0.011 . While promising, this simply indicates that the Mo(110) planes of the bottom electrode are highly aligned out of plane; it does not provide direct knowledge of the in-plane orientation or any surface features relevant to growing a wurtzite nitride on top of the Mo.
Sixfold symmetry is observed in the pole figures of the 4H-SiC (10 110), Al 0:94 B 0:06 N (10 15), and Mo (200
The diffraction peaks for the observed reflections occur at the same f value, meaning that an in-plane orientation relationship exists between the substrate, bottom electrode, and ferroelectric layers, resulting in a [10 10] wurtzite//[100]Mo//[10 10] 4H-SiC epitaxial relationship. The Mo bottom electrode grew with a strongly preferred (110) OOP orientation. A single crystal BCC structure with (110) OOP would result in a twofold symmetry for an off-axis (200). However, since six peaks occur in Fig. 1(g), Mo grew with three distinct in-plane orientations on the 4H-SiC substrate. We expect that this twinning is limited by the independent nuclei formation and lattice mismatch. It is clear that the hexagonal 4H-SiC underlayer is responsible for providing the structural sixfold template. Therefore, the in-plane relationship between a twinned, in-plane-oriented electrode and the subsequently grown wurtzite would be limited to at least the same constraints as the 4H-SiC/Mo. The sixfold symmetry of Al 0:94 B 0:06 N (10 15) in Fig. 1(f ) along with the grain morphology in Fig. 2(b) suggests a lack of single-crystal wurtzite. Hence, the wurtzite is crystallographically aligned with the substrate but with significant structural defects due to the Mo BE interlayer. I. The c-axis and a-axis lattice parameters for the Al 0:7 Sc 0:3 N unit cell are larger than those of their Al 0:94 B 0:06 N counterparts, consistent with the larger size of Sc atoms than either Al or B. For visualization purposes, the characterized surfaces for the crystal structure of each layer in these heterostructures are provided in Fig. 1(h). The prepared 4H-SiC surface, compatible symmetry, and relatively low surface energy of the Mo (110) plane provide a robust and reliable pathway for growing pristine bottom electrode Mo films on a bare substrate, and these highly textured electrode layers in turn lead to highly textured wurtzite nitride layers.
Figure 2 reveals the surface morphology of the deposited layers. The 100 nm thick Mo bottom electrode exhibits a Root Mean Square (RMS) value of 0.338 nm as supported by the faint thickness fringes appearing on the Mo(110) peak in Fig. 1(a). The fast Fourier transform of the Mo SEM micrograph is shown in the inset of Fig. 2(a) and reveals a hexagonal shape in reciprocal space based on the Mo grain structure. This sixfold FFT structure is a result of the grain
TABLE I. Unit cell parameters for the ferroelectric heterostructures. Parameter (Å) Al 0.7 Sc 0.3 N Al 0.94 B 0.06 N Mo 4H-SiC a 3.169 3.093 3.146 3.082 c 5.018 4.962 3.149 10.083 u 0.383 0.379 … … Unit cell vol. 41.713 41.103 31.175 82.964 Journal of Applied Physics . 27 The lateral grain diameters at the film surface of the 160 nm thick Al 0:7 Sc 0:3 N film range between 35+11 nm, whereas grains of the 400 nm thick Al 0:94 B 0:06 N film are about 3X wider (133+24 nm).
The compositions of the Al 0:7 Sc 0:3 N and Al 0:94 B 0:06 N films were measured using x-ray photoelectron spectroscopy (XPS) to investigate changes in the chemical states and quantify the atomic concentrations. Survey spectra collected after the removal of surface contaminants (C and O) are presented in Fig. 3(a) along with the identified electronic transitions for each element. Prior to the initial Ar etch, both samples exhibit a surface oxide, which is removed after the first 3 min etch [Figs. 3(b) and 3(c)]. Assuming a uniform etch rate through the film, the surface oxide is 7.5 nm thick. In both samples, oxygen is present in the bulk of the film even after the removal of surface contamination with an average value of 4.1 and 5.7 at.% in the Al 0:7 Sc 0:3 N and Al 0:94 B 0:06 N films, respectively. The oxygen signals are significant and above the measurement noise floor throughout both nitride film samples, whereas oxygen is undetectable in the Mo and SiC layers of the depth profiles. Although oxygen is typically considered undesirable for electrical properties, these results and recent reports point toward reducing leakage currents and reducing switching barriers compared to oxygen-free AlScN. 28,29 While an Al 0:7 Sc 0:3 alloy target was used for depositing the nominally Al 0:7 Sc 0:3 N film, XPS results yield an average value of x ¼ 0:15. This discrepancy may be an effect of preferentially sputtering Sc over Al atoms during the depth profile as is prevalent for this technique in many compounds. [30][31][32] Although this technique may provide uncertainty in the absolute quantification of the Al 0:7 Sc 0:3 N film composition, it is clear that the Sc/Al ratio is consistent throughout the film thickness. Energy dispersive spectrometry (EDS) measurements in SEM result in x ¼ 0:26 (Al 0:74 Sc 0:26 N) (Fig. S2 in the supplementary material). Furthermore, a 4.8 at.% O content was detected with EDS, which is similar to the XPS findings. For Al 0:94 B 0:06 N, XPS detected an average y ¼ 0:06 through the film, which is on par with previous reports focused on optimizing the composition for reliable hightemperature ferroelectric properties. 15,33 Electrical measurements were performed on 50 μm diameter top Mo electrodes on both samples at 400 C (additional details provided in Sec. IV). The hysteresis loops for each film exhibit values similar to previous reports for polarization and coercive field (E c ). 14,33 Nested hysteresis P-E loops are scaled with increasing maximum electric field of the applied triangular waveform, which was either conducted at 2 kHz for Al 0:94 B 0:06 N or 10 kHz for Al 0:7 Sc 0:3 N [Figs. 4(a) and 4(b)]. The Al 0:94 B 0:06 N sample polarization saturates at 150 μC cm À2 , and the Al 0:7 Sc 0:3 N sample saturates at 124 μC cm À2 . The values for Al 0:94 B 0:06 N are slightly inflated due to leakage contribution during the hysteresis measurements, which is primarily indicated by the negative slope of the hysteresis loop after saturation. Nevertheless, the P r for Al 0:94 B 0:06 N is higher than Al 0:7 Sc 0:3 N, which is also supported by the u parameter as given in Table I. The þE c for Al 0:7 Sc 0:3 N at 400 C is 3.48 MV cm À1 , whereas the ÀE c is À3.01 MV cm À1 . Al 0:94 B 0:06 N exhibits a þE c of 4.03 MV cm À1 and ÀE c of À3.47 MV cm À1 . This asymmetry may be due to multiple phenomena (e.g., stress, wake-up, imprint, and dissimilar dynamics), [34][35][36] and identifying the mechanism(s) is outside of the scope of this report. Although these films appear to contain roughly 5 at.% oxygen, the average E c for Al 0:7 Sc 0:3 N and Al 0:94 B 0:06 N closely follows the thermally activated switching process after accounting for the different excitation frequencies. 15 The influence of oxygen on the ferroelectric behavior in nitride wurtzite materials is not well-established, but in general, oxygen is accepted to diminish the performance since previous studies on the piezoelectric properties are reduced while leakage increases with increasing oxygen content. 37,38 In contrast, these ferroelectric results show saturated polarization reversal at 400 C, a temperature with 10 4 higher leakage contribution than room temperature. Figure 4(c) shows an example of Positive-Up-Negative-Down (PUND) pulse sequence with temporal resolution that was applied to the capacitors at 400 C. The PUND method provides information on the switching, capacitive, and leakage current contribution to polarization for each voltage step. The current density trace shows strong peaks for ferroelectric switching events ("P" and "N" pulses) near the onset of the voltage pulse maximum, and the current density post-switching reveals a finite amount of charge leakage collected at the pulse voltage. For non-switching pulses ("U" and "D"), a finite amount of charge leakage occurs again, which closely matches the leakage in the "P" and "N" pulses. P* (-P*) describes the total polarization during the "P" ("N") pulse (integrated from the sensed current) immediately prior to the pulse voltage returns to zero. Similarly but for a non-switching "U" ("D") pulse, P^represents the polarization collected. The difference in polarization (dP) between the same bias pulses ideally results in a pure switched polarization. The evolution of polarization through this PUND sequence is also plotted in 4(c) to illustrate the changes in P due to various mechanisms: ferroelectric switching, linear dielectric response, and leakage. Here, the pulse width was set to 10 μs and the voltage was increased incrementally.
From the nested PUND sequences with increasing voltage, the polarization vs electric field profile can be constructed from the pulsed measurements for the Al 0:94 B 0:06 N and Al 0:7 Sc 0:3 N samples [Figs. 4(d) and 4(e)]. For E , 2:5 MV cm À1 , there is a lack of switching and relatively low amounts of leakage during the pulse at 400 C. As E increases, there is an increase in switched P as tracked by the PÃ traces. The average slope between P ¼ 0 and P sat is 3X higher for Al 0:7 Sc 0:3 N than for Al 0:94 B 0:06 N, which suggests a broader distribution of activation energies for switching Al 0:94 B 0:06 N. The relatively large increase in the P^trace for Al 0:7 Sc 0:3 N is due to incomplete switching during the P pulse, allowing additional ferroelectric switching to occur on the U pulse. Further insight into this partial switching phenomenon is provided in Fig. S3 in the supplementary material, which reveals more switched polarization during the U pulse than P pulse based on the area under the J vs t curve. This behavior is not observed for the N and D pulses both because of the switching asymmetry shown in the hysteresis loops and the fact that the preset pulse affects the state of the capacitor prior to each PUND cycle. Lastly, the saturated 2P value for Al 0:94 B 0:06 N is 252 μC cm À2 compared to 237 μC cm À2 for Al 0:7 Sc 0:3 N. This distinction is a feature of the lattice cell u parameter, which determines the magnitude of the spontaneous dipole in wurtzite structures.
In summary, epitaxial ferroelectric AlMN (M = Sc or B) have been grown on (110) oriented Mo electrodes on 4H-SiC substrates with <1 nm RMS roughness, grain diameters of 133 nm for AlBN compared to 35 nm for AlScN, and nearly 5 at.% O through the film bulk. An in-plane alignment exists between each of the deposited layers and the substrates following a [10 10] Al 1Àx M x N (M=Sc or B)// [001] Mo // [10 10] SiC relationship. Ferroelectric measurements at 400 C show well-saturated P-E hysteresis and PUND response with 2P r 250 μC cm À2 .
Integrating AlN-based ferroelectric materials with SiC platforms can introduce new capabilities into high temperature electronics, and these results demonstrate a route to growing epitaxially aligned films with low leakage current during polarization switching, even at elevated temperatures. Accessing this level of structural quality ensures routes to controlling the microstructure across the fiber-textured to epitaxial spectrum depending on the bottom electrode quality. However, the growth of epitaxial ferroelectric Al 0:7 Sc 0:3 N and Al 0:94 B 0:06 N films is not expected to be the only feature that governs the leakage current; oxygen content is also assumed to play an important yet poorly understood role in the conduction processes of AlN-based films. Understanding the limits of oxygen solubility and the effects of this species on the electrical performance of ferroelectric nitride materials is important to control the growth process and performance geared toward hightemperature applications.
Prior to film deposition, the 4H-SiC substrate was cleaned in an ultrasonic acetone bath and then rinsed with methanol and isopropanol. The 100 nm thick Mo bottom electrode was sputter deposited after reaching a base pressure of ,5 Â 10 À7 Torr. Following this, the ferroelectric wurtzite nitride layer was deposited onto the Mo/SiC substrate without breaking vacuum. The sputter parameters are provided in Table II. After blanket deposition of the first two layers (Mo and Al 1Àx M x N), the 100 nm thick Mo top electrodes were lithographically patterned to create discrete capacitors for electrical measurements. The top electrode deposition condition followed the bottom electrode conditions apart from no intentional substrate heating.
XRD scans (2θ À ω, ω rocking curves, f, and RSMs) were performed with a Panalytical MRD 3XL XRD instrument with a Cu Kα1 (1.540 598 Å) source and a PIXcel 3 D detector. A 1/32 divergence slit was used for 2θ À ω and ω rocking curve scans with a line detector (1.913 mm active length) and a point detector (0.055 mm active length), respectively. For f and RSMs, a 1/2 divergence slit and open detector (14.025 mm active length) and 1D frame, respectively.
TABLE II. Deposition parameters for the Al 0.7 Sc 0.3 N/Mo and Al 0.94 Sc 0.06 /Mo heterostructures sputtered on 4H-SiC substrates. The pulsed DC (pDC) was set to 100 kHz and 1.5 μs pulses for the Al 0.7 Sc 0.3 and Al targets. Parameter Mo electrode Al 0.7 Sc 0.3 N Al 0.7 Sc 0.3 N Pressure (mTorr) 2 2 2 Temperature (°C) 500 300 300 Gas flow (SCCM) Ar (40) N 2 (20), Ar (20) N 2 (20), Ar (20) Throw distance (cm) 10 10 10 4 in. Target Mo Al 0.7 Sc 0.3 Al + BN Power supply DC pDC pDC + RF Power (W) 600 600 600 + 200
A Bruker ICON atomic force microscope (AFM) was used in contact mode to investigate the surface morphology and roughness of the deposited layers. Areas of 1 μm 2 were scanned in ambient conditions. A Zeiss AURIGA 60 field emission scanning electron microscope (SEM) instrument was operated at 1 kV with an in-lens detector to study the Mo surface microstructure.
For electrical characterization at 400 C, the samples were placed in a Lakeshore high-temperature vacuum probe station. A Radiant PiezoMEMs Analyzer was used for performing P-E loops and PUND pulse sequences where the drive and sense were connected to the bottom electrode and top electrode, respectively.
Composition analysis was carried out with x-ray photoelectron spectroscopy (XPS) and energy dispersive x-ray spectroscopy (EDS) techniques. A VersaProbe III by Physical Electronics Inc. (PHI) XPS instrument with a monochromated Al Kα source provided a 100 μm diameter, 25 W x-ray beam. Survey spectra and higher resolution elemental scans were collected with a 224 and 55 eV pass energy, respectively, at a take-off angle of 45 . An electron neutralizer and low-energy Ar ion neutralizer were used to mitigate surface charging. Spectra were analyzed with MultipPak 9.9.0.8 (PHI), and peaks were fit with a Gauss-Lorentz peak shape and Smart background. To prevent extraneous oxidation of the sample surface, the O 1s region was scanned immediately following the etch process (99.9995% Ar) first. Supporting composition analysis was performed with an EDAX Octane Elect Super energy dispersive spectroscopy (EDS) system at 8 kV to validate the large presence of oxygen in the nitride film and verify the Al/Sc ratio. The collected spectra were quantified in TeamEDS (EDAX) software with eZAF corrections.
See the supplementary material for additional figures of the XRD scans on the AlScN/Mo/SiC structure, a discussion on the EDS scan of the AlScN film, and further explanation of the ferroelectric PUND pulse tests.
the interface originating c-axis instability in highly Sc enhanced AlN for piezoelectric micro-electromechanical systems," J. Appl. Phys. 122, 035301 (2017).
J. Appl. Phys. 137, 204102 (2025); doi: 10.1063/5.0256146
© Author(s) 2025