Unveiling interfacial dead layer in wurtzite ferroelectrics

Abstract

Wurtzite ferroelectrics hold immense promise to revolutionize modern micro- and nano-electronics due to their compatibility with semiconductor technologies. However, the presence of interfacial dead layers with irreversible polarization limits their development and applications, and the formation mechanisms of dead layers remain unclear. Here, we demonstrate that dead layer formation in ScAlN, a representative wurtzite ferroelectric, originates from a high density of nitrogen vacancies in combination with interfacial strain. Atomic-scale investigations using scanning transmission electron microscopy (STEM) and electron energy loss spectroscopy (EELS), supported by first-principles calculations, reveal that compressive strain near the ScAlN/GaN interface reduces the formation energy of nitrogen vacancies, promoting their generation. These vacancies degrade dielectric properties and raise the ferroelectric switching barrier, the latter further exacerbated by compressive strain. These combined effects suppress polarization reversibility near the interface. This work elucidates the microscopic origin of interfacial dead layers and highlights the significance of defect and strain engineering in wurtzite ferroelectrics, which are essential to advancing their integration and scalability in next-generation electronic devices.

The authors reveal that interfacial dead layers in ferroelectric ScAlN originate from nitrogen vacancies and strain, highlighting their critical role in unlocking the full potential of nitride ferroelectrics for advanced device architectures.

Introduction

The discovery of ferroelectricity in wurtzite semiconductors, such as ScAlN, ScGaN, YAlN, and AlBN, has sparked tremendous interest in recent years, particularly for applications spanning electronics, piezoelectronics, acoustics, and photonics^{1, 2–3}. Those wurtzite ferroelectrics boast a unique combination of enhanced piezoelectricity, a robust ferroelectric phase, tunable coercive fields, substantial remanent polarization, excellent potential for thickness scaling, and high Curie temperatures^{4, 5, 6, 7, 8–9}. Additionally, they can be grown by using techniques like sputtering deposition^10,11, molecular beam epitaxy (MBE)^12,13, metal-organic chemical vapor deposition (MOCVD)^14,15, and pulsed laser deposition (PLD)¹⁶, ensuring seamlessly compatible with Si and GaN-based semiconductor technologies, as well as complementary metal-oxide-semiconductor (CMOS) back-end-of-line (BEOL) processes. These characteristics position wurtzite ferroelectrics as powerful contenders for next-generation micro- and nano-electronics, with applications ranging from field-effect transistors and high-frequency resonators to non-volatile memories, high-density storage, and high-speed computing architectures^{8,17, 18, 19, 20, 21, 22, 23–24}.

Despite significant advances in material growth and device fabrication, the development of wurtzite ferroelectrics remains challenged by issues such as difficulties in thickness scaling^25,26, field cycling fatigue^1,27, and pronounced electric leakage²⁸. Similar challenges have been encountered in hafnia-based ferroelectrics, where they are often attributed to the formation of an interfacial dead layer with unswitchable polarization, leading to collapse in the polarizability of ferroelectric capacitors^{29, 30–31}. In ferroelectric ScAlN, this phenomenon has been further implicated in severe fatigue, characterized by a marked degradation in remnant polarization following field cycling. Interfacial dead layers are likely the main reason for the limited endurance and reliability in wurtzite ferroelectrics. Their formation was empirically linked to defects and interfacial strain^{27,32, 33–34}, however, experimental evidence for these mechanisms is still absent.

In this study, the mechanisms of interfacial dead layer formation in wurtzite nitride ferroelectrics are clarified using scanning transmission electron microscopy (STEM) and electron energy loss spectroscopy (EELS). Through atomic-scale investigation of single-crystalline ferroelectric Sc_0.18Al_0.82N (ScAlN), we identify the emergence of a dead layer near the ScAlN/GaN interface, arising from a high density of nitrogen vacancies (V_N) and interfacial compressive strain. These vacancies are inherent to the as-deposited material and not a consequence of subsequent electrical poling. First-principles studies reveal that the compressive strain in ScAlN near the interface lowers the formation energy of V_N, facilitating their generation. The resulting increase in V_N introduce significant defect states in the bandgap and elevates the ferroelectric switching barrier—an effect further intensified by compressive strain. Consequently, leakage current increases, breakdown strength diminishes, and polarization reversibility is severely suppressed, ultimately leading to the formation of an interfacial dead layer. Our findings highlight the critical importance of defect and strain management in wurtzite ferroelectrics to unlock their full potential in advanced electronic, optoelectronic, photonic, and acoustic devices.

Results

Ferroelectric ScAlN/GaN heterostructure

As illustrated in Fig. 1a, a 70-nm-thick Sc_0.18Al_0.82N film was grown on 2-inch GaN/sapphire templates by MBE. The ScAlN film exhibited an atomically smooth surface and a single crystalline c-axis texture, with a full width at half maximum (FWHM) of 320 arcsec for the (0002) plane X-ray diffraction (XRD) rocking curve (Fig. S1). This crystallinity is comparable to that of state-of-the-art epitaxial ScAlN films^15,35. Figure 1b presents a cross-sectional high-angle annular dark field STEM (HAADF-STEM) image of the ScAlN/GaN heterostructure, revealing an abrupt and clean interface. Unless otherwise specified, all STEM images in this study were captured along the $⟨11\overline{2}0⟩$ zone-axis, with the film normal aligned to the <0001> direction. The uniform incorporation of Sc into the AlN matrix is confirmed by energy dispersive X-ray spectroscopy (EDS) elemental mapping for Ga, Sc, and Al, as shown in the insets of Fig. 1b. The EDS measurements showed that Sc content is 18%. Left of Fig. 1c shows an atomically resolved HAADF-STEM image of the ScAlN/GaN heterointerface, demonstrating an atomically sharp interface and a highly ordered wurtzite stacking sequence. The corresponding annular bright field STEM (ABF-STEM) image illustrates that the ScAlN was grown along the [0001] axis on GaN, possessing a uniform metal-polar (M-polar) lattice, as shown on the right of Fig. 1c.

[See PDF for image]

Fig. 1

Single crystalline ferroelectric ScAlN/GaN heterostructure.

a Schematic of the MBE-grown ScAlN/GaN heterostructure, along with the electrical configuration used for ferroelectric measurements. b Cross-sectional HAADF-STEM image of the ScAlN/GaN heterostructure. Insets show corresponding EDS elemental maps for Ga, Sc, and Al, respectively. c Atomic-resolution HAADF-STEM image (left) and ABF-STEM image (bottom right) of the ScAlN/GaN interface growing along the [0001] direction. The inset (top right) displays the atomic model for [0001]-orientated wurtzite M-polar III-nitride lattice. Metal (M, yellow) and nitrogen (N, light blue) atoms are embedded in the ABF-STEM image to highlight the inherited M-polar lattice of ScAlN from underlying GaN. dP-E and corresponding J-E hysteresis loops, recorded at 1 kHz, from a typical ScAlN capacitor shown in (a) confirming ferroelectricity.

Ferroelectric polarization switching in the ScAlN/GaN heterostructure was investigated through electrical measurements on fabricated metal-ferroelectric-semiconductor (MFS) capacitors (Fig. 1a). In all electrical measurements conducted in this study, the voltage pulses were driven from the top electrodes. Figure 1d shows the characteristic polarization versus electric field (P-E) hysteresis loop, along with the corresponding current density versus electric field (J-E) loop, recorded from a ScAlN MFS capacitor. Reversable polarization and discernible displacement currents at high electric fields unequivocally demonstrate ferroelectric polarization switching in the ScAlN, with a remanent polarization of about 100 μC/cm² and an average coercive field of 4.5 MV/cm. The asymmetry observed between forward and backward switching currents is attributed primarily to the asymmetric metal-semiconductor electrode configuration^15,36. Under fatigue testing, the remanent polarization remained stable up to 10⁵ cycles, followed by a gradual decline and eventual breakdown around 10⁷ cycles (Fig. S2).

Atomic displacement and interfacial dead layer

Figure 2a shows a standard positive-up-negative-down (PUND) measurement performed on a fresh ferroelectric ScAlN capacitor. Distinct displacement currents were observed without any pre-cycling, indicating a wake-up-free nature for this ferroelectric ScAlN. Additionally, the first displacement current was observed following the first positive voltage pulse, rather than the initial negative pulse, indicating that the spontaneous polarization in the as-grown ScAlN film can only be reversed by a downward external electric field, as marked by arrows in Fig. 2a. This phenomenon suggests that the pristine M-polar ScAlN has an upward intrinsic spontaneous polarization, aligning with the recent understanding of spontaneous polarization in III-nitrides^{9,37, 38–39}. Similarly, the appearance of negative displacement current upon applying the ‘N’ pulse suggests a downward spontaneous polarization for the positively switched ScAlN film. Notably, only leakage current was observed during the ‘U’ and ‘D’ pulses, confirming the complete reversal of polarization after the ‘P’ and ‘N’ pulses.

[See PDF for image]

Fig. 2

Atomic displacement in ferroelectric ScAlN.

a Transient current-voltage profiles during a typical PUND measurement for a fresh ScAlN/GaN capacitor. Brown and blue arrows indicate the polarization (P) and external electric field (E) orientation at each stage. The upward and downward arrows represent orientations pointing to the ScAlN surface and the ScAlN/GaN interface, respectively, as indicated in the inset. b–e ABF-STEM images of fresh ScAlN/GaN capacitors after poling with various pules of PUND sequence, as indicated by solid blue circles in (a): overview ABF-STEM image of ‘P’ and ‘U’ pulses poled ScAlN (b), and atomic-resolution ABF-STEM image captured from the red square location in (b) for pristine (c), ‘P’ and ‘U’ pulses poled (d), and full ‘PUND’ pulses poled (e) ScAlN. Metal (yellow) and nitrogen (light blue) atoms are superimposed to highlight the atomic stacking sequence, showing M-polar, N-polar, and M-polar lattice for (c–e), respectively. The polarization orientations are indicated with brown arrows. The white dashed lines in (b) sketch the outline of a representative cone-like domain. f ABF-STEM image of a field-cycled ScAlN device, showing high density of cone-like domain patterns in the polarization-reversed region on the left, while the pristine film on the right exhibits a uniform diffraction contrast.

The high crystallinity of the MBE-grown ScAlN enables a systematic investigation of the atomic-scale structural evolution following electrical poling. To this end, we first examined the atomic structure of fresh ScAlN capacitors poled with partial PUND voltage sequences. In ferroelectric ScAlN films, it is energetically more favorable for inverted domains to nucleate near the top electrode during poling, with subsequent domain growth extending toward the ScAlN deposition interface at the bottom^15,32. Therefore, we firstly focused on characterizing the atomic structure of ScAlN at about 20 nm distance from the top electrode, the red squared region in Fig. 2b. ABF-STEM images were captured from three ScAlN capacitors: pristine ScAlN (Fig. 2c), positively switched ScAlN using ‘P’ and ‘U’ pulses for poling (Fig. 2d), and negatively switched-back ScAlN via full ‘PUND’ pulses poling (Fig. 2e), as indicated in Fig. 2a. Comparing the atomic stacking sequence with the atomic model shown in Fig. 1c, it is evident that the M-polar lattice in pristine ScAlN was uniformly reversed to a N-polar one in the positively switched ScAlN, while the negatively switched-back ScAlN fully reverted to the initial M-polar lattice. Following the polarization orientation discussed in the PUND measurements, the polarization direction was determined in the ABF-STEM images, indicated by brown arrows. The direct observation of reversible lattice-polarity at the unit cell scale provides compelling evidence for intrinsic ferroelectricity in epitaxially grown single crystalline ScAlN.

Subsequently, the ferroelectric switching behavior of a field-cycled (100×) ScAlN capacitor was further investigated at the atomic scale. Before preparing the TEM specimen, ‘P’ and ‘U’ voltage pulses were applied to reverse the lattice to N-polarity. Figure 2f shows a typical ABF-STEM image of the field-cycled ScAlN capacitor, revealing a distinct cone-like domain diffraction contrast in the ScAlN region beneath the top electrode (left side), which is absent in the pristine ScAlN region (right side). The outline of those cone-like domains is sketched in in Fig. 2b, and S3a, S3b. This observation suggests that a high density of grain boundaries formed after electrical cycling. In comparison, a few dark diffraction contrasts observed in the pristine region is attributed to pre-existing columnar structures and low-angle grain boundaries, which locally modulate the electron diffraction conditions⁴⁰.

Figure 3a illustrates the ABF-STEM image of the field-cycled ScAlN (the left part of Fig. 2f). Based on the diffraction contrast, the ScAlN film is divided into three distinct regions, labeled ‘I’, ‘II’, and ‘III’ in the accompanying schematic. Figure 3b shows a typical ABF-STEM image recorded in region I, exhibiting a uniform N-polar lattice that extends ~50 nm from the top electrode into the ScAlN bulk. This observation is consistent with the appearance of the N-polar lattice following unipolar positive poling in a fresh ScAlN capacitor (Fig. 2d), further confirming the robust ferroelectricity of wurtzite ScAlN. The lattice-polarity of ScAlN in the dark contrast areas within region I was also examined, revealing a well-aligned N-polar lattice (Fig. S3d). This indicates that those dark contrasts represent grain boundaries rather than inverted domain boundaries. Similar features have been observed in both sputtered and MOCVD-grown ferroelectric ScAlN films^15,41. In contrast, the ABF-STEM images captured from ScAlN/GaN interface and region III show a M-polar lattice, as shown in Fig. 3c, suggesting that the ScAlN in region III is not switched. This unswitchable layer is a feature of the dead layer commonly observed in ferroelectrics. The formation mechanism of this dead layer will be discussed later.

[See PDF for image]

Fig. 3

Interfacial dead layer in ferroelectric ScAlN.

a ABF-STEM image of a field-cycled ScAlN (left) and the corresponding schematic of cone-like domain structures (right). Based on diffraction contrast, the ScAlN film is divided into three regions: the top layer (~50 nm from the surface), the transition region (10 ~ 15 nm away from the ScAlN/GaN interface), and the near interface layer, labeled as ‘I’, ‘II’, and ‘III’, respectively. b–d Atomic-resolution ABF-STEM image of region I (top ScAlN layer, N-polar) (b) region III (bottom ScAlN layer, M-polar) (c) and region II (transition region, mixed-polar) (d). The polarization orientations are indicated with brown arrows. The white dashed lines in (d) sketches the outline of the lattice transition region. e iDPC-STEM image (top), atomic model (middle), and intensity profile of atom columns (bottom) for the side-by-side lattice-polarity transition shown in (d). f iDPC-STEM image (left), atomic model (middle), and intensity profile of atom columns (right) for the face-to-face lattice-polarity transition shown in (d). The intensity profiles of atom columns in (e) and (f) are recorded along the white dashed arrows. Following the atomic stacking sequence, each peak in the intensity profile is assigned to metal or nitrogen atom columns. Blue and red letters (‘M’ and ‘N’) indicate atoms belonging to the M-polar and N-polar unit cells, respectively, while the black ‘N’ represents indistinguishable nitrogen atoms in the transition regions. Metal atoms (yellow) and nitrogen atoms (light blue) are superimposed in (b–f) to highlight the atomic stacking sequence.

Figure 3d shows a typical atomic-resolution ABF-STEM image captured from region II, located between the top N-polar and bottom M-polar regions. This area shows a nonuniform lattice-polarity transition zone, where both side-by-side and face-to-face lattice-polarity transitions are easily observed, as indicated by polarization arrows. In addition, some inclined transition boundaries are detected, comprising atomic-scale stepped side-by-side and face-to-face transition regions (Fig. S3e).

Top of Fig. 3e displays a zoomed-in integrated differential phase contrast STEM (iDPC-STEM) image of a typical side-by-side lattice transition region, where the M-polar domain on the left gradually transformed across five crossed M-N dumbbells into an N-polar lattice on the right. And the doublet M-N dumbbell spots for the M-polar and N-polar lattice are replaced by triplet spots in the transition region. This results from the crossed alignment of the M-polar and N-polar lattices along the projection direction of the STEM image, specifically the $\left[11\overline{2}0\right]$ zone-axis. An atomic schematic illustrating the side-by-side domain transition is shown in the middle of Fig. 3e. It is obvious that the metal and nitrogen atomic planes switched positions along the c-axis across the transition region. The intensity profile of atom columns across the transition region was evaluated along the center of the nitrogen atoms in the left-side M-polar unit cells, as shown at the bottom of Fig. 3e. Clearly, the intensity of nitrogen atom column in the left M-polar unit cells (labeled as blue ‘N’) gradually decreases across the transition region and disappears in the right N-polar ones. Meanwhile, the intensity of metal atom column (labeled as red ‘M’) shows an opposite trend, ultimately dominating in the right N-polar ones. A similar evolution behavior was observed when recorded the intensity profile along the center of the metal atoms in the left-side M-polar unit cells (Fig. S4). These observations suggest that the transition region consists of two wedge-shaped domains with inverted lattice-polarity, which is similar as that reported by Wolff et al. in MOCVD-grown ScAlN¹⁵.

Left of Fig. 3f shows a high-resolution iDPC-STEM image for the face-to-face transition region, where the N-polar domain on top gradually transformed across six crossed M-N dumbbells into a M-polar lattice at the bottom. Similar to the side-by-side transition, the doublet M-N dumbbell spots on the top and at the bottom changed into triplet spots in the transition region, indicating a superposition of N-polar and M-polar unit cells. In the triplet spots, two metal atoms have a splitting distance of ~0.13 nm, while that for nitrogen atoms are indistinguishable. An atomic model for this face-to-face domain transition is illustrated in the middle of Fig. 3f. The intensity profile of atom columns across the transition region was evaluated along the center of one vertical atom chain, as depicted on the right of Fig. 3f. It is obvious that the intensity of metal atom columns in N-polar unit cells (labeled as red ‘M’) decreases from top to bottom, while the intensity of metal atom columns in M-polar ones (labeled as blue ‘M’) increases simultaneously. Combining with the observations in Fig. 3e, this suggests that the transition region composes of two inclined wedge-shaped domains with opposite lattice-polarities. This finding corresponds well with the cone-like growth of N-polar domains from the top surface, as schematically shown in Fig. 3a.

The domain wall structure identified here agrees with established models of horizontal and inclined inversion domain walls previously reported in ferroelectric AlBN, ScGaN, and ScAlN^2,3,15,42. Those domain walls originate from the intermixing of M-polar and N-polar lattices, which leads to atomic column misalignments when viewed along the $⟨11\overline{2}0⟩$ zone-axis. Notably, the lattice-polarity transition region II nearly vanished after poling with a negative pulse exceeding the coercive field, resulting in a highly uniform M-polar lattice across the entire ScAlN region (Fig. S5). Furthermore, we thoroughly examined the lattice-polarity evolution of the ScAlN after the first positive pulse poling, using the same capacitor shown in Fig. 2d. A similar lattice-polarity transition region as illustrated in Fig. 3 was observed in nearly the same location (Fig. S6). These observations indicate that the interfacial dead layer originates predominantly from intrinsic structural factors rather than being induced by the poling process. Moreover, we find that the thickness of the dead layer can be modulated by tuning the growth conditions, as demonstrated in Figs. S7–S9.

Impact of nitrogen vacancy

In traditional ferroelectrics, formation of dead layer has been attributed to the presence of parasitic low-permittivity interfacial capacitor layers, which arise either from the joining of ferroelectric and electrode or from imperfections in thin film processing^31,43. Recent studies have shown that V_N, acting as conductive defects, could seriously alter the dielectric properties of ScAlN, leading to electric leakage path formation and the subsequent degradation of ferroelectricity^{28,33,44, 45, 46–47}. To uncover the underlying physics, first-principles density-functional-theory (DFT) calculations were employed to investigate the effect of V_N on ferroelectricity of ScAlN. We randomly generated unit cells for Sc₃Al₁₃N₁₆ (Sc_0.1875Al_0.8125N) with varying Sc coordinates (Fig. S10a–c). Figure 4a illustrates the most stable one, which was selected as the Sc doped crystal model for further analysis. By removing one nitrogen atom from the pristine Sc₃Al₁₃N₁₆ lattice, corresponding to a nitrogen vacancy concentration ([V_N]) of 6.25%, we constructed all alternative configurations with different V_N sites to identify the most favorable locations for vacancy formation. As illustrated in Fig. 4b, the V_N sites surrounded by pure Al atoms have high formation energy. When examining the four M-N bonds around each nitrogen atom in the pristine Sc₃Al₁₃N₁₆ (Fig. S10e), the bond lengths of these N sites are nearly identical, with the minimal standard deviation (SD), as presented in Fig. 4c. It takes significant energy to disrupt such an equilibrium bonding environment. For N sites with both Al and Sc neighbors, the larger deviation in M-N bond lengths reduces bonding stability, thereby facilitating V_N formation. In particular, the N8 and N13 atoms, which are bonded to two Sc atoms, exhibit larger bond length SD and lower formation energies for V_N8 and V_N13. This analysis suggests that V_N energetically prefer to form around Sc atoms.

[See PDF for image]

Fig. 4

Effect of nitrogen vacancy on ferroelectric ScAlN.

a Atomic structure of the Sc₃Al₁₃N₁₆ unit cell with the lowest total energy used in DFT calculations. Nitrogen atoms are color-coded based on the number of Sc atoms they are bonded to. b Formation energy of V_N at different sites. c Bond length SD for each nitrogen atom. d DOS of Sc₃Al₁₃N₁₆ with varying [V_N]. The zero of energy is positioned at the VBM. e Polarization switching energy profiles of Sc₃Al₁₃N₁₆ containing V_N, plotted as a function of normalized ferroelectric displacement (λ), where +1 and −1 correspond to M-polar and N-polar states, respectively. f Charge density difference of Sc₃Al₁₃N₁₆ containing a V_N8 defect, visualized at an isosurface level of 0.0045 e/ų. The cross-sectional view (right) is taken in the $\left(1\overline{1}00\right)$ plane containing the V_N8 defect, showing electron accumulation (red) and depletion (blue). g, h EELS spectra of Al-L_2,3 edge for pristine (g) and field-cycled (h) ScAlN films, extracted from the regions marked by red (top layer) and blue (near interface layer) rectangles in the ABF-STEM images on the right.

To elucidate the influence of V_N on the electrical properties of ScAlN, we analyzed the density of states (DOS) for ScAlN with varying [V_N]. As shown in Fig. 4d, V_N introduces multiple defect states near the valance band maximum (VBM) and conduction band minimum (CBM), as well as some deep-level defect states, and the density of these states increase with [V_N]. These defect states are the primary source of electric leakage observed in ScAlN, arising predominantly from defect-assisted Poole-Frenkel (P-F) emission and trapping-hopping processes^28,48,49. They can also lead to premature dielectric breakdown in ScAlN⁴⁹. To evaluate the impact of V_N on polarization switching, we calculated the polarization switching energy profile of the ferroelectric displacement in Sc₃Al₁₃N₁₆, as depicted in Fig. 4e. The presence of V_N steepens the double-well energy landscape of wurtzite ScAlN, with the polarization switching barrier increasing from 60 meV/formula unit (f.u.) in the defect-free crystal to 63 meV/f.u. for 1.56% V_N and 72 meV/f.u. for 6.25% V_N. These results suggest that the switching barrier increases gradually with [V_N]. Therefore, V_N not only introduces defect states that degrade dielectric properties, but also elevates the switching barrier. All of these restrict lattice and polarization reversibility.

In addition to V_N, other point defects such as Al vacancies (V_Al), Sc vacancies (V_Sc), and substitutional oxygen (O_N) may also be present in ScAlN. However, our calculations show that the formation energies of V_Al and V_Sc are nearly twice that of V_N (Fig. S11), indicating these metal vacancies are energetically less favorable in ScAlN⁵⁰. In contrast, O_N exhibits comparable formation energy than V_N⁴⁹, making it another prevalent point defect in ScAlN⁵¹. DOS calculations (Fig. S12a) reveal that while V_Al, V_Sc, or O_N introduce defect states near the band edges, they contribute minor to deep-level defect states, suggesting a limited role in dielectric breakdown, although they may still influence leakage current⁴⁹. It is known that the formation energies of V_Al and V_Sc are reduced under N-rich conditions, whereas V_N becomes less favorable⁴⁹. Experimental measurements reveal a substantial reduction in leakage current in ScAlN films grown under N-rich conditions (Fig. S7), suggesting that V_N, rather than V_Al or V_Sc, is the dominant native defect species. Moreover, a recent experimental study reported that oxygen incorporation can suppress leakage current without compromising ferroelectricity in ScAlN⁵². Taken together, these findings support the conclusion that V_N plays the predominant role in degrading dielectric performance in ScAlN, contributing to both increased leakage and premature dielectric breakdown.

Despite extensive theoretical predictions, direct experimental identification of V_N remains challenging. In a defect-free III-nitride lattice, metal atoms exhibit a +3-oxidation state, forming strong covalent bonds with nitrogen. The creation of V_N removes electron-accepting nitrogen atoms, leading to partial reduction of nearby metal cations from M³⁺ to M²⁺, and thereby altering their electronic structure. As illustrated in Fig. 4f, charge density difference for Sc₃Al₁₃N₁₆ containing a V_N8 defect show distinct electron accumulation around adjacent Al and Sc atoms. EELS provides a sensitive means of probing such electronic modifications by capturing the excitation of core electrons into unoccupied states⁵³. This approach has previously been used to identify oxygen vacancies in perovskite ferroelectrics, such as BTO, where the presence of Ti³⁺ in a Ti⁴⁺ matrix reduces peak splitting in the Ti-L_2,3 edge^54,55. In ScAlN, the Al-L_2,3 edge (~73–83 eV) serves as an ideal fingerprint for V_N detection for two reasons: (i) Al is the dominant cation species in ScAlN, and every V_N is coordinated by Al atoms; and (ii) the Al-L_2,3 edge is well-separated from the overlapping Sc-L_2,3 and N-K edges (~402–407 eV), minimizing spectral interference.

Simulated EELS spectra for Al-L_2,3 edges in Sc₃Al₁₃N₁₆ with various point defects are shown in Fig. S13. While the defect-free lattice features clear peak splitting in Al-L_2,3 edge, the introduction of V_N diminishes this characteristic. By contrast, other defects such as V_Al, V_Sc, or O_N primarily cause peak shifts or modest intensity variations without significant broadening. Partial DOS calculations further confirm that V_N introduces more pronounced modifications to the Al electronic states than other defects (Fig. S12b). These alterations in electronic structure, together with the emergence of defect states, can increase the number of final states for core-level excitation, manifesting as reduced peak splitting in Al-L_2,3 edge. This effect parallels the behavior observed in Ti-L_2,3 edges of oxide ferroelectrics containing oxygen vacancies^54,55. Thus, the suppression of the Al-L_2,3 splitting serves as an indirect yet reliable signature of V_N formation in ScAlN.

Figure 4g, h display EELS spectra of the Al-L_2,3 edges acquired from pristine and field-cycled ScAlN, respectively. In both cases, spectra were collected from the top and near interface layers of ScAlN, corresponding to regions I and III in Fig. 3a. As shown in Fig. 4g, the Al-L_2,3 edge recorded from the top layer of pristine ScAlN exhibits a distinct peak splitting, suggesting that Al³⁺ cations are predominant in this region. However, the spectrum from the near interface layer shows diminished peak separation, indicating the presence of Al²⁺, consistent with the presence of V_N. A similar spectral evolution is observed in the field-cycled sample (Fig. 4h). The peak splitting in the top region remains largely intact, suggesting minor V_N accumulation in this layer after electrical cycling. This behavior will be investigated in future studies. Although quantitative analysis of [V_N] remains challenging, the suppression of peak splitting in the Al-L_2,3 edge indicates a high density of V_N in the interfacial ScAlN layer in both pristine and field-cycled capacitors. Notably, the Al-L_2,3 peak splitting in the top ScAlN layer is also moderately weakened compared to simulated defect-free spectra, suggesting a lower density of V_N may also be present in this region. While the contribution of other point defects, such as O_N, to the modifications of Al-L_2,3 edge cannot be entirely excluded, their influence on spectral broadening appears limited and may partially account for the observed deviation between experimental and theoretical spectra.

The formation of V_N in ScAlN, despite N-rich growth conditions, may be attributed to several factors such as thermal desorption of N adatoms, local nitrogen deficiency at the growth front, lattice destabilization due to point defects, and non-equilibrium growth kinetics. The distinct regional distribution of V_N indicates that its formation energy has significant difference in region I and III. Since the ScAlN layer was grown by using the same N-rich conditions and has a uniform Sc distribution along the growth direction, the difference in formation energy does not result from compositional fluctuation. Considering the formation energy of V_N is highly linked to the bonding environment, as discussed in Fig. 4b, c, strain is most likely a crucial factor for V_N formation.

Correlation between strain, nitrogen vacancy, and ferroelectric switching

Subsequently, to explore the strain distribution, we investigated the evolution of lattice parameters along the growth direction in as-deposited ScAlN/GaN heterostructure. Figure 5a presents HAADF-STEM images captured at the ScAlN/GaN interface and from a region about 30 nm away from the interface, corresponding to region III and I that shown in Fig. 3a, respectively. Using these images, we extracted atomic coordinates and obtained lattice parameters by fitting the HAADF intensity of atom columns with two-dimensional Gaussian functions. The mean values for each half unit cell row are plotted in Fig. 5b. It reveals a noticeable variation in lattice parameter a, c, and c/a ratio between regions III and I. The mean lattice constant a for the top ScAlN layer is 3.201 Å, consistent with the experimentally statistical values for relaxed Sc_0.18Al_0.82N⁵⁶, as well as our theoretical predictions (Fig. S10). In contrast, the near interface ScAlN layer has a smaller lattice constant a of 3.182 Å. These results suggest that the initial ScAlN was coherently grown on GaN, exhibiting ~0.57% in-plane compressive strain. This strain likely originates from lattice and thermal expansion coefficient mismatches between ScAlN and GaN, as well as the initial nucleation dynamics during epitaxy^51,56. Notably, in as-grown ferroelectric ScAlN films, dark diffraction contrasts have been observed in the top region, while the interfacial layer exhibits uniform contrast (Fig. 2f, and S8). These features are attributed to the formation of columnar structures and low-angle grain boundaries, which serve as effective pathways for strain relaxation. The onset of these structures, which approximately aligns with the upper boundary of the interfacial dead layer, likely marks the beginning of strain relaxation (Supplementary Note 2).

[See PDF for image]

Fig. 5

Correlation between strain, nitrogen vacancy, and ferroelectric switching in ScAlN.

a Atomic-resolution HAADF-STEM images of the as-grown ScAlN/GaN interface (left) and ScAlN at about 30 nm from the interface (right). Blue dots represent atomic coordinates obtained by fitting the HAADF intensity of atom columns with two-dimensional Gaussian functions. b Statistical analysis of the extracted lattice parameters from (a), including in-plane lattice constant a (left), out-of-plane lattice constant c (middle), and the corresponding c/a ratio (right). The error bars denote the standard deviation for each half unit cell (0.5 u.c.) row, while the red dashed lines indicate the mean values of each region. The position of the ScAlN/GaN interface is set to zero, indicated by black dashed lines. c Effect of biaxial strain on V_N formation energy and the bond length SD at the N8 site. d Variations in the polarization switching barrier in relation to strain and [V_N]. The black dashed lines in (c) and (d) represent the strain level (~0.57%) observed experimentally.

To examine the impact of strain on V_N formation, we applied in-plane biaxial strain in our calculations, wherein negative values correspond to compressive strain, while positive values indicate tensile one. The V_N8 configuration, with the lowest formation energy, was chosen to investigate the effect of biaxial strain on its formation energy. As shown in Fig. 5c, compressive strain significantly reduces the formation energy of V_N8, indicating that strain promotes V_N formation. This effect arises from the strain-induced modification of the local bonding environment (Fig. S14). Specifically, as indicated by the red curve in Fig. 5c, compressive strain increases the bond length SD, destabilizing the lattice and thereby lowering the defect formation energy. Similar trends are observed for the V_N13 configuration (Fig. S15), which has the second lowest formation energy, as well as for ScAlN with 1.56% V_N8 (Fig. S16), confirming that compressive strain facilitates V_N generation. These findings are consistent with the EELS and STEM analyses presented above. In parallel, as shown in Fig. 5d, the polarization switching energy barrier increases nearly linearly with compressive strain in both defect-free and defective ScAlN. A 0.57% compressive strain alone raises the polarization switching barrier by ~3 meV/f.u., comparable to the ~3 meV/f.u. increase induced by 1.56% V_N in strain-free ScAlN. However, when V_N and compressive strain coexist, their effects are additive: the switching barrier increases by ~6 meV/f.u. and becomes more pronounced at higher [V_N], reaching ~16 meV/f.u for 6.25% V_N under the same strain level. This synergistic elevation in switching barrier further suppresses polarization reversibility near the interface.

Discussion

In-plane compressive strain alone has been experimentally reported to increase the coercive and breakdown fields in ScAlN without inducing irreversible polarization⁵⁷. In contrast, V_N not only raises the switching barrier but also introduces defect states that degrade dielectric performance, as evidenced by increased leakage currents in ScAlN films grown under relatively N-poor conditions and in samples with thicker interfacial dead layers (Figs. S7, S8). By integrating experimental observations with theoretical calculations, we identify a high density of V_N near the ScAlN/GaN interface as the primary origin of interfacial dead layer formation. Compressive strain, while not independently sufficient to induce unswitchable polarization, acts cooperatively by reducing the formation energy of V_N and further steepening the energy landscape for polarization switching. The combined effects of the degraded dielectric properties and elevated switching barrier severely restrict lattice and polarization reversibility near the interface, culminating in the formation of an interfacial dead layer. These findings highlight the critical role of V_N and strain in regulating the performance of wurtzite ferroelectrics.

To mitigate the interfacial dead layer, extremely N-rich growth conditions present a promising strategy. In such conditions, the increased nitrogen chemical potential suppresses V_N formation, while strain relaxation is concurrently promoted through modulation of the growth dynamics. This approach has been validated in our MBE-grown ScAlN films, where the dead layer thickness was reduced from ~30 nm to less than 10 nm (Fig. S8). Nevertheless, complete elimination of the dead layer remains a substantial challenge. Further exploration of complementary strategies, including alloy engineering, interlayer inserting, strain engineering, interface decoupling, controlled doping, etc., is needed.

In summary, we have identified that nitrogen vacancies and interfacial strain are the underlying origin of dead layer formation in wurtzite ferroelectric ScAlN. Atomic-scale STEM and EELS characterizations reveal a high density of nitrogen vacancies in the interfacial ScAlN region, where compressive strain is also present. First-principles calculations show that such strain lowers the formation energy of nitrogen vacancies, promoting their generation. These vacancies introduce massive defect states and steepened the energy landscape, the latter further exacerbated by the compressive strain. Consequently, degraded dielectric properties and elevated ferroelectric switching barrier collectively suppress lattice and polarization reversibility near the interface. These findings offer fundamental insight into the microscopic origins of interfacial dead layers and provide a framework for tailoring the properties of wurtzite ferroelectrics via defect and strain engineering. This work paves the way for advancing the integration and scalability of nitride ferroelectrics in next-generation electronic, optoelectronic, photonic, and acoustic devices.

Methods

Molecular beam epitaxy of ferroelectric ScAlN/GaN heterostructure

The ScAlN samples studied in this work were grown utilizing a SVTA MBE system, which is equipped with a Veeco Unibulb radio frequency (RF) plasma source for providing active nitrogen (purity 99.99999%). While Ga (purity 99.99999%), Al (99.99995%), and Sc (purity 99.999%) were evaporated from standard Knudsen cells. The background pressure in the growth chamber was 2 × 10⁻¹¹Torr prior to initiating the growth process. 2-inch GaN/sapphire templates, with a 5-µm-thick GaN layer prepared by MOCVD, were used as substrates. After undergoing a standard chemical cleaning procedure, the substrate was loaded into the MBE system and outgassed at 300 °C for 60 min in the preparation chamber, followed by outgassing at 600 °C for 30 min in the growth chamber. Subsequently, a 200-nm-thick Si-doped n-type GaN layer (with a residual electron concentration of 5 × 10¹⁹cm⁻³, corresponding to a conductivity of ~10³µS/cm) was deposited at 700 °C, followed by the growth of 70-nm-thick ScAlN film. A nitrogen flow rate of 1.0 sccm and an RF forward power of 400 W were maintained for the plasma source throughout the growth, yielding a deposition rate of 5 nm/min for GaN under slight metal-rich conditions. The ScAlN was grown under N-rich conditions to prevent the formation of cubic, perovskite, and intermetallic phases. The beam fluxes of Sc and Al were carefully controlled to achieve a Sc content of 18% (Sc_0.18Al_0.82N), selected to balance ferroelectric properties with crystalline quality. While higher Sc contents can facilitate polarization switching by reducing the energy barrier, they also destabilize the wurtzite phase, leading to increased structural disorder and defect density. Such degradation in material quality complicates detailed microstructural characterization, particularly at the atomic scale.

Fabrication of MFS capacitors

Circular pads of 50 nm Ti/100 nm Au, with a diameter of 20 to 40 µm, were lithographically patterned on the ScAlN surface to serve as top electrodes. The underlying n-type GaN layer was utilized as the bottom electrode.

Structural and electrical characterizations

Surface morphology was characterized in situ using reflection high energy electron diffraction (RHEED) and ex situ using a Bruker ICON atomic force microscope (AFM). XRD 2θ-ω scans were performed using a Philips PANalytical X’pert high-resolution XRD system with a Cu K_α1 radiation X-ray source (1.5406 Å, acceleration voltage 40 kV). Ferroelectric properties, including P-E loop and PUND measurements, were characterized using a Radiant Precision Premier II ferroelectric tester, with the voltage applied from the top electrode. STEM specimens were prepared using a ThermoFisher Helios G4 UX focused ion beam (FIB). HAADF-STEM and ABF-STEM images were captured using a spherical aberration-corrected FEI Titan Cubed Themis G2 300 TEM. The operating voltage is 300 kV with collection angles of 5–17 and 44–230 mrad for ABF and HAADF STEM imaging, respectively. EDS was also performed with the same TEM at 300 kV. EELS measurements were conducted using a high-energy resolution, aberration-corrected Nion U-HERMES 200 TEM, with a convergence angle of 20 mrad, a collection angle of 25 mrad, and an integration time of 2 minutes per pixel.

Theoretical calculations

See details in the Supplementary Information.

Acknowledgements

J.W., Y.L., R.W., Q.L., and H.Ye contributed equally to this work. This work was supported by the National Key Research and Development Program of China (No. 2023YFB3610400), the National Natural Science Foundation of China (NSFC) (No. 62321004, 62374002, 62374010, 62304008, and 12304218), the Natural Science Foundation of Beijing Municipality (No. Z230024), the Research Fund of Suzhou Laboratory (No. SK-1202-2024-012), the China Postdoctoral Science Foundation (No. 2023T160016), Shanghai Pujiang Program (No. 23PJ1402200), and the State Key Laboratory of Artificial Microstructure and Mesoscopic Physics at Peking University.

Author contributions

J.W., P.W., and X.W. conceptualized and designed the project. H.Ye and P.W. conducted the growth and XRD analysis of the ScAlN films. R.W. performed the electrical measurements. J.W., X.X., and T.W. conducted the STEM and EELS measurements. J.W., P.W., T.W., and X.Y. analyzed the STEM and EELS data. Q.L. and X.L. carried out the EELS spectra calculations. Y.L., W.T., and C.D. performed the DFT calculations and theoretical analysis. Y.T., B.Sheng, B.Shen, F.L., L.Y., H.Yang and X.W. contributed to the analysis of the experimental results. J.W., P.W., W.T., and X.W. wrote the manuscript. All authors contributed to discussions of the results and provided feedback on the manuscript at all stages. P.W. and X.W. supervised the research.

Peer review

Peer review information

Nature Communications thanks Sandeep Vura, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

All data in this work are available within the main text and Supplementary Information files, as well as available from the corresponding author.  are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-025-61291-2.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

1. Kim, K; Karpov, I; Olsson, RH; Jariwala, D. Wurtzite and fluorite ferroelectric materials for electronic memory. Nat. Nanotechnol.; 2023; 18, pp. 422-441.2023NatNa.18.422K1:CAS:528:DC%2BB3sXovVGrtrg%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/37106053][DOI: https://dx.doi.org/10.1038/s41565-023-01361-y]

2. Calderon, S et al. Atomic-scale polarization switching in wurtzite ferroelectrics. Science; 2023; 380, pp. 1034-1038.2023Sci..380.1034C1:CAS:528:DC%2BB3sXht1CgsLbF [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/37289886][DOI: https://dx.doi.org/10.1126/science.adh7670]

3. Wang, D et al. Electric-field-induced domain walls in wurtzite ferroelectrics. Nature; 2025; 641, pp. 76-82.1:CAS:528:DC%2BB2MXhtVehu7bM [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/40240596][DOI: https://dx.doi.org/10.1038/s41586-025-08812-7]

4. Wang, P et al. Dawn of nitride ferroelectric semiconductors: from materials to devices. Semicond. Sci. Technol.; 2023; 38, 043002.2023SeScT.38d3002W1:CAS:528:DC%2BB2cXitFGisrzE [DOI: https://dx.doi.org/10.1088/1361-6641/acb80e]

5. Akiyama, M et al. Enhancement of piezoelectric response in scandium aluminum nitride alloy thin films prepared by dual reactive cosputtering. Adv. Mater.; 2009; 21, pp. 593-596.1:CAS:528:DC%2BD1MXitFSjt74%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/21161988][DOI: https://dx.doi.org/10.1002/adma.200802611]

6. Mikolajick, T et al. Next generation ferroelectric materials for semiconductor process integration and their applications. J. Appl. Phys.; 2021; 129, 100901.2021JAP..129j0901M1:CAS:528:DC%2BB3MXmtlGjtLY%3D [DOI: https://dx.doi.org/10.1063/5.0037617]

7. Fichtner, S; Wolff, N; Lofink, F; Kienle, L; Wagner, B. AlScN: A III-V semiconductor based ferroelectric. J. Appl. Phys.; 2019; 125, 114103.2019JAP..125k4103F [DOI: https://dx.doi.org/10.1063/1.5084945]

8. Wang, R et al. Composition-graded nitride ferroelectrics based multi-level non-volatile memory for neuromorphic computing. Adv. Mater.; 2025; 37, 2414805.1:CAS:528:DC%2BB2cXislWltrzF [DOI: https://dx.doi.org/10.1002/adma.202414805]

9. Ye, H et al. Experimental determination of giant polarization in wurtzite III-nitride semiconductors. Nat. Commun.; 2025; 16, 1:CAS:528:DC%2BB2MXhtlalt7fI [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/40274792][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12022134][DOI: https://dx.doi.org/10.1038/s41467-025-58975-0] 3863.

10. Yasuoka, S et al. Effects of deposition conditions on the ferroelectric properties of (Al_1-xSc_x)N thin films. J. Appl. Phys.; 2020; 128, 114103.2020JAP..128k4103Y1:CAS:528:DC%2BB3cXhvVymt7zE [DOI: https://dx.doi.org/10.1063/5.0015281]

11. Xi, J et al. Realization of ferroelectricity in sputtered Al1-xScxN films with a wide range of Sc content. Mater. Today Commun.; 2024; 39, 108966.1:CAS:528:DC%2BB2cXpvVemtrc%3D [DOI: https://dx.doi.org/10.1016/j.mtcomm.2024.108966]

12. Hardy, MT et al. Epitaxial ScAlN grown by molecular beam epitaxy on GaN and SiC substrates. Appl. Phys. Lett.; 2017; 110, 162104.2017ApPhL.110p2104H [DOI: https://dx.doi.org/10.1063/1.4981807]

13. Wang, P et al. Fully epitaxial ferroelectric ScAlN grown by molecular beam epitaxy. Appl. Phys. Lett.; 2021; 118, 223504.2021ApPhL.118v3504W1:CAS:528:DC%2BB3MXht1ejsrnK [DOI: https://dx.doi.org/10.1063/5.0054539]

14. Leone, S et al. Metal-organic chemical vapor deposition of aluminum scandium nitride. Phys. Status Solidi (Rrl) – Rapid Res. Lett.; 2020; 14, 1900535.2020PSSRR.1400535L1:CAS:528:DC%2BC1MXitFSgsLvO [DOI: https://dx.doi.org/10.1002/pssr.201900535]

15. Wolff, N et al. Demonstration and STEM analysis of ferroelectric switching in MOCVD-grown single crystalline Al_0.85Sc_0.15N. Adv. Phys. Res.; 2024; 3, 2300113. [DOI: https://dx.doi.org/10.1002/apxr.202300113]

16. Zeng, Y et al. High quality epitaxial piezoelectric and ferroelectric wurtzite Al_1-xSc_xN thin films. Small Methods; 2025; 9, 2400722.1:CAS:528:DC%2BB2cXhslKrt73I [DOI: https://dx.doi.org/10.1002/smtd.202400722]

17. Hao, Z. et al. Single crystalline ScAlN surface acoustic wave resonators with large figure of merit (Q × k_t²). 2019 IEEE MTT-S International Microwave Symposium (IMS). https://doi.org/10.1109/MWSYM.2019.8700824 (2019).

18. Kim, K et al. Scalable CMOS back-end-of-line-compatible AlScN/two-dimensional channel ferroelectric field-effect transistors. Nat. Nanotechnol.; 2023; 18, pp. 1044-1050.2023NatNa.18.1044K1:CAS:528:DC%2BB3sXhtVOhtbjL [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/37217764][DOI: https://dx.doi.org/10.1038/s41565-023-01399-y]

19. Wang, D; Yang, S; Liu, J; Wang, D; Mi, Z. Perspectives on nitride ferroelectric semiconductors: challenges and opportunities. Appl. Phys. Lett.; 2024; 124, 150501.2024ApPhL.124o0501W1:CAS:528:DC%2BB2cXosVOjtL4%3D [DOI: https://dx.doi.org/10.1063/5.0206005]

20. Xie, Z et al. Nonvolatile and reconfigurable two-terminal electro-optic duplex memristor based on III-nitride semiconductors. Light.: Sci. Appl.; 2024; 13, 78.2024LSA..13..78X1:CAS:528:DC%2BB2cXntFait74%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/38553460][DOI: https://dx.doi.org/10.1038/s41377-024-01422-4]

21. Wang, D et al. Ultrathin nitride ferroic memory with large ON/OFF ratios for analog in-memory computing. Adv. Mater.; 2023; 35, 2210628.1:CAS:528:DC%2BB3sXms1Oqsr8%3D [DOI: https://dx.doi.org/10.1002/adma.202210628]

22. Zhang, Y; Zhu, Q; Tian, B; Duan, C. New-generation ferroelectric AlScN materials. Nano-Micro Lett.; 2024; 16, 1:CAS:528:DC%2BB2cXhsVCrtbrN [DOI: https://dx.doi.org/10.1007/s40820-024-01441-1] 227.

23. Casamento, J. et al. FerroHEMTs: high-current and high-speed all-epitaxial AlScN/GaN Ferroelectric Transistors. International Electron Devices Meeting (IEDM) (2022).

24. Wang, R. et al. Enhancing reliability in ferroelectric ScAIN non-volatile memory with AlOx layer insertion. Ieee Electron Device Lett., 99, 1–1 (2025).

25. Wang, D et al. Thickness scaling down to 5 nm of ferroelectric ScAlN on CMOS compatible molybdenum grown by molecular beam epitaxy. Appl. Phys. Lett.; 2023; 122, 052101.2023ApPhL.122e2101W1:CAS:528:DC%2BB3sXitlyjtbo%3D [DOI: https://dx.doi.org/10.1063/5.0136265]

26. Wang, D et al. Ferroelectric switching in Sub-20 nm aluminum scandium nitride thin films. Ieee Electron Device Lett.; 2020; 41, pp. 1774-1777.2020IEDL..41.1774W1:CAS:528:DC%2BB3MXhtleju7w%3D [DOI: https://dx.doi.org/10.1109/LED.2020.3034576]

27. Kim, KD et al. Evolution of the ferroelectric properties of AlScN film by electrical cycling with an inhomogeneous field distribution. Adv. Electron. Mater.; 2023; 9, 2201142.2023lars.book...K1:CAS:528:DC%2BB3sXksFantLo%3D [DOI: https://dx.doi.org/10.1002/aelm.202201142]

28. Chen, L. et al. Leakage mechanism and cycling behavior of ferroelectric Al_0.7Sc_0.3N. Materials. 17, 397 (2024).

29. Noheda, B; Nukala, P; Acuautla, M. Lessons from hafnium dioxide-based ferroelectrics. Nat. Mater.; 2023; 22, pp. 562-569.2023NatMa.22.562N1:CAS:528:DC%2BB3sXpsVygu7c%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/37138006][DOI: https://dx.doi.org/10.1038/s41563-023-01507-2]

30. Nukala, P et al. Reversible oxygen migration and phase transitions in hafnia-based ferroelectric devices. Science; 2021; 372, pp. 630-635.2021Sci..372.630N1:CAS:528:DC%2BB3MXhtVGnu7zO [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/33858991][DOI: https://dx.doi.org/10.1126/science.abf3789]

31. Chang, L; Alexe, M; Scott, JF; Gregg, JM. Settling the “dead layer” debate in nanoscale capacitors. Adv. Mater.; 2009; 21, pp. 4911-4914.1:CAS:528:DC%2BD1MXhs1WltL3O [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/25377595][DOI: https://dx.doi.org/10.1002/adma.200901756]

32. Wolff, N et al. Atomic scale confirmation of ferroelectric polarization inversion in wurtzite-type AlScN. J. Appl. Phys.; 2021; 129, 034103.2021JAP..129c4103W1:CAS:528:DC%2BB3MXhsl2qsrc%3D [DOI: https://dx.doi.org/10.1063/5.0033205]

33. Tsai, S et al. Field cycling behavior and breakdown mechanism of ferroelectric Al_0.78Sc_0.22N films. Jpn. J. Appl. Phys.; 2022; 61, SJ1005.1:CAS:528:DC%2BB2cXkslWiurc%3D [DOI: https://dx.doi.org/10.35848/1347-4065/ac54f6]

34. Kim, KD et al. Impact of operation voltage and NH3 annealing on the fatigue characteristics of ferroelectric AlScN thin films grown by sputtering. Nanoscale; 2023; 15, pp. 16390-16402.1:CAS:528:DC%2BB3sXitVGhsr%2FP [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/37791415][DOI: https://dx.doi.org/10.1039/D3NR02572A]

35. Wang, D et al. Impact of dislocation density on the ferroelectric properties of ScAlN grown by molecular beam epitaxy. Appl. Phys. Lett.; 2022; 121, 042108.2022ApPhL.121d2108W1:CAS:528:DC%2BB38XhvFKks7vP [DOI: https://dx.doi.org/10.1063/5.0099913]

36. He, Y et al. Al_0.68Sc_0.32N/SiC-based metal-ferroelectric-semiconductor capacitors operating up to 1000 °C. Nano Lett.; 2025; 25, pp. 4767-4773.1:CAS:528:DC%2BB2MXmt1Ghurk%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/40094734][DOI: https://dx.doi.org/10.1021/acs.nanolett.4c06178]

37. Fichtner, S; Yassine, M; Van de Walle, CG; Ambacher, O. Clarification of the spontaneous polarization direction in crystals with wurtzite structure. Appl. Phys. Lett.; 2024; 125, 040501.1:CAS:528:DC%2BB2cXhslenurjM [DOI: https://dx.doi.org/10.1063/5.0217301]

38. Wang, D; Wang, D; Yang, S; Mi, Z. Rethinking polarization in wurtzite semiconductors. Appl. Phys. Lett.; 2024; 124, 263502.1:CAS:528:DC%2BB2cXhsVems7vJ [DOI: https://dx.doi.org/10.1063/5.0212653]

39. Yassine, M et al. Modeling of polarization reversal-induced interface sheet charge in wurtzite-type AlScN/GaN heterostructures. J. Appl. Phys.; 2024; 135, 155702.2024JAP..135o5702Y1:CAS:528:DC%2BB2cXoslKrs7o%3D [DOI: https://dx.doi.org/10.1063/5.0190885]

40. Wang, P et al. Ferroelectric nitride heterostructures on cmos compatible molybdenum for synaptic memristors. Acs Appl. Mater. Interfaces; 2023; 15, pp. 18022-18031.1:CAS:528:DC%2BB3sXmtFajsbg%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36975150][DOI: https://dx.doi.org/10.1021/acsami.2c22798]

41. Chen, L. et al. Bipolar and unipolar cycling behavior in ferroelectric scandium-doped aluminum nitride. 2022 IEEE International Symposium on Applications of Ferroelectrics (ISAF). https://doi.org/10.1109/ISAF51494.2022.9870042 (2022).

42. Wolff, N et al. Electric field-induced domain structures in ferroelectric AlScN thin films. J. Appl. Phys.; 2025; 137, 084101.1:CAS:528:DC%2BB2MXkvFymurc%3D [DOI: https://dx.doi.org/10.1063/5.0244102]

43. Rabe, K. M., Ahn, C. H. & Triscone, J. Physics of Ferroelectrics: A Modern Perspective. 105, (Springer Science & Business Media, 2007).

44. Wang, Q et al. Understanding effect of distortions and vacancies in wurtzite AlScN ferroelectric memory materials: Vacancy-induced multiple defect state types and relaxation dependence in transition energy levels. Aip Adv.; 2022; 12, 125303.2022AIPA..12l5303W1:CAS:528:DC%2BB38XjtVersrzN [DOI: https://dx.doi.org/10.1063/5.0126651]

45. Wang, Q et al. Computational understanding role of vacancies and distortions in wurtzite ferroelectric memory materials: implications for device miniaturization. Mater. Adv.; 2022; 3, pp. 5532-5539.1:CAS:528:DC%2BB38XhsFCitrrE [DOI: https://dx.doi.org/10.1039/D2MA00044J]

46. Yang, J; Meng, X; Yang, C; Zhang, Y. Influence of sputtering power on crystal quality and electrical properties of Sc-doped AlN film prepared by DC magnetron sputtering. Appl. Surf. Sci.; 2013; 287, pp. 355-358.2013ApSS.287.355Y1:CAS:528:DC%2BC3sXhs1yns7bP [DOI: https://dx.doi.org/10.1016/j.apsusc.2013.09.155]

47. Liu, C. et al. Multiscale Modeling of Al_0.7Sc_0.3N-based FeRAM: the Steep switching, leakage and selector-free array. 2021 IEEE International Electron Devices Meeting (IEDM). https://doi.org/10.1109/IEDM19574.2021.9720535 (2021).

48. Liu, X et al. Aluminum scandium nitride-based metal–ferroelectric–metal diode memory devices with high on/off ratios. Appl. Phys. Lett.; 2021; 118, 202901.2021ApPhL.118t2901L1:CAS:528:DC%2BB3MXhtFaqtLvJ [DOI: https://dx.doi.org/10.1063/5.0051940]

49. Lee, C; Din, NU; Brennecka, GL; Gorai, P. Defects and oxygen impurities in ferroelectric wurtzite Al_1-xSc_xN alloys. Appl. Phys. Lett.; 2024; 125, 022901.1:CAS:528:DC%2BB2cXhsFyjsLfO [DOI: https://dx.doi.org/10.1063/5.0211892]

50. Rahman, MH; Mannodi-Kanakkithodi, A. Defect modeling in semiconductors: the role of first principles simulations and machine learning. J. Phys.: Mater.; 2025; 8, 022001.1:CAS:528:DC%2BB2MXjsF2qtbs%3D

51. Wang, P et al. Oxygen defect dominated photoluminescence emission of Sc_xAl_1-xN grown by molecular beam epitaxy. Appl. Phys. Lett.; 2021; 118, 032102.2021ApPhL.118c2102W1:CAS:528:DC%2BB3MXhvVWjtL8%3D [DOI: https://dx.doi.org/10.1063/5.0035026]

52. Islam, M. R. et al. Oxygen doping in ferroelectric wurtzite-type Al_0.73Sc_0.27N: improved leakage and polarity control. Adv. Electron. Mater. 11, 2400874 (2025).

53. Egerton, RF. Electron energy-loss spectroscopy in the TEM. Rep. Prog. Phys.; 2009; 72, 016502.2009RPPh..72a6502E [DOI: https://dx.doi.org/10.1088/0034-4885/72/1/016502]

54. Yang, Y. et al. Atomic-scale fatigue mechanism of ferroelectric tunnel junctions. Sci. Adv. 7, eabh2716.

55. Shao, Y; Maunders, C; Rossouw, D; Kolodiazhnyi, T; Botton, GA. Quantification of the Ti oxidation state in BaTi_1-xNb_xO₃ compounds. Ultramicroscopy; 2010; 110, pp. 1014-1019.1:CAS:528:DC%2BC3cXpsFSltLs%3D [DOI: https://dx.doi.org/10.1016/j.ultramic.2010.05.006]

56. Motoki, K et al. Improved crystallographic order of ScAlN/GaN heterostructures grown at low temperatures under metal rich surface conditions. J. Appl. Phys.; 2024; 135, 135105.2024JAP..135m5105M1:CAS:528:DC%2BB2cXotV2ntrg%3D [DOI: https://dx.doi.org/10.1063/5.0176344]

57. Yasuoka, S et al. Tunable ferroelectric properties in wurtzite (Al_0.8Sc_0.2)N via crystal anisotropy. Acs Appl. Electron. Mater.; 2022; 4, pp. 5165-5170.1:CAS:528:DC%2BB38Xis12nsr%2FN [DOI: https://dx.doi.org/10.1021/acsaelm.2c00999]