Scandium-Doped Aluminum Nitride (AlScN)

• Scandium-doped aluminum nitride is a wurtzite III-nitride alloy that transforms basic piezoelectric properties into multifunctional behavior including ferroelectricity and enhanced acoustic transduction.
• Sc alloying alters the lattice geometry by reducing the c/a ratio and bond energy differences, thereby lowering coercive fields and enabling strong electromechanical coupling.
• Versatile integration via sputtering, MOCVD, and MBE makes AlScN suitable for RF devices, photonic circuits, ferroelectric memories, and extreme-environment transducers.

Scandium-doped aluminum nitride, usually written $\mathrm{Al_{1-x}Sc_xN}$ and abbreviated AlScN or ScAlN, is a wurtzite III-nitride alloy in which substitution of Al by Sc converts aluminum nitride from a conventional polar piezoelectric into a multifunctional material platform spanning ferroelectricity, enhanced piezoelectric transduction, high-$k$ nitride heterostructures, and second-order optical nonlinearity. In the literature represented here, AlScN is used in bulk and surface acoustic resonators, guided-phonon waveguides, GaN electronic heterostructures, electro-optic and hybrid photonic circuits, ferroelectric memories, and extreme-environment transducers. Its recurring technological appeal is the combination of CMOS compatibility, scalable thin-film deposition, large remanent polarization, strong electromechanical coupling, and useful functionality in regimes ranging from ultrathin ferroelectrics to $\sim 18$ GHz RF filters and operation at $1000^\circ\mathrm{C}$ (Fichtner et al., 2018, Izhar et al., 2024, Gaddam et al., 2024).

1. Crystal chemistry and the origin of functionality

AlScN is the solid solution $\mathrm{Al_{1-x}Sc_xN}$, and its central materials significance is that Sc alloying continuously distorts the parent wurtzite AlN lattice toward a layered-hexagonal limit. In the first report of ferroelectric switching in a III-V semiconductor based material, increasing Sc content was described as increasing lattice “softness” and raising the internal parameter $u$ toward $1/2$, where the structure approaches the nonpolar layered-hexagonal configuration; tensile residual stress was shown to drive the same trend and to reduce the coercive field linearly in $\mathrm{Al_{0.73}Sc_{0.27}N}$ when the residual stress was varied from about $-0.8$ GPa to $+0.5$ GPa (Fichtner et al., 2018).

Later combinatorial work formalized two complementary Sc-driven mechanisms. First, Sc has a much larger ionic radius than Al, which changes the lattice geometry, especially by reducing the $k$0 ratio. Second, Sc is less electronegative than Al, so Sc–N bonding is more ionic, weakening directional covalency and easing atomic displacement during switching. In the AlScN and AlScBN composition space, increasing Sc makes $k$1 decrease and $k$2 increase, lowering $k$3, while high-throughput XPS charge-transfer analysis showed that the energy difference $k$4 decreases from about $k$5 eV at $k$6 Sc to about $k$7 eV at $k$8 Sc, correlating bond ionicity with coercive-field reduction (Messi et al., 10 Jun 2026).

The same alloying that enables ferroelectricity also amplifies other functional responses. Compared with AlN, AlScN is described as having enhanced second-order nonlinear and piezoelectric properties while maintaining a relatively large bandgap, and in RF-acoustic contexts it combines a substantially enhanced piezoelectric response with a relatively high dielectric constant, reduced mechanical compliance, low dielectric loss, and CMOS compatibility (Xu et al., 2024, Gubinelli et al., 2024). This coexistence of polar semiconductor character, strong piezoelectricity, and switchable polarization is the defining feature that distinguishes AlScN from both conventional III-nitrides and oxide ferroelectrics.

2. Thin-film growth, texture control, and epitaxial integration

A notable feature of AlScN research is the breadth of viable growth routes. Reactive sputtering underlies many BEOL-compatible ferroelectric and acoustic implementations, including AlScN deposited at about $k$9 for a MoS$\sim 18$0 ferroelectric field-effect transistor and at $\sim 18$1 for a ferroelectric tunnel junction memristor grown directly on Si/Pt, as well as $\sim 18$2 deposition for ultrathin AlBScN capacitors (Liu et al., 2020, Liu et al., 2020, Tong et al., 11 Nov 2025). Sputtering is also used for AlScN directly on SiC wafers, an important route for Sezawa-mode SAW devices and laterally confined phononic waveguides (Du et al., 2023, Deng et al., 23 Mar 2025).

For epitaxial nitride electronics, both MOCVD and MBE have established compositionally controlled AlScN/GaN integration. MOCVD growth with trimethylaluminum, $\sim 18$3, hydrogen carrier gas, and bis(methylcyclopentadienyl)scandiumchloride $\sim 18$4 achieved Sc incorporation up to $\sim 18$5, with XPS identifying $\sim 18$6 and cross-sectional S/TEM confirming preservation of the wurtzite framework in both films and AlScN/AlN/GaN heterostructures (Vangipuram et al., 14 Oct 2025). Plasma-assisted MBE, meanwhile, identified a lattice-matched composition window on GaN at $\sim 18$7, with an experimentally fitted relation $\sim 18$8 and a lattice-matched composition of $\sim 18$9; using that condition, pseudomorphic ten- and twenty-period AlScN/GaN multilayers were realized with excellent structural and interface properties (Nguyen et al., 2024).

These reports also make clear that the lattice-matched condition is not a universal scalar. One MOCVD study states that AlScN can be lattice matched to GaN in the $1000^\circ\mathrm{C}$0 Sc range, whereas the MBE study ties lattice matching to the growth method and growth temperature and uses that dependence to explain why earlier reports ranged broadly from $1000^\circ\mathrm{C}$1 to $1000^\circ\mathrm{C}$2 Sc (Vangipuram et al., 14 Oct 2025, Nguyen et al., 2024). A common outcome across methods is strong $1000^\circ\mathrm{C}$3-axis texture or oriented wurtzite growth, which is crucial because the dominant ferroelectric, piezoelectric, and electro-optic responses are all tied to the polar axis.

3. Ferroelectricity, switching kinetics, and coercive-field engineering

Ferroelectric AlScN was first established through nearly box-like $1000^\circ\mathrm{C}$4-$1000^\circ\mathrm{C}$5 hysteresis loops in $1000^\circ\mathrm{C}$6 with $1000^\circ\mathrm{C}$7 to $1000^\circ\mathrm{C}$8, polarity-specific wet etching after polarization inversion, stable switching polarization over $1000^\circ\mathrm{C}$9 s, and frequency-independent switched polarization over more than two orders of magnitude in frequency. In that work, the coercive field was reported as $\mathrm{Al_{1-x}Sc_xN}$0 MV/cm at $\mathrm{Al_{1-x}Sc_xN}$1 and $\mathrm{Al_{1-x}Sc_xN}$2 MV/cm at $\mathrm{Al_{1-x}Sc_xN}$3, with high remanent polarization of $\mathrm{Al_{1-x}Sc_xN}$4, an almost ideally square-like hysteresis, and a paraelectric transition temperature in excess of $\mathrm{Al_{1-x}Sc_xN}$5 (Fichtner et al., 2018).

Subsequent studies frame the main materials limitation as the unusually high coercive field. AlScN is described as combining a wide band gap, good thermal stability, and very large remanent polarization, with reported values around $\mathrm{Al_{1-x}Sc_xN}$6, nearly square $\mathrm{Al_{1-x}Sc_xN}$7–$\mathrm{Al_{1-x}Sc_xN}$8 loops, and stability up to $\mathrm{Al_{1-x}Sc_xN}$9, but with a coercive field on the order of MV/cm—two to three orders of magnitude above many oxide ferroelectrics (Zheng et al., 31 Aug 2025). Atomistic analysis in that study argues against a purely collective M-polar $u$0 nonpolar hexagonal $u$1 N-polar pathway as the dominant mechanism in realistic films: for $u$2, the concerted transition was found to induce nearly $u$3 $u$4-axis strain, whereas pre-existing $u$5 domain walls provide a strain-relief route. The field dependence is correspondingly split: at low fields, switching is governed mainly by gradual domain-wall propagation and is well described by the Kolmogorov–Avrami–Ishibashi model, while at high fields additional nucleation events appear and the simultaneous non-linear nucleation and growth model gives the best description. The same work reports that the switching field $u$6 and the activation field $u$7 both decrease monotonically with increasing Sc content across roughly $u$8 to $u$9 Sc (Zheng et al., 31 Aug 2025).

Interfacial engineering has emerged as a direct route to reduce the ultrathin-film switching penalty. In $1/2$0 capacitors, a Sc bottom electrode was shown to reduce the coercive field by over $1/2$1 compared to an Al bottom electrode, including a $1/2$2 reduction in $1/2$3 at $1/2$4 kHz and a $1/2$5 reduction at $1/2$6 kHz. The same study used the KAI relation $1/2$7, finding low-frequency exponents of $1/2$8 for Sc and $1/2$9 for Al bottom electrodes, while at higher frequencies the Al-electrode device showed a much larger exponent of $\mathrm{Al_{0.73}Sc_{0.27}N}$0 and the Sc-electrode device remained at $\mathrm{Al_{0.73}Sc_{0.27}N}$1; SEND strain mapping linked this behavior to reduced lattice mismatch and lower in-plane residual strain in the Sc-template stack (Zhang et al., 11 Jun 2025). This suggests that AlScN coercive-field engineering is not only a compositional problem but also an interface- and strain-management problem.

4. Acoustic resonators, filters, and guided phonons

The most mature non-ferroelectric exploitation of AlScN is as a high-coupling piezoelectric for GHz acoustics. In overmoded bulk acoustic resonators, the material enables operation beyond the thickness-scaling limits of conventional FBARs by preserving useful coupling when the acoustic field is intentionally distributed across both the piezoelectric layer and the electrodes. A $\mathrm{Al_{0.73}Sc_{0.27}N}$2 GHz $\mathrm{Al_{0.73}Sc_{0.27}N}$3 Sc-doped ScAlN OBAR using a Pt/ScAlN/AlSiCu stack and a $\mathrm{Al_{0.73}Sc_{0.27}N}$4 nm ScAlN layer demonstrated second-overtone resonance at $\mathrm{Al_{0.73}Sc_{0.27}N}$5 GHz with $\mathrm{Al_{0.73}Sc_{0.27}N}$6 to $\mathrm{Al_{0.73}Sc_{0.27}N}$7 and $\mathrm{Al_{0.73}Sc_{0.27}N}$8; the best measured fundamental tone gave $\mathrm{Al_{0.73}Sc_{0.27}N}$9 GHz with $-0.8$0, and the measured figure of merit was about $-0.8$1 for both tones. The same work emphasizes the OBAR tradeoff through $-0.8$2 and $-0.8$3, and shows that pentagonal geometries suppress transverse spurious modes more effectively than circular ones (Gubinelli et al., 2024).

At still higher frequency, periodically poled AlScN bulk acoustic structures use thickness-wise polarity engineering to avoid impractically thin films. A four-layer AlScN periodically poled piezoelectric film operating in the fourth thickness-extensional mode reached $-0.8$4 GHz with $-0.8$5, $-0.8$6, $-0.8$7, and $-0.8$8. Resonator-derived filters at $-0.8$9 GHz achieved insertion losses of $+0.5$0 dB for a three-element filter and $+0.5$1 dB for a six-element filter, $+0.5$2 dB bandwidths of $+0.5$3 MHz and $+0.5$4 MHz, and in-band IIP3 values of $+0.5$5 dBm and $+0.5$6 dBm, respectively (Izhar et al., 2024). At lower frequency but larger fractional bandwidth, an Al$+0.5$7Sc$+0.5$8N two-dimensional resonant rods resonator reached $+0.5$9 GHz with $k$00, $k$01, and $k$02, supporting projected fifth-order ladder filters with $k$03 fractional bandwidth, $k$04 dB insertion loss, and $k$05 dB out-of-band rejection (Zhao et al., 2022).

AlScN on SiC has also become a distinct platform for surface and guided acoustics. Sezawa-mode SAW resonators in $k$06 reached $k$07 GHz at $k$08 wavelength with $k$09 and $k$10, while the maximum measured $k$11 and $k$12 were reported at $k$13 GHz (Du et al., 2023). Moving beyond slab acoustics, two-dimensionally confined AlScN/SiC phononic waveguides with a $k$14m-thick AlScN film at $k$15 Sc support Rayleigh-like and Sezawa-like guided modes; at $k$16, the Sezawa-like mode has velocity about $k$17 m/s, resonance at $k$18 GHz, simulated $k$19, experimentally extracted $k$20, slab loss $k$21 dB/mm, and straight-waveguide loss $k$22 dB/mm (Deng et al., 23 Mar 2025). A plausible implication is that AlScN’s role in acoustics is no longer limited to resonators: it now spans overmoded RF transducers, Sezawa-mode SAW, and laterally routed phononic circuitry.

5. GaN heterostructures, polarization-engineered transport, and memory devices

In nitride electronics, AlScN often functions as a polarization-engineered barrier rather than only as a ferroelectric. MOCVD-grown AlScN/AlN/GaN heterostructures on GaN templates showed clear 2DEG formation at the AlScN/AlN–GaN interface, with C–V-extracted sheet densities of $k$23 cm$k$24, $k$25 cm$k$26, and $k$27 cm$k$28 for $k$29, $k$30, and $k$31 nm barriers, respectively. Hall measurements gave larger values of approximately $k$32, $k$33, and $k$34 cm$k$35, with mobilities of $k$36, $k$37, and $k$38 cm$k$39/Vs; the paper attributes the Hall/C–V discrepancy to surface oxidation, buffer conduction, and mercury-contact-area uncertainty, but both methods show that thicker barriers yield higher sheet charge (Vangipuram et al., 14 Oct 2025).

MBE studies extend this concept to multichannel transport. One report identified lattice-matched AlScN/GaN multilayers and measured net mobile charge densities of $k$40 for ten periods and $k$41 for twenty periods, with charge scaling approximately with the number of AlScN/GaN periods (Nguyen et al., 2024). A later transport-focused study then introduced GaN/AlN interlayers to improve conductivity in single-channel structures and leveraged the result in multi-channel heterostructures. In the best single-channel sample, room-temperature mobility reached $k$42 with $k$43, while a five-channel structure reached $k$44, $k$45, and $k$46 at $k$47 K, falling to $k$48 at $k$49 K (Asteris et al., 13 Nov 2025). These results place AlScN/GaN alongside state-of-the-art multi-channel nitride systems for RF and high-power electronics.

Ferroelectric memory implementations use a different aspect of the same material system. A post-CMOS-compatible AlScN/MoS$k$50 ferroelectric field-effect transistor employed a $k$51 nm $k$52 gate dielectric deposited at about $k$53, producing an ON/OFF ratio of $k$54, a memory window of about $k$55 V, a normalized memory window of $k$56, and stable two-state retention for up to $k$57 s (Liu et al., 2020). A CMOS-back-end-of-line-compatible ferroelectric tunnel junction memristor based on $k$58 nm $k$59 deposited at $k$60 showed diode-like polarity-dependent switching, rectification ratio $k$61 at $k$62 V, On/Off ratio $k$63, and retention $k$64 s at $k$65 K, with transport best described by Poole–Frenkel tunneling (Liu et al., 2020). Together these studies establish that AlScN supports both polarization-induced channel formation in GaN heterostructures and nonvolatile ferroelectric state variables in memory devices.

6. Photonics, electro-optics, and multifunctional transduction

AlScN has become attractive in integrated photonics because the same Sc alloying that enhances piezoelectricity also amplifies the second-order nonlinear response relevant to the Pockels effect. In silicon-integrated electro-optics, sputtered $k$66 on insulator was used as both the light-guiding medium and electro-optic medium in micro-ring resonators. The devices showed a maximum in-device effective EO coefficient of $k$67 pm/V at $k$68 GHz, a minimum $k$69 of $k$70 at $k$71 GHz, and a $k$72-dB modulation bandwidth of approximately $k$73 GHz; the same work notes that the $k$74 coefficient of $k$75 is about $k$76 that of AlN and that $k$77 reaches about $k$78 enhancement (Xu et al., 2024).

A central challenge for direct ScAlN photonics has been optical loss. Hybrid $k$79-ScAlN circuits address this by confining the optical mode in etched $k$80 while retaining the functional ScAlN layer underneath. In a monolithic sapphire / $k$81 nm AlN / $k$82 nm $k$83 / $k$84 nm $k$85 stack, the best Euler racetrack resonator reached $k$86 at $k$87, corresponding to a propagation loss of $k$88, compared with previously reported direct ScAlN photonic losses typically above $k$89 (Liu et al., 1 Aug 2025). This architecture is significant because it separates low-loss waveguiding from the functional nitride layer rather than requiring the ScAlN itself to be the etched optical core.

Multifunctional transduction appears most clearly in resonant IR detection. A plasmonically enhanced flexural-mode AlScN nanoplate resonator using $k$90-doped AlScN and a gold cross-shaped metasurface absorber was reported as an uncooled, ultrafast IR detector with $k$91 MHz, $k$92, experimental $k$93, thermal time constant $k$94, and normalized IR responsivity $k$95; the estimated thermomechanical-noise-limited NEP was $k$96 (Venditti et al., 26 Jun 2025). The device is notable because it uses AlScN simultaneously as the electromechanical transducer and as the dielectric in the plasmonic absorber, reinforcing the broader view of AlScN as a platform material rather than a single-function piezoelectric.

7. Reliability, extreme environments, and boron-modified descendants

The main materials liabilities of AlScN are now documented as oxidation, high coercive field, and endurance limitations. Operando HAXPES on $k$97 showed that Sc doping weakens the polar AlN bonds enough to enable ferroelectric switching below dielectric breakdown, but also increases the tendency toward oxidation. The study found replacement of nitrogen by oxygen, preferential oxidation of Sc rather than Al, an oxidation-related N$k$98 feature in the N $k$99 spectrum, and no confirmation of self-limiting oxidation; a $\sim 18$00 nm W cap suppressed oxidation effectively, with the capped sample remaining chemically stable up to about $\sim 18$01 V, whereas the uncapped sample oxidized further even under $\sim 18$02 V bias (Rehm et al., 2024). This has direct implications for ferroelectric capacitor and transistor reliability, since top-electrode chemistry and ambient exposure become part of the functional materials problem.

At the same time, AlScN exhibits exceptional thermal robustness. In TaSi$\sim 18$03/$\sim 18$04/TaSi$\sim 18$05 capacitors, functional operation was reported up to $\sim 18$06. Over this range, the coercive field decreased from $\sim 18$07 MV/cm at room temperature to $\sim 18$08 MV/cm at $\sim 18$09, the effective longitudinal piezoelectric coefficient increased from $\sim 18$10 pm/V at $\sim 18$11 to $\sim 18$12 pm/V at $\sim 18$13, and the electromechanical coupling coefficient was calculated to rise from $\sim 18$14 at room temperature to $\sim 18$15 at $\sim 18$16; S/TEM showed no detectable electrode diffusion into the AlScN layer after thermal cycling (Gaddam et al., 2024). A plausible implication is that AlScN’s reliability envelope is unusually bifurcated: chemically fragile under oxygen exposure, but structurally and functionally robust under extreme thermal load when the interfaces are properly engineered.

Much recent alloy engineering treats AlScN as the baseline ferroelectric nitride and seeks to alleviate its switching and endurance penalties by boron incorporation. A combinatorial HiPIMS study covering $\sim 18$17 unique samples and about $\sim 18$18 ferroelectric devices found that in AlScBN the remanent polarization remains $\sim 18$19, while the coercive field drops from $\sim 18$20 MV/cm to $\sim 18$21 MV/cm and endurance extends to $\sim 18$22 cycles without sample recharging; the same study concludes that B co-alloying lowers the amount of Sc needed to reduce $\sim 18$23 and links the endurance improvement to reduced defect density (Messi et al., 10 Jun 2026). In the ultrathin limit, $\sim 18$24 nm sputtered AlBScN capacitors showed ferroelectric switching at $\sim 18$25 MV/cm in C–V and symmetric polarization reversal near $\sim 18$26 MV/cm in PUND with $\sim 18$27s pulses, together with about two orders of magnitude lower leakage current than $\sim 18$28 nm AlScN and a breakdown-to-coercive-field ratio $\sim 18$29 (Tong et al., 11 Nov 2025). These derivative systems do not displace AlScN conceptually; rather, they clarify which of AlScN’s constraints are intrinsic to Sc-stabilized ferroelectric nitrides and which can be retuned by quaternary alloy design.