Strongly enhanced second-order optical nonlinearity in CMOS-compatible Al 1− x Sc x N thin films

Silicon photonics has enabled large-scale production of integrated optical devices for a vast array of applications. However, extending its use to nonlinear devices is difficult since silicon does not exhibit an intrinsic second-order nonlinearity. While heterogeneous integration of strongly nonlinear materials is possible, it often requires additional procedures since these materials cannot be directly grown on silicon. On the other hand, CMOS-compatible materials often suffer from weaker nonlinearities, compromising efficiency. A promising alternative to current material platforms is scandium-doped aluminum nitride (Al 1 x Sc x N), which maintains the CMOS compatibility of aluminum nitride (AlN) and has been used in electrical devices for its enhanced piezoelectricity. Here, we observe enhancement in optical second-order susceptibility ( χ (2) ) in CMOS-compatible Al 1 x Sc x N thin films with varying Sc concentrations. For Al 0.64 Sc 0.36 N, the χ (2) component d 33 is enhanced to 62.3 5.6 pm/V, which is 12 times stronger than intrinsic AlN and twice as strong as lithium niobate. Increasing the Sc concentration enhances both χ (2) components, but loss increases with a higher Sc concentration as well, with Al 0.64 Sc 0.36 N exhibiting 17.2 dB/cm propagation loss at 1550 nm and Al 0.


I. INTRODUCTION
Integrated photonics is a growing field that manipulates light signals using compact, chip-sized devices. 1 Far-reaching applications range from telecommunication to sensing to optical computing. 2While passive devices have proved useful in guiding light, nonlinear devices are capable of controlling interactions between different light signals in more complex processes, 3 widening the capabilities of integrated photonic chips.
Second-order nonlinearities (χ (2) ) are used in frequency conversion, modulation, and optical signal processing.Frequency converters shift light sources to particular wavelengths; [4][5][6] some can tune the source to a desired wavelength within a limited range. 7Supercontinuum generators broaden spectral width using a combination of nonlinear processes. 8By shifting signals to desired frequencies, frequency converters can enable communication between systems that operate at different frequencies. 9Nonlinear interactions can also be observed at a quantum level, 10 allowing single-photon manipulation; in particular, spontaneous parametric down conversion has been used to produce quantum photon pairs. 11Conversion between RF frequencies and optical frequencies through the linear electro-optic effect can also be considered a χ (2) nonlinear process; applications include quantum transduction 9,12 and integrated electro-optic modulators. 13Optical signal processing utilizes nonlinear processes such as difference frequency generation for optical frequency mixing. 14Furthermore, cascaded χ (2) processes can act as stronger effective χ (3) processes for applications that require higher order nonlinear processes, widening their utility.While these processes can be performed with commercial free-space optical components, nonlinear integrated photonic chips reduce device footprint and enable scalability for complex systems.
Since the crystal structure determines χ (2) values, nonlinear device performance is often limited by material choice.Additional material limitations include bandgap, absorption loss, ease and cost of fabrication, and compatibility with other materials.An important subset of photonics uses silicon to successfully produce devices on a large scale at low cost using techniques from CMOS facilities.Other CMOS-compatible materials can be grown directly on silicon or electrical chips, decreasing fabrication cost and complexity.Thus, CMOS compatibility is another important consideration that can impact ease and cost of fabrication, particularly on a large scale.
Unfortunately, conflicting limitations can complicate material choice.Silicon is clearly CMOS-compatible, but its centrosymmetric structure precludes it from exhibiting χ (2) processes; its bandgap of 1.1 eV also results in absorption of visible wavelengths. 15While many devices use lithium niobate (LN) due to its large χ (2) , with 16 it is not compatible with CMOS processes and cannot be directly grown on silicon platforms.Devices that utilize lithium niobate with silicon often utilize wafer bonding, making large-scale production difficult. 17III-V materials, including GaAs with one of the largest χ (2) values of d 36 = 119 pm/V, 16 can be grown on silicon, but most face a multitude of challenges due to lattice mismatch, different thermal expansion coefficients, and high growth temperatures. 18Several solutions include using buffer layers, 19,20 growing silicon on III-V substrates, 18 and wafer bonding. 21However, these methods are not always CMOS-compatible and require more complicated fabrication procedures.An exception to this difficulty is the III-nitride class, which could provide a CMOS-compatible option due to their low-temperature sputtering process. 22However, these materials exhibit lower χ (2) , with AlN's d 33 = 4 pm/V 23 and GaN's d 33 = 10.3 pm/V. 24Furthermore, high-quality GaN samples often use a thin buffer layer of AlN grown on Si first to reduce strain due to differing thermal coefficients. 19,20lloys provide a promising solution to tune and increase χ (2) while maintaining CMOS compatibility.In particular, substitutional scandium (Sc) in Al 1−x ScxN alloy thin films adjusts the crystal structure and properties, notably increasing its piezoelectric response. 25Al 1−x ScxN's development for electrical devices has resulted in a robust, low-temperature fabrication process that produces high-quality, wafer-scale samples.Unlike many other nonlinear materials, Al 1−x ScxN is grown at low enough temperatures to be CMOS-compatible and retains a relatively large bandgap.However, its nonlinear optical properties in the visible and nearinfrared (NIR) regions have remained largely unexplored.Here, we report enhanced optical χ (2) of Al 1−x ScxN thin films based on second harmonic generation (SHG) measurements.

II. FABRICATION
The Al 1−x ScxN films were deposited in a pulsed DC physical vapor deposition system (Evatec CLUSTERLINE ® 200 II) via cosputtering from separate 4-in.Al and Sc targets at 350C in a pure N 2 environment with a gas flow of 20 SCCM.The power on the Al target was held at 1000 W while the power on the Sc target was adjusted to reach the desired Sc concentration.Calibration of Sc cathode power to Sc concentration was confirmed using energy dispersive x-ray spectroscopy (EDS) in a scanning electron microscope (SEM).High accuracy for the SEM/EDS method is obtained by using a reference sample previously verified by Rutherford Backscattering Spectrometry (RBS) for the k-factor analysis. 26The samples were grown on 100 mm diameter (100) Si wafers coated with 20 nm of (0001) Ti and 100 nm of (111) Pt.The bulk Al 1−x ScxN was deposited with c-axis orientation perpendicular to the substrate, targeting a total thickness of 500 nm; the thickness of each sample was confirmed by cross-sectional SEM [Fig.1(a)].
We confirmed the quality of our samples, even at a high Sc concentration, by AFM imaging Fig. 1(b) and measuring surface roughness and full-width-half-maximum (FWHM) from x-ray diffraction (XRD) rocking curves Table I.The AFM images confirm the lack of anomalously oriented grains, 27 while low roughness over a large, 5 × 5 μm 2 area and consistently sharp XRD curves indicate smooth, highly oriented films. 25These Al 1−x ScxN films are nitrogen polar asdeposited as they are readily etched using KOH, 28 unlike metal polar films.Furthermore, large piezoelectric electromechanical coupling has been observed in these Al 1−x ScxN materials, 29,30 which would be diminished were there alternating domains oriented along the zaxis.As such, these films are strongly oriented and do not exhibit competing domains of opposite polarity.The films are still polycrystalline since in-plane orientation of crystal domains is random despite strong orientation along the direction perpendicular to the substrate.While our samples were grown directly on Pt to enable low-loss reflection SHG, the fabrication process can also be adapted for dielectric substrates, including silicon and sapphire, for low-loss optical and optoelectronic applications.

III. THIN FILM REFLECTION SHG
To determine absolute χ (2) components, we utilize the relationship between the peak intensity of the fundamental light (Iω) and SHG (I 2ω ), 31 o its nonzero χ (2) components are d 33 , d 31 = d 32 = d 15 = d 24 using condensed notation and Kleinman symmetry. 32In reflection geometry, shown in Fig. 2(a), input polarization (p or s) determines how each component contributes to d eff , where θω (θ 2ω ) is the angle of refraction for fundamental (SH) light and f terms are polarization-dependent correction factors that account for interference and multiple reflections, as defined in Sec.I of the supplementary material.In the absence of interference, these correction factors each equal one, recovering the typical equation for effective χ (2) components in wurtzite films (Ref.23).Although the samples are polycrystalline in the crystal plane, the in-plane polarization components of pand s-polarized fundamental light contribute to the SH signal, which is p-polarized in both cases.
L is calculated using cross-sectional SEM thickness measurements in Table I and angle of refraction.The effective refractive index was determined via ellipsometry and prism coupling (Table I); additional details are provided in Sec.II of the supplementary material.Samples with a higher Sc concentration have a slightly higher refractive index, which is consistent with other measurements of the refractive index. 33Dispersion is minimal as the difference in the refractive index at the fundamental and SH wavelengths is ∼0.02-0.03.Birefringence likewise has a negligible effect; additional details are provided in Sec.III of the supplementary material.By comparing the relative strength of the SH signals induced by pand s-polarized fundamental light, we extract d 33 and d 31 using Eqs.( 1

)-(3).
We measured the polarization-dependent reflected SHG from each sample using the setup shown in Fig. 2(b).Source light from an optical parametric amplifier pumped by a Yb:KGW laser amplifier (PHAROS, Light Conversion) at 1550 nm with a 159 fs pulse duration and 3 kHz repetition rate was incident upon our samples at an angle of 45 ○ .The focused beam size at the sample surface was 0.01 mm 2 based on knife edge measurements.The polarization was controlled to be either por s-polarized using two crossed polarizers.The reflected light signal at 775 nm was directed to a calibrated power sensor for absolute power characterization or a spectrometer for spectral measurements.All measurements were taken under the same conditions, accounting for sample differences like film thickness and refractive index.We confirmed the quadratic power dependence of the signal for each sample to ensure that the signal was due to perturbative SHG, as shown in Fig. 3(a).The SHG in each sample exhibits p-polarization for both sand p-polarized fundamental light, as shown in the supplementary material, Fig. 5, corroborating the above SHG analysis.SH spectra in Fig. 3 II.For intrinsic AlN, we calculate d 33 = 5.1 ± 0.4 pm/V and d 31 = 0.07 ± 0.006 pm/V.Other experimental χ (2) measurements for AlN range from 4 23 to 11 pm/V 34 for d 33 and smaller values ranging from 0.04 23 to 0.31 pm/V for d 31 . 35Thus, our calculations provide reasonable values for d 33 and d 31 of AlN.For Al 0.64 Sc 0.36 N, we observe the largest d 33 at 62.3 ± 5.6 pm/V, over twice that of LN; its d 31 = 4.5 ± 0.3 pm/V is also strongly enhanced compared to that of intrinsic AlN.Additionally, simulated SH power is strongly dependent on the particular model of Pt utilized.For the analytical calculation described in the Sec.I of the supplementary material, we utilized data from Werner et al 2009. 43As our substrates are sputtered Pt, this material may not perfectly match the model, introducing additional uncertainty that cannot be accounted for in error propagation.

IV. DISCUSSION
The χ (2) enhancement can be attributed to changes in the crystal structure.Al 1−x ScxN's enhanced piezoelectricity is attributed to Sc destabilizing AlN's wurtzite structure. 25In SHG and piezoelectricity, a material polarization is induced by an optical or a mechanical perturbation, respectively.As such, the softened wurtzite structure due to substitutional Sc is expected to cause a large χ (2) response as well.We expect χ (2) enhancement to continue until centrosymmetric cubic ScN grains begin to form at Sc concentrations around 43%. 25 It is also possible that this softening in the crystal structure affects the relative strength of χ (2) components, with a higher Sc concentration inducing a lower ratio of d 33 compared to d 31 .Young's or the elastic modulus along the z-axis softens more quickly with the increasing Sc concentration than that along the x-axis, indicating that the Sc concentration affects in-plane and out-of-plane crystal axes to different degrees. 36Furthermore, the Al 1−x ScxN piezoelectric coefficients, also written as d 33 and d 31 , increase at different rates with a higher Sc concentration. 36As piezoelectric and χ (2) coefficients are similarly enhanced by the softened crystal structure, it is possible that this softening could also explain the varying ratios of d 33 compared to d 31 depending on the Sc concentration.Further research should explore the effects and determine the optimal Sc concentration to use.
While we directly measured the second-order nonlinear optical susceptibility at optical frequencies, the enhancement from the strained wurtzite structure could persist at lower frequencies in the microwave regime.Since the Pockels coefficient can be expressed in terms of χ (2) at different frequencies, 31 Al 1−x ScxN could exhibit a stronger Pockels effect than intrinsic AlN, with potential applications in efficient EO modulators or integrated quantum transducers.
It has also been found that the Sc concentration affects absorption properties in the visible to ultraviolet wavelength range.While

ARTICLE scitation.org/journal/apm
AlN exhibits a large 6 eV bandgap, Al 1−x ScxN has a smaller bandgap, with high quality Al 0.59 Sc 0.41 N displaying a 4.5 eV bandgap and higher absorption at subband gap energies. 33Sample quality also has a significant effect, since Al 0.66 Sc 0.34 N was reported to have a bandgap of 2.94 eV. 37This was attributed to cubic ScN grain formation at Sc concentrations as low as 20%, which could be due to the differences in the fabrication methods. 33,37Furthermore, AlN has a direct bandgap, while Al 1−x ScxN's bandgap is expected to become indirect past concentrations of 23% Sc based on a larger measured optical bandgap compared to theoretical predictions. 33pplications that utilize longer wavelengths would be minimally affected by the change in the bandgap.Measurements of Al 0.85 Sc 0.15 N photonic waveguides indicate losses of 9 ± 2 dB/cm at 1550 nm; however, optimizing the etching process is expected to significantly reduce sidewall roughness and loss. 38We performed prism coupling measurements of propagation loss for a wafer of Al 0.64 Sc 0.36 N on sapphire, which indicate losses around 17.2 dB/cm at 1550 nm, and higher loss above 30 dB/cm at 633 nm.For an Al 0.80 Sc 0.20 N sample, the loss at 1550 nm is much lower, around 8.2 dB/cm, with loss above 30 dB/cm at 633 nm as well.Although relatively high, these propagation loss measurements provide an upper estimate of loss from absorption and may be improved by optimizing grain structure.Thus, Al 1−x ScxN can be used in waveguides and other on-chip optical elements; further improvements in fabrication techniques could lead to reduced propagation loss.For applications at shorter wavelengths, lower Sc concentrations could enhance χ (2) with less effect on absorption.Further investigation is necessary to determine loss in Al 1−x ScxN structures at different Sc concentrations and how its absorption affects applications in different wavelength regimes.
Another consideration is quasi-phase matching.While modal phase matching can enable high performance devices, it requires precise waveguide design and adds further constraints.Thus, many devices opt instead to use ferroelectric nonlinear materials, such as LN, which can be periodically poled to achieve quasi-phase matching.While periodic poling of non-ferroelectric materials is possible through doping and other fabrication techniques, 39 it requires additional complicated techniques.Devices with AlN have typically used modal phase matching, but Al 1−x ScxN is ferroelectric even at low Sc concentrations, 40,41 which could enable periodic poling in Al 1−x ScxN thin films.The films studied in this work were grown using identical processes to Al 1−x ScxN films exhibiting ferroelectric switching with poling parameters given in Refs.28 and 42, indicating the possibility of periodic poling for quasi-phase matching in these films.

V. CONCLUSION
In conclusion, we measured SHG from Al 1−x ScxN alloy thin films and reported significant enhancement of χ (2) values with the increasing Sc concentration, up to a factor nearing 12 for Al 0.64 Sc 0.36 N. Films were fabricated using a low temperature, CMOScompatible procedure; characterization with AFM and XRD confirms high-quality, well-oriented polycrystalline films.Despite the decrease in the bandgap with the addition of Sc, we expect this material to still be applicable for devices operating at NIR wavelengths.Furthermore, controlling the Sc concentration tunes material properties such as χ (2) and loss to suit individual applications.As a CMOS-compatible material, Al 1−x ScxN could more easily combine cutting-edge nonlinear technology with mature silicon photonics and enable large-scale production of integrated nonlinear devices, ranging from on-chip frequency converters to singlephoton sources.Future studies will improve etching processes to produce high-quality passive photonic elements, reduce absorption and waveguide loss, and produce devices including nonlinear waveguides and modulators.

SUPPLEMENTARY MATERIAL
See the supplementary material for the full derivation of absolute χ (2) value, as well as additional information on ellipsometry, dispersion, and birefringence.Figures of SH spectra for both input polarizations, SH polarization, SEM images, and AFM images for all samples are also included.

FIG. 1 .
FIG. 1. Thin film characterization.(a) Cross-sectional SEM image of the Al 0.64 Sc 0.36 N film, with a thickness of 485.6 nm shown in green.(b) AFM image of the Al 0.64 Sc 0.36 N film; no anomalously oriented grains are present.See the supplementary material for SEM and AFM images of all samples.

FIG. 2 .FIG. 3 .
FIG. 2.Schematic of the SHG measurement technique.(a) SHG reflection geometry in the crystal frame of reference.The sample thickness is t, refracted angle of SH light is θ 2ω , and polarization is labeled p and s.The SHG is p-polarized regardless of input polarization due to crystal symmetry.Multiple reflections are shown as their interference has a non-negligible effect.(b) SHG measurement apparatus.ND: neutral density filter, P: polarizer, L: lens, S: Al 1−x ScxN sample, FM: flip mirror, F: filter, and PM: power meter.The SH signal is separated from fundamental light using a high-pass filter or a prism for spectra and power measurements, respectively.
(b) indicate significant dependence on Sc concentration.The measured d 33 and d 31 values for all samples are shown in Fig. 4 and Table

TABLE I .
Properties of Al 1−x ScxN films studied in this work.Both rms roughness (Rq) and average roughness (Ra) were measured over a 5 × 5 μm 2 area and are low for all samples, indicating smooth films for optical measurements.Films are relatively thin, ranging around 440-490 nm.The low full-width-half-max (FWHM) of x-ray diffraction (XRD) curves indicates highly oriented films.Refractive indices at 1550 nm (nω) and 775 nm (n 2ω ) slightly increase with the Sc concentration.The effective nonlinear coefficient d eff is determined by crystal symmetry.Al 1−x ScxN exhibits a wurtzite structure (6 mm group),

TABLE II .
(2))values for each sample with different percentages of Sc concentration, where d 33 = χ Although d 31 is much smaller than d 33 , its contribution dominates the SH signal induced by p-polarized input light at 45 degree incidence. As  result, the calculated d 33 value depends strongly on d 31 , resulting in a higher uncertainty.To better determine d 33 values in the future, a similar experiment with a larger incidence angle would lower the effect of d 31 uncertainty on d 33 values.