1. Introduction

Aluminum nitride (AlN) has a large bandgap (E_g) of 6.2 eV and a broad transparency window spanning from ultraviolet to mid-infrared [1–2]. The non-centrosymmetric crystal structure of AlN contributes to its piezoelectric properties (d₃₃ = 5.53 pm/V and d₃₁ = −2.65 pm/V) as well as its strong second-order nonlinear susceptibility (χ⁽²⁾ = 4.7 pm/V) [3–5]. The second-order nonlinear optical property enables applications such as second harmonic generation (SHG) and parametric down-conversion [6]. Furthermore, its thermal stability and compatibility with the complementary metal oxide semiconductor (CMOS) process enable large scale, low cost and highly reliable wafer-level manufacturing of integrated photonic devices.

The doping of scandium (Sc) into AlN alters its band structure and optical properties, exhibiting a tunable E_g regarding various Sc concentrations [7–9]. In addition, Sc doping increases the lattice asymmetry of AlN, potentially enhances the piezoelectric coefficients and second-order optical nonlinearity such as optical birefringence [10]. When the doping concentration is up to 43%, the piezoelectric coefficient d₃₃ is four times higher than AlN films, and the piezoelectric coefficient d₃₁ is also increased significantly. Al_1−xSc_xN has been conventionally used in micro-electro-mechanical systems (MEMS) and radio-frequency (RF) devices due to its outstanding piezoelectric properties. The unique properties of Al_1−xSc_xN confer its significant potential for photonic applications in fields such as sensing, communication and information processing. However, the application of Al_1−xSc_xN in integrated photonics remains largely unexplored mainly due to two reasons: the limited research on optical properties of the material and the incomplete photonic device library.

The exploration of the optical properties of Al_1−xSc_xN holds immense importance in the realm of integrated photonics [11]. Although the optical properties such as E_g, refractive indices (n) and extinction coefficients (k) of Al_1−xSc_xN have been reported, to the best of our knowledge, its anisotropic properties of those parameters have not been extensively studied yet [9]. The optical anisotropy of Al_1−xSc_xN exhibits a pronounced effect on near-band-edge transitions and excitonic effects. Obtaining in-depth knowledge on the optical anisotropy provides valuable guidance for the design and optimization of photonic devices. Besides, as for the Al_1−xSc_xN photonic device library, there are few devices reported in recent decade [12–13]. The optical waveguides and waveguide bends are essential components of photonic integrated circuits. With a relatively high propagation loss of 9 dB/cm at x = 0.108 reported by Zhu et al. at λ = 1550 nm [13], the research on Al_1−xSc_xN waveguides and bends is still in its infancy.

In this paper, we present the investigation of the anisotropic properties of E_g, n and k for Al_1−xSc_xN at various Sc concentrations. The band structure of wurtzite Al_1−xSc_xN is analyzed quantitatively. In addition, Al_1−xSc_xN photonic waveguides and bends are designed according to the material characterization results. Finally, the photonic devices are fabricated on an 8-inch Si substrate using CMOS-compatible fabrication platform. For x = 0.087 and 0.181, the lowest propagation losses at λ = 1550 nm are extracted to be 5.98 ± 0.11 dB/cm and 8.23 ± 0.39 dB/cm and the 90° bending losses with 50 μm bend radius are 0.05 dB/turn and 0.08 dB/turn, respectively.

2. Optical properties of Al_1−xSc_xN

The 400 nm-thick Al_1−xSc_xN film was deposited on a 3 μm-thick silicon dioxide (SiO₂) layer which had been grown on an 8-inch Si (100) wafer. The Al_0.913Sc_0.087N and Al_0.819Sc_0.181N films were deposited from AlSc targets with nominal Sc contents of 9.6 at.% and 20 at.%, respectively, by an RF magnetron sputtering system. In the X-ray diffraction (XRD) characterization, the peak of Al_1−xSc_xN (002) exhibits a relatively narrow full width at half maximum (FWHM) of 1.84°. In addition, there is no other characteristic peak of Al_1−xSc_xN appearing within the angular range of the 2θ scan. The presence of a single (002) diffraction peak typically indicates a consistent orientation of the crystallites in the sample towards the (001) crystal plane. This indicates that the film has a good c-axis orientation which is perpendicular to the substrate [14].

Spectroscopic ellipsometry (SE) is a powerful optical measurement technique that relies on characterizing changes in the polarization state of linearly polarized light reflected from a thin film sample [15–17]. This non-destructive method plays a crucial role in the characterization of various materials and devices. The SE spectra amplitude ratio φ and phase difference Δ were measured to describe the changes in the polarization state of linearly polarized light after reflection from the Al_1−xSc_xN films [18]. The parameters φ and Δ satisfy the equation [19]:

The SE spectra of Al_1−xSc_xN films with different Sc concentrations were measured by a variable-angle spectral ellipsometer at room temperature. The optical properties were investigated by scanning the photon energy (hv) from 1.24 eV to 5.905 eV at the incidence angle θ = 60°, 65°and 70°. It is necessary to model the Al_1−xSc_xN film as a uniaxial anisotropic material to accurately fit the SE data [20–21]. The SE spectra were fitted using the anisotropic dispersion equations in the completeEASE software by Mueller matrix measurement. The n and k for extraordinary and ordinary lights were determined by minimizing the mean square error, with the results plotted in Figs. 1(a) and 1(b), respectively. The measured VASE data are presented in the Fig. S1 of the Supplement 1. The minimized mean square error (MSE) values for x = 0, 0.087, 0.181 are 21.1, 17.6, and 20, respectively.

 

Fig. 1. The dispersions of (a) n_o, n_e and (b) k_o, k_e for Al_1−xSc_xN films at photon energy ranges from 1.24 to 5.905 eV. The (c) Δn and (d) Δk plotted as a function of hv for Al_1−xSc_xN films.

Download Full Size | PDF

The electric field (E) of propagating light in wurtzite Al_1−xSc_xN can be separated into two components: parallel to c-axis (E∥ c) and perpendicular to c-axis (E⊥c) which are commonly known as the extraordinary and ordinary configurations, respectively [20,22]. The n and k of ordinary light are represented as n_o and k_o, while those of extraordinary light are denoted by n_e and k_e, respectively. To accurately characterize the anisotropy of the Al_1−xSc_xN films, the differences of the n and k are represented by birefringence Δn = n_e – n_o and dichroism Δk = k_e – k_o, respectively.

Semiconductors generally have relatively high stimulated intrinsic absorption coefficients and exhibit continuous absorption spectra [23]. When the energy of photons is sufficiently high (equal to or greater than the E_g), photons can interact with electrons in the valence band, causing electron transitions to the conduction band. This process is known as intrinsic absorption and serves as the primary mechanism for optical absorption in semiconductors. In the region where the photon frequency (ν) is less than the E_g divided by the Planck constant (h), it is difficult to excite electrons from the valence band to the conduction band, resulting in a small absorption coefficient. The n_o and n_e can be fitted using the Cauchy dispersion formula or empirical formulas like the Sellmeier equation [11,24]. Figure 1(a) illustrates the increase in the n with increasing Sc doping concentration. The spectra of all samples exhibit a clear distinction between the transparent region and the absorption region with a critical point that closely approximates the E_g. As the energy of photons approaches the E_g, more electrons can transit from the valence band to the conduction band, resulting in an increase in the optical absorption coefficient (α). There is a positive correlation between a and k. As the photon energy increases, distinct extinction coefficient values appear near the E_g of Al_1−xSc_xN, as shown in Fig. 1(b). In the region where hν > E_g, the n undergoes a sudden change, exhibiting anomalous dispersion. It is worth noting that in this region, the n_o and n_e exhibit two distinct peaks, resulting in the intersection of refractive indices and leading to the disappearance of birefringence at particular wavelengths.

There is a relationship between E_g and α for semiconductors [25],

 

Fig. 2. (a) The direct E_g of ordinary and extraordinary lights are extracted by linear fitting for Al_1−xSc_xN films (x = 0.087 and 0.181). (b) E_g of Al_1−xSc_xN as a function of x in Al_1−xSc_xN. The purple dashed line corresponds to the theoretical E_g reported by Zhang et al. [7]. The orange dashed line corresponds to the calculation formula reported by Baeumler et al. [9]. Blue dots and red dots are the direct E_g for this work. (c) Band structure and respective optical transitions between valence bands and conduction band in wurtzite Al_1−xSc_xN. (d) Δ_cr as a function of x in Al_1−xSc_xN.

Download Full Size | PDF

The Δn and Δk are investigated to provide more insights for the anisotropy of Al_1−xSc_xN thin films. The extrema in the Δn and Δk are typically associated with optical transitions governed by the polarization selection rules [20]. As shown in Fig. 1(c), two extrema are observed in the birefringence Δn curve near the E_g and the minimum and maximum values for x = 0.087 occur at 5.79 eV and 5.56 eV, respectively. For x = 0.181, the minimum and maximum Δn values are observed at 5.34 eV and 5.06 eV, respectively. As the peak of n_o exceeds the limit of the spectrometer, the minimum Δn value cannot be obtained for AlN (x = 0). The minimum and maximum of Δn curve align with the positions of peaks in the n_e and n_o. These extrema can be attributed to the energy of optical transitions. Similarly, Fig. 1(d) illustrates an obvious peak in the Δk near the E_g. As x increases, the peaks of Δn and Δk also increase, which means that the optical birefringence phenomenon contributed by its non-centrosymmetric crystal structure of Al_1−xSc_xN becomes more pronounced. The noticeable Δn and Δk observed in Al_1−xSc_xN at short wavelengths make it promising for applications in ultraviolet photonic devices.

The E_g of a semiconductor material can be determined by analyzing its absorption spectrum. By substituting α = 4πk/λ into Eq. (2) and plotting the value of (αhν)² on y-axis against the value of hν on x-axis, Eq. (2) can be regarded as a linear equation. The value of E_g can be obtained by the intercept of the line on x-axis. By performing linear regression on the linear portion of the graph, a linear equation can be obtained as demonstrated in Fig. 2(a). When x = 0.087 and x = 0.181, the determined band gap energies for ordinary light E_g,o in Al_1−xSc_xN are 5.73 eV and 5.32 eV, respectively. For extraordinary light, the band gap energies E_g,e are determined to be 5.50 eV and 5.06 eV. In the previous papers, the E_g that is not distinguishable between ordinary light and extraordinary light is denoted as E_g,eff [7,9]. Within the range of 0 < x < 0.25, Baeumler et al. extended Vegard's law by incorporating a nonlinear term to describe the bending of the E_g with respect to x in theoretical predictions for wurtzite Al_1−xSc_xN [9]. An empirical relation is derived to describe the composition dependence of the E_g which is indicated by the orange dashed line in Fig. 2(b). In addition, the E_g reported by Zhang et al. are indicated by the purple dashed line [7]. Both two dashed lines exhibit decreasing trend of the E_g with increasing x. The E_g obtained in this work are marked with dots. As compared to E_g,e, it is worth noting that E_g,o values are closer to the E_g,eff values from literature. It is conjectured that this is because luminescence detections approach E⊥c condition instead of E∥ c condition, resulting in the determined E_g,eff that are closer to E_g,o [20–21].

The band structure of wurtzite Al_1−xSc_xN is illustrated in Fig. 2(c). Due to the hexagonal crystal-field and spin-orbital splitting acting on the triply degenerate p-like valence band, the top of the valence band in Al_1−xSc_xN is divided into three bands: $\mathrm{\Gamma}_{7v}^u$, Γ₉v and $\mathrm{\Gamma}_{7v}^l$. The corresponding interband transitions from these valence bands to the bottom of the conduction band Γ_7c are represented by transitions labeled as A, B, and C [20,22].

Anisotropy is attributed to the ordered valence band at the Brillouin zone and a significant crystal-field splitting is observed based on the fitting data. According to the previous findings by Suzuki et al. and Jiang et al. [20,26], it has been established that the spin-orbit splitting Δ_so in AlN is very small. Therefore, based on the band structure, it is inferred that the crystal-field splitting energy Δ_cr is approximately equal to E_g,e − E_g,o. The Δ_cr are found to be −0.23 eV and −0.26 eV for x = 0.087 and x = 0.181, respectively. According to the data reported by Jiang et al. [20], the Δ_cr of AlN is approximately −0.2 eV. Thus, it can be inferred that the Δ_cr increases with the increase in Sc concentration, as plotted in Fig. 2(d). Due to the negative Δ_cr of Al_1−xSc_xN, the polarization of extraordinary light selectively allows the A transition (E∥c). In contrast, for ordinary light polarization with the E perpendicular to c-axis (E⊥c), both B and C transitions are allowed [20].

3. Waveguide design

The refractive index of a core material plays a crucial role in simulating single mode conditions of a photonic waveguide, which ensures efficient light transmission and mode confinement. Based on the SE measurement, the n of the Al_1−xSc_xN thin films were further fitted by the Cauchy dispersion model in the wavelength range from 250 to 2000nm, as illustrated in Fig. 3(a). There is an approximately linear relationship (n = ax + b) between the refractive indices and the Sc doping concentration with x < 0.25. The slope and intercept of this linear relationship are illustrated in Fig. 3(b). When λ > 900 nm, these two values tend to stabilize. The relationship between refractive indices and x at λ = 1550 nm can be expressed by the following equations:

 

Fig. 3. (a) The n_o and n_e at different Sc concentrations are fitted using the Cauchy dispersion formula from λ=250 to 2000nm. (b) When λ > 900 nm, the refractive indices have an approximately linear relationship with the Sc concentration. (c),(d) The n_eff simulation of Al_1−xSc_xN with x = 0.087 and 0.181.

Download Full Size | PDF

To minimize modal dispersion, the geometric structure of Al_1−xSc_xN channel waveguides is designed to support only the first-order mode. Based on the fitted n, the effective refractive indices (n_eff) of 400-nm-thick Al_1−xSc_xN waveguides are calculated using the Ansys Lumerical MODE solver. The sidewall angle of the simulated waveguide is fixed at 70 degrees based on fabrication results. The n of the SiO₂ cladding is 1.44. For TE mode, the E field of propagating light is perpendicular to c-axis (E⊥c), thus the n_o was chosen for the simulation. On the contrary, if the E of propagating light is parallel to c-axis (E∥c), the n_e was chosen to simulate the transverse magnetic (TM) mode [27].

Figures 3(c),(d) present the simulation results of the n_eff as a function of the top waveguide width (W) for TE₀, TE₁ and TM modes for Al_1−xSc_xN waveguides at x = 0.087 and 0.181 at λ = 1550 nm. The height of the waveguide is fixed at 400 nm. Light is better confined within the Al_1−xSc_xN waveguides as W increases, leading to an increase in the n_eff. When W is further increased, high-order modes will be generated. In addition, the n of Al_1−xSc_xN increases as x increases, which results in a smaller W required for the single mode behaviors. The W for single TE mode condition is simulated to be 0.84 μm when x = 0.087, whereas this value decreases to 0.78 μm at x = 0.181.

4. Device fabrication and characterization

To fabricate the Al_1−xSc_xN photonic devices, an i-line photolithography stepper was used to pattern the 400 nm-thick Al_1−xSc_xN thin films. After that, the Al_1−xSc_xN channel waveguides were formed by chlorine-based inductively coupled plasma (ICP) etching process. A 2 μm-thick SiO₂ cladding was deposited using plasma enhanced chemical vapor deposition (PECVD) after the photoresist stripping process. The three-dimensional (3D) schematic of the as-fabricated device is illustrated in Fig. 4(a). The SEM measurement results of critical dimension and a schematic diagram of the design layout are shown in Fig. S2 of Supplement 1. The 45-degree tilted SEM image of the waveguide is presented in Fig. S5 of Supplement 1.

 

Fig. 4. (a) 3D schematic of the fabricated Al_1−xSc_xN waveguide. The thickness of the Al_1−xSc_xN waveguide and the upper cladding are 0.4 μm and 2 μm, respectively. (b) Photograph of the 8-inch Al_1−xSc_xN photonic device wafer. (c) XSEM image of an Al_0.819Sc_0.181N waveguide. (d),(e) Optical microscopy images of devices to test the waveguide propagation loss and the bending loss, respectively.

Download Full Size | PDF

Figure 4(b) shows a photograph of the Al_0.913Sc_0.087N photonic devices on an 8-inch wafer. In general, the fabricated optical waveguides are not perfect rectangles due to the dry etching process. Figure 4(c) displays a cross-sectional scanning electron microscopy (XSEM) image of an Al_0.819Sc_0.181N waveguide. The waveguide has a sidewall angle of around 70°. As compared to the designed values, the top waveguide width W is reduced by approximately 70 nm for x = 0.087 and around 100 nm for x = 0.181 due to process errors. Figures 4(d),(e) show the optical microscopy images of the testing structures used for measuring waveguide propagation loss and bending loss. The spiral waveguides with lengths of 0.4, 0.8, 1.2 and 1.6 cm are employed to measure waveguide propagation loss. The structures shown in Fig. 4(e) are used to evaluate bending loss with the number of 90° bends being 6, 18, and 30. The design parameters and coupling loss for grating couplers are provided in Table S1 of Supplement 1. The simulated grating coupling efficiency as well as the impact of process error on it are presented in Fig. S3 of Supplement 1. The measured data of the grating coupler is shown in Fig. S4 of Supplement 1.

The propagation loss and bending loss of the channel waveguides were tested. A tunable laser in the Keysight 8164B Mainframe was used to emit the light. The light was passed through a polarization controller and was then coupled into the waveguide using a single mode optical fiber. The grating couplers used in the experiment were designed for the TE polarization mode. The optical power was obtained from the Keysight 8164B Lightwave Measurement System.

At λ = 1550 nm, the propagation losses of the Al_1−xSc_xN channel waveguides with x = 0.087 and x = 0.181 were obtained by the cut-back methods, as shown in Figs. 5(a) and 5(b), respectively. The W used in the figures are the designed widths rather than the actual widths of the fabricated waveguides. The lowest propagation losses for the TE single mode waveguides are 5.98 ± 0.11 dB/cm and 8.23 ± 0.39 dB/cm, respectively. The error bar represents the error in linear fitting. It can be observed that the propagation losses for x = 0.181 are higher than the ones for x = 0.087, which might be attributed to the larger surface and sidewall roughness of the Al_1−xSc_xN waveguides with a larger x as the uneven surfaces enlarge the scattering of the optical field. After optimizing the etching process, it is possible to suppress the propagation loss significantly [28]. The transmission spectrums and propagation losses for x = 0.087 and 0.181 in C band are shown in Fig. S6 and Fig. S7 of Supplement 1.

 

Fig. 5. (a),(b) The P vs L characteristics of the Al_1−xSc_xN waveguides at λ = 1550 nm for x = 0.087 and 0.181. The waveguide propagation losses were extracted by linear fitting. (c) Benchmarking of Al_1−xSc_xN waveguide propagation loss on Si substrates at λ = 1550 nm. (d),(e) The bending losses of the Al_1−xSc_xN waveguides with x = 0.087 and x = 0.181 extracted by linear fitting.

Download Full Size | PDF

Under TE single mode conditions, the propagation loss decreases as the waveguide width increases. When x = 0.087, considering the 70 nm process error, the designed maximum W is reduced from 0.9 μm to 0.83 μm. This value is smaller than the single mode condition of 0.84 μm obtained from the simulation. Consequently, under the TE mode at λ = 1550 nm, the structures remain in the single mode condition. The propagation loss decreases as W increases due to the smaller sidewall scattering effect.

However, when x = 0.181, the waveguide width of 0.9 μm exhibits the highest propagation loss. This might be due to that the higher mode is generated for the devices with W = 0.9 μm as the actual waveguide width is around 0.8 μm, which just exceeds the limit of the simulated single TE mode condition (0.78 μm) but is not enough to well confine the high-order mode. In this case, the scattering of high-order mode during propagation is large, resulting in increased propagation loss.

The optical propagation losses of Al_1−xSc_xN waveguides on Si substrates at λ = 1550 nm are benchmarked with the results from literature [13,29–31], as shown in Fig. 5(c). The propagation losses of AlN (x = 0) waveguides are also summarized for comparison. It can be observed that the average waveguide propagation loss for AlN is approximately 0.6 dB/cm [29–31], which is much smaller than those of the Al_1−xSc_xN waveguides.

The high propagation loss in Al_1−xSc_xN is caused by multiple factors, including scattering and absorption. Generally, a higher Sc concentration may lead to a larger optical propagation loss of the waveguide.

The scattering loss could be mainly attributed into three parts. Firstly, the abnormal grain growth, which is induced by complexion formation at crystallite interfaces, becomes more severe in Al_1−xSc_xN thin films with higher Sc concentration. This results in the formation of the abnormally oriented grains (AOGs) which enhance the scattering of light [32]. Secondly, the reduction in c-axis texture of the film with increasing Sc content may affect the propagation path of light within the crystal, making it more prone to deflection and scattering, further increasing the propagation loss [33]. Thirdly, in the fabrication process of the waveguide, the high hardness of Al_1−xSc_xN makes etching difficult, and the immaturity of the etching process leads to rough sidewalls, resulting in significant scattering loss.

In optical waveguides, the absorption characteristics of the material can cause gradual reduction of light power during propagation. Due to the imperfect growth process of Al_1−xSc_xN thin films, the introduction of Sc into AlN may introduce new defects or impurity levels, which can act as centers for light absorption, further increasing absorption loss.

To reduce waveguide losses, it is necessary to optimize the etching process such as adjusting the dry etching parameters and improving the selection of etchants. Additionally, optimizing the growth conditions of Al_1−xSc_xN thin films, including adjusting growth parameters and incorporating post-growth annealing treatments, which can enhance crystallinity and mitigate the impact of abnormal grain orientation.

In addition, it is observed that the propagation loss values from this work are lower than that from Ref. [13]. The propagation loss might be further suppressed by optimizing the thin-film deposition, lithography and dry etching process conditions, or by implementing high temperature annealing process after waveguide formation, which are subjects of our future works.

Furthermore, we have investigated the bending losses of Al_1−xSc_xN waveguides at λ = 1550 nm. Measurements were conducted on 90° bends with varying numbers of 50 μm radius in TE mode. The results are plotted in Figs. 5(d) and 5(e). It can be observed that the bending losses exhibit slight fluctuations for the waveguides with different W. In addition, the bending losses for x = 0.087 and x = 0.181 were measured to be 0.05 dB/turn and 0.08 dB/turn, respectively. The greater bending loss could be attributed to the increased scattering losses caused by the clustering of Sc and the rough surfaces of waveguides made from Al_1−xSc_xN with higher values of x. Although increasing the radius beyond 50 μm could further decrease the bending loss slightly, it will also increase the footprint of the photonic devices significantly. Therefore, we selected a bending radius of 50 μm for the actual design in this experiment. In the future, we will conduct systematic experimental studies on the bending loss as a function of bending radii.

5. Conclusion

In summary, the anisotropic optical properties of Al_1−xSc_xN films are investigated and Al_1−xSc_xN photonic devices are demonstrated on 8-inch Si wafer. The E_g, n and k of ordinary light and extraordinary light in Al_1−xSc_xN with various x are investigated for the first time. In addition, the propagation loss of Al_1−xSc_xN waveguides and the waveguide bending loss are characterized. The single mode waveguides with x = 0.087 and 0.181 exhibit propagation losses of 5.98 ± 0.11 dB/cm and 8.23 ± 0.39 dB/cm, respectively. The 90° bending losses with a radius of 50 μm are 0.05 dB/turn and 0.08 dB/turn at λ = 1550 nm, respectively, for x = 0.087 and 0.181. This work paves the way for large-scale integrated Al_1−xSc_xN photonic circuits on Si platforms.