Over the past decades, photonic integrated circuits (PICs) have attracted widespread research interest due to their advantages of scalability, low power consumption, and potential for cost-effective mass production [1,2]. PICs hold promise to achieve chip-scale systems with integrated functionalities. Electro-optic (EO) modulators that manipulate the amplitude or phase of light carriers with electrical signals are one of the key components in PICs [3], finding broad applications in high-speed optical communications [4], optical interconnections [5], quantum information processing [6], computing [7], and so on. The escalating demand in these fields has propelled the development of high-performance EO modulators.

Silicon-integrated EO modulators have been widely reported and can be categorized as follows: (i) silicon (Si) modulators exploiting the plasma dispersion effect [8,9]—they often require complex doping architectures and additional thermo-optic tuning heaters while confronting strict modulation speed, efficiency, and carrier absorption trade-offs; (ii) two-dimensional (2D)-material-based modulators (e.g., graphene) relying on the Pauli-blocking effect [10]—due to the complexity of material transfer techniques, wafer-level fabrication is currently not realized and results in substantial cost increase; (iii) ferroelectric oxides [e.g., lithium niobate (LN) and barium titanate (BTO)] modulators utilizing the Pockels effect [11–15]—compared with the plasma dispersion effect, the Pockels effect offers ultrafast, linear, and pure real refractive index modulation across a broad wavelength range without additional absorption losses. Therefore, LN and BTO modulators have been extensively studied in recent years. Unfortunately, lithium niobate-on-insulator (LNOI) thin films on 8-inch silicon substrates realized by the smart-cut method are costly. The preparation conditions for single-crystal BTO thin films on insulator substrates are even more stringent. The prevailing molecular beam epitaxy (MBE) technology suffers from low deposition rates. Furthermore, BTO’s Curie temperature is only 125°C, which would bring potential process difficulties and make it unsuitable for high-temperature operation. Finding a cost-effective EO material suitable for large-scale integration remains a challenge.

Aluminum nitride (AlN) thin films on the insulator substrate fabricated by a mature complementary metal-oxide-semiconductor (CMOS)-compatible sputtering process also exhibit the Pockels effect. This renders them a compelling EO material platform, offering cost-effective alternatives to LN and BTO. Waveguide-based AlN modulators operating in the communication band have been experimentally demonstrated showing an EO coefficient of 1 pm/V and a modulation speed of 4.5 Gb/s [16]. A recent study has experimentally verified that the optical second-order susceptibility (χ(2)) components d13 and d33 of scandium-doped aluminum nitride (AlScN) thin films increase significantly with increasing Sc concentration [17]. Specifically, the d33 of Al0.9Sc0.1N is enhanced by a factor of 3, while that of Al0.64Sc0.36N is increased by a factor of 12 compared to undoped AlN films. Since the EO coefficient rijk is proportional to the χ(2) [18], the Pockels effect is predicted to be enhanced by about 2.7 and 10 times for scandium-doped aluminum nitride films with concentrations of 10% and 36%, respectively. AlScN maintains a relatively wide bandgap [19], ensuring a broad transparency window. It is also compatible with various substrates such as silicon [20,21], silicon dioxide [22], and sapphire [19], rendering it a promising platform for integrated photonic applications. Table 1 in Appendix A lists the detailed comparison of some common EO materials with AlScN, specifically parameters related to high-frequency response, power handling capability, and stability of the modulator.Table 1.

Comparative Analysis of the AlScN with Several Common EO Materials

In this work, we present EO modulators on the AlScN-on-insulator (AlScNOI) platform. Micro-ring resonators (MRRs) are demonstrated to achieve effective EO modulation while minimizing footprint and reducing power consumption. The AlScN films serving as the light-conducting medium are fabricated using a CMOS-compatible sputtering process, providing the feasibility of large-scale manufacturing of EO modulators. The in-device effective EO coefficient of the AlScN is measured to be −1.1pm/V under direct current (DC) conditions and 2.86 pm/V under high-frequency conditions (12 GHz), and the measured EO bandwidth is up to 22 GHz. Our findings show the potential of AlScN-based modulators for realizing nonlinear optical applications.

A critical initial step in fabricating an integrated AlScN modulator is to prepare an AlScN thin film with high-quality and well-aligned crystal orientation, since the Pockels effect strongly relies on the relative alignment of the applied electric field, the optical polarization, and the crystal orientation. Using radio frequency magnetron reaction sputtering technology [23], 400-nm-thick AlScN thin films (with an ordinary refractive index no=2.13 and an extraordinary refractive index ne=2.2 at 1550 nm) are deposited on a thermally grown silicon dioxide substrate. The atomic percentages of Al and Sc are 90.4% and 9.6%, respectively. In our recently published work [24–26], a symmetric (2θ=θ) scan of X-ray diffraction (XRD) for the same AlScN film confirms the deposition of the AlScN film along the (0002) crystal orientation, with its c-axis normal to the substrate surface. The sharp rocking curve further ensures consistency in the c-axis orientation of the AlScN film. This c-axis-oriented film ensures the maximum utilization of the EO coefficients r33 and r13.

The modulator consists of grating couplers for coupling light into and out of the chip and MRRs with a quality factor (Q) of ∼10,000 to balance modulation efficiency and EO bandwidth. A fully etched single-mode waveguide design is employed to maximize the electro-optic interaction. As illustrated in Fig. 1(a), a single transverse-electric (TE0) optical mode is well confined in the AlScN waveguide with a thickness of 400 nm and a typical width of 1 μm. The calculated effective refractive index neff at 1550 nm is 1.73. The waveguide patterns are defined using electron beam lithography (EBL) and etched with a Cl2/Ar-based inductively coupled plasma (ICP) reactive ion etching (RIE) process (Appendix B).

Figure 1(b) presents an optical micrograph of the fabricated device. The scanning electron microscope (SEM) image in Fig. 1(c) displays a sidewall angle of 76° for the etched waveguide cross-section. The propagation loss of the waveguide is evaluated by the cutback method. Spiral waveguides of different lengths (0.5 cm, 1.2 cm, and 1.6 cm) are fabricated. Figure 1(d) presents the transmission spectra of each spiral waveguide in the wavelength range from 1520 nm to 1580 nm. Through multiple measurements and the error bar method, the averaged fitted curve at 1550 nm is presented in Fig. 1(e), revealing a propagation loss of 7.45±0.21dB/cm. The loss is mainly attributed to rough sidewalls and Sc-based non-volatile compounds produced during the etching process [27], which can be significantly reduced by optimizing the etching recipe. The intercept of the fitted curve in Fig. 1(e) is −19.01±0.26dB, attributed to the coupling losses from a pair of grating couplers and the bending losses of the spiral waveguides.

Using the refractive index ellipsoid method to describe the Pockels effect [28], the applied electric field must be parallel to the c-axis of the AlScN film to extract the maximum EO coefficients r33 and r13 [28,29]. Therefore, the electrodes are designed to be directly above the waveguide, with ground electrodes (GNDs) located on both sides of the waveguide to maximize the utilization of the out-of-plane electric field (Ez). As shown in the fabrication process flow in Appendix B, the metal electrodes are fabricated by two steps of lift-off processes on the SiO2 cladding to ensure the effective application of the electric field to the device via ground-signal-ground (GSG) probes.

Light from the tunable laser (Keysight 8164B) is coupled into/out of grating couplers via polarization-maintaining (PM) fibers for characterizing the transmission properties of the MRRs. Figure 2(a) displays the transmission spectrum of the add-drop ring with a diameter of 80 μm. Coupling gaps are 250 nm for the pass port and 300 nm for the drop port. The free spectral range (FSR) is ∼4.4nm; the maximum extinction ratio (ER) is 30 dB. The loaded Q value is determined to be 7781 through Lorentz fitting, as depicted in Fig. 2(b). Similarly, Fig. 2(c) illustrates the transmission spectrum of an all-pass ring with a diameter of 200 μm, a gap of 150 nm, and an FSR of ∼1.7nm. It exhibits a maximum ER of 25 dB, with a loaded Q of 20,305, as depicted in Fig. 2(d).

When the doping concentration of Sc is 9.6%, the crystalline phase of AlScN exhibits a hexagonal wurtzite structure similar to AlN, both possessing non-centrosymmetric crystal structures. The electro-optic tensor of AlScN has only five non-zero components: r23=r13, r33, r42=r51. The ellipsoid equation of the AlScN crystal after applying the electric field E can be simplified as (1no2+r13Ez)x2+(1no2+r13Ez)y2+(1ne2+r33Ez)z2+2r42Eyyz+2r42Exxz=1,(1)where Ex and Ey represent the in-plane components of the electric field, and Ez denotes the out-of-plane component. The values of EO coefficients r13 and r33 in AlN are similar, while r42 is observed to be quite small. To extract the r13 and the r33 of AlScN, the applied electric field must be parallel to the c-axis of the AlScN film, i.e., Ez=E and Ex=Ey=0. In this case, the change of the refractive index with respect to E can be expressed as $$\begin{gathered} Δnx,y=−12no3r13E, \\ Δnz=−12ne3r33E. \end{gathered}$$(2)nx,y denotes the effective refractive index of TE-polarized optical modes transmitted in the waveguide, while nz is the effective refractive index for TM modes.

The change in waveguide effective refractive index due to the Pockels effect is manifested as a linear shift of the resonance dip. As shown in Fig. 2(e), when the applied DC voltage is swept from 0 V to 30 V, the resonance wavelength is shifted by ∼5.2pm from the initial wavelength (λ0) of 1544.09 nm [Fig. 2(e)]. The voltage-induced effective refractive index change Δneff is calculated from the measured resonant wavelength drift Δλ by applying a DC voltage to the micro-ring modulator (Δneff/ng=Δλ/λ0), where ng is the group index. Then the in-device effective EO coefficient of the AlScN waveguide at the operating wavelength is derived to be −1.1pm/V from r13,eff=−2neffΔneff/(no4∬WEzEx,op2dxdz/∬HEx,op2dxdz). Here, Ex,op is the optical field, and Ez is the vertical RF field under the modulation voltage. The integral area W represents the field over the waveguide region, and the integral area H is the entire modulation region [30–32]. The simulated optical mode profile and electric field distribution are obtained through the finite element method (FEM). The oxide cladding tox is fixed to 1.8 μm, the electrode spacing g between the signal and ground electrodes is set to 1.6 μm, and the signal width above the waveguide is set to 0.2 μm wider than the bottom width of the waveguide (Appendix C). Our measurements indicate that the effective EO coefficient of Al0.904Sc0.096N is comparable to AlN (r33=r13=1pm/V), which is smaller than the value of 2.7 pm/V predicted by the linear relationship between optical second-order susceptibility d and EO coefficient r [17,18].

The high-frequency response of the fabricated modulators is analyzed utilizing the sideband measurement technique. Sinusoidal RF signals of different frequencies are applied to the device, and the first-order modulation sidebands can be observed at corresponding frequencies separated from the carrier [33]. By measuring the power ratio of the main peak to the first sideband, the modulation index m can be extracted using Jacobi-Anger expansion of the modulated optical electrical field [34,35]. In the case where m≪1, its zeroth-order and first-order Bessel functions can be approximated as J0(m)∼1 and J1(m)∼m/2, respectively. Then, the modulation index m can be extracted as [J0(m)J1(m)]2=(2m)2.(3)

Here, effective intensity modulation is achieved when the operating wavelength is chosen at a non-resonant point. MRRs feature a Lorentzian transfer function T(θ), where θ denotes the round-trip phase shift. By selecting the operating wavelength at its maximum slope, the half-wave voltage Vπ can be effectively decreased by a factor of 1/(2|dT/dθ|max) compared to a single-arm-driven Mach–Zehnder (MZ) modulator with the same EO interaction length L. The optimal operating wavelength of the micro-ring modulator can thus be determined. Then, we calculate the value of 2|dT/dθ| by fitting the measured transmission profile, and subsequently extract the effective EO coefficient and Vπ using Eq. (4) [36–38]: m=πVppVπ=2πno4r13,effL|dT/dθ|(∬WEzEx,op2dxdz/∬HEx,op2dxdz)λneff.(4)

The experimental setup for sideband measurements is illustrated in Fig. 3(a). RF signals generated by an arbitrary waveform generator (AWG, Keysight M8194A) are amplified by an RF amplifier (AMP) before being applied to the device. The optical signal is directly coupled to the device from a tunable laser through an external polarization controller and a PM fiber. The modulated optical signal is then fed into an OSA (Yokogawa AQ6370D), where the optical power of the sidebands observed at each RF frequency is recorded. To ensure the accuracy of measuring the EO response of the device, we calibrate the cables and maintain a relatively constant amplitude Vpp of the applied RF signal across different frequencies.

For evaluating the high-frequency EO performance of micro-ring modulators with various structures, two configurations are measured: an all-pass-type micro-ring (referred to as device A) and an add-drop-type micro-ring (referred to as device B), each having a radius of 40 μm. By applying RF signals ranging from 12 GHz to 30 GHz with the same Vpp of 6.32 V to both devices, as shown in Fig. 3(b) and Fig. 3(c), several pairs of sidebands are sequentially spaced with respect to the optical carrier, clearly demonstrating the generated intensity modulation. The effective in-device EO coefficient at 12 GHz for devices A and B is r13,eff=2.47pm/V and r13,eff=2.86pm/V, respectively. Furthermore, according to first-principles calculations, an increase in Sc concentration is predicted to yield a 2.5-times enhancement in the intrinsic EO coefficient of single-crystal Al0.7Sc0.3N compared to Al0.904Sc0.096N (Appendix D), which is expected to enhance the modulation efficiency.

Notably, EO coefficients measured at DC are slightly different from those measured at high frequencies. There are two factors that may contribute to the observed phenomena. The first is the shielding effect. Defects in the plasma-enhanced chemical vapor deposition (PECVD)-SiO2 cladding trap charges, altering the charge distribution of charges within the material and affecting the distribution of charges in the surrounding space. This alteration can lead to the shielding effect on charges located outside the oxide layer and impact the DC response of devices [39]. Second, there is a strong correlation between the EO coefficient and the frequency of the applied electric field. In the frequency range from DC to the first mechanical resonance (∼10MHz), the crystal is unclamped (stress-free). In this condition, the electronic, ionic, and piezoelectric responses jointly affect the value of the EO coefficient. At high frequencies (106MHz>f>10MHz), the crystal is clamped (strain-free), and the influence of the piezoelectric response dissipates, leading to frequency-dependent variations in the EO coefficients [40,41]. It is noteworthy that some of the contributions to the overall EO coefficient have opposite signs and may work counter to one another.

Then, the RF modulation efficiencies Vπ·L of the micro-ring modulators are analyzed. Four devices with different structures are selected and measured: device A [all-pass (AP), radius=40μm, Q=15,696, ER=15dB], B [add-drop (AD), radius=40μm, Q=19,580, ER=27dB], C (AP, radius=100μm, Q=21,410, ER=14dB), and D (AD, radius=100μm, Q=18,641, ER=24dB). As depicted in Fig. 4(a), within the measurement range of 12–30 GHz, device B shows a minimum Vπ·L of 3.12V·cm at 14-GHz modulation frequency and the minimum Vπ·L of devices A, C, and D is 8.81, 17.37, and 14.6 V · cm, respectively. The enhancement in modulation efficiency can be explained through straightforward metrics. Micro-ring modulators with larger ER and Q (i.e., larger |dT/dθ|) tend to possess higher modulation efficiency. This is due to the fact that a modulator with larger |dT/dθ| can generate a small resonant offset sufficient to achieve the desired modulation depth with a relatively low driving voltage. Also, high Q values promote adequate electro-optic interactions in modulators, thus yielding high modulation efficiencies. It is worth noting that devices C and D have longer modulation lengths than devices A and B, leading to a more significant decline in their modulation efficiencies at higher frequencies. This can be attributed to the larger cavity, resulting in an increased capacitance that needs to be driven by the applied source. To address this issue, potential improvements can be achieved through optimizing the design of traveling-wave electrodes and making slight increasement to electrode thickness. Additionally, modulation efficiency degrades with increasing RF frequencies across all four devices, as the ohmic loss becomes more severe at high frequency, reducing the voltage drop across the modulation region.

The 3-dB EO bandwidth of our micro-ring modulators is determined by fitting the sideband powers at various applied RF frequencies starting from 12 GHz [38,42–44], primarily constrained by the resolving accuracy of the optical spectrum analyzer (OSA). Figure 4(b) depicts the EO frequency response of device A under applied Vpp of 3.99 V and 2.62 V. The 3-dB EO bandwidths are approximately 22 GHz. Generally, the EO bandwidth can be estimated using the simplified models: 1/f=1/fRC+1/ft. The first term is relative to the RC time constant of the device and the second term is influenced by the lifetime of photons in the cavity. Attributed to the short modulation length and small capacitance of the lumped-element micro-ring modulator, the cutoff frequency is mainly limited by the photon lifetime, expressed as f3dB=ft≈c/λQ, thereby restricting the maximum achievable Q-factor.

Here, optical peaking is experimentally observed, as shown in Fig. 4(b), which helps to extend the EO bandwidth beyond the photon lifetime limitation. The peaking phenomenon is related to transient responses in the optical domain and can be explained using a small-signal model based on perturbation theory [45–47]. The voltage applied to the micro-ring modulator induces changes in the refractive index, which further causes phase shifts. With the phase shift accumulating sufficiently, it switches from destructive to constructive interference between the resonant light circulating in the cavity and the input modulated light from the waveguide. This transient response leads to an overshoot in the output power, which can be observed as the peaking effect.

The modulation frequency at which the peaking occurs is related to the detuning between the resonant frequency of the cavity fr and the optical carrier frequency f0. When the detuning f0−fr increases, the peaking occurs at higher modulation frequencies, resulting in a higher EO bandwidth (Appendix E). However, this enhancement comes at the cost of modulation efficiency; the EO response power decreases as the detuning increases, necessitating the use of amplifiers with higher gain in the test link. Therefore, we fixed a detuning of 0.195 nm, as this wavelength point yields the maximum |dT/dθ| for device A. As illustrated in Fig. 4(b), with the increase of Vpp, the overall power of the EO response shifts upwards. This phenomenon is attributed to a larger modulation voltage inducing a greater frequency shift in the resonator, which in turn amplifies the perturbation amplitude within the cavity. All curves were mathematically fitted, showing agreement with theoretical predictions (Appendix E).

In summary, we demonstrate the EO micro-ring modulator based on the AlScN-on-insulator platform. At a modulation frequency of 14 GHz, the device exhibits an RF modulation efficiency Vπ·L of 3.12V·cm. Moreover, the optical response of the modulator decays by 3 dB at a modulation frequency of 22 GHz. Leveraging the key advantages of AlScN thin films, which are the high-quality wafer-level deposition achieved through a low-cost fabrication process, fully CMOS-compatible AlScN modulators are poised to emerge as promising candidates for electro-optic signal processing on silicon-integrated photonics platforms.