1. Introduction

Phase shifter, which enables electrically tuning the effective refractive index of a waveguide, is an essential component for photonic circuits. For deposited photonics such as Si₃N₄, the thermo-optic (TO) effect is commonly utilized by employing an external metal heater above the waveguide. However, the TO modulation is inherently power hungry and slow [1]. Moreover, a relatively thick cladding layer is required in order to reduce the thermal crosstalk, making it difficult to integrate multiple photonic layers on silicon.

Recently, deposited aluminum nitride (AlN) has emerged as an alternative material for deposited photonics because of its excellent properties such as low optical loss over broad wavelength spectrum, similar refractive index as Si₃N₄, small stress, and CMOS compatibility [2–5]. Moreover, it offers large piezoelectric effect and nontrivial Pockels and Kerr coefficients, thus enabling for realization of various active optical devices [4–7]. Especially, an AlN ring modulator utilizing its Pockels effect has been demonstrated [4, 5]. However, the reported modulator has both electrodes placed above the AlN waveguide. Such a co-planar electrode structure requires a non-planar process for fabrication and the electro-optic overlap is relatively small because the maximum electric field created by a voltage applied on the electrodes does not go through the AlN waveguide.

Here, a straightforward structure is proposed as shown in Fig. 1(a) schematically. It is a standard parallel plate capacitor by introducing a thin metallic bottom electrode such as TiN below the AlN waveguide. Such a vertical double plate electrode structure can be fabricated by the standard planar complementary metal-oxide-semiconductor (CMOS)-compatible process and is feasible for integration of multiple photonic layers on Si. Upon applying a voltage, an almost constant electric field parallel to the y-axis (i.e. $E\approx E_y$) is created between the bottom and upper electrodes, as shown in Fig. 1(b). The electric field inside the AlN waveguide equals approximately to $E_y\approx V/[(t_{upper}+t_{bottom})\times {\epsilon }_{AlN}/{\epsilon }_{SiO2}+t_{AlN})]$, where V is the applied voltage, t_upper, t_bottom, and t_AlN are thicknesses of the upper SiO₂, the bottom SiO₂, and the AlN core, respectively, and ε_SiO2 ( = 3.9) and ε_AlN (≈10 [8]) are dielectric constants of SiO₂ and AlN, respectively. The electro-optic overlap approximately equals to the ratio of optical power confined in the AlN core.

 

Fig. 1 (a) Schematic AlN EO phase shifter, (b) the y-component electric field (E_y) distribution along y-axis, and (c) the calculated E_y at 40-V bias and optical absorption due to the electrodes for 1550-nm TE and TM modes versus the cladding SiO₂ thickness (t_upper = t_bottom) for t_AlN = 0.4 µm.

Download Full Size | PDF

To reach a large electric field at a given voltage, one need reduce the cladding SiO₂ thickness. However, it will increase the propagation loss because the metallic electrode is usually a highly lossy material. For a typical 1-µm × 0.4-µm AlN waveguide, Fig. 1(c) plots the calculated E_y at 40-V bias and the metal-induced propagation loss as a function of the cladding SiO₂ thickness. With the tradeoff between the electric field and metal-induced loss, the optimized cladding SiO₂ thickness is ∼1.4 µm for the transverse electric (TE) mode and it is >2 µm for the transverse magnetic TM mode because the TE mode can be confined within the AlN core more tightly than the TM mode.

2. Fabrication and measurement

The proposed devices were fabricated on Si using the standard CMOS back-end process. The starting were SiO₂-covered Si wafers. First, 120-nm TiN and 50-nm Si₃N₄ were deposited and patterned to form the bottom electrode, followed by 2-μm SiO₂ deposition and chemical mechanical polishing (CMP) to planarize the surface. Second, 400-nm AlN was deposited by sputtering using a pure Al target, followed by 200-nm SiO₂ deposition. The AlN layer was patterned by dry etch down to the SiO₂ layer by using the 200-nm SiO₂ layer as the hard mask. Third, 2-μm SiO₂ was deposited, followed by CMP to planarize the surface. After opening the contact holes to the TiN electrode, 2-µm Al was deposited and patterned. Here, the upper electrode is Al itself for simplification. For integration of multiple photonic layers, the upper electrode can be also a thin TiN layer. Since the standard planar processes with temperature not exceed 400°C were used, it is feasible to fabricate other photonic layer either beneath or above this AlN photonic layer for multilayer photonics integration. Moreover, some wafers before the Al deposition were additionally annealed at temperatures ranging from 500 to 1000°C for 5 min. The TiN/SiO₂/AlN structure remains stable up to 900°C annealing, whereas the 1000°C anneal damages the TiN film seriously. To converter phase variation into intensity modulation, both add-drop rings and asymmetric MZI modulators were fabricated, as shown in Figs. 2(a) and 2(b) respectively. Figure 2(c) shows the cross sectional transmission electron microscopy (XTEM) image of the fabricated phase shifter. The cladding SiO₂ thickness is ∼1.5 µm. Figure 2(d) is a zoom-in XTEM image of the AlN core. One sees it has a columnar micro-grain structure with c-axis orientation, in consistence with our previous X-ray diffraction (XRD) measurement.

 

Fig. 2 (a) Add-drop waveguide ring resonator modulators with radius of 40 µm, (b) Asymmetric MZI modulators with the arm length of 1.4 cm, (c) XTEM of the fabricated device, (d) Zoom-in XTEM of the AlN waveguide, and (e) and (f) numerically calculated TE and TM modes at 1550 nm.

Download Full Size | PDF

Numerical simulation shows that the effective mode index (n_eff) of the fabricated AlN waveguide at 1550 nm is 1.62 for TE and 1.54 for TE, as shown in Figs. 2(e) and 2(f) respectively. The electrode-induced absorption is calculated to be 0.1 dB/cm for TE and 5.6 dB/cm for TM. The dn_eff/dn_AlN value is calculated to be 0.7 for TE and 0.43 for TM, close to the ratio of optical power confined in the AlN core. Therefore, the fabricated devices are more suitable for the TE mode. For usage at TM mode, the AlN thickness should be increased.

After dicing, the chips were measured using the conventional fiber-waveguide-fiber method by edge coupling with inverted couplers. Devices with and without the additional annealing were measured. They exhibit similar performances. Therefore, only the results measured from the devices without additional annealing are presented here. The results for the TM mode are also presented to extract the AlN material properties at the TM mode although its performance is poor due to the large metal-induced loss.

3. Results and discussion

The propagation losses are measured from the straight waveguides with different lengths using the cutback method. The results are plotted in Figs. 2(a) and 2(b) for the waveguides without the electrodes. The propagation loss is ∼2.0 dB/cm for TE and ∼3.3 dB/cm for TM, close to the previous results [2]. The waveguides with the electrodes exhibit similar loss for TE whereas much larger loss for TM due to the electrode-induced absorption. Figures 3(c) and 3(d) plot spectra measured from add-drop ring resonators at the TE and TM modes, respectively. The Q-value is ∼1.1 × 10⁴ for TE and ∼2.0 × 10³ for TM. The group indices (n_g) are extracted to be 2.33 for TE and 2.12 for TM. The insertion loss (excludes the coupling loss, which is ∼2 dB per facet) is ∼0.5 dB for TE and ∼5 dB for TM.

 

Fig. 3 (a) Output power vs. waveguide length measured from waveguides, (b) propagation loss vs. wavelength, (c) TE transmission spectra measured from a ring with R of 40 µm, and (d) TM transmission spectra measured from a ring with R of 100 µm.

Download Full Size | PDF

When a voltage is applied, the spectrum of the ring shifts, as shown in Figs. 4(a) and 4(b) for TE and TM respectively. The shift of the resonant wavelength (λ_r) depends on voltage almost linearly from −40 V to + 40 V, as shown in Fig. 4(c). The dλ_r/dV value is deduced to be ∼0.26 pm/V for TE and ∼0.14 pm/V for TM, which corresponds to dn_g/dV of ∼3.9 × 10⁻⁷ V⁻¹ for TE and ∼2.0 × 10⁻⁷ V⁻¹ for TM.

 

Fig. 4 (a) Transmission spectra of TE measured on a 100-μm-R ring modulator under −40, 0, and 40 V bias, (b) Transmission spectra of TM measured on a 100-μm-R ring modulator under −40 and 40 V bias, and (c) the extracted resonant wavelength vs. voltage for TE and TM modes.

Download Full Size | PDF

Figures 5(a) and 5(b) plot spectra measured from an asymmetric MZI modulator with 1.4-cm long arm under −40, 0, and 40 V for the TE and TM modes. The TE spectrum exhibits typical behavior as it shifts almost linearly with the voltage increasing from −40 V to 40 V, from which the product of V_π⋅L_π is extracted to be ∼240 V⋅cm. The insertion loss (excludes the coupling loss) is ∼3 dB, which is contributed by the waveguide loss and the splitting/combining loss. The TM spectrum exhibits abnormal behavior because of the large electrode-induced loss. The measured insertion loss is ∼8 dB. Nevertheless, we can still estimate the product of V_π⋅L_π to be ∼320 V⋅cm approximately from the voltage-induced spectrum shift. The dn_eff/dV value is estimated from the V_π⋅L_π values to be ∼3.2 × 10⁻⁷ V⁻¹ for TE and ∼2.4 × 10⁻⁷ V⁻¹ for TM, which is consistent with that extracted from the ring resonators after taking the difference in n_eff and n_g into account. The dn_AlN/dV value can be extracted by taking the calculated dn_eff/dn_AlN value into account. Both TE and TM modes give a similar dn_AlN/dV value of ∼5.0 × 10⁻⁷ V⁻¹. Based on the equation $\Delta n_{AlN}\approx -0.5r_{13(or33)}{n}_{AlN}{}^{3}E$, where r₁₃ (r₃₃) is the AlN’s Pockels coefficient for TE (TM) mode and E is the electric field within the AlN waveguide (which can be read from Fig. 1(b)), the r₁₃ and r₃₃ values can be extracted. Our experimental results show that the deposited AlN has similar r₁₃ and r₃₃ values to be ∼1.0 pm/V, and they are stable up to 900°C annealing.

 

Fig. 5 Transmission spectra measured an asymmetric AlN MZI modulator under −40, 0, and 40 V bias (a) TE and (b) TM.

Download Full Size | PDF

It is expected that the EO phase shifter offers high speed because it is only limited by the resistance-capacitance (RC) delay. To estimate the transient property, a 0–80-V pulsed voltage with 1 MHz frequency is applied on the MZI modulator. The input light is a 1549-nm TE light from a tunable laser source and the output light is recorded by a photodetector as a function of time. The result is plotted in Fig. 6. Both the rise time (τ_r) and fall time (τ_f) is read to be ∼0.005 µs, then, the cutoff frequency defined as $f\approx 1/[2\pi \times Max({\tau }_r,{\tau }_f)]$ is estimated to be ∼30 MHz. However, this speed is mainly limited by the switching time of our slow voltage supplier. It is expected that the fabricated MZI modulator should have the cutoff frequency much larger than the above apparent value.

 

Fig. 6 Output waveform measured from by an AlN MZI modulator with 1549-nm TE input light and under 0–80-V pulsed voltage bias.

Download Full Size | PDF

The main issue of the present AlN phase shifter is its low modulation efficiency due to the relatively small Pockels coefficient of AlN. A method to improve the modulation efficiency is to increase the Pockels coefficient of AlN. It is argued that the Pockels coefficient of AlN may depend on its crystallinity degree [9]. Therefore, the Pockels coefficient may be improved by optimizing the AlN deposition process such as additional high temperature annealing and/or adding a thin buffer layer. Since our experimental result shows that 900°C annealing does not improve the Pockels coefficient, the annealing temperature for the possible crystallinity improvement needs to be larger than 900°C.

The other method is to increase the electric field inside the AlN waveguide at a given voltage. From Fig. 1(b), we can see that the electric field within the AlN waveguide is much smaller than that within the cladding SiO₂ layer because AlN has much larger dielectric constant than SiO₂. Therefore, one solution is to adopt a dielectric with large dielectric constant but small refractive index as the cladding dielectric. One possible candidate is Al₂O₃, which has dielectric constant of ∼10 [10] and refractive index of ∼1.7 [11]. Another possible solution is to employ a subwavelength grating (SWG) structure for the cladding layer [12]. The optical mode can be isolated by such a SWG structure whereas the overall equivalent oxide thickness (EOT) can be reduced so that the electric field within the AlN waveguide at a given voltage can be increased significantly.

4. Conclusion

In conclusion, an AlN EO phase shifter with a parallel plate capacitor structure is demonstrated on Si using the planar process, which is suitable for multilayer integration. Some new evidences are experimentally clarified: (1) the deposited AlN has similar Pockels coefficient of ∼1.0 pm/V for TE and TM modes, (2) the Pockels coefficient of the deposited AlN keeps stable up to 900°C annealing, and (3) the refractive index of AlN depends on the applied voltage almost linearly over a large range at least from −40 V to 40 V. Although the present phase shifter suffers from low modulation efficiency due to the relatively small Pockels coefficient, approaches for further improvement exist. Owing to the extremely low DC power consumption, high modulation speed, and feasibility for multilayer integration, the proposed phase shifter will play a key role for deposited photonics.