The last decade has witnessed remarkable progress in the field of silicon photonics. Taking advantage of mature semiconductor manufacturing infrastructure, silicon-on-insulator (SOI) waveguides have become the basic component in complex, mass-produced photonic integrated circuits (PICs). An essential property for such high-level integration is low propagation loss: foundry-manufactured SOI waveguides can now achieve losses less than 1 dB/cm at telecom bands [1]. However, at wavelengths shorter than the silicon bandgap wavelength at 1100 nm, SOI waveguides become highly absorptive, prohibiting shorter wavelength PICs. This restriction is an obstacle for fields that require higher photon energies, such as atomic physics and augmented or virtual reality (AR/VR) displays. A promising solution to this problem is to use weakly confined silicon nitride (${\rm{S}}{{\rm{i}}_3}{{\rm{N}}_4}$) waveguides, which recently exhibited ring resonator quality factors over 55 M at visible red wavelengths [2]. However, no foundry-based such ${\rm{S}}{{\rm{i}}_3}{{\rm{N}}_4}$ devices have been reported at visible, and no high-Q devices have been demonstrated for blue (450–490 nm) or violet (400–450 nm) wavelengths. While a photonic platform in this regime could be used to access important atomic transitions (${\rm{S}}{{\rm{r}}^ +}/{\rm{Sr}}$ at 422 and 461 nm, ${\rm{Y}}{{\rm{b}}^ +}$ at 411, 436, and 467 nm, etc.) and support essential functions in commercial applications, low loss in the blue and violet spectrum is much more challenging to achieve compared to longer visible wavelengths: Rayleigh-like scattering increases as ${\lambda ^{- 4}}$, and material absorption increases rapidly as wavelengths approach the materials’ bandgaps. Consequently, the highest Q’s among blue and visible integrated optical resonators in previous integrated photonic platforms were all below 0.5 M.

In this work, we present a crucial step in silicon photonics by demonstrating large-scale PIC technology at blue and violet wavelengths. Using a 200-mm-diameter foundry process, we achieve quality factors up to 6 M at 453 nm, which is an order of magnitude higher than the previous blue wavelength record. Near 405 nm, we measure a record low loss of ${\lt}{1.0}\;{\rm{dB/cm}}$, putting violet wavelength performance on par with state-of-the-art SOI telecom waveguides. These demonstrations open the door to large-scale manufacturing of short-wavelength PICs.

Waveguides are patterned on a 200-mm wafer using a process similar to that in Ref. [3] in a CMOS foundry. A 24-nm-thick ${\rm{S}}{{\rm{i}}_3}{{\rm{N}}_4}$ layer is deposited on a thermally oxidized Si substrate. The upper and lower claddings comprised 2-µm and 6-µm ${\rm{Si}}{{\rm{O}}_2}$, respectively. The foundry process ensures that the waveguides have minimal surface roughness and high material quality. The high aspect ratio reduces mode overlap with the ${\rm{S}}{{\rm{i}}_3}{{\rm{N}}_4}$ and the sidewalls, minimizing scattering losses (as in Ref. [3]) and taking advantage of lower material loss in the cladding.

Compared to the ultrahigh-Q ${\rm{S}}{{\rm{i}}_3}{{\rm{N}}_4}$ resonators demonstrated at red [2], which used a centimeter-scale bending radius to reduce bending and scattering loss, the selection of thickness here enables devices with tight bends and drastically reduced device areas. Finite-element-method mode simulations indicate that bending losses do not play a significant role at these wavelengths. As shown in Fig. 1(b), geometry limited Qs remain above 100 M for wavelengths below 450 nm and radii as low as 200 µm, which is essential for manufacturing high-complexity PICs.

 

Fig. 1. Blue–violet CMOS ${\rm{S}}{{\rm{i}}_3}{{\rm{N}}_4}$ photonics. (a) Patterned 200-mm wafer. (b) Simulated bending-loss limited Q for blue ${\rm{S}}{{\rm{i}}_3}{{\rm{N}}_4}$ waveguides. (c), (d) High aspect ratio waveguide mode profiles. ${\alpha _b}$ represents the bending loss.

Download Full Size | PDF

To determine waveguide loss for violet light, cutback measurements were performed using a 405-nm Fabry–Perot (FP) diode laser (QPhotonics QFLD-405-20S). The propagation loss of the 0.8-µm-wide waveguide was extracted based on the transmission of spirals with different lengths, as shown in Fig. 2(a). A record-low value of 0.93 dB/cm was measured. To the best of the authors’ knowledge, this structure is also the only single-mode (TE only) waveguide demonstrated in the violet or blue wavelength regime. For a wider waveguide (2.8 µm), the propagation loss remains almost the same (0.9 dB/cm), suggesting that the sidewall roughness is not the primary loss mechanism.

 

Fig. 2. Measurement at blue and violet. (a) Cut-back measurement results for 0.8-µm-wide waveguide spirals of around 405 nm. (b) Measured transmission spectra of a high-Q mode at 453 nm. Photos of (c) spiral waveguide under test at 405 nm and (d) loaded ring resonator at 453 nm.

Download Full Size | PDF

For blue wavelengths, the cutback measurement was repeated on the same structures with a 450-nm FP diodes laser (QPhotonics QFLD-450-10S). The measured loss was four times lower compared to violet, which is likely due to lower material absorption. Further, the waveguide losses are highly geometry dependent: cutback measurements of around 450 nm show that the loss increases from 0.22 dB/cm to 0.35 dB/cm when the waveguide width changes from 2.8 µm to 0.8 µm. This increase suggests that the sidewall roughness is a significant contributor to loss at this wavelength.

Ring resonators were also characterized using a tunable narrow-linewidth laser (Toptica DL Pro). Fig. 2(b) shows the transmission spectrum of a resonator with 1531-µm radius (20-GHz free spectral range) and 5-µm-wide waveguide width. Loaded Qs of 5.85 M were measured, with intrinsic Q reaching 6.0 M, which is one order of magnitude higher than previously reported results.

Figure 3 shows a comparison of reported high-Q resonator results for wavelengths below 480 nm. Notably, previous results displayed have all been achieved on non-CMOS platforms such as AlN [10] and ${\rm{A}}{{\rm{l}}_2}{{\rm{O}}_3}$ [12]. Thus, the platform reported here offers both superior performance and greater potential for large-scale manufacturing.

 

Fig. 3. Blue–UV Q comparison. Comparison of blue–UV resonator platforms [4–12]. The ${\rm{S}}{{\rm{i}}_3}{{\rm{N}}_4}$ ${{\mu}}$ ring Q at 405 nm is a conservative estimate based on loss.

Download Full Size | PDF

In conclusion, we have shown compelling results for blue–violet silicon photonics. A low-loss ${\rm{S}}{{\rm{i}}_3}{{\rm{N}}_4}$ waveguide platform was demonstrated by leveraging a CMOS foundry process. Such a demonstration is a milestone for silicon photonics; the unprecedented low loss and relatively tight bending radius combined with the high-volume production capability will open up new opportunities in atomic physics, sensing and other commercial applications. We expect that by combining this platform with hybrid/heterogeneous integration approaches and implementing nonlinear optical capabilities for frequency conversion, PICs at blue and violet with high complexity and versatile functions can be realized in the future.