Broadband couplers for hybrid silicon-chalcogenide glass photonic integrated circuits

: We report on the design, fabrication and testing of three types of coupling structures for hybrid chalcogenide glass Ge 23 Sb 7 S 70 -Silicon (GeSbS-Si) photonic integrated circuit platforms. The first type is a fully etched GeSbS grating coupler defined directly in the GeSbS film. Coupling losses of 5.3 dB and waveguide-to-waveguide back-reflections of 3.4% were measured at a wavelength of 1553 nm. Hybrid GeSbS-to-Si butt couplers and adiabatic couplers transmitting light between GeSbS and Si single-mode waveguides were further developed. The hybrid butt couplers (HBCs) feature coupling losses of 2.7 dB and 9.2% back-reflection. The hybrid adiabatic couplers (HACs) exhibit coupling losses of 0.7 dB and negligible back-reflection. Both HBCs and HACs have passbands exceeding the 100 nm measurement range of the test setup. GeSbS grating couplers and GeSbS-to-Si waveguide couplers can be co-fabricated in the same process flow, providing, for example, a means to first couple high optical power levels required for nonlinear signal processing directly into GeSbS waveguides and to later transition into Si waveguides after attenuation of the pump. Moreover, GeSbS waveguides and HBC transitions have been fabricated on post-processed silicon photonics chips obtained from a commercially available foundry service, with a previously deposited 2 μ m thick top waveguide cladding. This fabrication protocol demonstrates the compatibility of the developed integration scheme with standard silicon photonics technology with a complete back-end-of-line process.

nonlinearity, which are very important features for all optical signal processing and wavelength conversion [5]. However, these applications are limited by the large two-photon absorption (TPA) of Si [6]. One way to overcome this problem is to integrate chalcogenide glasses into the SOI platform. Chalcogenide glasses, which contain one or more of the chalcogen elements including S, Se and Te and are covalently formed with network formers such as Ge, Sb or As, have large third order nonlinearities and a much smaller TPA coefficient than that of Si [7]. Thanks to their remarkable optical properties such as large and tailorable refractive index, low loss, broad transparency region, and high optical nonlinearity, chalcogenide glasses have been demonstrated to be promising materials for a broad range of applications such as sensors [8,9], unconventional substrate integration [10,11], and alloptical signal processing [12]. Rekindled interest in nonlinear-based transmission by means of the nonlinear Fourier transform [13] may also increase interest in manipulation of solitons with highly nonlinear fibers [14,15] or on-chip waveguides [16]. In another optical communications application, phase sensitive amplification based on parametric amplification and its on-chip implementation has recently gained great interest [17]. Hybrid integration of As 2 S 3 devices on silicon has enabled stimulated Brillouin scattering (SBS) [18], which allows bringing a number of applications such as SBS enabled tunable time delays [19] and SBS enabled ultrahigh-resolution spectroscopy [20] on chip [21].
While previous works have investigated hybrid waveguides composed of both chalcogenides and Si, for example by infiltration of a Si slot waveguide, thus also obtaining enhancement of the optical intensity inside the chalcogenide [22], here we completely transition between Si and GeSbS waveguides in order not to be penalized by Si TPA inside the GeSbS waveguides. One of the main issues of integrating chalcogenide glasses into the SOI platform in this way is to couple light to and from the chalcogenide devices. Here, we focus on the efficient coupling of light into GeSbS chalcogenide glass waveguides hybridly integrated into the SOI platform by post-processing of chips fabricated in a standard SiP foundry, Singapore's Institute of Microelectronics (IME). Another issue is to define a fabrication flow and device geometries that allow for post-processing of fully fabricated SiP chips with a fully fabricated back-end-of-line (BEOL) stack. Indeed, it is our objective to integrate GeSbS based devices in a fully functional standard SiP platform with electro-optic functionality. This prevents, for example, relying on additional etch-stop layers other than those already present in the process.
At a wavelength of 1550 nm, GeSbS has a refractive index of 2.17, a third order nonlinear index on the order of 18 2 1 0 m / W − , and negligible TPA [23,24]. It is important to note that reported values for these properties can vary widely depending on material form (bulk or planar film), measurement wavelength and specific characterization method. The linear loss of single-mode GeSbS waveguides can be as low as 0.5 dB/cm [25,26]. Compared to arsenicbased chalcogenide glasses, GeSbS has low toxicity.
In this paper, three types of coupling structures have been designed, fabricated and analyzed. First, we present the design and experimental validation of a fully etched GeSbS grating coupler (GC) allowing direct coupling of light into GeSbS waveguides (Section 2), which can be advantageous for the first signal processing stage, e.g. if optical power initially exceeds levels that can be transported in Si waveguides without incurring high TPA induced losses. Next, we demonstrate two couplers between GeSbS and silicon nanowire waveguides, a GeSbS-Si hybrid butt coupler (HBC), Section 3, and a GeSbS-Si hybrid adiabatic coupler (HAC), Section 4. These approaches are benchmarked against each other in terms of insertion losses and back-reflections.
In all three cases, GeSbS waveguides are fabricated out of a 500 nm thick film deposited on top of SiP chips after post-processing, enabling their co-fabrication in a common hybrid technology platform. Coupling structures provide connectivity to 900 nm wide, fully etched GeSbS waveguides, either directly from a standard single-mode optical fiber with a 10.4 μm mode field diameter (MFD) in the case of the GC, or to and from fully etched, 400 nm by 220 nm Si wavegu anomalous di light generati structures, Ge possible in or µm in the GC GeSbS bus w monitor wave of the SOI w Bending radii losses -as v experimental

GeSbS gr
We first inve waveguides. T GeSbS GCs directly inject light first tran Si hybrid cou devices after n

GeSbS
The focusing angle θ of 11 chalcogenide using single-s ion etching w waveguides a to let the beam GC pair, conn be found in [

GeSbS-Si hybrid butt coupler
GeSbS-Si hybrid couplers are required to transmit light between GeSbS and Si single-mode waveguides. One intuitive hybrid coupler structure is the GeSbS-Si HBC, in which forward tapered GeSbS waveguides are butt-coupled to inverse tapered Si waveguides. This is similar to inverse couplers used for fiber-to-chip edge coupling [33] but further adapted to enable local BEOL removal. The goal here is to apply them to fully processed SiP chips.

GeSbS-Si hybrid butt coupler design and fabrication
In the HBC loop, light is coupled in and out of the chip through silicon GCs. The silicon GCs are followed by SiO 2 clad silicon waveguides with a thickness of 220 nm and a width of 400 nm, which are connected to a 500 nm thick GeSbS waveguide via the HBCs [ Fig. 3(a)]. To avoid scattering at the onset of the etch stop layer, the silicon waveguide is adiabatically transformed from a fully etched ridge into a deep-etched rib waveguide. Over a length of 15 µm, the silicon waveguide slab is tapered up to a total width of 3 µm. At the same time, the silicon waveguide core is tapered down from 400 nm to 200 nm in order to expand the mode and match its dimensions to the GeSbS waveguide. The waveguide core (but not the 90 nm etch stop layer) is terminated 1 µm before the nominal GeSbS taper facet position to accommodate lithographic overlay tolerances. As a result of this process, a residual nominally 1 µm wide layer of deep-etched silicon remains between the end of the silicon rib waveguide and the facet of the GeSbS taper. A 40 µm long GeSbS taper adiabatically narrows the GeSbS waveguide from a parameterized initial width (see below) to the 900 nm standard single-mode waveguide width used in this work.
As shown in Fig. 3(a), all-silicon alignment loops (SAL) with SiO 2 cladding are fabricated on the chip besides the HBC loops. The SALs have the same silicon GCs as the HBC loops, which are connected with a 140 µm long silicon waveguide resulting in negligible additional losses. They are used as reference structures to obtain the silicon GC losses needed to deconvolve the HBC losses.
The silicon GCs and waveguides were fabricated within the IME SiP platform with 248 nm deep UV lithography [34]. The GeSbS HBC devices were fabricated by EBL and RIE during post-processing of the chips, with fabrication details in [25].

GeSbS-Si hybrid butt coupling loss
Figure 4(a) shows the transmission spectra of an SAL and an HBC loop, both with an input power of 0 dBm. This HBC loop has a GeSbS taper width of 2.5 µm. The GeSbS waveguide loop is coupled to a RR with a radius of 15 µm, which results in an FSR of 9.9 nm. Since the minimum insertion loss of both loops is obtained at 1552 nm, we use it as the reference wavelength for the loss comparison (since it is also the wavelength at which data analysis is the most robust against GC peak wavelength variations). The total insertion losses of the SAL and HBC loops are respectively 6.8 dB and 12.2 dB at 1552 nm.
By fitting the resonance of the GeSbS RR, we extract losses of 5.5 dB/cm. Therefore, the optical loss resulting from the 223 µm long GeSbS waveguide loop is also negligible. Thus, by subtracting the insertion loss of the SAL from the insertion loss of the HBC loop, we get the HBC loss ( )-( ) = 2

Insertion Loss HBC Loop Insertion Loss SAL Hybrid Coupler Loss
(1) Figure 4(b) shows the loss spectrum of the HBC, with 2.7 dB losses at 1552 nm. The 1 dB and 3 dB optical bandwidths of the SAL and HBC loops are both 28 nm and 49 nm, respectively. Thus, the optical bandwidth of the HBC loop is determined by the silicon GCs. We can see in Fig. 4(b) that the 1 dB bandwidth of a single HBC is larger than 100 nm, which is the measurement range limit of our equipment.
The loss spectrum of the HBC contains 1.6 dB Fabry-Perot ripples, which is attributed to the back-reflection between the GeSbS-Si coupling facets as also corroborated by an FSR of 2.1 nm correlating with the 223 μm GeSbS waveguide length [note that Fig. 4(b) shows half the ripple strength since the transfer function of the loop, in dB, has been halved to normalize the losses to a single GC]. By analyzing the Fabry-Perot resonances [32], we evaluate the HBC reflection to be 9.2%.
A series of HBC loops with varying taper width have been fabricated to characterize its influence on the HBC loss. Figure 5 shows the simulated and measured HBC losses at 1552 nm as a function of the GeSbS taper width. In the measured data, the HBC has a minimum loss of 2.7 dB with a GeSbS taper width of 2.5 µm. Overall, good agreement is found between simulated and measured HBC losses; in particular, dependency on GeSbS taper width is predicted well.

GeSbS-S
In order to mi designed and SiP edge cou waveguide em such as polym light between coupling loss thicknesses co show the po performance w easier to fabr performance w Here, both ed circuit 6(b) and 6(c) show a magnified microscope image as well as a schematic view of an HAC. A 500 nm GeSbS film is deposited on top of the chip and subsequently etched to form GeSbS waveguides. A GeSbS waveguide with a parameterized initial width overlays the entire Si taper. After termination of the Si taper, the GeSbS waveguide is tapered down to the standard 900 nm width over a length of 40 µm. Figure 6(d) shows a scanning electron microscope (SEM) image of the GeSbS-Si overlay. It can be seen that the GeSbS film forms a bump where it overlays the Si waveguide. The cross section of the GeSbS bump is close to an isosceles trapezoid with upper and lower widths of respectively 400 nm and 608 nm. Since the SEM image was taken at an angle, the height of the bump cannot be directly extracted from the micrograph. However, based on rescaling of the known 500 nm height of the GeSbS film, the height of the bump was verified to be approximately 220 nm, coinciding as expected with the Si waveguide height. Figure 6(e) shows the cross section of the coupler at the point where the overlay starts, as assumed in the 3D-FDTD simulation. As the Si waveguide is tapered down along the main axis of the coupler, the GeSbS bump is also tapered down by the same amount, with a constant offset maintained between the waveguide width and the lower width of the isosceles trapezoid. The differences between the simulated HAC coupler losses with and without the GeSbS bump are negligible. As shown in Fig. 6(a), a pure silicon alignment loop with air cladding (SAL air ) is fabricated on the chip besides the HAC loops. The SAL air comprises the same Si GC as in the previous section, as well as a 140 µm long silicon waveguide. However, insertion losses and peak wavelength are changed due to the top air cladding.

GeSbS-
Here too, the Si PICs were fabricated within the IME SiP platform with 248 nm deep UV lithography [34] and subsequently post-processed with EBL after GeSbS film deposition.  Figure 7(a) shows the transmission spectra of a SAL air and a HAC loop with an input power of 0 dBm. This HAC loop has an initial GeSbS taper width of 1.5 µm. As for the HBC loops, the GeSbS waveguide of the HAC loops is coupled to a 15 µm RR with an FSR of 9.9 nm. The minimum loss of the SAL air is obtained at 1523 nm, at which the insertion losses of the SAL air and of the HAC loop are respectively 9.1 dB and 10.5 dB. By fitting the resonance of the GeSbS RR, we extract losses of 5.6 dB/cm. Similarly to the HBC chip, the waveguide losses inside the SAL air and HAC loop are negligible. The HAC loss is half of the insertion losses of the H 0.7 dB inserti The 1 dB respectively. and 57 nm, re wavelength ra HAC. As for the measurem The loss (small ripples which means loop.  µm. Good agreement is observed between simulated and measured losses for taper widths larger than 1.2 µm. However, for smaller GeSbS taper widths, measured losses become significantly larger than simulated losses. After visual inspection of the chips in an optical microscope, it appears this is caused by the GeSbS waveguide having defects in the region overlaying the Si waveguide for small GeSbS taper widths; such defects were not found for the wider tapers. Figures 8(b) and 8(c) show microscope images of GeSbS-Si HACs with GeSbS taper widths of 0.5 µm and 1.5 µm, respectively. We can see that in the GeSbS-Si overlay region, the HAC with 0.5 µm taper width is missing a piece of GeSbS waveguide, while the HAC with the 1.5 µm taper is free of defects.

GeSbS-Si hybrid adiabatic coupling loss
Since the HAC couplers appear the superior devices in terms of back-reflection and insertion losses, it would be desirable to also implement them with a device configuration and a process flow compatible with fully fabricated SiP chips including a full BEOL process, as available from commercial foundry platforms. This would, however, be considerably harder for the HACs than for the previously described HBCs. While one could also envision replacing the fully etched ridge waveguides used in the HACs by partially etched rib waveguides in order to provide an etch stop at the bottom of the BEOL stack, this would only partially resolve the problem due to the resulting topology of the etch stop layer (since for the HACs the Si waveguide has to extend into the region where the BEOL is deprocessed). Wet SiO 2 etches can be highly selective to a Si etch stop layer, however they cause their own set of difficulties, such as undercutting of the protected areas. Dry etches typically used to etch SiO 2 based on CHF 3 or fluorocarbons (C x F y ), while providing some selectivity [36], also attack Si, so that the Si waveguiding layer appears likely to be damaged in such a process in the absence of other non-standard etch stop layers not available in standard SiP chips. The Si etch stop layer used for HBC fabrication on the other hand does not serve for waveguiding, is planar, and is completely removed as part of subsequent etches. Consequently, etch selectivity is much less of an issue there and the optical quality of resulting structures thus not as severally impacted.
The overall resilience of the utilized GeSbS material across the range of process conditions applied in this paper is also noteworthy, in particular its ability to maintain its optical properties and to remain relatively free of defects. This is all the more remarkable since other chalcogenide materials such as As 2 S 3 or As 2 Se 3 have historically suffered from large variability in their post-deposition properties as well as increased defectivity.

Conclusions
Three types of coupling structures to GeSbS waveguides compatible with hybrid integration in a silicon photonics platform have been shown. The fully etched GeSbS grating coupler has a loss of 5.3 dB and results in 3.4% of the light being back-reflected into the GeSbS waveguide when it is coming from the latter. This grating coupler also allows injecting light directly into GeSbS, without a first Si based waveguide segment limiting the power, when nonlinear processing is targeted in a first device stage. Two alternatives, hybrid butt couplers and hybrid adiabatic couplers have been applied to transition between GeSbS and Si singlemode waveguides. The HBC has a 2.7 dB loss and 9.2% of the light is back reflected when injected from the GeSbS waveguide. The HAC has a reduced loss of 0.7 dB and negligible back-reflection.
While the HAC has better performance, both in terms of insertion losses and backreflection, we were able to fabricate the HBCs on standard chips on which the back-end-ofline stack had been locally deprocessed. The HACs on the other hand were fabricated on chips from a wafer without back-end-of-line stack. In view of combining chalcogenide waveguides with a fully functional silicon photonics platform, the ability to easily deprocess the back-end-of-line is a big advantage. All devices have been designed and fabricated with the same GeSbS film thickness, so that the GeSbS-to-Si waveguide transitions can be cofabricated with the grating couplers. For example, initial light injection into GeSbS waveguides via grating couplers can be combined with a downstream transition into Si waveguides. Compatibility of waveguide transition geometries with back-end-of-line deprocessing is expected to be instrumental to combine GeSbS waveguides with the entire set of functionalities already available in standard silicon photonics platforms.