GaAs-based photonic integrated circuit platform enabling monolithic ring-resonator-coupled lasers

Gallium arsenide (GaAs)-based diode lasers can provide direct light emission in the wavelength range from 630 nm to 1180 nm.^1–3 This spectral region is relevant for various applications, including spectroscopy,^4,5 quantum photonics,^6,7 and bio-sensing/bio-imaging.⁸ Most of the aforementioned applications require tailored sources beyond standard Fabry–Pérot type lasers to achieve narrow linewidths,⁹ multiple wavelengths,¹⁰ or tunable wavelength emission.^11,12 However, meeting these demands necessitates the integration of one or more additional waveguide components, such as wavelength-selective elements of the distributed feedback (DFB)^13,14 or distributed Bragg reflector (DBR)^15,16 type, compact waveguide bends,^17,18 couplers,^10,19 and thermo- or electro-optic phase shifters.^17,19,20

Although all of these components have already been realized in different GaAs-based devices separately (cf. Refs. 10 and 13–20), there is no established GaAs photonic integrated circuit (PIC) platform that can provide optical gain, passive waveguides, and the aforementioned functionalities on a single chip. In addition, most GaAs-based waveguide components have not yet reached the level of maturity seen in InP PICs, where a well-established foundry ecosystem exists.^21–23 The difference in maturity between GaAs- and InP-based PIC technologies has both economic and technical reasons. The economic reason is due to the high demand for telecom-related applications that require light emission in the O-band (around 1310 nm) and C-band (around 1550 nm). The technological reason is the more complicated multi-step epitaxy process for the GaAs/AlGaAs material system, which is highly susceptible to oxygen at the regrowth interface and has a high growth velocity of crystalline defects. However, significant progress has been made in recent years to advance and mature the two-step epitaxial growth of GaAs-based lasers.^9,12,24 This progress includes devices emitting not only around 1 µm but also at a wavelength of 778 nm.²⁵ 

Based on this progress, this work presents a GaAs-based PIC platform that integrates active gain sections emitting around 1050 nm and passive waveguides through two-step epitaxy. In addition, the platform features both shallow- and deep-etched waveguides on the same chip. Figure 1 shows the resulting waveguide cross-sections and mentions their potential use cases. The most notable difference to standard GaAs lasers is the now available high lateral (in-plane) index contrast and the resulting high modal confinement of the deep-etched waveguides, which is essential for the realization of compact integrated waveguide components. Building on this, we demonstrate a high-quality racetrack ring resonator together with several other integrated waveguide components, such as semiconductor optical amplifiers (SOAs) providing optical gain, low-loss passive waveguide routing, spot-size converters, and couplers, all monolithically integrated. To demonstrate the use of these waveguide components, ring-resonator-coupled lasers are fabricated and characterized.

In the context of PIC-based lasers, high-quality ring resonators can provide frequency-selective feedback to stabilize and tune the emitted radiation, resulting in narrow linewidth and widely tunable lasers.²⁶ At emission wavelengths around 1050 nm, these lasers are of interest for applications such as optical coherence tomography (OCT)⁸ and airborne and space-based light detection and ranging (LIDAR) systems.¹² 

In this section, we provide a detailed explanation and introduction to the design and concept of our GaAs-based PIC platform, which features a newly developed vertical structure, various waveguide types, and different integrated waveguide components.

The introduced PIC platform utilizes four different types of waveguides, as shown schematically in Fig. 1. From left to right, these are a shallow-etched waveguide with an active region used in sections where optical gain is required, a shallow-etched and low-loss waveguide without an active region for low-loss waveguide routing, and a deep-etched waveguide with and without an active region. These deep-etched waveguides provide high in-plane optical confinement, which is essential for compact waveguide bends.

The shallow-etched waveguides are designed to support only the fundamental waveguide mode. This is achieved by combining a 2.5 µm wide ridge with an effective index step Δn_eff of 1 × 10⁻³, which is obtained by etching index guiding trenches from the p-side down to a residual layer thickness above the active region of d_res,s = 0.3 µm; see Fig. 2(a). The corresponding waveguide cross-section, including information on the refractive index distribution and the fundamental TE mode intensity, is shown in Fig. 2(c).

In contrast, the deep-etched waveguides are designed to be 2.0 µm wide and are etched through the complete vertical waveguide into the AlGaAs n-cladding layer, resulting in d_res = −0.8 µm. The corresponding waveguide, as shown in Fig. 2(d), supports the first three TE modes. However, reducing the ridge width further to limit the number of supported modes would increase the spatial overlap of the fundamental mode, with the etched ridge sidewalls potentially increasing scattering losses. Therefore, it is essential that all deep-etched waveguide components are designed to avoid modal cross-coupling.^27,28

The waveguide types introduced above are based on a vertical structure that was designed to minimize losses, provide strong vertical optical confinement, and reduce parasitic reflections at the active/passive interface. The structure is based on a 596 nm thick undoped GaAs waveguide core that embeds a Ga_0.52In_0.48P etch-stop layer placed directly beneath an In_0.28Ga_0.72As/GaAs double quantum well using GaAs_0.85P_0.15 spacer layers similar to the structure realized in Ref. 9. However, unlike in the cited work, both the n- and p-cladding layers were designed to offer a large index step of Δn_n = 0.15 and Δn_p = 0.23 by using Al_xGa_1−xAs with x = 0.25 and x = 0.40 having layer thicknesses of 2030 nm and 1030 nm, respectively. On the p-side, a 330 nm thick Al_0.2Ga_0.8As intermediate layer was added between the waveguide core and the p-cladding to increase the spatial overlap of the vertical mode with the p-side. This design reduces the dependence of the modal index on the residual layer thickness in the etched regions, which is essential to reliably fabricate the shallow-etched waveguides. The resulting profiles of the refractive index and the free carrier absorption for the chosen doping profile are shown as black and red lines in Fig. 2(a). The corresponding profiles of the vertical fundamental mode and modal indices are shown in panel (b) as red and blue lines, respectively. In both panels, the results of the passive waveguide are represented by dashed lines. Here, the active region is removed between the first and second growth steps, which will be described in Sec. III. As a consequence of the quantum well removal, the passive waveguide is ∼65 nm thinner compared to the active one, resulting in a slightly narrower vertical mode profile.

The active and passive vertical waveguide modal losses caused by the free-carrier absorption profile [see Fig. 2(a)] were calculated by determining the overlap of the corresponding modes and the doping profile, leading to 0.08 cm⁻¹ and 0.09 cm⁻¹, respectively. These 1D results are in good agreement with the values obtained by 2D simulations for the deep-etched waveguides. The losses of the shallow-etched waveguides are slightly lower but differ only in the third decimal place. The modal distributions and the resulting loss figures presented in this section were obtained using Photon Design FIMMWAVE (for more information, see Subsection  1 of the  Appendix). This tool was also used to obtain the simulation results presented in Sec. II B.

To demonstrate the performance and versatility of our GaAs-PIC platform, we realized an integrated racetrack ring resonator consisting of several waveguide components. A schematic view of the ring resonator is shown in Fig. 3(a). Its design follows the idea to combine compact, deep-etched U-bends with straight and low-loss shallow-etched waveguides combined by mode spot-size converters.^29,30

To connect the deep- and shallow-etched waveguides, the ring resonator contains four spot-size converters, which adiabatically transform the mode profile shown in (a) into that of (b) and vice versa. The used design is schematically shown on the left side of Fig. 3(a) and is similar to those used in Refs. 27 and 32. Here, the waveguide width and the shallow-etched part next to the ridge are tapered linearly from w_s = 2.5 µm to w_d = 2.0 µm and from w_t = 10 µm to 0 µm, respectively. The corresponding simulated fundamental mode transmittance as a function of taper length L_t is shown in Fig. 3(c) for three different values of the residual layer thickness d_res. For spot-size converters longer than 80 µm and d_res,s = 0.3 µm, the transmittance T_{1,1} is larger than 99%. In addition, the coupling into higher modes T_{1,m} of the deep-etched waveguide was calculated to be smaller than −20 dB.

The coupling between the ring resonator and its bus waveguides was realized by directional couplers formed by shallow-etched waveguides. A cross-section of the coupler is shown on the left side of Fig. 3(a). The results of the calculated crossover lengths obtained by performing a super-mode analysis are shown in panel (d). It is evident that a coupler gap in the order of 1 µm is sufficient to obtain a crossover length of below 1 mm. The coupler region is formed by the straight waveguide inside the ring resonator and two cascaded sine S-bends, which are each 750 µm long and cover a lateral offset of 10 µm. Shallow-etched waveguides have been used to keep the critical dimensions of the coupler gap in the order of 1 µm, which allows its fabrication using i-line lithography. In contrast, the strong lateral confinement of the deep-etched waveguide would have required coupler gap dimensions in the order of 100 nm. To account for unavoidable etch depth uncertainties during fabrication, a variation in the coupler gap has been realized, spanning from 0.5 µm to 2.5 µm.

The ring resonators investigated in this work were based on Euler U-bends with R_eff = 75 µm, resulting in an individual bend length of L_U = 341.5 µm. The spot-size converter was chosen to be L_t = 150 µm long, and the straight waveguide was chosen to have a length of 892 µm. This results in a total ring circumference of L_R = 3067 µm. The gap of the directional couplers used to couple the light in and out of the resonator was w_gap = 1.5 µm for all measured devices.

The vertical waveguide structure was grown by a two-step metal organic vapor phase epitaxy (MOVPE) process, as described in Refs. 9 and 11. The first growth step concludes with the double InGaAs quantum well and a GaAs top layer. Consequently, no AlGaAs cladding layer is exposed between the growth steps, which minimizes oxygen incorporation and oxidation at the regrowth interface avoiding the formation of deep-level traps and defects. Next, a positive resist is coated and exposed using i-line lithography to cover the active device sections. In the remaining area, the quantum wells and the GaAs top layer are removed by a wet chemical etch step. Following the resist removal, the GaInP etch-stop layer is removed using selective wet chemistry. The second growth process is performed in the same MOVPE reactor used before and involves in situ surface cleaning using CBr₄, followed by the completion of the full vertical structure as outlined in Sec. II. Figure 4(c) illustrates an active-passive interface fabricated by the process described above. It can be seen that the passive waveguide without an active region is ∼65 nm thinner compared to the active waveguide.

After the epitaxial growth, the waveguides were defined using i-line lithography and formed by dry etching trenches into the p-side. Here, a BCl₃/Cl₂-based reactive ion etching process was applied utilizing a parallel plate reactor (SENTECH SI 591) under a high bias voltage. The lithography and etch step has to be repeated twice to form the shallow- and deep-etched regions separately as in Ref. 10, requiring an etch depth of 1495 nm (d_res,s = 0.3 µm) and 2595 nm (d_res,d = −0.8 µm), respectively. It is important to note that the shallow-etch step is more critical regarding etch depth variations than the deep one since it directly influences the resulting effective refractive index Δn_eff; see Figs. 3(c) and 3(d). In contrast, the lateral index guiding of the deep-etched waveguide is practically independent of the etch depth as long as the vertical waveguide is completely etched through. After etching, a standard wafer rinse process was applied, followed by a PECVD deposition of a silicon nitride (SiN) insulating layer. As the next process step, the p-contacts were formed by selectively removing the SiN on top of the active ridge waveguides and depositing a Ti–Pt–Au layer. After thinning the GaAs wafer to a thickness of 130 µm, the n-contacts were formed by depositing a Ni–Au–Ge layer on the backside of the entire wafer.

Figure 4 presents the scanning electron microscope images of various waveguide cross-sections and elements, including an Euler U-bend and a mode-size converter. Panels (a) and (b) show only deep-etched waveguides, while (d) contains both deep- and shallow-etched waveguides connected by a spot-size converter. The insets of panel (a) and panel (b) demonstrate the smoothness of the waveguide surface realized by the process flow described above. The discontinuity at the taper tip visible in panel (d) is a consequence of the process steps involved. For example, the etch masks of the two etch steps must be precisely aligned, with the mask of the second etch step being applied to the topology of the shallow-etched waveguide. In addition, the critical dimensions of the taper tip are close to the resolution limit of the i-lithography used. As a result of these factors, it is difficult to precisely control regions where shallow- and deep-etched structures intersect.

Finally, the wafers were separated into smaller chips (laser bars). Facets of high optical quality were formed by cleaving along crystallographically defined surfaces. The results presented in Sec. IV A were obtained with uncoated facets. In contrast, thin-film coatings were applied to the devices measured in Secs. IV B and IV C to obtain a power reflectivity of 0% (30%) and 30% (0%) at the front (rear) facets, respectively.

To determine the internal parameters of laser diodes, such as internal efficiency η_i and absorption coefficient α_a, resonator length-dependent measurements of the differential efficiency η_d (or laser threshold I_th) can be fitted using analytical expressions.³³ 

To characterize the internal parameters of three different waveguide types enabled by this platform, suitable test structures have been processed as shown in Fig. 5(a). The first device (type I) contains a shallow-etched ridge waveguide with active region over its entire length, whereas device types II and III contain a 2 mm long active waveguide section and a 4 mm shallow-/deep-etched passive waveguide, respectively. In the latter devices, the shallow- and deep-etched sections are connected with a spot-size converter as introduced in Sec. II B.

Typical results of the power–current characteristics of these Fabry–Pérot lasers are shown in Fig. 5(b) for L = 6 mm. These measurements were repeated after successively shortening the resonator length L in 1 mm steps by re-cleaving the laser chips at the front facet. The resonator length dependence of the slope of the measured lasers is shown in Fig. 5(c) as inverse differential efficiency as a function of the total resonator length L (black error bars). The results of device type I have been fitted using the equations introduced above by setting R = 0.3, T_ap = 1, and L_p = 0, leading to an internal efficiency of η_i = 0.78 and an absorption coefficient of α_a = 3.60 cm⁻¹ (−15.6 dB/cm). Adapting L_p, the fits of device types II and III resulted in α_p = 0.42 cm⁻¹ (−1.8 dB/cm) and α_p = 0.51 cm⁻¹ (−2.2 dB/cm) for the shallow- and deep-etched passive waveguides, respectively. Here, α_a was fixed to the value of device type I. All results are summarized in Table I.

To characterize the Euler U-bend-based racetrack ring resonator design as introduced in Sec. II, test devices have been fabricated; see Fig. 6(a). These structures feature a 3.5 mm-long shallow-etched active waveguide section that is connected to a passive section containing a directional coupler that connects the ring resonator. This bus waveguide further leads to the front facet using a 3° tilted, 1 mm long outcouple bend.

To get first results, the active region was directly used to generate TE-polarized amplified spontaneous emission (ASE) to probe the TE ring resonances using an on-chip broadband source. The optical spectra were measured by collecting the emitted ASE on the front facet of the chip using an optical spectrum analyzer (Yokogawa AQ6373B). In this experiment, the device under test was protected from optical feedback using an optical isolator having an isolation of about 60 dB.

Figure 6(b) shows the measured optical ASE spectrum from 1049.5 nm to 1050.5 nm containing densely spaced power drops caused by the ring resonator. The corresponding free spectral range was measured to be 0.095 nm. In addition, it is apparent that no higher-order modes get excited. The black dotted line in the left panel of (b) shows a close-up of one resonance around 1050 nm. A Lorentzian fit of this resonance is depicted as a red line and reveals a spectral width at full width at half maximum of Δλ = 0.023 nm and an extinction ratio of 1.8 dB. However, these results are limited by the spectral resolution of the used OSA, which is 0.02 nm.

To obtain more accurate results, the ring resonators were additionally characterized using a single frequency DFB laser from TOPTICA EAGLEYARD emitting at 1064 nm (EYP-DFB-1064-00025-1500-BFY12-0002). In these measurements, the probe light was coupled into the device under test through the front facet. By sweeping the injection current of the DFB laser from 50 mA to 150 mA, its emission wavelength shifts linearly with 2.47 pm/mA; see Fig. 6(c). The optical signal passing the ring resonator at the through port was detected by measuring the photocurrent using the integrated active region as the monitor diode (MD). In addition, a polarization filter was used to probe the TE and TM response separately. More information on this method can be found in Subsection  2 of the  Appendix.

The resulting baseline corrected and normalized transmission spectra³⁵ are depicted in Fig. 6(d). The corresponding free spectral ranges were determined to be 95.6 pm and 92.0 pm for the TE and TM polarized light, respectively. By using the physical resonator length L of 3067 µm, the resulting group indices can be estimated to be n_g,TE = 3.86 and n_g,TM = 4.01.

The results obtained by this fitting procedure are summarized in Table II for both polarization modes. The fact that t > a indicates that the resonator is under-coupled, which explains the moderate extinction ratio ER of about 5.4 dB to 7.1 dB. From the a values, the averaged absorption coefficients of the ring resonator were calculated to be 0.89 cm⁻¹ (−3.9 dB/cm) and 0.74 cm⁻¹ (−3.2 dB/cm) for TE and TM modes, respectively. Using the definition Q_L = λ₀/Δλ, the spectral widths were determined to be 5.4 pm and 4.6 pm for the TE and TM polarized light, respectively. It is noteworthy to mention that these results were not limited by the measurement method; for more information, we refer to Subsection  2 of the  Appendix.

High-Q ring resonators are of interest for several applications. One of these is their use in the context of integrated laser devices as a wavelength-selective element.⁴⁰ A simple layout of such a ring-resonator-coupled laser is schematically depicted in Fig. 7(a). Here, the add-drop ring resonator is realized as an in-line optical reflector, which provides wavelength-selective feedback at its resonances. The laser cavity is formed between this ring mirror and the front facet, which was coated to have 30% reflectance. The waveguides at the anti-reflection coated rear facet were angled at 3° to the facet normal to avoid parasitic reflections.

Figure 7(b) shows the power–current characteristics of this device measured at the front and rear facets, where the latter contains the emitted power from both pass-ports. The laser threshold and the slope efficiency at the front facet were determined to be 65 mA and 0.14 W/A, respectively. At an injection current of 150 mA, a power of 12.5 mW is achieved. The discontinuities of the power–current curve indicate longitudinal mode hops. However, in between these hops, the measured emitted spectrum indicates single-mode emission; see Fig. 7(c). At an injection current of 140 mA, the side-mode suppression ratio (SMSR) is higher than 40 dB.

The free spectral range (FSR) of 47 pm visible in the optical spectra corresponds to a resonator round trip length of about 6 mm using $FSR=λ02Lng$ and applying the TE-mode group index of n_g,TE = 3.86 determined in Sec. IV B. This length, however, neither corresponds to the ring circumference (L_r = 3067 µm) nor the physical resonator round trip length L_prr = 2L_a + 2L_pr + 0.5L_r = 10 064 µm, where L_a = 2250 µm and L_pr = 2015 µm are the gain section and passive waveguide routing lengths, respectively. Rather, the emission wavelength is defined by the ring-resonator-coupled cavity FSR described by the optical Vernier effect.^26,41 This hypothesis is supported by intracavity reflectometry measurements. For this purpose, a high-resolution spectrum below laser threshold (I = 60 mA) was measured with a double echelle monochromator (LTB DEMON); see Fig. 7(e). The corresponding resonator lengths can be determined using a Fourier transform based analysis; see Subsection  3 of the  Appendix for more information. The results of this procedure are shown in Fig. 7(f) and show a dominant contribution from the coupled cavity (Vernier cavity) and a minor effect from the physical cavity length marked by L_ver and L_res, respectively.

Based on the waveguide and ring resonator figures of merit determined in Secs. IV A and IV B and summarized in Tables I and II, the effective reflectivity R_eff of the in-line ring resonator mirror can be calculated following the approach presented in Ref. 26. The resulting power reflectivity |R_eff|² is shown in Fig. 7(d) having a peak value of 14%. The impact of this ring mirror on the laser linewidth Δν = Δν_FP/F² is described by the chirp reduction factor F.^26,42 Here, Δν_FP is the intrinsic linewidth of the Fabry–Pérot gain section using 30% and |R_eff|² as the front facet and ring mirror reflectivities. The calculated F as shown in Fig. 7(d) has a value of 3.8 at resonance, indicating a possible linewidth reduction of factor 14.4.

The results of the waveguide loss characterization presented in Sec. IV A lead to absorption coefficients of 3.60 cm⁻¹ and 0.42 cm⁻¹ (1.82 dB/cm) for the 2.5 µm wide single-mode shallow-etched active and passive waveguides, respectively. The results from other active/passive GaAs-based waveguide platforms having a similar emission wavelength (1030 nm and 1064 nm) and ridge width (3 µm and 4 µm) show with α_a = 3.44 cm⁻¹ similar losses for the active waveguide ¹² but with α_p = 0.6 cm⁻¹ and α_p = 4.05 cm⁻¹ higher losses for the passive waveguide.^9,12 In addition, the losses of the deep-etched passive waveguides of our platform have been measured to be 0.51 cm⁻¹ (2.21 dB/cm), only 21% higher than those of the shallow one.

The passive waveguide losses achieved in this work are comparable to those reported for commercially available InP-PICs, which range from 2 dB/cm to 4 dB/cm.^21,22,43 In comparison, the losses of open-access silicon-on-insulator (SOI) based waveguide platforms are in the order of 1 dB/cm to 2 dB/cm.^23,44 It should be noted that most SOI platforms combine both, a high index contrast and single-mode waveguides. In the case of the proposed GaAs-PIC platform, single-mode shallow-etched and multi-mode deep-etched must be combined to achieve this. Recently, multi-mode SOI waveguides have been shown to achieve very low losses in the order of 0.1 dB/cm.⁴⁵ However, it is to be noted that the mentioned loss figures were obtained for wavelengths around 1550 nm. Assuming a scattering loss dependence of ∝ λ⁻³ as suggested by the normalized Payne–Lacey model,⁴⁶ waveguides at 1050 nm are about 3.2 times more susceptible to scattering losses than those at C-band wavelengths.

In Sec. IV B, the results of the ring resonator characterization are presented. The intrinsic quality factors were measured to be 2.6 × 10⁵ and 3.2 × 10⁵ for the TE and TM polarized fundamental mode, respectively. These values exceed the Q-factors achieved for micro-ring resonators fabricated on various InP platforms.^47–50 In comparison, Q-factors above 10⁶ have been reported for an all-passive AlGaAs-on-insulator waveguide platform at C-band wavelengths.⁵¹ In the same wavelength range, a thick SOI platform, using a resonator design very similar to the one used in this work, has yielded Q factors exceeding 10 × 10⁶.³⁰ Furthermore, Q > 10⁶ has recently been reported using a commercially available low-loss SiN waveguide platform, but at a much shorter wavelength of 630 nm.³⁸ 

The propagation losses of the presented platform extracted from the resonator characterization are 3.63 dB/cm (TE) and 3.20 dB/cm (TM). Assuming TE polarized light, negligible insertion losses of the spot-size converters, and using the losses measured for the straight waveguides, the additional distributed losses caused by the bends is 1.68 dB/cm. The upper limit for the insertion losses of the investigated Euler U-bends with R_eff = 75 µm can therefore be estimated to be 0.13 dB. This value is significantly lower than the insertion losses of 0.9 dB reported in Ref. 18 for an Euler U-bend with R_eff = 114.5 µm at λ = 1.3 µm and are on par with typical values for InP-based generic foundry platforms.⁴³ In conclusion, the presented results indicate that the achieved Q-factors are limited by losses caused by the waveguide elements within the resonator (U-bends and spot-size converters). A possible reason for losses caused by the spot-size converters is the discontinuity at the taper tip, as shown in Fig. 4(d). In order to resolve this issue, a further optimization of the process technology appears to be necessary.

The laser results presented in Sec. IV C have demonstrated the ability of the presented GaAs-PIC platform to implement fully integrated ring-resonator-coupled lasers. It reaches 14 mW single-sided power emission at 150 mA, which is on par with state-of-the-art heterogeneously integrated ring-based lasers.^26,52,53 Higher powers could be achieved by increasing the reflection of the in-line ring reflector by transitioning from under- to over-coupled ring resonators by increasing the bus waveguide to ring cross-coupling. The discontinuities in the power–current characteristics indicate unwanted mode hops. Although high SMSR values of above 40 dB were obtained at a current of 140 mA, future designs need additional wavelength-selective elements. Lasers based on multiple, slightly detuned coupled ring resonators are capable of realizing narrow linewidth emission and can be tuned over a wide wavelength range.^26,52,54

In this paper, we presented a GaAs-based PIC platform, which supports low-loss passive waveguides as well as monolithically integrated optical gain sections achieved via selective quantum well removal and two-step epitaxy. In addition to standard shallow-etched waveguides, the platform also supports deep-etched ones, enabling the realization of compact curved waveguides.

The losses of the passive waveguides have been determined to be 1.8 dB/cm and 2.2 dB/cm for the shallow- and deep-etched passive variants, respectively, based on cavity length-dependent laser measurements. These values are at the same level as those of commercially available InP-PIC platforms. To demonstrate the performance of the GaAs-PIC platform, we realized racetrack ring resonators. These resonators are formed using deep-etched Euler U-bends, which are connected to a shallow-etched coupler section via intracavity spot-size converters. The characterization revealed high intrinsic optical quality factors of 2.6 × 10⁵ and 3.2 × 10⁵ for the TE and TM polarized fundamental mode, respectively.

To demonstrate the use of ring resonators in the context of monolithic GaAs-based laser sources, a ring-resonator-coupled laser was fabricated and characterized. The front facet power was measured to be 14 mW at an injection current of 150 mA. Furthermore, single-mode operation was achieved with up to 40 dB side mode suppression. However, to fully assess the impact of the ring resonator on the laser linewidth, laser frequency noise measurements are required.

The presented results pave the way for the realization of monolithic integration of GaAs-based narrow linewidth and widely tunable lasers based on coupled ring resonators. To achieve this, the insertion loss of the waveguide elements inside the ring should be further reduced. This might be achieved by increasing the effective radius of the Euler U-bends. Nevertheless, the PIC platform presented here opens the possibility of higher integration densities and more on-chip functionalities than ever before on a GaAs-laser platform.

The work presented here was primarily supported by internal funding. We are indebted to Professor Günther Tränkle for this opportunity. In addition, this work was practically supported by the Research Fab Germany (FMD) under Grant No. 16FMDQ2.

The authors have no conflicts to disclose.

Jan-Philipp Koester: Conceptualization (lead); Formal analysis (lead); Funding acquisition (equal); Investigation (lead); Methodology (equal); Project administration (equal); Software (equal); Validation (lead); Visualization (lead); Writing – original draft (lead); Writing – review & editing (lead). Hans Wenzel: Conceptualization (equal); Formal analysis (supporting); Funding acquisition (supporting); Methodology (supporting); Project administration (supporting); Software (equal); Supervision (supporting); Validation (supporting); Visualization (supporting); Writing – review & editing (supporting). Jörg Fricke: Conceptualization (supporting); Investigation (supporting); Methodology (equal); Writing – review & editing (supporting). Matthias Reggentin: Formal analysis (supporting); Investigation (supporting); Methodology (equal); Writing – original draft (supporting); Writing – review & editing (supporting). Pietro Della Casa: Investigation (supporting); Writing – review & editing (supporting). Poojitha Sammeta: Investigation (supporting); Writing – review & editing (supporting). Olaf Brox: Investigation (supporting); Writing – review & editing (supporting). Michael Ekterai: Formal analysis (supporting); Investigation (supporting); Writing – review & editing (supporting). Mario Kohlbrenner: Investigation (supporting); Writing – review & editing (supporting). Andreas Renkewitz: Investigation (supporting); Writing – review & editing (supporting). Christof Zink: Investigation (supporting); Writing – review & editing (supporting). Thomas Tenzler: Investigation (supporting); Writing – original draft (supporting); Writing – review & editing (supporting). Jos Boschker: Investigation (supporting); Writing – original draft (supporting); Writing – review & editing (supporting). Markus Weyers: Investigation (supporting); Resources (supporting); Writing – original draft (supporting); Writing – review & editing (supporting). Andrea Knigge: Conceptualization (supporting); Funding acquisition (equal); Project administration (equal); Resources (lead); Supervision (lead); Writing – review & editing (supporting).

The data that support the findings of this study are available from the corresponding author upon reasonable request.

The modal profiles shown in Fig. 2 were calculated using the commercial software FIMMWAVE from Photon Design.⁵⁵ The quasi-TE solver used is based on the finite difference method, and a constant spatial grid of Δx = 0.1 µm and Δy = 0.02 µm was used. The modal transmittance data presented in Sec. II were obtained using the FIMMPROP module based on the mode expansion method. Here, 60 quasi-TE modes were considered. The taper algorithm used to evaluate longitudinally non-uniform waveguide segments (spot-size converters and U-bends) was set to its default settings (integrationOrder: 0, Tolerance: 0.01, minStepsizeFrac: 0.01). All simulations have been performed at a wavelength of 1064 nm.

Since the processed ring resonators have a spectral width on the order of or below the resolution of most optical spectrum analyzers, they cannot be used to directly measure the all-pass ring resonator filtered spectra. Alternatively, we measured the ring resonances by scanning the emitted wavelength of an external probe laser whose peak wavelength shift was previously characterized to be 2.27 pm/mA [see Fig. 6(c)], similar to what was done in Ref. 56.

The emission wavelength of the DFB laser used (TOPTICA EAGLEYARD, EYP-DFB-1064-00025-1500-BFY12-0002) was changed from 1064.01 to 1064.25 nm by varying its injection current from 50 to 150 mA in 0.1 mA steps. This leads to a wavelength resolution of 0.25 pm. Before coupling into the front facet of the device under test, the probe light was polarization-controlled to distinguish between TE and TM polarizations. The active region behind the resonator was used as a monitor diode, where the resulting photocurrent was measured using a Keithley 486 Picoammeter. The measured ring-resonator-filtered signal was baseline corrected using the asymmetric least squares smoothing method.³⁵ 

In general, the resulting data are the convolution of the measured ring resonances and the spectral line shape of the DFB laser used. However, since the linewidth of the used laser of 2 MHz is about 600 times narrower than the measured ring resonances with 1.2 GHz, its shape can be assumed to be delta-like.

Reflectometry is a well-known method for the investigation of diode lasers and integrated waveguide devices.^57,58 Here, an adaptation of these methods is described that was used to obtain the results shown in Fig. 7(f). First, the measured optical spectrum OS is interpolated to be equidistant in frequency f space. Next, the optical spectrum is Fourier transformed using FFT(OS(f)). Finally, the FFT(OS(f)) is plotted against the spatial grid Z defined from 0 µm to N · Δz in steps of $Δz=c0Δtng$⁠, where Δz is the spatial resolution, $Δt=1NΔf$ is the temporal resolution, and N is the number of discrete entries of OS(f). It can be seen that the spatial grid depends on the group index n_g. Therefore, TE and TM spectral data must be scaled using n_g,TE and n_g,TM, respectively.