1. Introduction

The burgeoning demand for compact, high-performance, and energy-efficient laser sources in photonic integrated circuits (PICs) has captured significant research interest. These laser sources find applications in the fields as varied as optical communication, environmental sensing, spectroscopy, and quantum photonics [1–4]. Among various photonic platforms, silicon photonics offers an outstanding pathway to realize large-scale, highly functional photonic chips due to its tight light confinement and already existing mature fabrication technology, providing a reliable and scalable platform for large-scale integration and commercialization [5–7]. Silicon-based components like waveguides, resonators, modulators, photodetectors, (de)multiplexers, couplers/splitters, and switches [8–14] have seen significant advancement. On-silicon laser sources have also been demonstrated, such as those based on Raman gain [15], Si nanocrystals in SiO₂ [16], germanium [17], and rare earths [18], although these approaches find difficulty in the improvement of laser performances.

The heterogeneous integration of III-V semiconductors is one of the most promising approaches to realizing high-performance lasers on Si. Direct heteroepitaxy on Si has been examined as a scalable approach in which quantum dots (QDs) are often considered a promising gain medium [19–22] due to their resilience to material defects. However, precisely controlled crystal growth is required to integrate bright QDs on Si. Wafer bonding has also been utilized to integrate III-V semiconductor lasers coupled to silicon waveguides [23–25]. However, most of the bonded materials are removed for patterning the active devices, resulting in low material usage efficiency. Hybrid integration approaches have also been intensively investigated. A standard and commercialized technique is flip-chip bonding [26,27], which, however, in general, cannot achieve high-density integration and suffers from low integration throughput. Photonic wire bonding [28] and butt coupling [29] are also often utilized but hold the same issues as flip-chip bonding.

Transfer printing [30] is a promising hybrid integration method that can potentially overcome the challenges posed by the on-Si integration. This technique is based on pick-and-place assembly using an elastomer stamp and enables high through-put, high-density, and high precision integration of fully processed III-V semiconductor devices on Si. Only a small area of the III-V material is transferred, thus permitting a significant increase in the material usage efficiency. The technique is basically free from the issue of thermal expansion mismatch since tiny, thin structures are transferred at room temperature. The combination of all these advantages makes transfer printing a promising technique for the hybrid integration of III-V semiconductor lasers on silicon substrates.

In recent years, transfer printing has been employed for the fabrication of various hybrid integrated devices [31–33]. In [34], the transfer printing of a 40 × 970 µm² InP coupon with AlInGaAs quantum wells was transferred over Si waveguide structures covered with 20-nm benzocyclobutene (BCB), to make a III-V-on-silicon distributed feedback (DFB) laser. In [35], a completed semiconductor optical amplifier (SOA) InP coupon of 45 × 950 µm² is transferred over SiN on Si to make a hybrid III-V DFB laser. Ref [36] adopted a different approach, transferring 6.6 mm long InGaAs quantum cascade lasers onto Si-on-Sapphire substrates through an SU-8 adhesive layer and demonstrating optical coupling to Si waveguides. Ref [37] details the fabrication of directly modulated membrane distributed reflector (DR) lasers on a Si waveguide using transfer printing and successfully demonstrated data transmission at 50Gbps/s in Non-Return to Zero (NRZ) modulation.

Compared to the above examples using quantum wells, the application of transfer printing for the hybrid integration of QD lasers is less mature despite the superior properties of QDs as the gain medium, such as low threshold current, high-temperature stability, low linewidth enhancement factor and resilience to optical feedback [38,39]. Recently, a transfer-printed O-band DFB InAs QD laser was demonstrated by transferring 45 µm × 1400 µm InP coupons on a Si photonics platform [40]. While the DFB laser in reference [40] offers single-mode operation around 1.3 µm and a high side-mode-suppression-ratio of 44 dB at 40°C, distributed Bragg reflector (DBR) lasers could potentially provide an advantage in terms of easier wavelength tunability [41] – a vital parameter for versatile photonic applications – beyond what is achievable with DFB lasers. Additionally, DFB lasers require precise control over the grating period and depth, making them more challenging to fabricate than DBR. O-band edge-emitting QD lasers were also integrated with waveguiding platforms with coupon lengths of up to 2.4 mm [42]. However, there largely remains in the development of transfer printing integration of QD lasers on Si in terms of design and fabrication processes. Indeed, so far, there is no report on DBR QD lasers assembled on Si photonics by transfer printing.

Herein, we report a hybrid integrated InAs/GaAs QD DBR laser by transfer printing a 1.6 mm x 147 µm QD SOA over a Si waveguide. Laser oscillation was achieved through the QD SOA gain combined with the optical feedback from the DBR mirrors patterned in both ends of the Si waveguide. The following two key structures were carefully designed to achieve mutual optical coupling in an efficient manner. For stable transfer printing, we prepared the airbridge GaAs-based QD SOAs by adhesive bonding a source QD wafer on another GaAs substrate using a spin-on glass (SOG). The cured SOG, the sacrificial layer, was etched using vapor-phase hydrofluoric acid (VHF). We also employed a dedicated elastomer stamp with a shape adapted to transferring a ridge waveguide laser. We observed laser oscillation around the wavelength of 1250 nm with a threshold current of 47 mA at room temperature and stable operation up to 80°C.

2. Device design and simulations

Our hybrid device comprises two vertically stacked waveguides made of distinct materials. The upper GaAs waveguide consists of an active region based on InAs QDs, while the lower layer consists of a Si waveguide integrated with DBRs at both ends (Fig. 1). The Si waveguide has a height ${h_{\textrm{Si}}} = $ 300 nm and etched depth ${h_{\textrm{Si ridge}}} = 210$ nm. Both the GaAs and Si central waveguides are $L = $ 1 mm long and flanked by spot-size converters (${L_{\textrm{taper}}} = $ 300 µm) for efficient light coupling. The central Si waveguide has a width ${w_{\textrm{Si cwg}}} = $ 300 nm and widens to ${w_{\textrm{Si}}} = $ 3 µm, whereas the GaAs central waveguide is $w = $ 5 µm wide at the top and narrows to sharp tip around both ends. The DBRs are designed by periodically modulating waveguide width between 1 and 3 µm with a period $\mathrm{\Lambda } = $ 214 nm. The output and mirror DBR have ${N_1} = $ 100 and ${N_2} = $ 200 periods, respectively. The Si waveguide is patterned on thermally grown SiO₂ with a thickness of 2 µm. Between the Si and GaAs waveguides, there is a layer of BCB with a thickness of ${h_{\textrm{BCB}}} = $ 20 nm.

 

Fig. 1. Schematics of the hybrid integrated laser: GaAs waveguide with QDs placed over a Si waveguide terminated with DBRs. The top panel shows a top view, and the bottom panels show a cross-section at the center of the structure.

Download Full Size | PDF

To simulate transmission and reflection spectra of the DBRs, we performed 3-dimensional finite difference time domain (3D-FDTD) simulations using the commercial software Ansys Lumerical FDTD. The excitation source, spanned over a wavelength range from 1.15 to 1.35 µm, was guided in the fundamental TE mode of the waveguide with ${w_{\textrm{Si}}} = $ 3 µm and injected to the DBRs. The obtained spectra are presented in Fig 2(a) (2(b)) for transmitted (reflected) light. There are multiple valleys (peaks) in the transmission (reflection) spectra, which become deeper (taller) with increasing the number of DBR periods N. The deepest dip for transmission was found around 1253 nm. The transmittance values at 1.25 µm wavelength are 8.1% and 2.7% for $N = $ 100 and 200, respectively. The reflectance is 92.9% at the mirror end DBR and 86.7% at the output DBR. We estimate that there is 5.2% (4.4%) scattering loss in the DBR of $N = $ 100 ($N = $ 200).

 

Fig. 2. (a) DBR transmission and (b) reflection spectra for $N = $ 100 and $N = $ 200.

Download Full Size | PDF

To assess coupling efficiency through the spot-size converters, an eigenmode expansion (EME) solver software (Ansys Lumerical Mode) was utilized. We evaluated the light coupling from the fundamental TE mode in the GaAs waveguide (Fig. 3(a)) to that of the underlying Si waveguide (Fig. 3(b)). We took into account the tilted side wall of the GaAs waveguide with an angle of 100° with respect to the horizontal plane (see Fig. 1). This angle is chosen because the dry etching method used to define the SOA structures gives this result (see Fig 5(a)). The GaAs waveguide height is ${h_{\textrm{GaAs}}} = 2.44$ µm with an etched dept of ${h_{\textrm{GaAs ridge}}} = 2.34$ µm. Here we assumed the tip width of the tapered GaAs waveguide to be ${w_{\textrm{taper}}} = 100$ nm, reflecting the parameters measured in the experimentally realized structure discussed later. Efficient energy transfer from the GaAs layer to the Si layer can be confirmed in the lateral cross-sectional view plotted in Fig. 3(c). The coupling efficiency between the two TE fundamental modes is calculated to be 93%. A path of loss is the conversion into the TE third order mode, to which 3.4% of photons are converted. The high reflection and low loss design is advantageous to efficient and low threshold lasing. The coupling efficiency decreases to 87% with SOA misalignment of 280 nm with respect to the Si waveguide center and drops sharply afterward.

 

Fig. 3. GaAs-Si waveguide evanescent coupling simulations for the combined GaAs and Si spot size converter. (a) Mode in the GaAs waveguide. (b) Mode in the Si waveguide. (c) Lateral cross-sectional view showing the mode transition from GaAs to Si.

Download Full Size | PDF

3. Laser fabrication procedure

Figure 4 summarizes the entire processes that we developed for the hybrid integration of QD SOAs on the Si waveguide by transfer printing. We started with the preparation of a GaAs wafer with QDs suitable for device fabrication. The QD active layers are an 8-layer stack of InAs/GaAs QDs, each layer having a thickness of 40 nm. The quantum dot active region has been modulation doped with Be in the quantum dot capping layer. P-modulation doping of the quantum dot active region can notably enhance the performance of QD lasers at high temperatures and increase their temperature stability [43].

 

Fig. 4. Fabrication procedure of an InAs(GaAs) Quantum Dot hybrid DBR laser on Si by transfer printing: (a) InAs(GaAs) QD laser wafer grown with a p-GaAs substrate and its epitaxial structure; (b) SOG coating over a plain GaAs substrate; (c) wafer bonding both GaAs wafers and curing at 300°C; (d) substrate removal; (e) definition of the QD SOA by photolithography and dry etching; (f) patterning of a SiO₂ passivation layer; (g) patterning of a HfO₂ passivation layer; (h) patterning and deposition of the metal electrodes; (i) removing of n-GaAs around the lasers; (j) patterning the SOG layer; (k) patterning of the photo-resist capsule; (l) sacrificial layer etching with VHF; (m) attaching the PDMS stamp to the laser for picking-up; (n) quickly picking up the laser; (o) placing the QD SOA over the Si waveguide; (p) releasing the PDMS stamp; (q) O₂ plasma cleaning to remove the resist capsule.

Download Full Size | PDF

We conducted photolithography to pattern the GaAs layer into the mm-scale QD SOAs in the form of an airbridge. Finally, we transfer printed SOAs on the Si waveguide to complete the device structures. Here, a key development was the use of SOG-mediated wafer bonding prior to the SOA fabrication, which facilitates the sacrificial layer etching and successful transfer printing. Another key point is found in the elastomer stamp with a tailored structure for firm adhesion to the SOA coupons. With these developments, we succeed in the transfer-printing based integration of the hybrid laser structure. In the following, we explain each process one by one.

The process begins with wafer bonding of a p-doped GaAs wafer with a QD SOA layer onto another GaAs wafer (Fig. 4(a-c)). Wafer bonding onto a wafer of the same material largely mitigates the issue of thermal expansion mismatch. The QD SOA layer was grown upside down and consists of an AlGaAs etch stop layer, a 300 nm p-GaAs cap layer, a 1.44-µm p-AlGaAs clad, the QD active layers, a 200-nm n-AlGaAs etch stop layer and a 100-nm n-GaAs contact layer. The crystal growth was performed at QD Laser Inc. The QD wafer was bonded upside down onto the GaAs wafer spin-coated with an SOG material (Dow Corning, FOx-15). After bonding, the merged wafer was annealed at 300 °C for curing. Then, we covered the edge of the bonded QD wafer by photoresist and performed chemical wet etching of the p-GaAs substrate [44,45] first using a non-selective H₃PO₄– H₂O₂ (3:7 v = v) solution and second using a 50%wt. citric acid–H₂O₂ (4:1 v = v) solution. The wet etching stops at the Al_0.7Ga_0.3As etch stop layer, which was subsequently removed by a room-temperature hydrofluoric (HF) aqueous solution with a concentration of 10 vol% (Fig 4(d)), making the bonded wafer ready for SOA fabrication.

We note that SOG-mediated wafer bonding is helpful in the formation of airbridge QD SOAs suitable for transfer printing integration. When forming the GaAs airbridge structure, wet selective etching of an underlying AlGaAs sacrificial layer is widely used, which, however, often causes collapsing of the structures due to the surface tension of the wet solution. We considered using vapor HF (VHF) to address the problem; however, this approach leads to difficulties in releasing the AlGaAs sacrificial layer due to contamination from residues generated by the reaction with VHF. Therefore, we employed the sacrificial layer of SiO₂ by SOG, which can be selectively removed with VHF without any noticeable contamination of the device layer.

Next, we fabricated SOAs in the bonded QD layer on the GaAs wafer. We first sputtered an 800-nm-thick SiO₂ layer on top of the wafer. Then, the ridge waveguide structure with taper sections was patterned using photolithography and reactive ion etching. Dry etching with CF₄ and Ar gases was used for patterning the SiO₂ layer, which was used as a hard mask for etching the GaAs layer with a gas mixture of BCl₃ and Ar. The GaAs dry etching was timed to stop at the n-Al_0.7Ga_0.3As etch stop layer, which is later removed with a 10 vol% HF solution (Fig 4(e)). Afterward, a passivation layer for contact formation was created by sputter deposition of a 100 nm thick SiO₂ layer. Openings for p- and n-contacts were perforated using photolithography and buffered HF (BHF) etching (Fig 4(f)). Then, a second passivation layer is deposited using atomic layer deposition (ALD) of HfO₂ for protecting the SiO₂ layer during VHF etching, which can pass through the photoresist and erode the device layer. ALD HfO₂ is much less reactive to VHF than sputtered SiO₂ [46,47]. Openings for p- and n-contacts were re-perforated using BHF while carefully keeping all the SiO₂ layer edges covered with HfO₂ (Fig 4 (g)). Subsequently, a liftoff process was performed to form metal electrodes of AuGeNi/Au by electron-beam evaporation (Fig 4 (h)). Finally, the remaining n-GaAs layer was removed from around the laser structures by wet etching using an H₃PO₄– H₂O₂ (3:7v = v) solution. After this process, the SOG layer under the SOAs is exposed for etching (Fig 4(i)).

Now, we patterned the SOAs into air-suspended structures for transfer printing integration. First, we partially removed the SOG around each SOA by etching with BHF (Fig 4(j)) and then formed a resist capsule around the SOA. The photo-resist shell is anchored on the GaAs substrate with tethers (Fig 4 (k)). Finally, the SOG sacrificial layer under the SOAs is completely etched with VHF (Fig 4 (l)). We transfer printed the airbridge SOAs using a homemade polydimethylsiloxane (PDMS) stamp, a schematic of which is shown in Fig. 4(m). Instead of using a conventional flat stamp, we used a customized PDMS stamp with a 10-µm-height dual bar prepared by mold casting (fabrication details included in Supplement 1). This stamp structure avoids touching on the top of the GaAs ridge waveguide and homogeneously adheres to the flat regions next to the ridge, suppressing the delamination of the stamp during picking up. After attaching the stamp, we quickly pulled it to break the tethers and lift the SOA (Fig. 4(n)).

The final process step is to integrate the lifted SOAs on Si waveguides. The Si waveguides with DBRs were prepared on a SOI wafer with a 300-nm-thick Si layer on a 2-µm buried oxide layer. Electron beam lithography and dry etching were used to pattern the Si layer. The etching depth was measured to be 210 nm. After the waveguide patterning, the entire Si structure was covered with SiO₂ and planarized by chemical mechanical polishing. The polished surface was leveled with the top of the Si waveguide, by calibrating the time needed to remove the required amount of SiO₂. Prior to placing SOAs on the Si waveguides, the Si chip was spin-coated with a 20-nm BCB layer to improve the adhesion. We measured the surface roughness of the SOI sample. The Si surfaces exhibited a roughness of approximately 0.23 nm, whereas the SiO₂ surfaces had a roughness of about 2.1 nm. There was a height variation of up to 8.8 nm from the top of the Si surface to the top of the SiO₂ surface, with the Si surface being taller. We employed a homemade transfer printing machine [48,49] equipped with fine translational stages and a high-resolution optical microscope and precisely placed the lifted SOA onto a Si waveguide (Fig. 4(o)). As determined in previous work, the alignment accuracy of this setup can be as small as 100 nm [48]. Then, the PDMS stamp was retracted by a slow peel motion (Fig. 4(p)). The whole laser structure is completed after the removal of the resist capsule using O₂ plasma (Fig. 4(q)). We note that we facilitated the stamp release process by sliding the stamp before pulling it up [50]. The sheared force on the stamp accelerates the delamination of the stamp from the SOA. Despite the surface roughness challenge, transfer printing was successfully achieved. The utilization of uncured BCB was pivotal in overcoming these issues, providing a tacky surface that facilitated the laser's adhesion during transfer, demonstrating the method's adaptability to less-than-ideal surface conditions.

Figure 5 shows pictures of a fabricated hybrid laser. A cross-section of the GaAs ridge structure after the dry etching process is shown in Fig. 5(a), highlighting the waveguide sidewall angle of 101° with respect to the horizontal axis. Figure 5(b) shows the tip of the GaAs waveguide taper patterned by photolithography. Figures 5(c) and (d) show the hybrid laser structure near the output edge. A bubble under the 100 nm GaAs film can be observed from the output side, even though the waveguide appears stretched. During the transfer printing process, despite our meticulous efforts to ensure uniform contact, occasional bubbles can form. These are typically due to minute irregularities on the surface or slight variations in pressure during the bonding process. Good alignment of the QD SOA to the underlying Si waveguide can be confirmed with the latter two pictures. From other transfer attempts (not shown) we obtained a yield of 12.5%. This yield accounts for both the physical transfer failures, where the SOAs did not adhere to the substrate, as well as instances where the transferred lasers did not function as expected.

 

Fig. 5. (a) SEM picture of the etched GaAs waveguide ridge; (b) SEM picture of the etched GaAs waveguide inverted taper tip. (c) SEM and (d) optical microscope pictures of the hybrid-integrated laser recorded around an output port, confirming good alignment of the laser.

Download Full Size | PDF

4. Laser performance evaluation

Laser performance was assessed under pulsed current injection ranging from 0 to 250 mA at a frequency of 1 kHz and a duty cycle of 0.1%. We employed a Keithley 2520 pulsed laser diode test system for this analysis, which also recorded voltage across the laser's electrodes. Light output was monitored using a germanium photodiode, while a temperature control system maintained the laser temperature. Laser spectra were captured using an optical spectrum analyzer.

Figure 6(a) shows current-intensity curves measured under various temperatures. The hybrid laser exhibited a threshold current of 47 mA (threshold current density of 4.4 mA/µm²) at an ambient temperature of 22°C, signifying effective current injection into the QD layer. At an injection current level of 150 mA, we observed a spectral intensity peak at 1248.9 nm, with a linewidth of 2.8 nm (Fig. 6(b), red curve), corresponding to lasing from the ground state of the QD ensemble. Indeed, a photoluminescence spectrum of the InAs/QD wafer peaks at 1243.9 nm. We also found that the laser mode is sustained by the reflection with the DBRs, as evident with their transmission spectrum overlayed in Fig 6(b) (blue curve). Interestingly, we also noticed a secondary lasing peak at 1257.0 nm, spectrally separated by 8.1 nm from the first lasing peak. The energy difference between the two peaks is 6.4 meV and considering that the FWHM of the photoluminescence peak is 31 meV, larger than 6.4 meV, the two peaks are likely caused by inhomogeneous broadening due to the variation of QDs. We determine an SMSR of 14.4 dB. By linearly fitting the V-I characteristics (Supplement 1, Fig. S3), we obtain a differential resistance of 12.6 Ω.

 

Fig. 6. (a) Current versus light output curves measured at several temperatures from RT (22°C) to 80°C; (b) normalized laser spectrum with a peak wavelength at 1249 nm (red, the longitudinal modes on the laser spectrum are not resolved due to low resolution), overlayed with the transmission spectrum of the DBR of N = 100 (blue); (c) experimental threshold current dependence on the laser temperature (red dots). The black solid line is a linear fit for the data between 22°C and 60°C, resulting in T₀ = 134 K.

Download Full Size | PDF

We also examined the temperature stability of the hybrid laser (see Fig 6(a)). The device exhibited stable lasing up to 80°C although its threshold current was increased to 78 mA. We managed to observe lasing up to 115°C with a surged threshold current of 260 mA (not shown). At this high temperature, the behavior of the laser became unstable. Figure 6(c) shows a plot of threshold current as a function of operation temperature. Through fitting the data between 22 and 60°C, we deduced a temperature coefficient of the threshold current, T₀ to be 134 K. These results suggest the potential of the hybrid QD laser for high-temperature operation.

To compare laser performance, we refer to our previous work on reference [24], where the same epitaxial wafer growth was used to create DBR lasers wafer-bonded on a silicon DBR waveguide. In ref [24], lasing was observed at pulsed current injection, with sable operation at temperatures up to 115°C, and a threshold current density of 7.8 mA/µm². The higher temperature operation is degraded in this work, which may be due to the smaller laser dimensions 1.6 mm × 5 µm here as opposed to 5.2 mm × 20 µm in [24], reducing the laser surface and decreasing heat dissipation. Other contributions for lower maximum temperature operation are the presence of SiO₂ and BCB between the SOA bottom surface and the SOI substrate, as well as a non-optimal adhesion between the device and substrate. The threshold current density is improved in this work, which we attribute to improved fabrication techniques.

5. Conclusion

In this study, we demonstrated a hybrid InAs/GaAs QD laser on Si by transfer printing. The Si waveguide possessed DBR mirrors and sustained the laser oscillation with the amplification by the transfer printed QD SOA. To the best of our knowledge, the demonstrated device is the first hybrid QD distributed Bragg reflector laser realized using transfer printing. For the successful transfer printing process, we developed airbridge QD SOAs using SOG mediated bonding and VHF etching. The dual rail PDMS stamp was used for stably picking up the QD SOA. The transfer printable QD OSA can be co-integrated with many different photonic components on various platforms, thus leading to the plethora of new photonic integrated devices that take advantage of the superior properties of QDs, such as low threshold, high-temperature operation, narrow linewidth, and resilience to optical feedback.