III–V Microwires with Reversed Ridge Waveguides Selectively Grown on Pre-Patterned Si Substrates

Abstract

To construct functional photonic integrated circuits, an efficient and compact laser is expected to be incorporated into the complementary metal–oxide–semiconductor platform. Monolithic integration of III–V lasers on pre-patterned Si substrates by the aspect ratio trapping method is a promising solution. Here, microwires with reversed ridge waveguides (RRWs) on pre-patterned Si substrates were reported. By metal–organic chemical vapor deposition, high-quality InP microwires with RRWs were obtained, and InGaAs/InGaAsP multi-quantum-well structures with InGaAsP separate confinement hetero structure (SCH–MQW) were successfully achieved. The SCH–MQW structure was buried in the InP microwire, which was beneficial for transferring the heat generated in the active region. The micron size also contributes to the efficiency of thermal diffusion. Further, simulation results showed that the metal absorption loss could be less than 4 dB/cm by properly controlling the contact area between metal electrodes and microwires. This proposed structure opens up an alternative pathway for electrically driven III–V lasers seamlessly interfaced with Si-photonics.

1. Introduction

Compatible with the mature and standardized complementary metal–oxide semiconductor (CMOS) platform, silicon photonics is regarded as a significant promise to meet future requirements for power-efficient and high-density interconnects [1,2]. Several basic photonic components are available in monolithic SOI, including arrayed waveguide gratings, optical filters, Ge photodetectors, ring- and traveling-wave electro-refractive modulators, and so on [3,4]. Due to the indirect band structure of silicon (Si) materials, lasers on Si have always been one of the biggest challenges for silicon photonics. III–V materials have far better light emission efficiency, so integrating III–V materials on Si is considered a promising solution and has been extensively studied [5,6,7,8]. Hybrid bonding has reached the commercialization stage, but the process flow is complex, expensive, and incompatible with CMOS processing [9]. To fully exploit the potential of silicon photonics, direct heteroepitaxy III–V lasers on Si has been studied [10]. The key to direct heteroepitaxy is to engineer the generation and propagation of defects caused by the difference in materials properties between Si and III–V [11]. By introducing a thick buffer, high-quality III–V materials grown on a planar Si were obtained [12] and electrically driven lasers have been demonstrated [13,14]. However, the thick buffer makes it difficult to efficiently couple light from III–V lasers to Si waveguides and this problem is yet to be solved. As an alternative, selective growth of III–V materials on pre-patterned Si or SOI avoids the thick buffers. and confines defects [15]. By using the aspect ratio trapping (ART) technique and V Si grooves, high-quality III–V materials on Si or SOI such as InP, GaAs, InAs, and GaSb materials have been obtained [16,17,18,19], and optically pumped micro-nano lasers have been demonstrated [20,21,22]. However, electrically pumped lasers have not been demonstrated so far. Compared to optically pumped lasers, carrier absorption loss, heat accumulation, and metal absorption loss must be considered when fabricating electrically pumped lasers [23,24]. In order to solve these problems, different epitaxy methods and structures have been reported, for example, extending the material volume of III–V materials by lateral epitaxy [25] or increasing the size of the SiO₂ trenches [26].

Here, by the aspect ratio trapping (ART) approach, an InP microwire with reversed ridge waveguide (RRW) on Si substrate was demonstrated by metal–organic chemical vapor deposition (MOCVD). InGaAs/InGaAsP multi-quantum-well (MQW) structure with InGaAsP separate confinement hetero structure (SCH–MQW) were successfully inserted into InP microwires. The SCH layer was introduced to reduce carrier absorption loss and to confine the carriers and optical field. The SCH–MQW structure was buried in the InP microwire, allowing for efficient transfer of the heat generated in the active region, and the micron size also contributes to the efficiency of thermal diffusion. The reversed ridge waveguide gradually changes from top to bottom. The contact area between metal electrodes and microwires can be controlled by chemical mechanical polishing (CMP). Simulation results showed that the metal absorption loss was less than 4 dB/cm by properly controlling the contact area.

2. Materials and Methods

The epitaxial growth was performed by using MOCVD (AIXTRON 200, Herzogenrath, Germany) on the pre-patterned Si substrate, and the pressure was 55 mbar. Triethylgallium (TEGa), trimethylgallium (TMGa), trimethylindium (TMIn), arsine (AsH₃), and Phosphine (PH₃) were used as precursors. In the experiments, the Si wafers were patterned with [011]-oriented oxide stripes, wide openings were 500 nm, and the substrate declination angle was 0°. The detailed fabrication of the pre-patterned Si substrate has been previously reported [27]. The prepared sample was loaded into the MOCVD reactor, and then thermal cleaning in an H₂ ambient was carried out at 720 °C. While wafers were being cooled from the H₂ baking temperature, AsH₃ was introduced into the reactor to form one monolayer of As-terminated Si {111} surfaces. And then, GaAs buffer layers were grown by a two-step growth of the low-temperature (400 °C) nucleation layer and high-temperature (630 °C) high-quality layer. The next step was a 450 °C low-temperature InP layer and a 650 °C high-temperature InP layer growth. Last, the InGaAsP SCH layer, InGaAs/InGaAsP MQWs, and the InP capping layer were grown. The surface morphology of III–V microwires and the thickness of the SCH layers, quantum wells, and quantum barriers, as well as their chemical information, were obtained from scanning electron microscopy (SEM), transmission electron microscopy (TEM), and energy dispersive spectroscopy (EDS). Room temperature micro-photoluminescence (µ-PL) was performed to reflect the material quality.

3. Results and Discussion

3.1. InP Microwires on Pre-Patterned Si Substrate

The epitaxial growth of the InP microwires with RRWs on the pre-patterned Si substrates was demonstrated first. Under the chosen growth conditions and growth time, high-quality InP microwires with RRWs on the pre-patterned Si substrates were obtained by MOCVD. Figure 1a,b shows the cross-sectional SEM images of InP microwires with RRWs grown on the pre-patterned Si substrates. From Figure 1a, it is clear that the structure out of the SiO₂ trench is composed of four {111}, the total height and width of which are 1.81 µm and 1.85 µm, respectively. The micron size of the structure is beneficial to improve heat dissipation efficiency, and the micron size is sufficient to confine the light field to the structure outside the SiO₂ groove. Thus, scattering loss caused by InP/SiO₂ sidewalls can be avoided, and optical leakage loss caused by the Si layer can be ignored. Figure 1b shows an array of InP microwires with RRWs, the overall profiles of the structures are uniform. The plan-view SEM image of the InP microwires with RRWs is shown in Figure 1c; the arrays of the InP microwires show a defect-free region and uniform profiles along the <100> direction, which provides promise for preparing the lasers. The morphological uniformity of the InP microwires with RRWs also indicates the stability and reliability of epitaxial conditions. The SCH layers can shrink the light field to reduce carrier absorption loss and lower the barrier height of the MQW structure, which is beneficial for reducing the threshold current of the laser. Hence, the InGaAsP was introduced as the material of the SCH layers and the quantum barriers, and the InGaAs was introduced as the material of the quantum wells. The designed SCH–MQW microwire is shown in Figure 1d. The active region was buried in the InP materials out of the SiO₂ trench so that the heat generated in the active region could be better transferred. And the reversed ridge waveguide was formed by the structure inside the SiO₂ trench, and it was located directly below and very close to the SCH–MQW structure, introducing carrier restriction and light field restriction in a horizontal direction. For the electrically pumped laser, the contact between the bottom InP layer and the top InP layer will cause the current leakage; to avoid this, the bottom of the SCH–MQW structure was located in the SiO₂ trench.

3.2. SCH–MQW Microwires on Pre-Patterned Si Substrates

As designed in Figure 1d, the microwires with RRWs, including the InGaAsP SCH layers and InGaAs/InGaAsP MQW structures, were successfully obtained by MOCVD. Here, a thick InGaAsP calibration layer was first grown on the bottom InP layer instead of the MQW layer and the top InP layer. Then, the high-resolution X-ray diffraction (HRXRD) pattern was used to analyze whether the InGaAsP matched InP, and the PL was used to analyze the wavelength of InGaAsP. HRXRD curves and μ-PL spectra of InGaAsP matched to InP are shown in Figure 2a,b separately. The $\omega$ − $2\theta$ scanning method was used to scan the diffraction peaks around the (004) Bragg reflection of the Si substrate. As shown in Figure 2a, the InGaAsP material used for the SCH layer and the quantum barrier was almost strain-free. As reported [23], a wavelength of 1.3 µm was expected. From Figure 2b, the bandgap of InGaAsP was around 1.34 µm. The components of InGaAsP were theoretically calculated that the In fraction was around 0.71, the Ga fraction was around 0.29, the As fraction was around 0.63, and the P fraction was around 0.37. Similarly, the In fraction of the InGaAs used for the quantum wells was around 0.53.

Here, platinum (Pt) was deposited to protect the microwires, and then a focused ion beam (FIB) was used for the sample preparation. At last, TEM characterization, together with an EDS analysis, was performed to understand the geometrical and chemical information of the lattice-matched SCH layers and MQW structures. Cross-sectional scanning TEM (STEM) images of the overall epitaxial structure and the details of the SCH–MQW structure are shown in Figure 2. The bottom InP inside the SiO₂ trench serves as buffer layers for the growth of the SCH–MQW structure. It can be seen that the crystalline defects are localized in the V-groove trench, and no effects propagate through the GaAs/InP buffer layer to the SCH–MQW region. From Figure 3a, the profile is slightly different from that of the InP microwire, the reason is that the top InP is grown on the InGaAsP layer. The structure out of the SiO₂ trench was still composed of four {111}, and the active region was buried in the InP materials out of the SiO₂ trench so that the active region could be protected and the heat generated in the active region could be better transferred. In addition, the micron size also facilitates the transfer of heat in the active region. For the III–V microwire with RRW, it is formed by the structure inside the SiO₂ trench and outside the SiO₂. The reversed ridge waveguide, formed by the structure inside the SiO₂ trench, is located directly below and very close to the SCH–MQW structure, introducing carrier restriction and light field restriction in a horizontal direction. For the electrically pumped laser, the contact between the bottom InP layer and the top InP layer will cause the current leakage. To avoid this, the bottom of the SCH–MQW structure was located in the SiO₂ trench, as shown in Figure 3b. From Figure 3b, there is a (001) plane in the SCH–MQW region, which is grown on a ridge InP layer formed by near (001) and {111} surfaces. Figure 3c,d presents the magnified cross-sectional TEM images of the SCH–MQW area marked in the red dashed squared in Figure 3b and the detailed SCH–MQW top region, respectively. Every layer of the bottom InP layer, the InGaAsP SCH layer, InGaAs/InGaAsP MQW, and the top InP layer was clearly identified and no crystalline defects were observed in the SCH–MQW region. The interface between the bottom SCH layer and the first quantum well, marked by the red circle in Figure 3c is P rich, which is caused by gas flow switching, and it can be improved by properly tuning the gas flow. Due to the large volume of the target lamellae, the region marked in the red circle in Figure 3d probably is generated during the preparation of the lamellae, but the reason needs to be further confirmed by experiments.

The thickness of the layers that make up the SCH–MQW structure was also quantified. As shown in Figure 3c,d, the thickness of each layer along the {111} plane is uniform, and the thickness of the layer located in the top region is different from that located in {111} facets. Hence, three positions were selected for measuring the thickness of every layer. Black arrows (a1–a3) were also labeled in Figure 3b to indicate measurement positions.

Table 1. The thickness of SCH layer, quantum well layers, and quantum barrier layers at a1, a2, and a3, marked in Figure 3b. Along the growth direction, the SCH–MQW structure is composed of the bottom SCH, 1st well, 1st barrier, 2nd well, 2nd barrier, 3rd well, and the top SCH.

Measured Position	The Thickness of the SCH–MQW Structure (nm)
	Bottom SCH	1st Well	1st Barrier	2nd Well	2nd Barrier	3rd Well	Top SCH
a1	25	2.5	5	3.8	3.8	2.5	19.2
a2	55	15	4.2	12	2.1	12.5	17.5
a3	25	5	5	5	3	5	25

lists the thickness of the bottom SCH layer, quantum wells, quantum barriers, and the top SCH layer at the corresponding positions. For the bottom SCH layer and each quantum well, the thickness at the top near the (001) facet is much thicker than that at two {111} surfaces. On the contrary, the top SCH layer and the quantum barriers at the top region were thinner than that at two {111} surfaces. For III–V materials grown inside nano-scale trenches, a self-limiting growth profile, with a fixed size of convex {111} facets and the top (001) facets are developed to minimize the total surface energy during the selective area growth [28]. The self-limiting size of the (001) facet of InGaAsP, is smaller than that of InP, but larger than that of InGaAs [29]. Hence, the epitaxy of the bottom SCH layer and the quantum wells exhibited a huge growth preference along the vertical [001] direction to minimize the (001) facet, and the epitaxy of the quantum barriers and the top SCH layer was just the opposite. In addition, the optical confinement factor is calculated by comparing the optical energy in the MQW region with the total energy in the simulated region, and the optical confinement factor of the MQW structure shown in Figure 3a is 3.72%.

Figure 4a shows the superposition of the STEM image and the EDS linear scan in the SCH–MQW region. The scan range (cyan line) covered the whole SCH–MQW structure. The composition of As (blue line) in the quantum well was higher than that of the quantum barrier, and its content in the quantum barrier was higher than that of the SCH layer, while the P distribution (yellow line) was just the opposite. The ingredients of Ga (red line) in the quantum well were higher than that of the quantum barrier, and its content in the quantum barrier was higher than that of the SCH layer, while the P distribution (yellow line) was just the opposite. The distribution of Ga was consistent with As, and the distribution of In was consistent with P. The difference between the components of the barrier layer and the SCH layer may be due to the diffusion of the components between the quantum barriers and the quantum wells. A more intuitive illustration is the EDS surface scan. A cross-sectional element distribution of the grown material can be visualized by recording the appearance of each element in a two-dimensional area. The plots of the element distribution of In, Ga, As, and P in the SCH–MQW structure are shown in Figure 3b–e. On this profile, the interfaces are sharp enough to make a clear distinction between the quantum wells, the quantum barriers, and the SCH layers, and it is consistent with the results of the EDS linear scan.

To have the first feedback about the optical properties of the SCH–MQW microwires, the room temperature µ-PL measurement was investigated and the result is shown in Figure 5. The pump source was provided by a 633-nm He-Ne laser under continuous wave operation. The power of the pump source was about 8 mW, measured by a handheld plane detector at the emergent side of the objective lens. The peak wavelength and the full width at half maximum (FWHM) of the µ-PL spectrum were about 1510 nm and 119 nm, separately. The measured PL FWHM is mainly caused by thickness uniformity and the composition uniformity of the MQW structure. It was analyzed that the peak wavelength at 1.34 µm was caused by the SCH layer.

3.3. Potential Analysis of the Microwires for Silicon Photonics

To evalute the feasibility of microwires with RRWs as Si photonic light sources, the metal absorption loss and the couple from the microwire to Si waveguide were discussed separately by finite difference time domain (FDTD) method using Lumerical Solutions’ software. The layout of the designed layer grown on the pre-patterned Si substrates is shown in Figure 6a. The top of the Si layer and the top of the microwire act as electrode contact layers. The polyimide (PI) was carried out after epitaxial growth of the SCH–MQW microwire, and it acted as a planarization material; then, the (001) plane in contact with metal could be obtained by chemical mechanical polishing. The width of the part outside the SiO₂ trench varies from bottom to top, so the contact area between the metal and the microwire can be controlled by controlling the polishing time. Here, the metal absorption loss of the fundamental TE mode inside the microwire as a function of the reserved height of the microwire (h, shown in Figure 6a) was calculated by simulation, as shown in Figure 6b. For the microwire in the simulated structure, as shown in Figure 6a, the metal was assumed to be Au, the h₀ was 0.32 µm, and the w was 1.65 µm. The preserved height of the microwire (h) was set as a variable to optimize the structure, and its initial value was 1.63 µm. As shown in Figure 6b, the metal absorption loss of the fundamental TE mode decreased from 10.7 dB/cm to 0.06 dB/cm as the h progressively increased from 1.1 µm to 1.6 µm. When the h was 1.4 µm, the metal absorption loss was 2.52 dB/cm. This gave sufficient tolerance to the chemical mechanical polishing process. The insert of Figure 6b displays the profile of the fundamental TE mode with h₂ set as 1 µm. The results indicate the potential for microwires with RRWs to prepare electrically pumped lasers and the importance of the fine process implementation for electrically pumped lasers. For the microwire with RRW, the light field was confined into the structure outside the SiO₂ groove. Thus, when the V-groove Si trench was fabricated on the substrate Si layer and the SiO₂ trench was prepared on the buried oxide layer, the light emitting plane of the micron laser with RRW and the straight Si waveguide was on the same plane, which was beneficial for coupling light from micron lasers into straight Si waveguides. The results demonstrated by simulation show the potential for microwires with RRWs in silicon photonics platforms, even though lots of work on process preparation are needed.

4. Conclusions

In conclusion, we demonstrated III–V microwires with RRWs by MOCVD. A high-quality InP microwire with RRW was fabricated using the ART method, and the InGaAs/InGaAsP MQW structure with InGaAsP SCH layer was successfully inserted into the InP microwire. TEM investigation confirms that each layer has clear boundaries, and no defects exist in the SCH–MQW region. The theoretical calculations indicated that, by controlling the contact area between metal and microwire in perspective, it is possible to obtain a metal absorption loss of less than 4 dB/cm. By transferring the microwire with RRW to the bottom Si layer of the SOI substrate, we can solve the coupling issue between the microwire with RRW and the straight Si waveguide. These research findings open up a possible way for fully functional silicon photonics.