1. Introduction

The volume of global data is growing exponentially as a result of the development of technologies such as big data and cloud computing, putting larger demands on data centers’ power consumption, throughput, and transmission rate [1–3]. Optoelectronic integrated circuits (OEICs) using silicon photonics technology are promising to meet the above demands and have attracted extensive attentions [4,5]. However, due to the intrinsic properties of the indirect bandgap of silicon material, realizing reliable light sources on silicon substrates is a key challenge for silicon-based OEICs [6]. Monolithic integration of III-V semiconductor lasers on silicon has significant potential in the current implementations of silicon-based light sources. Among the techniques, direct epitaxy growth of III-V lasers on silicon substrates has been a more promising method due to the low cost and suitability for mass production [7].

However, the dissimilarities between the III-V and silicon materials result in the generation of various crystal defects such as threading dislocations (TDs), film cracking, and antiphase domains (APDs)—limiting device performance and reliability [8]. In recent years, by virtue of the higher tolerance to defects, quantum dot (QD) lasers directly grown on silicon have been attracting much attention [9–13]. The lifetime is now more than 22 years at 85 °C for the reported state-of-the-art QD lasers on silicon [14]. In contrast, the most prolonged lifetime of GaAs-based quantum well (QW) laser on silicon is only ∼200 h after decades of research [15–18]. Nevertheless, silicon-based QW lasers still have a unique advantage, namely, the direct transplantation of existing GaAs-based and InP-based QW-related technologies onto silicon platforms.

For silicon-based lasers, reducing the threading dislocation density (TDD) in buffer layers and optimizing the laser structure are the primary ways to improve performance. In recent years, great progress has been achieved in the former [19–21] while the development of the latter has stalled. It should be mentioned that the tolerance of the lasers to TDs can be significantly increased with an appropriate structure design. Selvidge et al. [22–24] recently proposed a new structure with trapping layers (TLs) to improve the device performance. They found that misfit dislocations (MDs) form during the post-growth cooldown due to the combined effects of thermal-expansion mismatch between III–V layers and silicon and the mechanical hardening in the QD active region. To tackle this issue, thin Indium-containing layers as trapping layers (TLs) were inserted above and below the QD active region to transfer the MDs away from the active region to the TLs. The performance improvements of this novel design for the QD laser structure are comparable to reducing the TDD by one order of magnitude, presented by 2 times reduction in threshold current, 60% increase in slope efficiency, and over 200 times improvement in the reliability of silicon-based QD lasers [14]. Up to now, TLs have been just applied in silicon-based QD lasers but not yet in silicon-based QW lasers. However, MDs were also found in the InGaAs active layer grown on silicon, as reported by E. Groenert [25], resulting in the serious deterioration of silicon-based QW lasers. Because the silicon-based QW lasers maintain the advantages of a wide band range and feasible transplantation of QW-related technologies, there is an urgent demand to improve the performance of silicon-based QW lasers. Therefore, it is necessary and meaningful to investigate the beneficial effects of TLs on the performance of silicon-based QW lasers through experiments.

In this work, the realization of electrically-pumped continuous-wave (CW) room-temperature InGaAs/AlGaAs multiple quantum well (MQW) lasers epitaxially grown on silicon by molecular beam epitaxy (MBE) is reported. The as-grown material was fabricated into Fabry-Perot (FP) broad-area laser chips with a cavity length of 1 mm and a stripe width of 20 µm. By inserting TLs into the AlGaAs cladding layers, MDs were blocked away from the active region. This significant effect results in a 2.7-fold reduction of threshold current density(from 4.6 kA/cm² to 1.72 kA/cm²) and a 7.2-fold increase in maximum output power(from 13.2 mW to 96 mW) under pulsed condition. The measured threshold current density was 2.7 kA/cm² under CW condition. The maximum single-facet output power of 45.3 mW and the slope efficiency of 0.143 W/A were achieved under an injection current of 980 mA.

2. Material growth and device fabrication

As shown in Fig. 1, two heteroepitaxial laser structures were fabricated on GaAs-on-Si (001) templates, including sample A and sample B where 7-nm doped In_0.15Al_0.85As trapping layers were used. Both structures were carried out in a solid-source MBE system.

 

Fig. 1. (a) Schematic of sample A without TLs. (b)Buffer structures with DFLs design. (c) Schematic of sample B with TLs inserted in the cladding layers.

Download Full Size | PDF

For the GaAs-on-Si template, samples A and B used the exact same growth conditions. The 2-inch Si (001) substrates with 4° miscut toward the [110] direction were adopted to prevent the formation of APDs when directly growing polar III-IV materials on nonpolar Si substrates. Before being loaded into the growth chamber, RCA clean method was performed on Si substrates to remove residual impurities. After 30 mins of deoxidation at 1100 °C under As flux, the GaAs buffer layers were initialed with a 30-nm GaAs nucleation layer at a temperature of 300 °C, followed by a 100-nm low-temperature (350 °C) GaAs layer, a 150-nm mid-temperature (500 °C) GaAs layer, and a 250-nm high-temperature (560 °C) GaAs layer. All the temperatures mentioned above are measured by thermocouple. To improve the material quality, four sets of strained-layer superlattices (SLSs) separated by 300-nm GaAs spacing layers, containing 5-period 10 nm In_0.17Ga_0.83As/10 nm GaAs, were grown as dislocation filter layers (DFLs) to drive TDs to glide along the interface, as shown in Fig. 2(a). After depositing 100-nm GaAs, the samples were removed from the chamber for characterization. As shown in Fig. 2(b), the GaAs/Si surface morphology was measured by atomic force microscopy (AFM), and the root-mean-square (RMS) roughness was 2.9 nm (10 µm × 10 µm) with no APDs observed. To estimate the TDD of the as-grown material, electron channeling contrast imaging (ECCI) was performed. As shown in Fig. 2(c), 193 TD-points were observed in the scanning area of 8.5 µm × 6.4 µm, which corresponds to the TDD of 3.5 × 10⁸cm^-2.

 

Fig. 2. (a) Cross-sectional transmission electron microscopy (TEM) image of the buffer layers. The 4-period DFLs are used to suppress the propagation of TDs. (b) AFM and (c) plan-view ECCI images of the surface after the deposition of 100 nm GaAs on the 4^th DFL. The RMS roughness and TDD are 2.9 nm (10 × 10 µm²) and ∼3.5 × 10⁸/cm², respectively.

Download Full Size | PDF

For the laser parts, the first structure started with a 700-nm Si-doped GaAs contact layer. Then 3-period 8 nm In_0.17Ga_0.83As/15 nm Al_0.2Ga_0.8As quantum wells were grown, which were sandwiched by symmetrical separate confinement heterostructure (SCH) structures including 1.5-µm Al_0.4Ga_0.6As cladding layers and 100-nm Al_0.2Ga_0.8As waveguide layers. For comparison, in sample B, a 7-nm n-doped In_0.15Al_0.85As strained layer and a 7-nm p-doped In_0.15Al_0.85As strained layer were inserted 100 nm above and below the active region, respectively. And the last structure ended with a 200-nm Be-doped GaAs contact layer. Selvidge et al. used In_0.15Ga_0.85As TL on p-side and In_0.15Al_0.85As TL on n-side to minimize band misalignment [22]. In this experiment, both In_0.15Al_0.85As TLs was used, and the trapping layers with different materials will be considered in the near future.

As shown in Fig. 3(a), the as-grown laser wafers were processed by standard wet etching and metallization techniques. The details of the fabrication procedure can be referred to our previous reports with proper changes [17,26]. All these lasers were mounted on copper heatsinks with C-mount packages. The schematic diagrams of the device are shown in Fig. 3, and the broad-area laser was composed of 20-µm stripe width and 1-mm cavity length. The mirror-like laser facet was formed by cleaving without any optical coating. As shown in Fig. 3(b), the cross-sectional scanning electron microscope (SEM) image of the device fabricated from sample B is presented.

 

Fig. 3. (a) Schematic image of the device structure. (b) Cross-sectional SEM image of the part of a device structure.

Download Full Size | PDF

3. Result and discussion

To investigate the effect of the TLs, the devices fabricated from sample A and sample B were tested. Figure 4(a) shows the light-current (L-I) properties of the devices from sample A and sample B, measured under pulsed operation (5 µs-pulsed width, 1%-duty cycle) at room temperature (25 °C). For the device of sample A, the threshold current is 746 mA, corresponding to a threshold current density of 3.73 kA/cm².In contrast, the threshold current of the device from sample B is 277.1 mA and the threshold current density is 1.39 kA/cm², which is 37.1% of that from sample A. In addition, when the injection current is 700 mA, a single-facet output power of sample B is 96 mW and the slope efficiency is 0.272 W/A. Under the same injection current, however, sample A has not yet reached the threshold for lasing. The lower threshold current density and higher output power can be due to a lower density of non-radiative recombination centers, presumably from the lower density of MDs in the active region. The effect of TLs could be considered the key to reducing the density of non-radiative recombination centers in the active region. In addition, another device of sample B with similar performance was tested at various temperatures under pulsed condition. The laser was able to operate up to 80 °C under pulsed condition, still producing an output power of 56 mW. The calculated characteristic temperature is 163.8 K from 10°C to 80 °C.

 

Fig. 4. (a) Typical L-I characteristics of broad-stripe lasers from sample A and sample B, respectively. (b) A TEM image of the active region. TL traps the TD successfully, leading to the formation of MD. (c) Schematic representation of sample A. MDs form at the interface of QW layer. (d) Schematic representation of sample B. Insertion of TLs displaces MDs away from QW layer to above and below.

Download Full Size | PDF

To further investigate the reason for the great difference in the performance of the lasers, the cross-sectional TEM of the active region is shown in Fig. 4(b). The InAlAs TL above the QW traps the TD, causing an MD to form far away from the active region. It is also found that TLs can merely change the direction of TD’s propagation slightly and cannot reduce the density of TD. Furthermore, the schematic representations of dislocation evolution for two structures are shown in Fig. 4(c) and Fig. 4(d) for further explanation, respectively. Supported by Selvidge’s theory [22], MDs will form at the top and bottom interface of the QW active region during the post-growth cooldown, due to the combined effects of the thermal-expansion mismatch and the hardening-effect of indium-containing alloy. The insertion of InAlAs trapping layers can replace the MD which is initially located in the active region as shown in Fig. 4(d). The aforementioned content proves that TLs is highly effective in Si-based QW lasers.

The laser fabricated with sample A failed to operate under the room-temperature (25 °C) CW condition. However, benefiting from the role of TLs, the room-temperature (25 °C) CW operation was achieved for the device from sample B. As shown in Fig. 5(a), the typical light-current-voltage (L-I-V) property of the same device of sample B was measured under the CW condition at room temperature. From the I-V curve, the extracted differential resistance of the laser is 9.5 Ω. From the L-I curve, the threshold current is 537 mA, corresponding to a threshold current density of 2.7 kA/cm². The highest single-facet output power of 45.3 mW and a slope efficiency of 0.143 W/A were achieved at the injection current of 980 mA. The same device was also aging at room temperature under a constant current of 1.25 times of threshold current. The time-to-failure(time to double initial threshold current) is about 21.7 mins. The laser spectra of the device from sample B at various injection currents under CW condition are shown in Fig. 5(b). The lasing wavelength gradually increases from 996 nm to 1004 nm as the injection current increases from 560 mA to 860 mA. This phenomenon indicates that the device has serious heating during operation, presumably from the relatively serious non-radiative recombination and high resistance, limiting the performance and reliability of the device. In future work to solve this problem, we will improve the crystal quality and optimize the device structure.

 

Fig. 5. (a) Typical L-I-V characteristics of broad-stripe lasers from sample B (b) Lasing spectra of sample B at various injection currents at room-temperature.

Download Full Size | PDF

4. Conclusion

In conclusion, we have demonstrated continuous-wave electrically pumped 980 nm InGaAs/AlGaAs multiple quantum well lasers directly grown on Si substrates with trapping layers. The whole structure was carried out in an MBE system, including a 2-µm GaAs buffer layer with DFLs and a laser structure with TLs. The FP broad-area laser chips with a cavity length of 1 mm and a stripe width of 20 µm were fabricated and tested. With the effects of TLs, a 2.7 times reduction in threshold current density and a 7.2 times increase in single-facet maximum output power were achieved under pulsed operation. The same device under CW condition at room temperature was also measured and the threshold current is 537 mA with a corresponding threshold current density of 2.7 kA/cm². The single-facet maximum output power is 45.3 mW when the injection current is 980 mA and the slope efficiency is 0.143 W/A. This work provides a feasible approach to improving the performances of GaAs-buffered InGaAs/AlGaAs quantum well lasers monolithically grown on silicon via InAlAs trapping layers.