Room-temperature continuous-wave operation in the telecom wavelength range of GaSb-based lasers monolithically grown on Si

Over the past few years, global IP traffic has been exponentially growing, driven by the explosion of requests for computing, cloud storage, and telecom. This brings difficulties in efficiently moving the information at each scale of dimensions, from worldwide links to chip-to-chip and intra-chip interconnections. In this context, silicon photonics has recently become of increasing interest due to its potential for co-integrating optical data links and data processing electronics on CMOS-compatible Si platforms. The key component that is difficult to realize in this technology, however is the laser source, due to the indirect Si band-gap. The most advanced approach is surely the heterogeneous integration of III-V gain chips on a Si platform.^1,2 Yet, these techniques suffer from issues related to achieving precise alignment and to large material waste, leading to increased complexity and cost. On a longer term, the direct epitaxial growth of III-V lasers on Si-photonics circuits appears as the most promising, cost-effective, and versatile approach.

III-V heteroepitaxy on Si however has proven difficult due to the conjunction of large lattice-, thermal-, and polarity- mismatches.³ Still, much progress has been made in the recent years on the understanding and mastering of these issues, and encouraging laser results have been reported.⁴ Notably, room temperature (RT) laser emission has been demonstrated in a continuous wave (cw) mode near 1.3 μm with GaAs-based quantum-dot lasers^5–7 and near 2 μm with GaSb-based quantum well (QW) lasers,⁸ all grown on off-axis (001) Si substrates. These wavelengths are of interest for telecom and sensing applications, respectively. Nevertheless, full deployment of Si-photonics for telecom applications also requires integrated-lasers emitting near 1.55 μm, the wavelength of choice for long-haul communication systems. This spectral range however is out-of-reach with GaAs quantum-dot lasers and alternative technologies have to be developed. Although we have previously achieved laser emission near 1.55 μm with GaSb-based QW lasers grown on Si substrates,⁹ these lasers suffered from high-threshold current densities, resulting in poor performances limited to pulsed operation. In this work, we demonstrate cw, RT operation of GaSb-based quantum well laser diodes grown on Si in the telecommunication wavelength range.

The first issue with 1.55 μm GaSb-based lasers is their band-structure design. In fact, this material system is best suited for mid-IR optoelectronics where high-performance laser diodes have been produced for a while.¹⁰ In contrast, only recently could high-performance 1.55 μm GaSb-based lasers diodes be achieved.¹¹ The laser heterostructure used in this work relies on this recent work. Figure 1 shows the bandstructure of the whole epitaxial stack calculated using the nextnano suite.¹² The active zone is made of three Ga_0.80In_0.20Sb/Al_0.68In_0.32Sb composite quantum wells (QWs) separated by 20 nm Al_0.35Ga_0.65As_0.03Sb_0.97 barrier layers.¹¹ Each composite QW consists in a 6-nm wide Ga_0.8In_0.2Sb QW with two insertions of 0.45-nm thin Al_0.68In_0.32Sb layers (inset in Fig. 1). This composite QWs exhibit a 1.35% strain with respect to GaSb. The Ga_0.80In_0.20Sb/Al_0.68In_0.32Sb composite QWs form type-I QWs with totally delocalized wave-functions for both electrons and holes (inset in Fig. 1) and a calculated electron-hole wavefunction overlap of 96.2%. The confined electron and hole levels are located 180 and 190 meV below the barrier level, respectively (Fig. 1). Although sufficient for the holes, this confinement energy is not high enough to totally suppress thermal escape of electrons out of the QWs at room temperature and above.¹³ 

The whole epitaxial stack of the laser heterostructure is displayed in Fig. 1. The composite QW system is embedded within 230-nm-thick Al_0.35Ga_0.65As_0.03Sb_0.97 undoped waveguide layers. The cladding layers are 1.5 μm-thick Al_0.9Ga_0.1As_0.07Sb_0.93 layers. Graded AlGaAsSb layers are inserted to smooth out the band profiles between the n-type GaSb buffer layer and the n-type Al_0.9Ga_0.1As_0.07Sb_0.93 bottom cladding on one hand, and between the p-type Al_0.9Ga_0.1As_0.07Sb_0.93 top cladding and the p-type GaSb contact layer on the other hand.

The laser structures have been grown by solid-source molecular-beam epitaxy (MBE) in a reactor equipped with valve-cracker sources for both As and Sb group-V elements. Quarters of two in. (001) silicon substrates with a 6° offcut towards the [110] direction were used. The 6° miscut has been chosen in order to limit the formation of anti-phase domains arising from the polarity difference between the substrate and the III-V materials.³ Prior to epitaxy, the substrates were prepared by applying both ex-situ and in-situ procedures described earlier in detail.¹⁴ At this stage, we used a very simple buffer layer. The growth was initiated by depositing 4 monolayers AlSb directly on the Si substrate at 450 °C, followed by the growth of the GaSb buffer while ramping the substrate temperature up to 500 °C.¹⁵ After 1-μm GaSb, the temperature was ramped down to 450 °C in order to grow a 150-nm thick n-doped InAs_0.92Sb_0.08 layer. This layer, lattice-matched to GaSb, plays two roles. It is an etch-stop layer during laser processing and it is used to form the n-side ohmic contact, as described below. Finally, an 800-nm thick n-type GaSb layer and the laser structure were grown at 470 °C. Be and Te were used as p- and n-type dopants, respectively.

Ridge laser diodes (LDs) were processed using standard photolithography and wet etching. The final LD geometry is depicted in Fig. 2, which shows both p- and n-contacts taken on the epitaxial structure. The p-contact was taken on the top ridge while the n-contact was taken on the InAs_0.92Sb_0.08 layer located within the buffer layer. This geometry avoids to drive the current through the highly defective III-Sb/Si interface and ensures higher LDs performances.⁸ We thus etched through the whole structure down to the InAs_0.92Sb_0.08 layer. Electrical insulation was obtained using the AZ1518 resist. Ti–Au and AuGeNi were used as contact metals for the p- and n-type contacts, respectively. Laser cavities were formed by simple cleaving of the chips. No optical treatment was applied to the facets. The devices were then soldered epi-side up with indium on Cu heat-sinks. The n-contact is typically about 40 μm away from the ridge.

Next, we performed electro-optical characterizations of the LDs. First characterization was done in a pulsed regime (21 kHz repetition rate, 1 μs pulse duration) at RT with 100-μm wide ridge and 1-mm long cavity broad area LDs. We show in Fig. 3 the L–I–V curves for a typical LD. The threshold current density is 1 kA/cm², a value which is comparable to similar lasers grown on their native GaSb substrate.¹¹ The series resistance R_s is 3 Ω and the turn-on voltage V_d is 0.8 V, which is very close to the bandgap.

We have subsequently characterized 10 μm × 1 mm narrow-ridge LDs in cw operation. The L–I–V curves taken between 15 °C and 35 °C for a typical LD are displayed in Fig. 4. The threshold current increases from 300 mA to 450 mA in this temperature range, corresponding to a characteristic temperature T₀ = 50 K, a value comparable to that of GaSb-based lasers grown on Si emitting in their natural wavelength range near 2 μm⁸ and to initial InP-based QW lasers grown on InP emitting near 1.55 μm with comparable carrier confinement.¹⁶ The external differential quantum efficiency η_d, obtained from the slope of the L–I curves, changes from 2.5% to 1% in the 15–35 °C temperature range. The output power per facet at 500 mA drive current is 3.5 mW at 15 °C and 3 mW at 20 °C. Note however, that the thermal roll-over has not been reached in Fig. 4 and that the LD facets were not treated, which indicates that higher power could be achieved. Finally, the inset in Fig. 4 depicts the spectrum obtained with the LD operating in CW at 15 °C, showing an emission peaked at 1.59 μm, i.e., in the C/L bands dedicated to high performance telecom systems.

Excellent reproducibility of the I–V characteristic was observed while the threshold current density of broad area LDs varied between 1 and 1.5 kA/cm² and the threshold current intensity of the narrow-ridge LDs varied between 300 and 500 mA, both at room temperature. At this stage, we ascribe these variations to different facet qualities of the LDs. Indeed, Si does not spontaneously cleave along the [110] crystal direction as the III-V semiconductors do to form natural facets.

Comparing both the threshold current and the output power of these LDs in cw operation with those of similar LDs grown on their native GaSb substrate¹¹ evidences the negative impact of structural defects arising from strain relaxation of the III-V material grown on Si and of the facet quality. Several routes are open to further improve the laser performances. Regarding the epitaxial step, the III-Sb-on-Si nucleation stage can still be refined and the buffer layer engineered to reduce both the density and the impact of structural defects. In addition, the active zone design can be further improved as previously done with InP-based telecom lasers while cavity fabrication can be improved. Finally, statistical studies on various cavity geometries will allow identifying and quantifying the origin of losses.

In conclusion, by employing a new design of the active zone, as well as new strategies for Si substrate preparation and molecular-beam growth nucleation as compared to previous work, we have demonstrated room-temperature, cw operation in the telecom wavelength range of GaSb-based laser diodes epitaxially grown on (001) Si substrates. This bridges a gap and opens the way to the realization of monolithically integrated C/L band Si-photonics systems.

This work has been partly supported by the French ANR (Project OPTOSi, No. ANR-12-BS03-002) and by the French “Investment for the Future” program (EquipEx EXTRA, No. ANR-11-EQPX-0016). E.T. acknowledges support from the Institut Universitaire de France (IUF).