1. Introduction

Nowadays laser technology is maturing rapidly and is finding applications in areas such as high resolution spectroscopy, medicine, optical telecoms, radar-lidar, metrology… where highly coherent tunable high power light sources emitting in the visible to mid-IR range are required in the continuous-wave (cw) regime at room temperature (RT). A highly coherent tunable laser design can be achieved using an extended-cavity quantum-well (QW) semiconductor surface emitting laser in a stable optical cavity, so called Vertical–External–Cavity–Surface–Emitting–Semiconductor–Lasers (VECSELs). In addition to compactness and low power consumption, semiconductor lasers have the advantage to cover a wide emission wavelength range at RT exploiting III-V semiconductor technologies. Diode-pumped VECSELs combine the approach of diode-pumped solid-state lasers and engineered semiconductor lasers, generating both low divergence circular diffraction limited output beams (TEM₀₀) and multiwatt powers in cw at RT [1, 2]. Indeed this high-Q vertical cavity design boosts the light coherence, generating continuously-tunable sub-MHz linewidth single-frequency laser with a linear polarization state, without any intracavity filter, as reported in [3, 4] at power levels ≤ 20mW. These features are intrinsic to the use of a high-Q air-gap relatively ”long” stable cavity with a ideal homogeneous gain behavior - as with QW’s -, where amplified spontaneous emission and non-linear mode interactions are negligible [4].

In a previous work [5], based on this design principle we demonstrated a highly efficient tunable low noise micro-chip VECSEL, generating 300mW output without any thermal management on GaAs materials. We took advantage of thermal lens–based stability to develop a short external cavity without the need of any intracavity filter. Mainly two solutions have been described in the literature to boost the power performances: first growing the structure upside down and removing the substrate to dissipate the heat through the Bragg mirror bonded on a highly conductive host substrate [1, 2, 6, 7]; or secondly removing the heat through the top of the semiconductor structure by optical bonding it to a thick window of high thermal conductivity and high optical quality (like sapphire, diamond or SiC) [8, 9]. This intracavity optical heatspreader introduces some birefringence, a parasitic Fabry-Perot etalon effect, thermal lensing and non-negligible optical losses. For the first solution using metallic bonding, different techniques could be used like solid-liquid inter-diffusion bonding with AuIn₂ [6, 10] on SiC substrate, or thick gold electroplating process [7] which potentially improves the uniformity.

In this work, we took advantage of a laser technology using a process based on lowtemperature thin film thermal AuIn₂ bonding onto SiC to improve the heat dissipation [6]. Other works reported on high power single frequency VECSEL operation at RT using diamond or SiC heat spreader and intracavity filters [6, 8, 9], up to 0.5W [6] at 283 K, but in bulky and relatively complex laser systems with low efficiency < 8%, no continuous tunability, and a broad free running linewidth > 100 MHz.

In this paper, we demonstrate that this compact short-cavity QW VECSEL design principle can be extended to multiwatt operation with square pump profile provided by high power multimode fibre-coupled diode laser at large incidence angle. We demonstrate high efficiency watt level single frequency operation of a low noise VECSEL at RT in cw. Exploiting membrane technology bonded on SiC for thermal management, we get the highest power ever published to our knowledge [6, 8, 9]. We show theoretically why high quality elliptical beam can oscillate in this single transverse mode plano-plano optical resonator. The far field phase map and beam quality factor (M²), the side-mode-suppression-ratio (SMSR), the polarization state, the relative intensity noise (RIN) and the frequency noise are studied. The design principles can be extended to any wavelength.

 

Fig. 1. (a) VECSEL gain structure design and technology. (b) Reflectivity and photoluminescence of the GaAs-based VECSEL gain structure bonded on SiC emitting at 1µm.

Download Full Size | PDF

2. High power VECSEL device design and technology

2.1. VECSEL gain structure design and technology

The GaAs-based VECSEL gain structure emitting at λ = 1 µm designed for ~800 nm pumping (Fig. 1), was grown by MOCVD in a D180-Veeco TurboDisc reactor using TMGa, TMAl, TMIn, and AsH3 at 60 mTorr and at a temperature of 700° C. It is composed of an epitaxial high-reflectivity (99.9%) bottom AlAs/GaAs Bragg mirror (27.5 pairs) and an active layer on top, designed with 6 strained balanced InGaAs/GaAsP QWs in a 13λ/2 GaAs long active region, avoiding crystal darklines (fast traps). A low excitation carrier lifetime τ_e > 10 ns was measured, ensuring complete bleaching of residual absorption in weakly-pumped laser areas. Under high injection at threshold, it goes down to τ_e ≃ 2−3ns due to radiative recombinaisons, as calculated [11]. The QW’s are distributed among the optical standing-wave antinodes with a distribution function 111010100100 (from top surface to bragg mirror) such as the excited carrier density is almost equal in the QW’s, to ensure a low laser threshold and a homogeneous gain [4]. The QW number is optimized for low threshold with 1−2% cavity losses. A 30 nm AlAs confinement layer and a 8 nm GaAs caping layer are added on top. The structure was grown in reverse order on a (100) GaAs substrate. An AlGaAs etch-stop layer was introduced between the substrate and the active region. A λ/2 thick GaAs phase matching layer was added on top of the bragg mirror to finish the growth and later enhanced the reflectivity (metal mirror).

To greatly improve heat dissipation, the VECSEL gain structure was bonded onto a 270 µm thick SiC substrate (thermal conductivity of 490W/m.K) on the Bragg mirror side by solidliquid inter-diffusion bonding with a sub-micron thin AuIn₂ [10] layer. The AuIn₂ alloy is formed with a weight concentration of 54% of indium and the initial thickness of the metal layers are calculated accordingly. A 30nm Ti layer is evaporated on both the epitaxial structure and the carrier substrate to improve adhesion. 150nm of Au are then deposited on top of the carrier substrate and 150nm Au, 600nm Indium and 20nm Au on top of the active structure. The 20nm of Au on the active structure are required to avoid the oxidation of the Indium layer prior to bonding. This thin bonding layer process provides low thermal impedance together with mechanical stiffness and flatness required for this membrane technology. The drawback here is a possible low quality bonding/thermal uniformity over large surface, as observed here. The AuIn₂ bonding is formed at 200°C, but its fusion temperature is 490°C; this allows further processing steps at elevated temperatures and does not induce any strain-related degradation of the optical properties of the structure. A pressure of about 250kg/cm ² is applied to ensure a good contact surface of the two wafers while the temperature is ramped to 200°C and maintained at this temperature during 2 hours. The pressure is kept constant during the cooling down process. Then both mechanical polishing and wet selective etching was utilized to remove the GaAs substrate. The remaining part of the substrate is removed by selective wet etching down to the AlGaAs etch-stop layer. Citric acid (50mg of C ₆ H ₈ O ₇, 50ml H ₂ O, 15 ml H ₂ O ₂) has been used that exhibits a very high selectivity. Finally, the etch-stop layer has been removed with HF 10%.

The technological process is finished by evaporating a λ/4 thick Si₃N₄ antireflection coating on the bonded VECSEL membrane, using PECVD at high temperature (300°C), to suppress microcavity effect and pump reflection. Thus 88% of the incident pump power P_p is absorbed in the active region thick GaAs barriers. Figure 1(b) shows the reflectivity and low excitation photoluminescence spectra of the SiC bonded structure, showing an optimized design with a good matching between the bragg mirror centre wavelength and the QW gain wavelength for 300K operation.

 

Fig. 2. (a) High power single frequency tunable VECSEL design, (b) Thermal lens-based cavity stability: experimental and calculated Gaussian waist w ₀ (@1/e ²) on the VECSEL gain mirror varying with L; in inset : experimental pump profile and temperature induced index profile (FEMLAB simulation) in the VECSEL structure (8 W pump power).

Download Full Size | PDF

2.2. VECSEL device: short plano-plano cavity principle pumping at large angle

The VECSEL device is formed by the VECSEL gain mirror, a millimeter air gap L = 3 – 10mm (round trip time τ_RT = 2L/c ≃ 67 ps for L = 10mm), and a commercial flat mirror with a reflectivity of 1-T_oc ≃ 99% [Fig. 2(a)]. The chip is soldered on a Peltier element to stabilize the chip temperature with a precision of 10⁻³ K. The external mirror is held by an ultrastable mirror mount (New Focus 9882), mounted on a small Piezoelectric Transducer (PZT) to tune the cavity length, thus the laser frequency. The VECSEL gain structure is optically–pumped in cw by a commercial fibre–coupled high power multimode GaAs laser diode (Unique Mode, P_p = 8W at 808 nm, 200µm core diameter). The pump beam was focused on a spot size of 2w_p = 232µm diameter (FWHM; top-hat like shape [Fig. 2(b)] with two commercial achromat lenses at a incidence angle of ~ 45°, leading to a strongly elliptical beam shape. The components are glued on a home made breadboard and inserted in a metallic box for thermal and acoustic noise isolation (excepted the pump diode chip).

To reach high power, the high conductivity host substrate provides efficient and relatively uniform heat removal from the VECSEL chip. We measured a device thermal resistance of R_th = 4.5±0.5K/W_inc relative to the incident pump power for this pump area, in rather good agreement with the 2D symmetric simulation (FEMLAB) [Fig. 2(b)]. The fraction of heat power generated is ≃ 0.35 here. R_th was determined under laser operation by measuring both the laser wavelength shift with incident power (Δλ/ΔP_inc) and the wavelength shift with temperature (Δλ/ΔT = 0.35nm/K).

To be in a single frequency light state, the laser has to operate on a single transverse and longitudinal mode, and light polarization state (linear along [110] here) [3]. First, in order to stabilize only the fundamental transverse mode in this short free space unstable plano-plano cavity, we took advantage of pump induced positive thermal lens in the VECSEL chip [5], without here the need of a concave mirror and even using an elliptical low quality pump beam shape (Fig. 2, see section 3). Thus, for homogeneous thermal bonding, the laser beam shape would follow the pump beam symmetry. To stabilize only one longitudinal mode in the gain bandwidth, we took advantage of the ideal homogeneous gain behavior of QW VECSEL [4], where non-linear mode interactions are weak. Then we chose a short cavity length L < 10 mm, without the need of any intracavity filter, to prevent from multimode operation, or mode hopping, due to technical perturbations, specially while using a multimode noisy pump [4–6]. We chose a cavity length short enough so that the characteristic time (≪1 ms here) to stabilize a single longitudinal mode [4] is shorter than the characteristic time of strong technical fluctuations (mechanical, thermal...), already potentially large in the kHz frequency range (see Fig. 6).

This high-Q external-cavity VCSEL design leads to low noise laser emission. This short cavity design allows a broad continuous frequency tuning [5]. These QW VECSEL design principles are valid at any wavelength by using suitable semiconductor materials.

 

Fig. 3. (a) Simulated far field phase map and horizontal intensity profile (dotted line) after 100 round trips for L = 7.5 mm; (b) Far field phase map at high power recorded with a wavefront sensor (field curvature zernike term removed); intensity profile (dotted line) at high power (2.1 W) recorded with a beam profiler (solid line : Gaussian fit).

Download Full Size | PDF

3. Single transverse mode stability and spatial coherence at high power: Fresnel simulation and experiment

We observed TEM₀₀ operation over a wide range of cavity lengths and pumping values. In the experiment, depending on the spot position on the chip, we observed either a quasi-circular laser beam shape, or a slightly elliptical one in both directions. Thus the thermal contact is not uniform over the pump beam and the index profile does not necessary follow the elliptical pump shape symmetry. Figure 2(b) shows the measured and calculated cavity stability diagrams for a Gaussian TEM₀₀ beam of waist w ₀ at the diffraction limit, assuming a plano-concave stable cavity here [12]. It shows that even using a unstable plano-plano cavity and a top-hat like shape poor quality elliptical pump beam, the laser is stabilizing a single fundamental mode in a plano-concave like cavity. Taking an average index change in the VECSEL gain mirror of dn/dT ≃ 2.7×10⁻⁴/K as measured in our structure, we simulated (FEMLAB) the induced temperature/index change radial profile for a circular top-hat like shape pump beam [Fig. 2(b)].

In this plano-plano cavity disturbed by a non ideal thermal lens, the optical cavity is unstable. Indeed, the index profile induced is not a parabolic shape, due to the launched top-hat-like pump shape. However, we simulated in the Fresnel diffraction approximation that after less than N=100 round trips, thus typically the photon lifetime τ_p ≃ τ_RT /T_oc ~ 100 × τ_RT (the characteristic time to reach the laser steady state [4, 12]), the laser cavity is stabilizing a quasi-Gaussian transverse TEM₀₀ mode. We calculated that our thermal lens-based cavity behaves like a single transverse mode optical resonators [13], where higer transverse modes can not oscillate.

For the simulation, we used the electric field transmission function t_V after one reflection in the VECSEL gain mirror, taking into account the simulated temperature induced index profile, as

where k ₀ = 2π/λ is the wavevector in free space, Δn(r) is the complex index change profile simulated [Fig. 2(b)], taking into account gain guiding with 1.2% modal gain here, and L_µc = 2.4µm is the wave penetration length in the VECSEL gain mirror. We propagated in the unstable cavity an ideal Gaussian beam field E ₀(r) = exp[−(r/w ₀)²] (with a near field waist w ₀ as in the experiment), launched on the flat mirror (plane wave), over N round trips passing through the VECSEL gain mirror. We used an iterative (m+1) Fresnel diffraction equation (r ² = x ²+y ²),

We thus injected the incident field on the VECSEL gain mirror E_m obtained after m round trip as the new entry. We numerically solved the 2D Fresnel Integral [Eq. (2)] thanks to a Fourier transform based method [14]. We used a (201×201) point matrix with a spatial resolution of 6.8 µm. The numerical simulation is ended by calculating the laser far field in the Fraunhofer approximation (Fourier transform of the near field). We obtained a similar theoretical far field intensity profile and phase distribution as in the experiment at high power.

The far field phase map at high power was recorded with a wavefront sensor based on lateral shearing interferometry technology (SID4-PHASICS). The intensity profile was also recorded with a beam profiler for more resolution. For this result, a quasi-circular far-field divergence angle of 0.14° (FWHM) was measured with a second moment leading to M² ≅ = 1.2 (Fig. 3) at high power. The second moment measurement and far field phase map shows that the beam is close to diffraction limit with a quasi-Gaussian shape, in spite of a poor quality elliptical multimode pump diode beam, in good agreement with the simulation.

4. Free running single frequency operation

4.1. Single Longitudinal Mode at high Output Power

Single frequency TEM₀₀ operation is observed at ~ 1.02µm and for cavity length ranging from L = 3 to 10 mm. The maximum output power measured is P_out = 2.1W at RT in cw [Fig. 4(a)] for P_p = 8W and L ≃ 5 mm, limited by the available pump power. The apparent high threshold and non-linear slope efficiency is due to bad pump to laser overlapping without thermal lens, leading to a > 100% actual efficiency as the laser waist reduces. Note that QW absorption is negligible and quickly bleached here, as laser waist is not much bigger than pump waist above threshold (~ 20 × transparency density at peak intensity) and laser intensity rapidly stronger than the absorption saturation intensity respectively. The SMSR is > 45dB (apparatus limited) at maximum power [Fig. 4(b)], while the calculated quantum limit gives 60 dB here [4]. The light polarization is linear along the [110] crystal axis thanks to gain dichroïsm in QWs [3, 4]. The orthogonal polarization extinction ratio is > 30 dB. We recorded a mode hope free tunability of ≃ 30 GHz (or 0.1 nm) for L=5mm, varying L with the PZT voltage. The measured laser (gain) tuning rate with temperature is ~ 120 GHz/K.

4.2. Intensity and frequency noise in free running operation

This high-Q VECSEL design leads to low RIN and frequency noise operation despite the use of multimode high power pump diode. The pump and the free running single frequency VECSEL were studied in terms of RIN spectrum (Fig. 5). We use a 50Ω adapted photodiode at a photocurrent level well above the equivalent thermal noise of the detection device (~ 2 mA). Below typically f=5 kHz, the VECSEL and pump RIN’s are influenced by a set contributions (mechanical, thermal and electronic) which can evolve over two consecutive measurements. Above f=5 kHz, the VECSEL RIN is pump limited up to the VECSEL cutoff frequency [f_c^L ~ 41MHz for L=7.5mm from Eq. (4)] where the fluctuation power spectral density reaches the shot noise level. The VECSEL RIN quantum limit RIN_Q for f < f_c^L is far below here, as calculated [15]

 

Fig. 4. (a) Single frequency VECSEL output power in cw at RT. (b) Laser spectrum at high power recorded with a high resolution confocal Fabry Perot (32MHz FWHM); in inset: spectrum recorded with an optical spectrum analyzer (15 GHz resolution).

Download Full Size | PDF

where η ≫ 1 is the pump rate above threshold relative to QW transparency, and β_sp ≃ 1.1 the spontaneous emission factor.

The dynamics is thus in the relaxation-oscillation-free regime, as the photon lifetime τ_p > τ_e the carrier lifetime, with a laser cutoff frequency f_c^L below 100 MHz. The laser is in a class-A regime and tends to a class-B regime if the cavity length is decreased [4, 12]. The theoretical pump to VECSEL RIN transfer function (Fig. 5) is given by [15]:

where ω = 2πf. The relative rms fluctuation in the 10 Hz – 300MHz frequency range is < 0.1% at 0.8W output power. Between laser cutoff frequency f_c^L and the cavity Free-Spectral-Range value (f=40MHz- 20 GHz), the laser intensity noise is at the shot-noise fundamental limit, leading to an ultra low noise source in this frequency range.

The VECSEL frequency noise was also measured using a home-made stable plano-concave Fabry-Perot interferometer (2 cm long, 350 finesse, f_c^FP = 21MHz cutoff frequency) as frequency discriminator [Fig. 6(a)]. The Fabry-Perot components are glued on a home made breadboard and inserted in a metallic box for thermal and acoustic noise isolation. The operating point is set at 60% of Fabry-Perot transmission maximum leading to a conversion slope of 5.3 %/MHz [inset of Fig. 6(a)]. The photodiode current was recorded with an optimizedsoftware driven acquisition card, leading to weak dead times (~20 %). This allows to perform noise spectra averaging while the laser stays near the choosen operating point of the Fabry-Perot interferometer, in spite of a quite large frequency-conversion slope and quite low start fequency (20 Hz).

Below 1 kHz, the frequency noise is limited by thermal and mechanical contributions (characteristic peaks around 200 Hz) and 1/f technical noise. The frequency noise spectral density is measured up to the cutoff frequency f_c^FP of Fabry-Perot interferometer (lower than f_c^L), where the frequency noise comes close to the fundamental quantum white noise limit (~ 0.1 Hz²/Hz up to 200MHz as calculated in our case) [4, 15]. However it is experimentally limited here by the VECSEL RIN level above 2 MHz. Above f_c^FP, the signal reaches the background level, with preeminent shot noise contribution.

 

Fig. 5. Pump RIN_p, VECSEL RIN at P_out = 800mW and theoretical RIN transfer function spectra.

Download Full Size | PDF

 

Fig. 6. (a) Frequency noise spectral density at 800mW output power (L_c=7.5 mm) : experiment (solid line) and theory without mechanical noise (dash line); in inset Fabry Perot transmission spectrum. (b) Laser power spectral density deduced from experiment showing a 37 kHz linewidth (FWHM) over 1 ms.

Download Full Size | PDF

As already discussed in our previous work [5], we believe that the main contribution of the frequency noise above f=200 Hz is the pump intensity noise induced thermal fluctuations. Indeed, a fraction of pump fluctuations is converted to thermal fluctuations in the VECSEL gain structure (thermal noise spectral density ≃ RIN_PP ² _pR ² _th), which in turns shifts the optical path length of the laser resonator (via optical index change dn/dT), leading to frequency fluctuations. For instance, in the f=10 kHz - 200 kHz frequency range, the frequency noise spectral density exhibits several narrow peaks which are characteristic to switching-power-supply-biased pump laser noise. The temperature to frequency fluctuations conversion coefficient (average in the VECSEL gain structure) is Γ₀ ≃ dn/dT × c/λ × L_µc/L [5], up to a thermal cutoff frequency of few kHz here. The thermal contribution to the frequency noise spectral density is thus given by

This macroscopic model is valid for f ≪ 10MHz. We simulated the pump to optical frequency noise spectral density transfer function (using FEMLAB; close to a 2^nd order short pass filter with a ~4 kHz cutoff frequency), which we multiplied by the pump RIN (smoothed here). The theoretical spectrum shows relative good agreement with the experimental one [Fig. 6(a)].

From the experiment, we obtain a rms laser frequency noise of 32 kHz (over 1 ms) for L ≃ 7.5mm without any active frequency stabilization. Finally, the linewidth given in Fig. 6(b) was performed from integral computation of the frequency noise spectral density [15]. This leads to a Gaussian–like shape and a linewidth of Δν_L ≃ 37 kHz (FWHM) over 1 ms, with a slight enlargement of wings due to relatively strong noise peaks in the frequency noise spectrum around f=100 kHz. This linewidth value is similar to what would be measured using a standard heterodyne technique.

5. Conclusion

We demonstrated a compact highly coherent and efficient high power QW semiconductor laser at the multiwatt level (2.1W, pump limited). The single frequency laser exhibits low intensity and frequency noise (sub-40 kHz linewidth) and is continuously tunable. It is based on a short plano-plano cavity VECSEL, where a VECSEL membrane is bonded on SiC for thermal management. Better bonding/thermal uniformity could be achieved with an alternative method based on Au electroplating to create a ~ 100µm thick gold substrate [7]. Work is in progress in this direction. We obtained diffraction limit quasi-Gaussian beam even pumping with a low beam quality (elliptical top-hat like shape) multimode diode laser. We showed that our thermal lens-based cavity behaves like a single transverse mode optical resonators.We showed that pump properties define the cavity design and laser coherence. The laser RIN and linewidth can be further reduced using noiseless excitation, like low noise pump or electrical pumping.

This design principle and the laser properties obtained can be extended to any wavelength by using suitable semiconductor materials, and is scalable to higher output powers using larger active diameters [5]. This low noise compact device shows higher quality than standard commercial laser diodes or solid-state lasers.