The membrane external-cavity surface-emitting laser (MECSEL) has recently emerged to solve some of the constraints limiting wavelength extension and power scaling of vertical external-cavity surface-emitting lasers (VECSELs) [1]. The key distinctive feature is that the gain element does not contain a monolithically integrated distributed Bragg reflector (DBR) as is the case in conventional VECSELs [2,3]. Second, the MECSEL enables the gain structure to be placed between two heat spreaders and, hence, potentially leads to a more efficient cooling and better power scaling capability. Pioneering work towards this achievement was made by Iakovlev et al. who proposed a very similar setup [4] in 2014. Furthermore, Yang et al. investigated a semiconductor membrane laser, also called DBR-free semiconductor disk laser [5], with the laser-active semiconductor membrane bonded onto one IC heat spreader. They also proposed the laser-active region arrangement between two heat spreaders as thermally beneficial. Recently, a MECSEL with diamond heat spreaders operating around 985 nm was reported [6]. Moreover, Yang et al. succeeded in demonstrating [7] a MECSEL double-side cooled using SiC emitting approximately at 1037 nm and also reported on the comparison with the single SiC heat spreader approach [8]. The use of SiC as an IC heat spreader has its benefits in the huge thickness homogeneity, the high degree of parallelism, and the comparably low price related to a single-crystal diamond.

In this Letter, we present a study comparing single-side pumping (SSP) and double-side pumping (DSP) of a MECSEL, as well as a detailed set of characterization data. The possibility of DSP is a built-in benefit of the MECSEL itself, enabling the pumping and treating of this type of active region membrane heat spreader package in a similar way, like classical solid-state (thin-disk) laser elements [9,10]. This study was performed with an active region optimized for 780 nm emission with the aim at extending the accessible wavelength range into the red/near-infrared transition region and to meet the need of high-quality beam sources in this wavelength range.

The MECSEL structure was grown on an undoped 2 in. GaAs $(100)\pm 0.5°$ wafer using a V80H-10 VG Semicon solid source molecular beam epitaxy system at a growth temperature of about 575°C. A GaAs buffer layer was first grown on the substrate followed by an $\sim 150\mathrm{nm}$ thick AlAs process layer and the active region, which was designed to be completely symmetric and resonant for 780 nm. The total thickness of the gain region was 577 nm [see the scanning electron microscope (SEM) picture inset in Fig. 1] and corresponded to a resonant $2.5\lambda$ subcavity. It consists of twelve ${\mathrm{Al}}_{0.09}{\mathrm{Ga}}_{0.91}\mathrm{As}$ quantum wells (QWs) with a thickness of about 7 nm each, arranged in four groups of three QWs each. The QWs are separated and embedded by layers of ${\mathrm{Al}}_{0.39}{\mathrm{Ga}}_{0.61}\mathrm{As}$—8 nm in between the QWs and 12 nm around the QW package. These four QW packages can be seen in the SEM inset in Fig. 1 as vertical stripes, which are a little bit brighter than the surrounding and equally spaced within the active region. As spacer material separating the QW groups, we employed ${\mathrm{Al}}_{0.58}{\mathrm{Ga}}_{0.42}\mathrm{As}$, which has an absorption edge at $\sim 550\mathrm{nm}$ [11]. The whole active region was then embedded between a 10 nm thick AlInP and a 10 nm thick GaInP layer. These layers helped block the electron diffusion and act as window layer at the semiconductor heat spreader interface.

 

Fig. 1. Detailed schematic drawing of the V-shaped MECSEL setup including the resonator with its mirrors M1, M2, and M3, the pump optics, and the pump beam path, as well as all devices that were used for characterization.

Download Full Size | PDF

In order to isolate the gain membrane, a piece of about $7\mathrm{mm}\times 7\mathrm{mm}$ of the as-grown structure was bonded with the epitaxy side onto one $\sim 350\mathrm{\mu m}$ thick and uncoated 4H-SiC wafer piece. The SiC wafer piece was of about the same size as the gain membrane sample and acts later as an IC heat spreader. After the substrate and AlAs layer removal process (details can be found elsewhere [1]), a second (uncoated) SiC heat spreader originating from the same SiC wafer was bonded on top of the SiC-membrane package. The whole sandwich was then clamped between water/glycol-cooled and Peltier-element temperature controlled copper plates. Indium foil was used to improve the thermal contact between the SiC heat spreaders and the copper plates. For all measurements presented here, the heat sink temperature of the gain element was set and stabilized to $T_{\mathrm{hs}}=20°\mathrm{C}$. The gain sandwich clamped between copper plates was mounted to a $XYZ$ stage, which was mounted onto a tilt/jar stage itself. The whole characterization setup was built around this gain element mount and is schematically shown in Fig. 1. Each of the copper plates had a conical aperture of about 1.5 mm in diameter (see photograph inset of Fig. 1) to enable vertical laser emission as well as pumping from both sides. As a pump laser, we used a Coherent Verdi-V18, which delivered up to 18.5 W of pump power at a 532 nm emission wavelength.

For an accurate comparison of SSP and DSP of the gain element, incident pump power $P_{\mathrm{inc}}$, reflected pump power $P_{\mathrm{refl}}$, and transmitted pump power $P_{\mathrm{trans}}$ were measured directly before or after the gain element and the absorbed pump power $P_{\mathrm{abs}.,\mathrm{pump}}$ could be calculated. For DSP, the pump beam was divided with a polarizing beam splitter (PBS). A $\lambda /2$ plate was used to adjust the polarization of the pump beam before the PBS to divide the pump beam in such a way that the different losses in the two different pump arms 1 and 2, as well as the polarization depending variation in reflection at the air–SiC interface, are compensated, and a symmetrically balanced pump absorption situation was created. The distribution of the absorbed pump light for the two incident directions related to the pump arms and the total absorbed pump light (100%) within the gain package was 50.03% for pump arm 1 and 49.97% for pump arm 2. For the high-power measurements with SSP, the PBS was removed, and only pump arm 1 was used. A pump spot diameter of $(88\pm 8)\mathrm{\mu m}$ in the tangential and $(91\pm 8)\mathrm{\mu m}$ in the sagittal plane was calculated from a measured pump lens distance of $(204\pm 1)\mathrm{mm}$. For the MECSEL characterization, a V-shaped resonator with a total length of $\sim 495.5\mathrm{mm}$ was used (see Fig. 1). The gain element was placed at the waist of the IC laser beam between mirrors M1 and M2 [both high reflective (HR)], which is located at a distance of approximately 98.5 mm from M1. Mirror M3 was either HR with $R_{\mathrm{M}3}>99.9\%$ (hereinafter referred to as the HR outcoupler) for the broadband tuning measurement or partly transmissive for the MECSEL’s wavelengths with $R_{\mathrm{M}3}=(95\pm 1)\%$ (hereinafter referred to as the 5% outcoupler) for all other measurements presented in this Letter. The calculated IC beam waist diameter was $\sim 80\mathrm{\mu m}$ in the tangential and $\sim 155\mathrm{\mu m}$ in the sagittal plane. Therefore, mode matching was achieved for the tangential plane during laser operation. In the sagittal plane, the mode diameter was pump spot limited.

The power transfer behavior was measured for both configurations, SSP and DSP, using a Thorlabs PM200 with S310C sensor head. The output power ($P_{\mathrm{out}}$) results are plotted in Fig. 2 as a function of $P_{\mathrm{out}}$ over $P_{\mathrm{abs}.,\mathrm{pump}}$. For DSP, a slightly smaller pump threshold of 0.69 W was determined compared to SSP via pump arm 1 (PBS was removed) with a threshold of 0.79 W. The maximum output power increased from 3.13 to 3.22 W, and the differential efficiency ${\eta }_{\mathrm{diff}}$ increased from 31.9% to 34.4% for the DSP configuration. ${\eta }_{\mathrm{diff}}$ was determined via linearly fitting the data points between 2.5 and 8 W of $P_{\mathrm{abs}.,\mathrm{pump}}$. In the DSP configuration, the MECSEL reaches its maximum power output slightly earlier than with SSP, but the thermal influence on the slope behavior starts for both configurations at about 8 W of absorbed pump power, from which the output values deviate from the expected linear increase. This was also recently observed by Yang et al. [7].

 

Fig. 2. Power transfer measurements of the free-running MECSEL in SSP (pump arm 1) and DSP configurations. The inset shows an enlarged view of the threshold region.

Download Full Size | PDF

An essential figure of merit enabling us to compare the pump configurations (DSP, SSP arms 1 and 2) and to determine the thermal behavior [12] of the system is the thermal resistance $R_{\mathrm{th}}$. Therefore, spectra were taken whilst power transfer measurements for all pump configurations at the same spot on the sample were performed, determining the spectral shift per change of dissipated power $\mathrm{\Delta }\lambda /\mathrm{\Delta }{P}_{\mathrm{diss}}$.

$P_{\mathrm{diss}}$ was calculated as follows: $P_{\mathrm{diss}}=P_{\mathrm{abs}.,\mathrm{pump}}-P_{\mathrm{out}}$. Furthermore, the thermal shift of the laser emission $\mathrm{\Delta }\lambda /\mathrm{\Delta }{T}_{\mathrm{hs}}$was measured under constant pumping of $P_{\mathrm{abs}.,\mathrm{pump}}=2.53\mathrm{W}$ with a duty cycle of 5% to avoid heating effects of the pump laser. The spectra were recorded with a StellarNet Blue-Wave spectrometer (resolution limit 0.8 nm). Figure 3 shows the spectral positions of the long wavelength flank at half-maximum of the laser emission plotted over dissipated power $P_{\mathrm{diss}}$ and heat sink temperature $T_{\mathrm{hs}}$, respectively. The linear fits applied to the data show the lowest spectral shift per dissipated power of 1.34 nm/W for the DSP configuration. For SSP via pump arms 1 and 2, the spectral shifts per dissipated power are both higher (1.43 nm/W and 1.52 nm/W for SSP arms 1 and 2), but differ slightly. We connect this difference with pump spot sizes that are not identical due to slight variations in the pump lens distances. The heat sink temperature tuning measurements reveal a spectral shift of 0.25 nm/K. The thermal resistance can now be calculated [12] to $R_{\mathrm{th},\mathrm{DSP}}=5.36\mathrm{K}/\mathrm{W}$ for DSP and, considering the average, $R_{\mathrm{th},\overline{\mathrm{SSP}12}}=5.9\mathrm{K}/\mathrm{W}$ for SSP.

 

Fig. 3. Spectral shifts of laser emission $\mathrm{\Delta }\lambda$ plotted over dissipated power $P_{\mathrm{diss}}$ (full squares, circles, and upwards pointing triangles) and heat sink temperature $T_{\mathrm{hs}}$ (full downwards pointing triangles) are plotted here.

Download Full Size | PDF

The spectral tuning was measured using IC birefringent filters and the results are plotted in Fig. 4. Two different filters, 0.5 mm and 2 mm in thickness, were used as the 0.5 mm thick filter with a free spectral range of $\mathrm{\Delta }{\lambda }_{780\mathrm{nm}}=136.6\mathrm{nm}$ that did not introduce enough losses to properly tune the laser with the 5% outcoupler. A wide tuning range of 44.5 nm from 767.0 to 811.5 nm was determined using the HR outcoupler and the 0.5 mm thick birefringent filter. With the 5% outcoupler and the 2 mm birefringent filter, a tuning range of 22.5 nm was reached, while the laser exceeded 1 W output from 776 to 790 nm with a maximum of 1.61 W at 781.3 nm.

As a further important laser parameter, the beam quality factor $M^2$ was measured. It revealed a value of $M^2<1.09$ ($M_X^2<1.134$ and $M_Y^2<1.040$, including device inaccuracy of 5%), performed at $\sim 2\mathrm{W}$ of absorbed pump power with a Thorlabs M2MS that was equipped with a Thorlabs BC106N-VIS camera. The emitted beam possesses a slight ellipticity of 1:1.19 originating from the V-shape of the resonator, but it was limited by the pump beam diameter as mentioned above.

Determining the spectral emission characteristics of the free-running MECSEL, a high-resolution spectrum was recorded (see Fig. 5) behind mirror M2. For this spectral measurement, an Ando AQ6317C optical spectrum analyzer with a resolution limit of 0.02 nm was used. The visible Fabry–Perot resonances, which are equally spaced by $\mathrm{\Delta }\lambda \approx 0.32\mathrm{nm}$, are caused by the single SiC heat spreaders. Owing to the excellent thickness homogeneity of the SiC wafer pieces, no beat-node was visible, in difference to other reports where diamond heat spreaders were used [1,6]. The spectral spacing can be connected to the wafer thickness, which is measured to be $\sim 350\mathrm{\mu m}$.

 

Fig. 4. Tuning measurements at $P_{\mathrm{abs}.,\mathrm{pump}}=7.58\mathrm{W}$, taken with the 5% outcoupler and HR outcoupler.

Download Full Size | PDF

 

Fig. 5. Typical high-resolution spectrum of the free-running MECSEL is shown here.

Download Full Size | PDF

Looking at the power transfer behavior of both pump configurations (Fig. 2) and assuming the pump light absorption follows Beer–Lambert’s law, one could explain the lower laser threshold, the improved differential efficiency by 2.5% points, and the slightly higher maximum output power by 90 mW for DSP by the more homogeneous distributed pump light within the active region membrane. A closer look at the thermal rollover point reveals a slightly earlier rollover for the DSP configuration. This could be understood, as the more homogeneous pump distribution also leads to a less strong temperature gradient within the active region, assuming heat is created where absorption takes place. This means that once the gain structure rolls over thermally, this happens more simultaneously within the structure in the DSP configuration. On the other hand, in the SSP configuration, where a stronger thermal gradient can be expected, this side of the gain membrane facing the pump light might experience a slightly earlier rollover than the side turned away from the pump light, and the laser maintains operation up to a slightly higher $P_{\mathrm{abs}.,\mathrm{pump}}$ before the thermal shutdown. A stronger thermal gradient could be beneficial for the heat dissipation within the semiconductor membrane and finally into the SiC heat spreaders. Also, sufficient charge carrier diffusion [13] from the hot to the cold side in the semiconductor might partly compensate negative thermal effects. Despite the fact that higher local temperatures inside the gain membrane are present [12] for SSP, which is confirmed by the $\sim 9\%$ higher thermal resistance, the determined differences comparing DSP and SSP in laser threshold, differential efficiency, maximum output power, and thermal rollover are rather small. This already indicates relative good heat removal conditions. The thickness of the gain membrane is obviously small enough, dissipating the introduced heat properly, while not experiencing a huge impact of the different pumping schemes. Due to this, finding an optimum gain membrane thickness containing more QWs, considering pump light absorption and heat dissipation out of the membrane itself, has an additional margin to power scale the presented system.

In summary, a MECSEL employing SiC heat spreaders for symmetrical cooling was experimentally examined using single/double-side barrier pumping. This MECSEL represents the first one operating at room temperature as well as in DSP configuration. A decrease of 9% in thermal resistance was determined for DSP explained by the 2.8% improved value in maximum output power of 3.22 W, the increase in differential efficiency to 34.4% by 2.5% points, and the reduced absorbed pump power by 12.7% for the laser threshold. The presented laser with a maximum operation range from 767.0 to 811.5 nm and excellent beam quality factor of $M^2<1.09$ extends the available range of vertically emitting high-power semiconductor lasers and paves the way for further research on power scaling of MECSELs, where DSP or even multi-pass pumping [10] might become relevant.