1. Introduction

The high-power 97x nm sources are attractive for nonlinear frequency conversion [1], laser pumping [2], material processing [3], life science [4,5], etc. Among various laser technologies, although the solid Yb³⁺ gain covers the wide emission wavelength of 950 nm-1200 nm, the lasing at 97x nm may have a lower efficiency and a higher threshold, caused by the gain competition with 1030 nm and strong amplified spontaneous emission [2]. Conventional semiconductor lasers, employing flexible bandgap engineering, can achieve high-performance emission at this band, but there are some other limitations: The edge-emitting lasers exhibit excellent power performance and efficiency, while the elliptical light spot and large divergence angle may limit further applications [6]; For surface-emitting laser, they are with better beam quality and spectrum features, but the relatively small gain volume may lead to smaller modal gains and limit the emission power. Fortunately, the above issues will be mitigated in the semiconductor disk lasers. They are well-suitable for applications requiring high power and high beam quality at specific wavelengths. Especially, the introduction of the external cavity can further offer flexible functionality, such as intracavity harmonic and terahertz generation [7,8], ultrafast pulse generation [9], etc.

The SDLs have captured increasing attention following the significant results reported by M. Kuznetsov and A. Mooradian [10]. To date, they have covered an emission range from ultraviolet to infrared [4–6], and the recorded maximum power has reached beyond 100 W [11]. For microcavity lasers like SDLs, the performance is predominantly limited by heating [12–14]: Without an anti-reflection coating, a Fabry-Perot microcavity will be formed between the semiconductor-air interface and the distributed Bragg reflector (DBR) [15]. To realize a higher gain, quantum wells (QWs) should be seated at the microcavity resonance peak and form the resonant periodic gain (RPG) structure. While the gain peak of QWs and the microcavity resonance are with different temperature shift rates, the heating may easily make the QWs away from the peak and invalidate the RPG structure [16]; Besides, with temperature increased, the QWs material gain will also be significantly decayed [17]. Additionally, non-radiative losses, such as Auger losses, also become more relevant at this time [13,18]. Thus, a higher pump is required to maintain the needed gain, this self-perpetuating process causes an intensified heat; Moreover, the carrier in the QWs also easily escapes into barriers at high carrier densities and temperatures, leading to increased non-radiative loss. All the above effects are superposed upon the device until the thermal rollover occurs. In particular, since the smaller energy bandgap difference between the barrier and QW, 920-980 nm SDLs may suffer from poor carrier confinement and are more affected by heat [19]. As a review pointed out [20], the recorded maximum output power of 920-980 nm SDLs is only about 40 W, and no new records have emerged in the past fifteen years.

Here, we demonstrate systematic strategies for the performance improvement of the 970 nm SDLs. In this exploration, the chip design, packaging process, resonator, pumping, and cooling strategies have been optimized: Firstly, for chip design, the thermal performance is improved by designing the wafer in a flip-chip configuration and using the strain-compensated multi-well RPG structure. Furthermore, since the gain peak and microcavity peak have different temperature shift rates, a detuning is predesigned to guarantee the RPG structure; For high-quality packages, the soldering process has been optimized, and the scanning acoustic microscopy images demonstrate that the optimized soldering interface has good uniformity. We further explore two different heat sinks to optimize heat conduction, and the higher thermal rollover power and lower thermal resistance in the diamond-soldered device confirm its superior heat conduction performance. To further increase the emission power, different resonator, pumping, and coolant settings are employed to optimize the performance. As a result, the 970 nm SDL can emit a CW power of 70.3 W, and a thermal resistance of 0.49 K/W is attained. The laser is highly efficient with a maximum slope efficiency of 58.2% and the pump threshold is only around 1.83 kW/cm². Considering that high energy/peak pulsed emissions are attractive for practical applications, the performance under QCW pumping is also explored. Setting the duty cycle to about 11%, the chips can emit a peak power of 138 W without thermal rollover, and the single pulse energy reaches about 13.6 mJ.

2. Structure optimization and fabrication

2.1 Structure design and optimization

As present in Figs. 1(a) and 2(a), the chip is designed in a flip-chip structure. The InGaP and Al_0.3Ga_0.7As layers are adopted to serve as the etch-stop layer and window layer for GaAs substrate removal and preventing the nonradiative recombination at the chip surface, respectively. Next, five-periods double In_0.17Ga_0.83As/GaAs QWs are designed for 97x nm emissions. The multi-well RPG structure can provide an optimized overlap between the QWs and microcavity and is more tolerant to the variation of growth, processing, and temperature [21]. Twenty-six pairs of AlAs/GaAs DBR are employed to achieve high reflectivity at the target wavelength. Employing the commercial PIC3D software (Crosslight), the QWs’ gain characteristic can be explored based on the k·p theory [22,23]. One can observe from Fig. 2(b) that the designed QWs have a gain peak of about 957 nm at a gain temperature of 300 K. To achieve higher stimulated emission intensity, all QWs should be located at the peaks of the microcavity standing wave, quantified by the longitudinal confinement factor (LCF) [24]. The LCF is equivalent to a built-in spectrum filter, which will modulate the QWs’ material gain. The typical temperature shift rate difference between the peaks of QWs gain and LCF is about ∼0.3 nm/K [16]. To guarantee the resonant periodic gain structure at a higher temperature, there should be a pre-offset. By adopting the method reported in Ref. [24], we can calculate and design the corresponding LCF in the chip. The adopted LCF is present in Fig. 2(b), One can find that the LCF exhibits a wavelength-dependent resonance peak at 965 nm, and the corresponding pre-offset is about 8 nm. The estimated optimal matching temperature between QWs and LCF is around 320 K.

 

Fig. 1. (a) The designed gain chip structure; (b) The packaged SDL structure.

Download Full Size | PDF

 

Fig. 2. (a) Distribution of refractive index and standing wave field in the designed gain chips; (b) The QWs gain spectrum and longitudinal confinement factor; (c) Measured temperature-dependent reflection spectrum of the chip front; Insert: an enlarged detail image; (d) Measured temperature-dependent PL spectrum; Insert: the variation of PL intensity and dip reflectivity with temperature; (e) Measured the pump spot sizes at different magnifications.

Download Full Size | PDF

2.2 Epitaxial fabrication and characterization

After design, the wafer is grown by metal-organic chemical vapor deposition technology. In our previous reports, some explorations have been carried out to improve the epitaxy growth quality [25,26]. After growing, the reflection spectrums are first measured under different temperature, the dip in the high-reflection band reflects the modulated quantum well absorption, and the dip depth reflect how well the QWs and resonant peak are matched [26]. It can be found from Fig. 2(c) and 2(d) that the dip reaches the deepest at ∼318 K, meaning the better matching at this temperature. Furthermore, it can also be observed that the stop band has a red-shift rate of ∼0.07 nm/K. Another important feature is photoluminescence (PL) spectrums, which reflect the quantum wells gain modulated by the Fabry-Perot microcavity [24]. As the temperature increased, present in Fig. 2(d), the wavelength was moved from 963.5 nm to 974.0 nm and the peak intensities reached the maximum at around 313 K. The change in PL peak intensity can also reflect the match between QWs and resonant peaks. Since the pump introduced temperature rise during the testing of the PL, the temperature of the highest PL intensity is slightly lower than the deepest dip. The above result demonstrates that the actual optimal matching temperature between QWs and LCF agrees with our design.

The packaged device structure is present in Fig. 1(b). The pump source is a fiber-coupled laser diode operating at a wavelength of 808 nm, and the pump spot is reimaged onto the chip at an angle of 30 °. Two flat-convex lenses with different curvatures are placed after the output fiber for pump collimation and focusing, and a tunable pump area (A_p) of 3.84 mm², 2.83 mm², and 1.36 mm² can be obtained by setting curvatures (as present in Fig. 2(e)). Since the pump is oblique incidence and not compensated for, the spot is elliptical. The cooled water is first used to cool the heat sink at 288 K. Then, to reach lower temperatures and remove several hundreds of watts in heat load, the coolant of pentafluoropropane is employed to cool the heat sink directly, controlled by an electrically driven control system (LNEYA, Sundi-1075WY). At the same time, to avoid condensation of water vapor on the chip at low temperatures, the test was conducted on a nitrogen-dry operating platform. Since, there is no anti-reflection coating on the SDL surface, which results in about 30% reflection loss of pump power [27].

3. Packaging process and thermal resistance

Optimizing heat dissipation capability is always an important research issue for high-power lasers, whether in semiconductor or solid-state lasers, etc [28]. For SDLs, the generated heat in the small pump area is mainly transferred by heat conduction mechanisms [29]. Some previous simulation works have detailed analyzed the SDLs’ thermal characteristics, and these results can greatly guide the improvement of heat dissipation capacity [30–33]. A typical measure proposed is to adopt diamonds with high thermal conductivity to optimize the heat dissipation capacity. At the same time, to make actual performance closer to the ideal situation, reasonable optimization and setting of process parameters should also be carried out to achieve good contact interface quality between diamond and gain chip. For example, in Ref. [27], Hou et al. adopted the Cu-Sn alloy for pre-metallization to improve the diamond-bonding interface quality, which achieve a thermal resistance of 1.25 K/W for 600 µm pump spot size. In our exploration, considering the differences in thermal expansion coefficients between gain chips and heat spreaders, high-purity Indium with good ductility (elastic modulus≈1.27 GPa) is used as the bonding solder to alleviate thermal stress [34]. Furthermore, the Ti/Pt/Au layers are been employed to pre-metalize the diamond heat spreaders. The role of the Au layer is bonding with indium during the diffusion process, and the Ti layer is prone to adhere to the semiconductor surface. The Pt layer provides a diffusion barrier between Ti and Au, which is necessary to strengthen the Ti-diamond bonding. Combined with previous analysis [27], Indium solder is easy to produce defects during reflowing. Thus, some process explorations have been carried out to improve interface quality, such as optimizing solder thickness, reflow temperature, welding fixtures, etc. For example, some different solder thicknesses (2.5 $\mu m$, 5 $\mu m$, etc.) have been attempted to explore the soldering quality, and it found that it is more likely to produce voids and defects under the thinner solder, resulting in poor interface quality. Figures 3 presents the soldering interface images and the corresponding output power distribution with different indium thicknesses. The interface images are achieved by non-destructive scanning acoustic microscopy (Sonoscan, D9650). The interface with a 2.5 $\mu m$. solder is with poor uniformity and even defects in the effective pumping area (blue dashed line). At this time, the power distribution is significantly fluctuating, and the devices are easy damaged. On the contrary, the interfaces can often have good uniformity and few obvious bubbles or voids with a 5 µm solder thickness, and a relatively uniform output power distribution can also be observed. For our process settings, the 5 µm solder thickness can achieve better interface quality and will be adopted in the following experiments. It should be emphasized that minimizing the thickness of solder is beneficial for device performance and reliability theoretically. The thermal conductivity of indium is only about 80 $W \cdot {m^{ - 1}}{K^{ - 1}}$ and an increase in solder thickness will increase the thermal resistance of the laser. In addition, its thermal expansion coefficient is also different from those of gain chips and heat spreaders, and the increase in thickness will further lead to an increase in thermal stress. Excessive thermal stress may affect the reliability and even lead to device damage. If a good interface quality can be achieved with thinner solder, maybe a better heat dissipation capacity can be attained. After packaging, to further improve the thermal performance, the wet etch mechanism will be employed to remove the substrate with poor thermal conductivity. To avoid lateral etching, we coated the chip edges with the photoresist to protect cleaved faces. After removing the substrate, the total thickness of the gain chip is only about 5 µm.

 

Fig. 3. (a) and (c) Soldering interface images with different indium thicknesses of about 2.5 $\mu m$ and 5 $\mu m$, respectively; (b) and (d) are the corresponding output power distribution.

Download Full Size | PDF

For comparison, we explored two different heat sinks to optimize the heat conduction capacities. The chemical-vapor-deposition diamond and high-purity copper are both obvious choices for distributing the heat, as their large thermal conductivities of ∼ 1800 W m⁻¹K⁻¹ and 400 W m⁻¹K⁻¹, respectively [14]. The polycrystalline diamond heat spreader is with a dimension of 4 × 5 × 1 mm. According to previous simulations [30], it should be noted that there is an optimal value for the diamond thickness, rather the thicker the better. To quantitatively estimate the different packaging materials, the thermal resistance method is adopted [35]. By measuring the wavelength shift with temperature and the dissipated power, the temperature rise can be attained as T−T₀= (δλ/δT)/Δλ, where T₀ is the heat sink temperature, and the dissipated power is the difference between the pump and the output [26,36,37]. Under CW operation, setting A_p= 2.83 mm², T₀= 288 K, we measure these spectrum characteristics with a spectral analyzer (Ocean Optics, HR4000), as present in Figs. 4(a) and 4(b). As the temperature increased, the central spectrum was moved from 968.5 nm to 970.9 nm with a redshift rate of ∼0.12 nm/K. The lasing spectrum with a shift rate of 0.057 nm/W and 0.16 nm/W versus the dissipated power for the two different packaged SDLs. Figure 4(c) shows the dissipated pump powers caused the temperature rise. Thermal resistance (R_th) can be attained by calculating the slope yielding 1.32 K/W and 0.49 K/W for copper and diamond packaged devices, respectively. The emission characteristics are present in Fig. 4(d), and an obvious better result can be observed in the diamond-soldered device. The similar slope efficiency indicates comparable optics loss in the two devices. The pump powers at the rollover point are about 34.5 W and 96.2 W, and the dissipative powers are 23.7 W and 62.7 W, respectively. The higher dissipative power at the rollover point in the diamond-soldered device confirms its superior performance. In addition, the corresponding QWs’ temperatures at the rollover point can also be attained as 319.3 K and 318.7 K, which agrees with the optimal matching temperature between QWs and LCF. Compared with the previous reports [14,27,30,38], we have made some progress in optimizing the thermal resistance and may be regarded as an extension and enrichment of the earlier works on this topic.

 

Fig. 4. (a) Emission wavelength under different coolant temperatures; (b) Emission wavelength under different dissipated power; (c) Measured output power under different package materials; (d) Relationship between the temperature increase and the dissipated pump power.

Download Full Size | PDF

4. External-cavity optimization and SDL performance

4.1 Resonant cavity and pump-to-mode ratio

Performance improvement does not only rely on heat dissipation but also depends on optimizing the external resonant cavity and pumping strategy [14]. Firstly, the spatial overlap between the pump and cavity modes determines the transverse mode gain, and the matching of pump-to-mode is crucial for the threshold and efficiency of the desired modes [39]. With a thin gain, the fundamental mode size on the chip depends on the cavity length (L_c) and curvature radius (R_c) of the output mirror. When the pump beam is reimaged onto the chip surface with an area of 2.83 mm², the calculated fundamental mode spot area and pump-to-mode ratio under different resonant cavity settings are present in Fig. 5(a). One can be observed that the calculated ratio decreases with the curvature, and reaches the minimum value at half of the curvature radius. Combining with the previous theoretical analysis in Ref. [39], increasing the ratio is beneficial for achieving higher output power. The experimental results in Fig. 5(b) further confirm this conclusion well, and higher output power can be observed at a smaller curvature of 40 mm. Another method to control the ratio is to adjust the cavity length, and the corresponding experimental results can be found in Fig. 5(d), a power of 40.4 W can be attained under a larger ratio. On the other hand, we have also attempted different transmittance to explore the emission performance, one can observe from Fig. 5(c) that the SDL with a 4% transmittance is better than that of the other mirror system. It should be emphasized that employing a large pump-to-mode ratio is not conducive to achieving high beam quality, since the larger ratio is more likely to excite higher-order transverse modes. That is why the reported records of the high-power SDLs are often with a large pump-to-mode ratio, and the SDLs are usually operated in multiple transverse modes [20,38,40]. As inserted in Fig. 5(b), the emission beam profile at a 40 mm curvature is also a poor-quality multiple transverse mode, and the beam quickly diverges within a short distance, making it difficult to compress and measure the M². One way to increase beam quality and maintain good power under a large pump spot is to increase the curvature of the mirror [41], and the pump-to-mode ratio is usually required to be less than 1 to achieve high beam quality [39]. In future work, some larger curvature can be adopted to increase the fundamental mode area and strive to achieve high brightness emission.

 

Fig. 5. (a) Calculated areas of the fundamental mode spot and the area ratio of the pump to the fundamental mode under different resonant cavity settings. (b) Measured output power under different curvature radius of mirrors; Insert: the emission beam profile; (c) Measured output power under different transmittance of mirrors; (d) Measured output power under different cavity lengths.

Download Full Size | PDF

4.2 Pumping sizes and CW performance

Secondly, changing the pumping size can also impact the emission performance, which is equivalent to varying the density of the heating. Under CW operation, setting T₀ = 288 K, the thermal performance under variations of the pump area is explored, as shown in Fig. 6(a). Employing the same method, the system’s thermal resistance can be measured as 0.95 K/W, 0.49 K/W^, and 0.41 K/W for the pump sizes of 1.36 mm², 2.83 mm² and 3.84 mm². The dissipated pump powers at the rollover point are about 31.0 W, 62.7 W, and 72.7 W, and the QWs’ temperatures are 317.5 K, 318.7 K, and 317.8 K, respectively. The data clearly show that the heat dissipation capacity is improved by enlarging the pump area. Similar results can be observed in the simulation by Heinen et.al, where increasing the pump spot area is beneficial for reducing thermal resistance [30]. Figure 6(b) shows the emission performance, and one can observe that although the SDL under a pump spot area of 3.84 mm² has a better heat dissipation capacity, obviously better emission power can be attained for the smaller spot area of 2.83 mm². Combined with previous reports, the large pump size may also greatly increase the gain length in the lateral plane, significantly intensifying the guided spontaneous emission and the unwanted optical loss [42,43]. Thus, for performance improvement, the pump spot has an optimal value, rather than the larger the better. Besides, according to Fourier's law [29], a simple method can be employed to estimate the thermal conductivity (k) of the entire gain chip: $k = h/{R_{th}}A$, where h represents the layer thickness and A is the thermal conductivity area in the vertical direction of heat flux.The calculated thermal conductivity is about 640 W/m·K for the pump sizes of 3.84 mm². It should be emphasized that there is a deviation in using the pump area for estimation, as heat transfer will undergo radial diffusion, increasing the actual area [30,33].

 

Fig. 6. (a) Measured temperature increases under different pump sizes; (b) Emission power under different pump sizes; (c) Emission powers with different coolant temperatures; (d) Central wavelength versus pump powers under different coolant temperatures.

Download Full Size | PDF

Alternatively, another strategy is to further cool the QWs, which can reduce waste heat generation. In a simple picture, the QWs’ material gain can be improved by decreasing the temperature, meaning the reduced waste heat. Under CW operation, fixing A_p= 2.83 mm², the emission powers under different coolant temperatures are present in Fig. 6(c) and 6(d). At the cooled water temperature of 288 K, the device has a power of 40.4 W and a slope efficiency of about 45.6%, the pump threshold is only around 1.44 kW/cm². To reach lower temperatures and remove several hundreds of watts in heat load, the coolant of pentafluoropropane is also employed to cool the heat sink. One can observe that the output power and efficiency are all improved with the temperature decreased, and the highest power of 70.3 W and an efficiency of 58.2% are attained. The higher efficiency proves less waste heat generation. On the other hand, the pump threshold density is increased from 1.44 kW/cm² to 1.83 kW/cm². Decreasing the temperature will enlarge the detuning, leading to an increase in the threshold [44]. From the spectral perspective, with changes in temperature or pump power, the central wavelength varies between 960 nm and 973 nm.

4.3 Pulse pumping and QCW performance

The high energy/peak pulsed emissions are also attractive for practical applications: In medicine, they increase efficiency and decrease damage to healthy tissue [45]; In laser adaptive telescopes, they allow for Rayleigh blanking and fratricide avoidance [46], etc. The SDL is further adjusted to be with the QCW pumping and the pulsed performances are explored. In other words, adopting pulse pumping is also equivalent to altering the time of heat generation. As presented in Fig. 7(a), the device can be operated with tunable repetition rates and duty cycles. Setting the duty cycle to about 11%, the measured pulse width and repetition rate are 109 $\mu s$ and 1 kHz (Fig. 7(b)). Under this pumping condition, a peak power of 111.2 W and single pulse energy of 10.6 mJ can be attained at a coolant temperature of 288 K (Fig. 7(c)). By cooling the temperature to 243 K, the peak power, efficiency, and energy are further improved to 138.1 W, 55.1%, and 13.6 mJ, respectively. It should be emphasized that thermal rollover did not occur, and the power is limited by the pump. Adopting a similar method, the thermal characteristic at this time is also analyzed. It can be observed from Fig. 7(d) that the wavelength has a much smaller shift rate of 0.24 nm/W with dissipated pump peak power, and the corresponding temperature rise rate is only 0.22 K/W. The results confirm better heat dissipation under this pumping condition. It should be emphasized that the temperature rise rate varies with the pump pulse width and duty cycle [47], and lower rates may be obtained by further optimizing the pump parameters.

 

Fig. 7. (a) Measured oscilloscope train with different repetition rates; (b) Pulse in the range of 900 $\mu s$ and 20 $ms$; (c) Emission QCW peak powers under different temperatures; (d) Central wavelength and calculated temperature increase.

Download Full Size | PDF

5. Concluding remarks and discussion

In summary, we demonstrate the systematic strategies for the performance improvement of the 970 nm SDL. In the exploration, the chip design, packaging process, resonator, pumping, and cooling strategies have been optimized: For chip design, the thermal performance is improved by designing the wafer in a flip-chip configuration and using the strain-compensated multi-well RPG structure. Considering the gain peak and microcavity peak have different temperature shift rates, a detuning is predesigned to guarantee the RPG structure; For high-quality packages, the soldering process has been optimized, and the scanning acoustic microscopy images demonstrate that the optimized soldering interface has good uniformity. We further explore two different heat sinks to optimize heat conduction, and the higher thermal rollover power and lower thermal resistance in the diamond-soldered device confirm its superior heat conduction performance. To further reduce the heat generation, different resonators, pumping strategies, and coolant temperatures are employed to optimize the performance. As a result, under CW operation, a CW power of 70.3 W is attained and the thermal resistance of the 970 nm SDL can be reduced to around 0.49 K/W. The laser is highly efficient with a maximum slope efficiency of 58.2% and the pump threshold is only about 1.83 kW/cm². Considering QCW emissions are attractive for practical applications, the SDL performance under QCW operation is also explored. Setting the duty cycle to about 11%, the chips can emit a peak power of 138 W without thermal rollover, and the corresponding pulse energy is as high as about 13.6 mJ. Table 1 presents a summary of leading developments concerning the 920-980 nm high-power single-chip multimode SDLs. To the best of our knowledge, we have reported the best results in terms of power in this wavelength region.

Table 1. Summary of the leading developments concerning the 920-980 nm high-power single-chip multi-transverse-mode SDLs.^a

View Table

The SDL integrates the merits of high power, high efficiency, and low threshold, which may be attractive for material processing, life science, etc. It should be emphasized that higher power may be realized by adopting a larger detuning between the gain peak and LCF, and this strategy has been widely adopted to realize high power emissions in the 1 $\mu $m region [11]. Although constructive, the large detuning also seems to bring a very high threshold and working temperature. Alternatively, to balance power and threshold, a relatively small pre-offset may be a good choice, but of course, the requirements for heat dissipation are higher. Here a small detuning is adopted, by optimizing heat dissipation, and the high-power (>70 W) and low-threshold (<1.83 kW/cm²) emissions are attained. Overall, we believe this exploration can help understand the thermal characteristics in high-power SDL and can also be considered as an extension and enrichment of the earlier works on this topic.