Modeling and experimental realization of modelocked VECSEL producing high power sub-100 fs pulses

Thanks to their semiconductor gain medium and external cavity, vertical external cavity surface emitting lasers (VECSELs) have a tremendous versatility in their designs and applications. Passive modelocking of VECSELs has been a particularly flourishing research topic in recent years. The growing interest in VECSELs is driven by the numerous inherent advantages of such systems when compared to other laser technologies (low cost, low complexity, high repetition rate, wavelength flexibility, high output power, low noise properties, high beam quality, etc.). Laser sources combining femtosecond pulses with high peak power can benefit many applications, such as multi-photon imaging,¹ high-resolution time domain terahertz spectroscopy,² or self-referenced gigahertz frequency combs.³ VECSELs operating in the multi-GHz regime are particularly promising for frequency combs as they would provide wide comb-tooth spacing and high power per mode, whereas other lasers using dielectric gain media generally suffer from Q-switching instabilities at a high repetition rate due to their long upper-state lifetimes.^4,5

The continuously improving performance of VECSELs in terms of pulse duration and power can certainly be linked to the deeper understanding of microscopic dynamics in the semiconductor during and after the passage of an intense pulse, when the system is driven far from equilibrium. The broad theoretical knowledge gathered in recent years^6–11 has already been exploited in the engineering of semiconductor structures¹² and cavity geometries,^13,14 with the objective of generating the shortest pulse duration or the highest peak power.

To date, the shortest pulse duration demonstrated is a 107 fs pulse with a repetition rate of 1.63 GHz and an average output power of 100 mW, which was externally compressed down to 96 fs,¹⁵ while the highest peak power of 6.3 kW was achieved with a repetition rate of 390 MHz.¹⁴ These performances have surpassed other semiconductor laser technologies and are interesting for the generation of self-referenced frequency combs, particularly in the GHz regime where existing laser systems (fiber, solid state) are not well suited. Combination of high power and short pulse duration is essential to achieve a spectral broadening in nonlinear media, allowing f-to-2f interferometric stabilization of the offset frequency. The required octave-spanning coherent supercontinuum generally requires amplification of the oscillator output power,³ making these systems relatively complex, expensive, and bulky. However, the continuing improvements of VECSEL performance and new designs of silicon nitride waveguide could enable direct supercontinuum generation without prior amplification, followed with a detection and stabilization of the offset frequency.

In this letter, we present a structure design optimized numerically using a nonlinear many-body model to produce high power sub-100 fs pulses. We detail the strategy employed to minimize the pulse duration, while keeping a high output power and we demonstrate the shortest pulse duration generated directly from a semiconductor laser to date. The modelocking simulation of the optimized nominal structures is compared to a simulation of actually grown structures. These simulations are in good agreement with the experimental results and show that further improvements are within reach with a better growth and coating accuracy. The output pulses are thoroughly characterized, revealing clean single pulse modelocking operation at high power.

The semiconductor gain structure was numerically optimized by modelling the pulse formation dynamics in a VECSEL cavity.⁸ In this model, the excited carriers from the pulse interaction with the quantum well (QW) based semiconductor gain and absorber media are treated via the semiconductor Bloch equations while the field propagation is described by Maxwell's wave equation. The structure design parameters that were varied in these simulations are: the number and placement of the QWs relative to the optical field antinodes, the length of the microcavity, and the type of coating applied to the surface of the chip. It was shown that a non-uniform distribution of multiple QWs (MQWs) per antinode is advantageous for short pulse generation as it increases the saturation intensity of the gain and broadens the spectral gain bandwidth. The length of the microcavity should also be kept at a minimum since it directly reduces the microcavity Q-factor, hence the spectral selectivity. However, a shorter active region also results in lower pump absorption and shorter strain-compensating barriers. To recycle the pump power, we use a distributed Bragg reflector (DBR) transparent at the pump wavelength followed by an additional metal reflector. By using a metallic mirror like gold, the reflectivity at both the signal and pump wavelength is enhanced. We can then reduce the number of DBR pairs significantly while keeping a high reflectivity,¹⁶ reducing the thermal impedance and widening the spectral bandwidth. Finally, a bi-layer dielectric coating is designed to provide a broad and spectrally flat group delay dispersion (GDD) with a value approaching zero fs² at the operating temperature and angle of incidence of the lasing field. The optimized design layout of a VECSEL structure emitting around 1030 nm is presented in Fig. 1.

The gain structure consists of a gold reflector followed by an InGaP phase matching layer, 16 pairs of quarter-wave AlGaAs/AlAs layers, and 12 InGaAs/GaAsP QWs distributed non-uniformly along the 3λ/4 long active region. It is completed with an InGaP cap layer and a bi-layer dielectric coating (Ta₂O₅/SiO₂). To ensure a good crystal quality of this structure grown by metalorganic chemical vapour deposition (MOCVD), the QW strain was cautiously compensated by the GaAsP barriers to limit the local integrated strain below ±20% nm and the total active region strain below 5% nm. The details of the individual layer composition and thickness are given in Table. I.

The semiconductor saturable absorber mirror (SESAM) used in this study is grown by molecular beam epitaxy. It consists of 29 quarter-wave layer pairs of AlAs/GaAs followed by a single InGaAs QW placed 5 nm from the surface to provide fast carrier recombination, and is coated by a single layer of Ta₂O₅ with a thickness of 127 nm.

Once the devices were grown, processed, and coated, we carefully measured, and fit the reflectivity and GDD spectra to assess the growth and coating deposition deviations from the nominal design. The actual structures were then used in our modelocking simulation. Figure 2(a) shows the GDD spectra of the gain structure measured at normal incidence and room temperature. The simulated GDD of the structure is also plotted. The simulated pulse duration as a function of the QW background carrier density, i.e., the pump power, for the nominal and actual structure is plotted in Fig. 2(b). The structures are simulated at an operating temperature of 375 K and at an angle of incidence θ of 30°, which correspond to the experimental angle giving the shortest pulse duration. The pulse intensities are also plotted to show the effect of pulse duration on the maximum peak power. This simulation assumes a total nonsaturable cavity loss of 1% [0.8% Output Coupler (OC) + 0.2% of scattering and mirror losses]. The Ta₂O₅/SiO₂ coatings of the nominal and actual gain structure have thicknesses of 113/112 nm and 94/97 nm, respectively.

For each structure, the pulses get shorter and more intense as the carrier density is increased, likely due to the broadening and increase in the unsaturated gain, and stronger saturation of the absorber. It is however limited by the stability of the single pulse modelocking regime, as unstable or multiple pulses appear at high carrier densities. These results also show that the accuracy of the coating deposition is critical. A small variation in the coating layers thicknesses can significantly affect the pulse characteristics, likely due to a sub-optimal GDD of the structures. Indeed, the minimal pulse duration varies from 69 fs with the nominal thickness to about 88 fs for our realized coatings even though their thicknesses only differs by a few nanometers. Experimentally, we can thus expect a minimal pulse duration around 88 fs. We should note that the peak power achievable also depends on the duration of the pulse. In particular, at high pump power, shorter pulses have a broader spectrum and deplete the gain medium over a broader carrier distribution, thus reducing the saturation fluence of the gain which in turn increases the peak power reached by the pulse. At a carrier density of 2.05 × 10¹² cm⁻², the pulse duration simulated with the actual structure is 94 fs, which is very close to the experimental value reported below. At this carrier density, the simulated optical spectral width is 14.4 nm FWHM, giving a time bandwidth product 1.22 times above the Fourier transform limit. Residual linear GDD and self-phase modulation are the two main factors stretching the pulse duration.

The VECSEL chip and SESAM were then tested in a ring cavity configuration, as illustrated in Fig. 3.

This ring cavity geometry provides a colliding pulse modelocking regime (CPM) and has the advantage of having the pulses passing only once on the gain chip per round trip while the gain is depleted twice by the counter-propagating pulses. This minimizes the cavity GDD since the gain chip is the main contributor of third order dispersion (TOD), while reducing the chances of harmonic modelocking. This cavity also presents other advantages due to the pulse interactions in the absorber, such as a reduced saturation fluence and a robust modelocking regime.^12,13 The gain medium is placed at a quarter of the total cavity length L = 136 mm from the saturable absorber, assuring an equal pumping duration and gain recovery for both pulses. The cavity is completed with a highly reflective concave mirror with a radius of curvature of 75 mm, and a flat output coupler with a reflectivity of 99.2%. This gives a mode waist (radius) of 152 μm on the VECSEL and 92 μm on the SESAM. The angle of incidence on the VECSEL was finely adjusted to minimize the pulse duration, as it significantly affects the GDD of the cavity. At an angle θ of 30° and a TEC temperature of 10 °C, we obtained the minimum pulse duration of 95 fs with an output power of 90 mW per output beam for an incident pump power of 24 W. This pulse duration of 95 fs obtained directly from a semiconductor laser compares with the previously reported pulse duration of 107 fs externally compressed to 96 fs.¹⁵ Figure 4(a) shows the non-collinear second harmonic generation (SHG) autocorrelation of the output pulse from one of the beams. It reveals a nearly perfect fit with a sech² pulse shape. The pulse duration and phase was also measured with a commercial SHG frequency-resolved optical gating interferometer (FROG). The retrieved pulse intensity and phase are plotted in Fig. 4(b). The pulse duration of 98 fs from the FROG measurement is in good agreement with the fitted value of 95 fs from the autocorrelation measurement and does not show any significant distortion.

Both output beams have identical physical characteristics (duration, spectrum, and power), as previously observed with CPM VECSEL.^12,17 It was also shown that the two output beams can be coherently combined,¹⁷ which could provide a total output power of 180 mW here. At higher pump power, the pulses become more distorted (more chirp) leading to longer duration and eventually to multi-pulsing.

To further characterize the modelocking regime, we recorded the microwave spectrum of one beam with a high bandwidth photodetector and spectrum analyzer. Figure 5 shows that all harmonics of the fundamental pulse repetition rate (2.2 GHz) are present and that the beatnote linewidth is extremely small (jitter limited) with an amplitude >85 dB above the background noise. This is a clear indication that all the modes are phase-locked.

The spectral phase and intensity of the pulse retrieved from the FROG measurement are plotted in Fig. 6. The retrieved optical spectrum shows a spectral width of 12.32 nm (FWHM) and a nearly quadratic spectral phase at the center of the spectrum, suggesting that the pulse duration could be further shortened with a simple linear compression stage.

The spectral intensity measured directly with a grating spectrometer is similar to the spectrum measured with the FROG, indicating that the output beam is purely pulsed, without any CW component. The pulse spectrum is centered at 1025 nm with a spectral width of 12.9 nm, corresponding to a time-bandwidth product of 0.35, i.e., 1.1 times the Fourier transform limit. Figure 7 shows the spectrum together with the GDD of the cavity.

Unsurprisingly, the optimal pulse duration was obtained when the optical spectrum was centered on the flattest part of the total GDD spectrum. We can also observe that the wings of the spectrum experience more third order dispersion (TOD) than the center of the pulse spectrum. This is mostly due to the limited stopband of the gain and absorber DBRs. The increase in TOD can also be observed in the spectral phase measurement (Fig. 6), where the deviation from a linearly chirped pulse is more significant on the wings of the spectrum.

To further reduce the pulse duration towards the theoretical limit of 69 fs, a better accuracy of the coating deposition seems necessary as a small thickness deviation significantly affects the achievable pulse duration. Ultimately, the DBR stopband will be the next limitation as the dispersion gets inevitably high towards the edge of the stopband. It was shown previously that with an artificially large DBR stopband, pulse durations as short as 20 fs could be supported by a MQW VECSEL structure.⁸ This would of course necessitate either materials with very high refractive index contrast (dielectric DBR) or a complete removal of the DBR itself such as in recently demonstrated membrane VECSEL devices (MECSELs).^18–20 

In conclusion, we presented the design strategy of a modelocked VECSEL device for the generation of ultrashort pulses. We detailed the gain structure design, discussed the influence of the placement of the QWs according to the field standing wave, and pointed out the importance of the antireflection coating for the compensation of the entire cavity dispersion at operating temperature and actual angle of incidence. We realized a full simulation of the modelocking regime, showing the influence of the coating on the resulting pulse duration and predicting a lower limit of 69 fs for an optimal structure and 88 fs for the actual structure. The structure grown by MOCVD was then tested in a CPM ring cavity geometry which was optimized for the shortest pulse duration. A pulse duration as short as 95 fs was demonstrated directly from the oscillator, with an output power of 90 mW per beam, which can potentially be coherently combined to give a total power of 180 mW. The thorough characterization of the pulse, suggested that the pulse duration is limited by the dispersion of the cavity, which could be partly improved by a better coating accuracy or more significantly alleviated with a membrane VECSEL.

We would like to thank Jason Jones and Robert Rockmore from the University of Arizona for helpful discussions. This material is based upon work supported by the Air Force Office of Scientific Research under Award No. FA9550-17-1-0246.