Emerging technologies based on cold atoms such as atomic clocks [1,2] rely on accurate measurements through locking a laser to a stable atomic transition. These high-accuracy devices, however, require stringent laser properties in terms of linewidth, stability, and power output; therefore, the laser systems are typically bulky, expensive, and power-hungry [1,2]. Semiconductor lasers offer good chip-scale integrability [3–5], low production cost, and reduced power consumption [6], which are all desirable properties for the realization of compact clock systems [7]. Of particular relevance is the GaAs/AlGaAs material platform because it can cover a wide wavelength emission range from 700 nm to 900 nm, which includes important cold atom transition lines such as ${^{87}{\rm Rb}}$ and ${^{133}}{\rm Cs}$ at 780.24 nm and 852 nm, respectively [8].

One of the critical issues for the development of high-power lasers on a GaAs/AlGaAs material platform is the aluminum-oxide formation on the air-exposed surfaces. Not only does such oxide formation decrease the power density threshold for catastrophic optical damage (COD) and mirror damage (COMD) [9,10], but it also degrades the laser output in terms of carrier lifetime, propagation losses, and self-heating [11–13]. Moreover, the aluminum fraction in the epilayer material poses several issues for epitaxially regrown structures where the oxygen contamination of the aluminum-containing layers can substantially decrease the fabrication yield. One of the possible solutions to mitigate the aluminum oxidation is the use of an aluminum-free active area [14]. From an optical gain point of view, the majority of the structures reported in the literature are designed with a large overlap between the optical mode and the gain region to minimize the laser threshold. Such design strategy is, however, not ideal for high-power and narrow-linewidth lasers as structures with low power density and low internal losses are desirable to both extend the power range with single-mode operation and reduce the linewidth in accordance to the Schawlow–Townes formula [15].

In this work, a distributed feedback (DFB) laser design is proposed to reduce the linewidth at the emission wavelength of 780.24 nm. An aluminum-free active area was designed with InGaAsP quaternary layers to reduce the risk of COD and COMD. In addition, a mode expander layer was introduced in the n-doped AlGaAs cladding to stretch the optical mode vertically, thereby decreasing the average power density and the internal propagation losses. Finally, a long cavity of 4 mm was employed to further decrease the average power density and reduce the laser linewidth. The fabricated DFB lasers exhibited stable emission wavelength of 780.24 nm with powers up to 60 mW and linewidths down to 612 kHz, a value suitable for the ${^{87}}{\rm Rb}$ atom cooling at the D2 transition (${\lt}{1}\;{\rm MHz}$ [16]).

The first aspect we considered in the laser optimization was the epilayer design of the active region. In order to limit the interaction of the optical mode with the aluminum-containing layers, InGaAsP quaternary compounds were chosen for the quantum wells (QWs) and barriers, and these layers were simulated using the Schrödinger–Poisson current solver Nextnano++ [17]. Two compressively strained 4-nm-thick ${{\rm In}_{0.413}}{{\rm Ga}_{0.587}}{{\rm As}_{0.44}}{{\rm P}_{0.56}}$ QWs inserted between 10-nm-thick ${{\rm In}_{0.27}}{{\rm Ga}_{0.73}}{{\rm As}_{0.44}}{{\rm P}_{0.56}}$ barriers provide laser emission around 780.24 nm. The active area is enclosed between two graded separate confinement heterostructure (GRINSCH) layers of AlGaAs with aluminum fraction gradually increasing from 0.4 to 0.55. Finally, the GRINSCH layers are sandwiched between two doped ${{\rm Al}_{0.55}}{{\rm Ga}_{0.45}}{\rm As}$ cladding layers. The use of an aluminum-free active area reduces the optical confinement factor of the optical mode within aluminum-containing layers by 10% in comparison to standard GaAs/AlGaAs designs [18] and gives more tolerance in defining the DFB gratings as the active area can be etched without introducing additional scattering losses or decreasing the power density threshold for COD and COMD [9,10].

The Schawlow–Townes formula is an effective tool to analyze the impact of the epilayer design on the laser linewidth [15]. For semiconductor DFB lasers, the formula can be expressed as

Equation 1 indicates that lasers with low $\Delta \nu$ require low propagation losses ${\alpha _i}$, equivalent mirror losses for DFB lasers ${\alpha _m}$, mode confinement in the active QW region ${\Gamma _\textit{QW}}$, and Henry factor ${\alpha _H}$. Also, linewidth narrowing can be obtained by increasing the output power ${P_{\rm out}}$ of the laser. It is worth noting that ${\alpha _i}$ and ${\Gamma _\textit{QW}}$ are predominantly related to the design of the gain material, while ${\alpha _m}$ depends on the grating design; hence, these contributions to the linewidth can be independently optimized. The linewidth enhancement factor ${\alpha _H}$ has a substantial impact on nonlinear effects such as spatial hole burning (SHB), and is critical for narrow-linewidth semiconductor lasers as it increases for high values of ${P_{\rm out}}$ [19], which explains the linewidth broadening experienced by high-power lasers. The other terms (i.e., $h\nu$ photon energy and ${r_\textit{sp}}/{R_\textit{sp}}$ spontaneous emission fraction in the lasing mode) effecting the linewidth in Eq. (1) can be assumed as fixed terms in this treatment and will not be considered in the following.

As reported in previous work for 1550 nm lasers [20], the introduction of a mode expander layer in the n-doped cladding decreases several parameters that play a role in defining the laser linewidth in Eq. (1): (i) the mode confinement ${\Gamma _\textit{QW}}$; (ii) the propagation losses ${\alpha _i}$; and (iii) the average power density in the transversal direction $\overline {{P_{\rm tran}}}$. The mode expander was implemented by introducing a layer in the lower cladding with a higher refractive index (i.e., ${{\rm Al}_{0.35}}{{\rm Ga}_{0.65}}{\rm As}$) in comparison with the surrounding layer (i.e., ${{\rm Al}_{0.55}}{{\rm Ga}_{0.45}}{\rm As}$), so as to pull the optical mode further into the n-doped cladding and expand the mode size. The magnitude of this effect depends on the mode expander thickness and on its distance from the active area; hence, the laser parameters ${\Gamma _\textit{QW}}$ and ${\alpha _i}$ can be engineered as a function of the mode expander design. The vertical mode profiles with and without the mode expander, the band structure, and the refractive index profile around the active area are reported in Fig. 1.

 

Fig. 1. Top left axis, the vertical mode profile; bottom left axis, the conduction and valence bands for the active area; bottom right axis, the refractive index for the mode expander. The zero position in the $x$ axis represents the center of the active area.

Download Full Size | PDF

The distance of the mode expander from the active area is a critical parameter because the lower energy bandgap of the mode expander layer can create an asymmetrical confinement for electrons and holes in the active area. From simulation of the energy levels and carrier confinements in the active area by Nextnano++ software, a distance between the active and the mode expander larger than 100 nm was found to be sufficient to ensure a symmetrical band structure in the active area. On the other hand, a larger distance would reduce the effect of the mode expander on the active area, which would demand a wider mode expander to achieve an equivalent mode attraction. The lower energy bandgap of the mode expander, however, acts as a potential barrier for electrons and holes; hence, it is preferable to limit the mode expander thickness to minimize the value of an additional series resistance through the epilayer. Based on these considerations, a mode expander distance of 100 nm was chosen for the epilayer design.

The fraction of the modal power overlapping with the active region ${\Gamma _\textit{QW}}$ and the propagation losses ${\alpha _i}$, given by the sum of the propagation losses in the active region ${\alpha _\textit{QW}} = {\Gamma _\textit{QW}}{k_p}$, ${ n}$-doped ${\alpha _n} = {\Gamma _n}{k_n}$_, and ${p}$-doped cladding ${\alpha _p} = {\Gamma _p}{k_p}$ [20], are simulated through a finite difference eigenmode (FDE) 1D-simulation tool (Lumerical Mode solution [21]). The factors ${k_{QW,n,p}}$ represent the normalized absorption coefficients of the QWs and claddings for doping levels $n{= 10^{18}}\;{{\rm cm}^{- 3}}$ and $p = 6 \times {10^{17}}\;{{\rm cm}^{- 3}}$, respectively, and are obtained considering the free carrier absorption and the intervalence band absorption as the main loss mechanisms [22–24]. Both ${\Gamma _\textit{QW}}$ and ${\alpha _i}$ are reported in Fig. 2 as a function of the mode expander thickness.

 

Fig. 2. Simulations of the mode confinement of the active region ${\Gamma _\textit{QW}}$ (left axis) and of the propagation losses ${\alpha _i}$ (right axis) as a function of the mode expander thickness for a mode expander distance of 100 nm from the edge of the active region.

Download Full Size | PDF

From Eq. (1), a lower value of ${\Gamma _\textit{QW}}$ provides a narrower laser linewidth; however, the reduction of ${\Gamma _\textit{QW}}$ also effects other laser parameters such as the threshold current and the output power ${P_{\rm out}}$ [25]. Therefore, a trade-off value of 1% was selected as a design parameter as this ${\Gamma _\textit{QW}}$ value is sufficient to achieve a threshold current density as low as $1\;{\rm kA}/{{\rm cm}^{- 2}}$ [26] while it narrows the linewidth by a factor 2.4 in comparison to an epilayer design without the mode expander layer. For a mode expander thickness of 250 nm, which corresponds to a ${\Gamma _\textit{QW}}$ value of 1%, Fig. 2 also indicates a decrease of the propagation losses from ${\alpha _i}$ from $5.4\;{{\rm cm}^{- 1}}$ to values below $4.1\;{{\rm cm}^{- 1}}$. This is a consequence of the larger mobility of electrons [23] compared to holes [24] in AlGaAs compounds, which translates into lower absorption in n-doped materials compared to p-doped materials. The mode expander design also has a good tolerance in terms of thickness and material composition. An inaccuracy of 10 nm in layer thickness or 1% in aluminum fraction, which is significant compared to the tolerances for QW growths, translates in a variation of ${\alpha _i}$ and ${\Gamma _\textit{QW}}$ lower than $0.1\;{{\rm cm}^{- 1}}$ and 0.1%, respectively. The decrease of ${\alpha _i}$ entails a further reduction of the linewidth according to Eq. (1). Moreover, the mode expander decreases the average power density in the transversal direction $\overline {{P_{\rm tran}}}$, which is linked to the power output by the equation

Regarding the Bragg grating design, the two key parameters to be optimized are the coupling factor $\kappa$ and the cavity length $L$. The product $\kappa L$ is the total grating coupling and determines the modal gain threshold ${\alpha _m}$ in Eq. (1). It should be noted that although the grating coupling depends only on the product $\kappa L$, gratings with the same values of $\kappa L$ but with different relative values of $\kappa$ and $L$ have different impacts on the laser properties. In fact, for the same product $\kappa L$, a longer cavity translates into a lower power density and, hence, a reduction of the nonlinear effects, which could broaden the linewidth. A longer cavity, however, has higher propagation losses per round trip ${\alpha _i}L$, which increases the lasing threshold and decreases ${P_{\rm out}}$; hence, its value has to be carefully optimized by considering all the laser design parameters simultaneously.

The coupling factor $\kappa$ was simulated through a FDE 2D-simulator with a longitudinal approximation and confirmed through a 3D-simulator EigenMode Expansion (EME) by Lumerical MODE software for a shallow-etched sidewall Bragg grating, with a width of 1.2 µm and 2.2 µm for the two sections of the ridge waveguide, as described in a previous work [27]. The considered etching depth for the sidewall grating is 1.6 µm, which corresponds to stopping the etch 150 nm above the active area. Although the coupling coefficient $\kappa$ is reduced by the effect of the mode expander layer, which pulls the optical mode out of the grating etched in the p-doped cladding, a lower value of $\kappa$ is functional for achieving narrow linewidth, provided that the total grating coupling $\kappa L$ is above 1 [28]. In order to have the Bragg condition at the target wavelength of 780.24 nm, the grating period at the first order is ${\sim}117\;{\rm nm}$_, which gives very low tolerances to fabrication inaccuracies. Hence, a third-order grating geometry with a 50% duty cycle and a 351 nm period was selected to relax the fabrication tolerances, providing a coupling factor of approximately $4\;{{\rm cm}^{- 1}}$ [27], which is consistent with the coupling coefficient measured from the stop band in the DFB spectrum. The cavity length was chosen to be 4 mm, which gives a $\kappa L$ product of 1.6, a value optimal to ensure stable DFB lasing operation under single-mode operation [28]. Although a longer cavity increases the threshold current and the total propagation losses, such an increase is less prominent in lasers with a small number of QWs and low mode confinement ${\Gamma _\textit{QW}}$, as for the epilayer design in this work [25]. Longer cavities are also expected to improve the thermal behavior of the laser as the dissipated heat can be spread over a longer length.

The epitaxial layers were grown by a commercial supplier on 650-µm-thick GaAs substrates. The sidewall DFB grating was patterned on a 500-nm-thick hydrogen silsesquioxane resist by electron beam lithography using a Vistec VB6 tool. The pattern was transferred to the sample by reactive ion etching using an Oxford Instrument ICP180 tool with a ${{\rm Cl}_2}/{{\rm BCl}_3}/{\rm Ar}/{{\rm N}_2}$ chemistry to etch through the upper cladding with a high aspect ratio, as shown in Fig. 3. The sample was then covered with a ${{\rm SiO}_2}$ passivation layer, and a narrow window on the top of the waveguide was etched to allow for current injection. The top (TiPtAu) and bottom (AuGeNi) contacts were evaporated after the sample was thinned down to a thickness of approximately 250 µm. Finally, the device was cleaved before antireflection (AR) and high-reflection (HR) coatings were deposited on the cavity facets to suppress any parasitic Fabry–Perot effects.

 

Fig. 3. A SEM picture of a sidewall DFB grating after the dry etch of the AlGaAs cladding. The high aspect ratio and sidewall smoothness prove the robustness of this critical fabrication step.

Download Full Size | PDF

 

Fig. 4. The power, voltage, and current characteristics of the DFB laser.

Download Full Size | PDF

The fabricated devices were fully characterized for light-voltage-current (LIV), optical spectrum, and linewidth characteristics under continuum-wave operation and at a fixed temperature of ${{20}^ \circ}{\rm C}$. The LIV characteristics shown in Fig. 4 indicate a threshold current of 140 mA with a voltage bias lower than 2 V up to an injection current of 250 mA, which demonstrates a good electrical conductivity of the epilayer. The DFB laser has kink-free emission between the threshold and an injection current of 240 mA, albeit the power output linearly increases until the injection current of 275 mA exceeds 60 mW. The laser spectrum was measured by an optical spectrum analyzer with 0.01 nm resolution for increasing values of the injected current in order to confirm single-mode operation and emission wavelength tuning (see Fig. 5). The DFB laser clearly exhibits single-mode behavior with no evidence of mode hopping and a side-mode suppression ratio (SMSR) exceeding 40 dB within the 180–250 mA current range and with a peak wavelength tunability of 0.3 nm.

 

Fig. 5. The spectra of the DFB laser for different injected currents at a temperature of 20°C. The peak wavelength shows a monotonous increase with the current and can be finely tuned about the target wavelength of 780.24 nm.

Download Full Size | PDF

Finally, the laser linewidth was characterized through a heterodyne detection technique [29]. The laser signal was beaten with a commercial 50 kHz Ti:sapphire laser source at 780 nm wavelength and coupled to a fast photodetector to generate a radio frequency (RF) signal. The generated RF beat note was then resolved by a RF spectrum analyzer in order to measure the full width at half-maximum (FWHM) of the convolved signals. A Voigt line shape, the convolution of a Gaussian and a Lorentzian line shape, was employed to fit the RF beat note. The natural Lorentzian-shaped laser linewidth was extrapolated from the fitted Voigt linewidth. The measured RF signal and its Voigt-shaped fit are shown in Fig. 6 with a measurement sweep time of 25 µs. The RF fitting indicates that the laser has a linewidth of 612 kHz, a value that is suitable for ${^{87}}{\rm Rb}$ atomic cooling applications [16].

 

Fig. 6. RF beat signal and its Voigt fit for the narrowest linewidth obtained at the current injection of 250 mA. The FWHM is 612 kHz (single sweep with a sweep time of 25 µs; resolution bandwidth, 200 kHz; video bandwidth, 30 kHz).

Download Full Size | PDF

In conclusion, an optimized, aluminum-free active region DFB laser with a mode expander and a long-cavity laterally coupled Bragg grating has been demonstrated at 780.24 nm wavelength with a 60 mW emission. The measured SMSR of 40 dB and linewidth of 612 kHz match the requirements for ${^{87}}{\rm Rb}$ D2 transitions for atom cooling applications.