1. Introduction

Vertical-cavity surface-emitting lasers (VCSELs) are steadily increasing in importance with yearly production in the tens of millions [1]. They offer advantages over other laser structures in terms of beam quality, power consumption, and manufacturability, with applications in spectroscopy, sensing, and optical communication. Despite many inherent advantages provided by the VCSEL geometry, the nominal cylindrical symmetry of their cavity results in near-degeneracy of the polarization-modes belonging to the fundamental transverse mode. In the absence of significant mechanism to provide polarization stability, small changes in the VCSEL operating conditions can cause an abrupt polarization switching (PS) from a polarization state oriented in one direction to the orthogonal direction [2]. The PS can be caused by changes in modal gain [3], phase anisotropy in the presence of birefringence [4], or the onset of higher-order transverse modes [5]. It can be detrimental to many applications requiring control of polarization state, precise emission wavelength, and low noise output [6], hence polarization control in VCSELs has been an active area of research since polarization variability was first reported in 1988 [7]. Many have reported different methods to add modal selectivity (see [8–11], or for a review Chapter 5 in [1]), for instance, employing polarization-dependent effects in the active region or mirrors, or fabricating resonators with deviations from cylindrical symmetry. Although the peak of the optical field intensity is located several microns beneath the surface, conventional methods of modifying a VCSEL mode are implemented by structuring the exit facet [12]. Directly patterning at the core of the device, however, would allow more significant modifications of modal birefringence and dichroism, offering more effective approaches to polarization control.

Here, we investigate the impact of intracavity (IC) patterning on polarization mode control in long wavelength (1310nm) VCSELs made by wafer fusion [13]. The wafer fusion process allows high-resolution patterning of structures at the interface between the gain and distributed Bragg reflector (DBR) sections, which after wafer fusion results in patterning of the refractive index at the core of the optical cavity. This method has previously demonstrated effective transverse mode control in VCSEL arrays [14] and single-transverse-mode VCSELs [15]. In the present work, we extend the approach to non-cylindrically symmetric intracavity patterns, and demonstrate their impact on the polarization states of the VCSEL beam. We present data from a large number of samples and identify new trends that are due to the patterning. This result is consistent with predictions that can be made from a commonly used theoretical model of polarization dynamics [4].

2. Devices and experiment

The VCSELs in this study were fabricated using a double-wafer-fusion technique [13], where the VCSEL InP-based active region is separately grown and later bonded to the bottom and top GaAs/AlGaAs DBRs [Fig. 1(a)]. The active region was grown on a (001) InP substrate, containing six AlGaInAs strained quantum wells, a tunnel junction (TJ) layer, contact layers, and cavity spacers. Midway through the growth process, the sample is removed from the reactor and the TJ layer patterned with shallow discs of 6-μm diameter in order to provide current and optical confinement. After overgrowth, the upper cavity spacer is patterned using electron beam lithography and dry etching [14] with a pattern that removes the azimuthal symmetry in the VCSEL cavity for selecting a single polarization mode [16]. The particular pattern investigated here consists of two shallow arcs with an inner diameter of 9 μm and outer diameter of 11 μm that are concentric with the TJ disc and subtend an angle of 90° [Fig. 1(b)]. The active wafer is then fused to the two DBRs at high temperature and pressure [13], and conventional VCSEL device fabrication completed. Wafer-bonding two different materials at high temperatures can produce undesired residual strain in the device, which has been associated with variable polarization characteristics in VCSELs [17]. Thus, the modifications introduced by intracavity patterning must overcome significant built-in variations in devices.

 

Fig. 1 The VCSEL with intracavity patterning. (a) Schematic cross-sectional view of VCSEL structure. (b) Top-down schematic of structure with crystal axes indicated and rotation angle θ defined. (c) Sub-threshold infrared image, light emission from TJ and scattering off of IC pattern. (d) VCSEL cross section SEM image with etched pattern at the top of the optical cavity indicated.

Download Full Size | PDF

We tested 66 VCSELs distributed across a quarter wafer: 28 devices had intracavity patterning and 38 were control devices with no patterning (aside from the TJ pattern). All had the same TJ diameter of 6 μm. In the sub-threshold, amplified spontaneous emission (ASE) near-field images of the patterned devices, some displayed bright arcs conforming to the patterning geometry [Fig. 1(b), (c)] while in others these features were very dim. To understand this observation, we selected three patterned devices at random and performed cross-sectional scanning electron microscopy (SEM) analysis using a focused ion beam tool. We found in two devices that the pattern has been etched to the designed 60 nm depth [Fig. 1(d)], but in the third, unintentional chemical etching extended to a depth of 250 nm [18]. We assume that the differing etch depth is responsible for the differing near-field patterns and refer to the devices with bright arc patterns as “deeply etched”, and those with dimly lit patterns as “shallowly etched”.

The samples were placed on a microscope stage and maintained at a temperature of 25 °C. We recorded an ASE image of each device to determine the pattern fidelity and classify them as shallowly or deeply etched. Polarization-resolved light and voltage versus current (LIV) curves were measured for currents up to 15 mA; examples for an unpatterned (control) device are shown in Fig. 2. From these measurements, several characteristics relevant to the polarization features were recorded: the current at polarization switching I_PS, the polarization angle of the light near threshold θ, the polarization mode (PM) wavelength splitting Δλ, and the change in output power ΔP when the PM switches from the threshold mode to the subsidiary.

 

Fig. 2 (a) Polarization resolved LI curves of a control VCSEL (no intracavity arc patterning). Green trace taken for polarizer aligned to threshold polarization mode (“y-polarization”), blue for polarizer rotated by 90° (“x-polarization”), and red for total power. (b) Detail of LI curve near bistability region, showing power drop after PS, ΔP. Arrows indicate the change of power with sense of current. (c) Sub-threshold emission spectrum taken at 1 mA bias current, showing distinct peaks for each polarization mode and their separation Δλ, and higher-order transverse modes HOM. (d) Distribution of polarization switching current for VCSELs with and without intracavity patterning.

Download Full Size | PDF

Figure 2(a) presents the LI curve for an unpatterned device as the current is increased to 15 mA and then decreased in 0.1-mA steps (green and blue traces, only). With a threshold current I_th of 1.2 mA, the VCSEL emits a maximum power of 2.3 mW at 14 mA (red trace). A complete PS occurs at I_PS = 3.6 mA for increasing current and 3.1 mA for decreasing current, thus defining a bistability region of 0.5-mA width. Higher-order transverse modes reach threshold near 8.5 mA, leading to kinks in the LI curve and producing a mixed polarization state. The total output power is seen to decrease after the polarization switch by an amount ΔP [Fig. 2(b)], indicating there is PM gain anisotropy within the VCSEL. The fundamental transverse mode emits at λ = 1311.8 nm, with two polarization modes separated by Δλ = 0.13 nm [Fig. 2(c)]; lasing first occurs on the short-wavelength (“y”) mode. Characterization done with the remaining samples followed the same general trend, but important improvements in polarization stability were observed with the patterned VCSELs, as highlighted below.

As a first evaluation of the effect of intracavity patterning, we plot a histogram of the measured I_PS (the upper edge of the bistability region) for the three different groups of devices [Fig. 2(d)]. Good overlap is seen between the distributions of the control devices and the shallowly patterned devices, with an average I_PS of 3.8 and 3.5 mA, respectively. The deeply etched devices show improved polarization stability, with an average I_PS of 5.1 mA.

To further analyze the reasons for the increased stability against PS of the patterned devices, we display in Fig. 3 the measured polarization angle versus PM splitting. In the unpatterned devices, Δλ varies from 0.02 to 0.24 nm with 12 devices having less than 0.1 nm splitting. In the patterned devices the splitting varies from 0.05 to 0.29 nm, with only 4 devices that display splitting less than 0.1 nm. Unpatterned VCSELs with large wavelength splitting tend to align in polarization parallel to the [110] crystal direction, whereas those with smaller splitting are rotated toward the [1–10] direction, consistent with unintentional mechanical strain producing birefringence through the elasto-optic effect [17]. The polarization angle of the patterned devices aligns within 10-20° from the [110] axis for all values of wavelength splitting, regardless of pattern etch depth. These results indicate that the intracavity pattern introduces deterministic birefringence that to a large extent sets the direction of the polarization mode, unlike the uncertainty in polarization angle exhibited by the unpatterned devices.

 

Fig. 3 Polarization angle θ versus polarization mode splitting for all VCSELs measured (see Fig. 1 for definition of θ).

Download Full Size | PDF

Figure 4 shows the variation of I_PS with the measured PM splitting Δλ. Generally, I_PS increases with Δλ. Devices with shallow etching show I_PS similar to the unpatterned devices near Δλ = 0.1 nm, but I_PS increases more quickly with increasing Δλ. This suggests a direct link between the polarization stability and the near field mode image of the device.

 

Fig. 4 Polarization switching current I_PS versus PM splitting for all VCSELs measured.

Download Full Size | PDF

The suppression of PM switching at currents below ~4mA for the deeply etched devices can be due to increase in overlap of the lasing PM with the TJ gain region and/or greater scattering losses from the intracavity pattern experienced by the competing PM. Insight into the difference in modal gains of the two PMs is provided by Fig. 5, which plots ΔP versus I_th. Here, we observe different trends for the patterned and control devices. The control devices show an average ΔP (gray line) that stays below 0.05 mW and is rather constant with threshold current (disregarding special cases where only one device had a given threshold value), whereas the patterned VCSELs show increasing average ΔP (yellow line) in devices with lower I_th. This observation suggests the patterning is increasing the modal gain for the PM at threshold. This trend is clearest in deeply etched devices, where the effective index steps would be greatest.

 

Fig. 5 Change in power after a PS versus threshold current for VCSELs with and without intracavity patterning. The yellow trend line shows the average value for the patterned VCSELs at each threshold current. Similarly, the gray line is the average for the control VCSELs.

Download Full Size | PDF

3. Theory and Modeling

Polarization behavior in such devices can be explained using the spin flip model (SFM) of VCSEL dynamics [4], so-called because it accounts for the total angular momentum of the recombining carriers to generate the orthogonally polarized light. The model includes the effects of cavity anisotropies and is based on normalized rate equations for two coupled electric fields, the total carrier population, and the difference in carrier sub-populations with opposite spin. Key SFM parameters that describe the behavior of our control VCSELs are defined in Table 1. The values listed derive from the device shown in Fig. 1. They were optimized to reproduce the correct PS and bistable regions, as well as its laser dynamics [18]. We are primarily interested in PS that occurs in single-transverse mode VCSELs, but the onset of HOMs can also cause the PS, and this could be accounted for using more advanced versions of the SFM [19].

Table 1. SFM Model Parameters of Control VCSEL [18]

View Table

Stability analysis of the rate equations (cf. Equation (44) in [20]) can identify regions where either of the two linear polarization states is selected to lase. The calculated stability diagram for a VCSEL with gain anisotropy is presented in Fig. 6. Starting with the values from Table 1, we sweep the PM splitting Δλ (proportional to γ_p) and injection current to obtain the regions where the x-polarization (blue region) and y-polarization (green region) (see Fig. 1) modes are stable. The upper limit to y-mode stability is shown with a solid black line. Gain anisotropy selects the higher frequency y-polarized mode at threshold, but as we increase the bias current, the x-mode gains stability. Only later does the y-mode lose stability, giving a small region of bistability (red region, designated xy). For some simulated values of bias current and birefringence, neither mode is stable, designated “none” in the figure. The minimum value of I_PS occurs near Δλ = 0.1 nm, similar to experimental data (see Fig. 4), with I_PS rapidly increasing for lower values of Δλ, and moderately increasing for larger values.

 

Fig. 6 Stability diagram for a VCSEL with significant gain anisotropy (γ_a = 2.5/ns), showing regions where the x- and y-polarized modes are stable (see color code). Contour lines indicate the upper edge of stability for the y-polarized mode similar to data in Fig. 4. Solid line: γ_a = 2.5/ns, dashed lines: γ_a = {1, 1.5, 2.0, 3} /ns.

Download Full Size | PDF

For Δλ>0.2 nm, we notice I_PS increases faster in the experimental data than what is shown in the calculated diagram. We observe that changing the gain anisotropy in the laser can begin to replicate this effect. Broken lines in Fig. 6 display the upper edge of the y-mode stability for different levels of gain anisotropy. When γ_a increases, so does I_PS, thus polarization stability is enhanced for any value of Δλ. For example, using the model presented in Table 1, but increasing γ_a by 0.5/ns increases the upper edge of y-mode stability by 0.5 mA. We observe that increasing γ_a will increase both the minimum value for I_PS, in line with the experimental results, as well as the rate at which the y-mode stability region increases with Δλ. This affect is seen in the devices designated as deeply etched in the experimental data. In the measured data, it was observed that devices having large PM splitting usually displayed a larger degree of gain anisotropy. In this first-order analysis, we assume constant dichroism (γ_a) and then run a series of two-dimensional scans to find the change in polarization stability. If we were to incorporate this correlation between Δλ and ΔP into our model, then Fig. 6 clearly shows that the theory will converge to the measured response, and the polarization stability of the y-mode will increase more rapidly with Δλ. A second detail that is suppressed in this analysis is that operation characteristics of a device vary across the wafer, and this could play a key role in aligning theory to experiment. For example, converting the measured data to normalized values requires knowledge of the device threshold current, gain, and photon lifetime [21], but these were not calculated for each device in this study. Despite these simplifications, the SFM is able to predict an enhancement of polarization stability that is due to the increased birefringence and dichroism that comes from the intracavity patterning of the VCSEL.

It is worthwhile to note that, even with strong dichroism between modes, the VCSEL does not have perfect stability. This can be attributed to the linewidth enhancement factor α. This parameter in a VCSEL can cause the PS whenever there is also birefringence [4, 20], but some stability is gained as α decreases. Our devices showed response consistent with a value near 7.5 [18], but others have reported smaller values in long-wavelength VCSELs [22].

4. Conclusion

In conclusion, we investigated polarization mode control in VCSELs by introducing anisotropic intracavity patterning of the refractive index, and thereby adjusting the modal birefringence and gain difference. Our experiments show that the current at which polarization switching occurs can be significantly increased and the polarization direction can be nearly fixed by introducing arc-shaped air gaps in close proximity to the VCSEL gain region. We present our data in terms of easily measured experimental parameters (Δλ, ΔP, I_th), and then compare the observed trends to the behavior expected from the spin flip model of VCSEL polarization dynamics. This allows us to isolate variables that may be coupled together, such as an increase in ΔP with I_PS, or variation of I_th with λ. We obtain good agreement with the measurements and gain insight into the role of birefringence and dichroism in polarization mode stabilization. Data is presented for devices with arc patterns whose inner diameter was 9 μm and outer diameter was 11 μm, but similar trends were also observed in other test patterns (arcs with 9-μm inner diameter / 10-μm outer, 26 devices, 13 deeply etched; or arcs with 8-μm inner diameter / 10-μm outer, 23 devices, 9 deeply etched). Our results give indications for producing VCSELs with better polarization mode stability.