3.1.

Two-Photon Polymerization Direct-Laser Writing

In recent years, additive manufacturing (3D printing) has proved to be a viable alternative to conventional planar fabrication technologies, with applications in photonics, biology, or lab-on-chip among others. 3D printing offers indeed more degrees of freedom on shapes that can be achieved, enabling new fabrication schemes. As a non-collective method, 3D printing is well suited to fabricate single objects, which is typically the case when creating a microlens on top of a single VCSEL chip. Compared with other additive manufacturing technologies, 2PP 3D printing attracts more and more attention in micro-optics field, as it enables the fabrication of freeform optical elements with high resolution ($<0.2\mu \mathrm{m}$) and high accuracy ($<\pm 0.5\%$). Considering the precision required for the fabrication of micro-optical elements, this technique offers appropriate characteristics^19–21 and was successfully applied to fabricate a complete miniature visible spectrometer,²² as well as for the precise fabrication of micro-optical elements on single-mode optical fibers²³ or LED devices.²⁴ It was also recently exploited to couple VCSEL arrays to multimode fibers for data transmission experiments²⁵ and to fabricate an external collimator for VCSEL used in 3D sensing systems.²⁶ However, direct fabrication on a VCSEL chip to be used for miniaturized optical spectroscopy has not been yet reported.

The 2PP writing principle used is shown in Fig. 4. The photosensitive material is polymerized by absorption of two photons simultaneously. This non-linear optical process occurs when an ultrashort pulse laser is focused tightly in a small area. The size of this voxel is related to the laser spot, the laser power, and the properties of the material itself. By scanning the laser beam throughout the material or by moving the sample with a fixed laser spot, a 3D structure can be polymerized. As shown in Fig. 4(a), the substrate is immersed into the photosensitive liquid resist. In our case, the substrate is a printed circuit board (PCB) with a mounted VCSEL chip. By slowly moving the sample up, the polymerization occurs layer by layer at the computed positions. Once the exposure is complete, a development solution is used to remove the uncured resist. The set-up used is photonic professional L-3D (Nanoscribe Gmbh) with a 780 nm femtosecond laser source ($\text{pulse}\approx 100\mathrm{fs}$; repetition rate = 80 MHz) and a writing intensity ranging from 0.982 to $1.547\mathrm{TW}/\mathrm{c}{\mathrm{m}}^2$ depending on the chosen exposure power.²⁷ In order to obtain a good compromise in terms of resolution and writing time, a $63\times$ objective was selected and used in dip-in laser lithography mode suited for high-resolution patterning on a non-transparent substrate. The $63\times$ microscope objective has a numerical aperture (NA) of 1.4, giving a working distance of $190\mu \mathrm{m}$. This configuration leads to a voxel size of 160 nm in $X/Y$ axes and about 700 nm in the $Z$ axis. The chosen photoresist was IP-Dip resist, as it is well suited to the fabrication of high resolution/high aspect-ratios optical elements with a high refractive index in the near-infrared spectral range.

Fig. 4

(a) Principle of the two-photon polymerization 3D printing on VCSEL chip (Nanoscribe Gmbh). The VCSEL mounted onto a PCB is dipped into the photoresist, which is in direct contact with the objective lens. (b) Zoom on a lensed-VCSEL chip mounted on the PCB.

3.2.

Microlens Design

ZEMAX optical modeling software was used to design a micro-optical element made of a cylindrical polymer pedestal surmounted by a hemispherical lens for collimating the VCSEL beam (Fig. 5). We used the ZEMAX merit function with GBPD operand to minimize the Gaussian beam paraxial divergence taking into account the optical properties of the lens material^28,29 and the possible ranges for lens dimensions, namely diameter $D$, sag $T$, and pedestal height $L$, which are limited by technological constraints. Since the emitting area is non-centered on the VCSEL chip [Fig. 6(a)], the lens diameter $D$ cannot be chosen larger than $80\mu \mathrm{m}$. For $80\mu \mathrm{m}$, the divergence can be theoretically reduced from 14.4 deg to 2.32 deg with a $150\mu \mathrm{m}$-thick pedestal, a lens sag of $14\mu \mathrm{m}$ and a radius of curvature (ROC) of $60\mu \mathrm{m}$. This corresponds to a beam size of less than $50\mu \mathrm{m}$ at 2 mm from the lens output surface, which fits well our sensor requirements. Note that this divergence value could be further reduced, for example, by increasing the pedestal height $L$. However, it is preferable to limit this parameter to $150\mu \mathrm{m}$ to maintain a reasonable aspect ratio for the entire micro-optical element ($\le 1:2$).

Fig. 5

(a) Lens design. (b) Spatial distribution of the beam irradiance calculated using ZEMAX (Gaussian beam propagation) at a distance of 2 mm for the reference without lens (top) and with lens (bottom). The corresponding theoretical divergence is reduced from 14.4 deg to 2.32 deg (full angle at $1/e^2$).

Fig. 6

(a) 3D CAD design of the pedestal + lens system to be fabricated on the VCSEL die. (b) Corresponding STL mesh comprising 22,284 faces and 11,414 vertices.

3.3.

3D Printing Optimization

Owing to the nature of 2PP technology, the pedestal and the micro lens can be designed as a single unit, with the additional benefit of being able to arbitrarily change all dimensions. This design can be done using a conventional 3D computer-aided-design (CAD) software. In our case, we used FreeCAD, an open-source parametric 3D CAD software. The resulting design is then exported as a.STL file, which is a file format commonly used for 3D printing. While the native parametric 3D CAD design is composed of primitives (i.e., a combination of solids such as cubes, cylinders, spheres…), the STL file is an approximated version of the parametric design that uses triangular planes to reproduce the surface geometry of the 3D model. The Nanoscribe set-up uses STL file format as an input, as such the settings for exporting the 3D parametric design to an STL file are critical: coarse settings would lead to large facets and would be detrimental to the overall print quality, whereas very refined parameter would lead to enormous files and much longer processing while not bringing additional benefits to the print quality. Our settings are a compromise between file size and print quality. An illustration of STL file for the pedestal and microlens is shown in Fig. 6.

As mentioned above, the optimal writing conditions result from a compromise between time and resolution. We tried several combinations of 100% exposure power with scan speeds of 30 and $50\mathrm{mm}/\mathrm{s}$, and 40% to 63% exposure powers with scan speeds of 10 and $30\mathrm{mm}/\mathrm{s}$ to optimize lens shape and morphology while keeping a maximal fabrication time of 1 h per lens. SEM images of final 3D lenses optimized for three different slicing resolutions (SR) are shown in Fig. 7(b). As expected, the surface of a lens fabricated by 3D-printing is less smooth than that fabricated with a non-contact dispensing method such as inkjet printing [Fig. 7(a)]. Layered patterns are visible, especially for large SR. However, the geometry of the pedestal and of the lens can be better adjusted and controlled with this method.

Fig. 7

SEM views of a microlens fabricated (a) by our former technique (successive soft-printing of three dry resist films + DLW+ inkjet printing), (b) by 3D-printing using three different SRs (SR = 0.5, 0.3, and $0.1\mu \mathrm{m}$) with scanning speeds of 30, 30, and $50\mathrm{mm}/\mathrm{s}$, and exposure powers of 63%, 50%, and 40%, respectively (maximal intensity for 63%: 1.547 TW/cm²).

3.4.

Lens Properties

Lens dimensions were measured using white-light optical profilometry (Bruker Contour GTI and S neox from Sensofar) [see typical images in Fig. 8(b)]. After a calibration step to compensate for the material shrinkage that occurs during 3D-printing (typically $\sim 4\%$ depending on the exposure power), the final lens diameter and sag were measured at $D=78.78\pm 1.67\mu \mathrm{m}$ and $T=13.66\pm 0.93\mu \mathrm{m}$, respectively, in good agreement with the desired values. The ROC measured for three different SR is shown in Fig. 9(a). As seen, the dispersion of the values increases for lower resolutions. Nevertheless, we measured for the least favorable resolution ($\mathrm{SR}=0.5\mu \mathrm{m}$) a ROC equal to $60.36\pm 1.67\mu \mathrm{m}$, which is also in good agreement with expectations.

Fig. 8

(a) 3D and 2D images of a lens printed with an SR of $0.3\mu \mathrm{m}$ acquired with a S neox optical profilometer. (b) Example of a plot acquired by Mach–Zehnder interferometry used to calculate the focal length and the RMS aberrations (before calibration).

Fig. 9

(a) ROC and (b) surface roughness Sq (root mean square height).

Lens surface roughness was also characterized using optical profilometry. These measurements confirmed the SEM observations: the lower the slicing value, the lower the roughness [Fig. 9(b)]. The minimal value was indeed obtained for $0.1\mu \mathrm{m}$, with a root means square height $S_q$ of 81.65 nm, of the same order of magnitude as that reported for standard 2PP-DLW techniques³⁰ However, in this case, the total writing time is 60 min, whereas it is much shorter for lower resolutions: 10 min and 5 min, for 0.3 and $0.5\mu \mathrm{m}$, respectively. Given the moderate difference in surface roughness observed between the 0.3 and $0.5\mu \mathrm{m}$ cases (98 versus 88 nm) and the significant gain in fabrication time (×2), the SR used for lens fabrication on the VCSEL chip was finally set at $0.5\mu \mathrm{m}$.

In addition, a lens array printed on a glass substrate using the average resolution ($\mathrm{SR}=0.3\mu \mathrm{m}$) was characterized using a Mach–Zehnder interferometer. With this set-up, it was possible to determine the focal length through the interpretation of the observed fringe pattern³¹ [see example for one lens in Fig. 8(b)]. A mean value of $98.6\pm 2.3\mu \mathrm{m}$ was determined (six lenses measured). The numerical aperture of the lens ($\mathrm{NA}=0.35\pm 0.02$) and the optical aberrations ($\mathrm{RMS}=0.143\pm 0.04$ lambda) were also derived from these measurements. The latter value is higher than $\lambda /14$, indicating that the lenses are not diffraction-limited. However, their quality seems sufficient for our non-imaging application. Additional measurements for the other resolutions are still in progress.

4.1.

Beam Divergence and Emitted Power

After lens fabrication, the profile of the VCSEL Gaussian beam was acquired for two different axial positions with a beam profiler camera placed on a moving stage. By transforming the measured intensity profiles using a Gaussian fit to extract the waist sizes at both positions and knowing the distance between them, the corresponding beam divergence could be determined and found to be $3.0\pm 0.1\mathrm{deg}$ (full angle at $1/e^2$) [Fig. 11(a)]. This value is comparable to the one previously obtained by the authors using UV direct laser writing in dry films coupled to inkjet dispense. It is higher than the theoretical value given by the model (2.32 deg). A better fit with the experimental value is found by taking into account in the model the fabrication uncertainties on lens dimensions and most importantly, by assuming that the initial beam waist is not located at the VCSEL surface but $3\mu \mathrm{m}$ beneath it (i.e., at the position of the quantum well active zone in the laser). This indicates that the lens design could be further optimized with the model and that a minimal divergence of 2.5 deg is experimentally achievable. Nevertheless, the spot size is estimated to be $55\mu \mathrm{m}$ at a distance of 2 mm (instead of $44\mu \mathrm{m}$ expected), which is sufficiently for the aimed application. Concerning the VCSEL optoelectronic behavior after lens integration, a slight modification of the threshold current is visible on the light-current curve. This is due to the change in the reflectance of the top mirror, linked to the presence of the IP-Dip lens polymer material at the VCSEL surface. The output emitted power level is, however, found to be similar to that of the reference device, showing no significant effect despite lower surface quality [Fig. 11(b)]. We also checked that the degree of polarization of the emitted beam was kept higher than 75%, which means the grating effect on polarization selection was preserved after lens integration.

Fig. 11

(a) Comparison of the normalized angular beam distributions of the reference and lensed VCSEL chips (solid lines: experimental, dotted lines: model). (b) Light–current curve measured for both devices.

4.2.

Available Tuning Range Under Current Variation

The evolution of the emission spectrum as a function of the applied current was also measured after lens integration (Fig. 12). The SMSR remained higher than 10 dB for injection currents below 5 mA, but a dual mode emission progressively appeared for upper currents. This led to a final single mode tuning range of only $\sim 3\mathrm{nm}$. This behavior can be attributed to the fact that the respective diameters of the polarization grating ($4\mu \mathrm{m}$) and of the buried oxide aperture ($6\mu \mathrm{m}$) of the used VCSEL chip are suited to achieve a single-mode operation with an output medium in the air and not in a polymer material.

Fig. 12

Emitted spectrum of the VCSEL chip after lens integration as a function of applied current. The tuning range with SM emission ($\mathrm{SMSR}>10\mathrm{dB}$) is reduced to $\sim 3\mathrm{nm}$.

To get a better understanding of this behavior, we performed optical simulations of threshold gain features in a VCSEL device with and without lens insertion (Fig. 13).³² For both cases, the calculated fundamental mode threshold gain as a function of internal maximum temperature is plotted in Fig. 14(a). The corresponding difference between the fundamental mode gain ($G_0$) and the second transverse mode gain ($G_1$) is defined as: $(1-G_0/G_1)\times 100$ and shown in Fig. 14(b). The reference device stays single mode over all the current range (blue solid curve), as measured. An internal spatial temperature profile with a peak of 150°C at the cavity section is needed to match the experimental wavelength variation, by assuming the standard temperature dependence of the refractive indices ($2.3{10}^{-4}$).^33,34 The thermal lensing diminishes the gain difference, which reaches an estimated value of 21% for the lens-less device at the internal temperature increase of 150°C. Integrating the lens, the gain threshold $G_0$ increases by 20% [see red solid curve in Fig. 14(a)], and the gain difference between the two modes diminishes. This is caused by the lower index step between the VCSEL facet and the polymer lens. Assuming for SM operation the computed value of 21%, we get for the lensed VCSEL an SM range of 3 nm, which is determined by the continuous red curve crossing the 21% gain difference (dotted line), as shown by the green arrow in Fig. 14(b). This value corresponds to the experimental observations.

Fig. 13

Compressed index cross section map (without repetitions) of the lensed-VCSEL structure (a) (diameter of grating relief: $4\mu \mathrm{m}$, diameter of oxide aperture: $6\mu \mathrm{m}$) and (b) new proposed design (grating relief: $8\mu \mathrm{m}$, oxide aperture: $2.2\mu \mathrm{m}$).

Fig. 14

(a) Calculated fundamental threshold gain $G_0$ for the reference (blue) and the lensed (red) device (solid lines) and new design (dashed lines) as a function of the temperature increase. (b) Corresponding modal gain difference $(1-G_0/G_1)\times 100$.

Although this reduced SM tuning range is still exploitable for our sensor provided a suitable design of the optical resonator with a free spectral range of less than 3 nm, it would be preferable to relax the technological constraints on the microring fabrication and keep the entire initial spectral tuning range available. A first possibility would be to restore the original index step at the VCSEL facet, by inserting an air gap between the VCSEL surface and the cylindrical lens, which would imply a suspended lens with external pillars. The fabrication of such a 3D microstructure is feasible using 2PP photopolymerization. ZEMAX optical modeling indicates that a beam collimation as efficient as with a standard lens on a bulk pedestal could be obtained in this case by simply choosing a smaller ROC. However, this approach would require the creation of an air gap with a thickness of at least few microns between the VCSEL and the lens bottom surface, to ensure an efficient removal of un-polymerized material after fabrication. This large air gap might render the lens mechanically more fragile compared to a bulk lens. A simpler way would consist instead in slightly modifying the VCSEL design without changing the epilayer nor the parameters of the grating that serve to fix the polarization. Namely, it would consist in a singlemode VCSEL not based on the surface relief technique, but rather on small oxide aperture with a transverse design suited for the aimed application. As shown in Fig. 14(b), by simply choosing a reduced oxide aperture size of $2.2\mu \mathrm{m}$ and a polarization grating area with a much larger diameter ($8\mu \mathrm{m}$), the latter can be regarded as infinite for the optical field. Our modeling results show that this device provides better threshold performances and polarization-stable single mode operation over the whole tuning range, as desired.