1. INTRODUCTION

Optical interconnects are indisputably needed for transmitting information in telecommunications, local area networks, and rack-to-rack links. Recently, migrating the optical domain to the board level has emerged as a potential solution for the physical limitations of electrical interconnections in terms of the $\text{bandwidth}\times \text{length product}$, channel density, and power consumption [1]. Various approaches were realized including embedded fibers [2], free-space optical links [3], and polymer multimode waveguides [4]. Embedded multimode waveguides were particularly favored thanks to their relaxed alignment tolerances and compatibility with printed circuit board (PCB) fabrication processes. However, single-mode waveguides are now the focus due to their coupling efficiency to on-chip single-mode integrated waveguides and single-mode fibers [5].

Since first reported in 1965 [6], three-dimensional direct laser writing (3D-DLW) has attracted much attention thanks to its unique capability of fabricating arbitrary three-dimensional submicrostructures [7,8]. Especially in integrated optics, 3D-DLW provides a promising solution to overcome the density and complexity limit of conventional planar integrated optics by guiding optical signals freely in three dimensions. The two main materials used in 3D-DLW are glass and photopolymer.

By two-photon absorption, 3D-DLW in glass generates a microplasma that permanently modifies the refractive index of the glass in the focal volume [9,10]. Moving the sample or the laser beam in a computer-programmed 3D path creates a free-form waveguide. Compared to glass, polymer materials offer greater flexibility and optical sensitivity at a lower cost. 3D-DLW in photopolymer utilizes both one-photon and multiphoton absorption via continuous laser irradiation and ultrashort laser pulses, respectively. The former demands less energy but offers limit confinement on the polymerization volume [11]. The latter, in contrast, allows for strong confinement of the multiphoton absorption volume, enabling freeform structures. Nevertheless, sufficient refractive index contrast in embedded polymer waveguides is still a great challenge. Unlike freeform waveguides for chip-scale interconnects [12,13], here, an acceptable index contrast must be achieved without wet developing unexposed structures. Several approaches have been demonstrated using the difference between radical and cationic monomers [14] or between thermal cure and photolytic cure [15]. Unfortunately, the main drawback of these approaches is that the polymerization can continue via thermal or one-photon processes which in turn leads to reduced structural stability and hence high optical losses (e.g., $1.3\mathrm{dB}/\mathrm{cm}$, [15]). Alternatively, 3D freeform waveguides have been fabricated using a microdispenser system to dispense core monomer into liquid cladding monomer via a needle [16]. However, this approach is only suitable for multimode waveguides as a result of the large core sizes.

In this paper, we report a novel fabrication technique utilizing external diffusion of a low-index guest monomer into the photopolymer matrix in order to increase the refractive index contrast. Specifically, we demonstrate the manufacturing of symmetric single-mode 3D waveguides with adjustable core dimensions in polymer via two-photon lithography. With our latest waveguides, we achieved a much lower average transmission loss of $0.37\mathrm{dB}/\mathrm{cm}$ at a transmission wavelength of 850 nm compared to the previous work [17]. Moreover, first prototypes of waveguide arrays were produced with high density.

2. MATERIAL AND APPROACH

To be able to fabricate single-mode polymer waveguides with sufficient refractive index contrast, we use local curing of the polymer by 3D-DLW and subsequent diffusion of a lower-index guest monomer into the uncured photopolymer matrix. Hence, a novel epoxy-based photopolymer has been developed for this fabrication concept. Its main component is an oligomer based on a bisphenol-A diglycidylether. In addition to a photoinitiator, the photopolymer also contains $\gamma$-butyrolactone (GBL) solvent which allows for the forming of defect-free films by spin-coating. The resist is highly transparent at 780 nm but sensitive to 365 nm light.

The impact of monomer diffusion on the refractive index of both cured and uncured resist was investigated (Fig. 1). To determine the dispersion curves of the differently treated resist layers, the Cauchy equation was fitted to refractive index values determined via m-line spectroscopy. Equation (1) presents the Cauchy equation for the cured photoresist:

 

Fig. 1. Refractive indices of the resist measured at 1.540, 0.988, 0.594, 0.450, and 0.402 μm wavelengths and their Cauchy fitting dispersion curves. The refractive index of the uncured resist (black, 1) reduces slightly after UV cure at 365 nm and hard-bake at 50°C (red, 2) but decreases more significantly if a low-index monomer was diffused into the photoresist matrix before UV cure and hard-bake (blue, 4). The refractive index of previously cured resist (red, 2) hardly changes after the monomer diffusion process (green, 3).

Download Full Size | PDF

The UV cure at 365 nm wavelength lowers the refractive index of uncured resist slightly by 0.003. This photo-induced index contrast is not sufficient for optical waveguides. Hence, external diffusion of an aliphatic guest monomer was utilized which results in a significant decline of the refractive index by 0.016 (line 4, blue). This reduction is attributed to the lower index of the guest monomer $n_D=1.445\pm 0.0005$ relative to the host oligomer $n_D=1.59\pm 0.0005$. This experiment also reveals that the index of the cured resist (line 2, red) barely changed after the diffusion process of the guest monomer (line 3, green). In other words, the gaseous monomer hardly diffused into the polymerized, crosslinked resist. This difference in index contrast generation in the uncured and cured resist is the precondition for fabricating 3D buried waveguides via multiple photon lithography.

3. DESIGN AND FABRICATION

The fabrication of free-form 3D waveguides and waveguide arrays uses a femtosecond laser to expose the waveguides in the resist layer. A monomer external diffusion process is followed by UV exposure to stabilize the structure thermally and chemically.

A. Three-Dimensional Two-Photon Lithography System

Two-photon lithography is a rapid prototyping technique enabling the fabrication of arbitrary submicron 3D structures in photoresists. Figure 2 shows the basic schematic of a two-photon lithography system from Nanoscribe GmbH [18]. An erbium-doped fiber laser emits ultrashort pulses of 100 fs at 780 nm wavelength with a repetition rate of 80 MHz. An acousto-optic modulator (AOM) is used to control the optical power to a maximum of 70 mW. The laser beam is expanded before being split into an objective and a detector setup. The former is an objective with a $63\times$ magnification and a numerical aperture of 0.75. The latter detects sample interfaces and corrects its tilt angles via the intensity of the reflected light. A piezoelectric 3D scanning stage drives the sample tracking the programmed coordinates given from the computer. The objective is shifted vertically along the optical axis (z-direction) and the sample holder is translated laterally ($x$, $y$-directions).

 

Fig. 2. Basic schematic of a 3D multiphoton lithography system (Nanoscribe GmbH). The femtosecond laser beam is focused into the photopolymer through an objective with a numerical aperture of 0.75. An AOM tunes the laser power allowing voxel scaling. A camera and an integrated confocal system map the sample position and tilt angles with respect to the laser beam. A QWP converts linearly polarized light into circularly polarized light for a laterally symmetrical voxel. A piezoelectric scanning stage drives the sample in 3D.

Download Full Size | PDF

To avoid the depolarization effect which induces lateral de-shaping of the effective volume in the focus (voxel), a quarter-wave plate (QWP) was installed converting linear polarization into circular polarization of the light [19]. Ultrashort pulses and a high repetition rate assure a high peak power and a fast light–matter energy transfer. The combination of numerical aperture and magnification of the objective guarantees a tight focus while allowing for noncontact writing in a thick layer of photoresist with a working distance of 1.5 mm.

The depth of laser focus inside the resist is calculated by taking into account the aberration correction and diffraction at the air/resist surface. It is crucial to accurately drive the laser focal point along the resist depth due to the thin thickness of the resist. Therefore, we derived the defocus factor ($D$), namely, the shift of the laser focal point due to the refractive index mismatch between air ($n_1$) and the photoresist ($n_2$), using Snell’s law for marginal rays as

B. Fabrication Process

The fabrication flow of the 3D embedded polymer waveguides comprises four basic steps as illustrated in Fig. 3: casting, 3D-DLW followed by thermal treatment, external monomer diffusion, and flood UV exposure with hard-bake.

 

Fig. 3. Fabrication flow of the 3D embedded polymer waveguides consisting of four main steps. (a) Spin-coating on a silicon wafer and prebaked for several hours. (b) 3D laser lithography to pattern the core trajectory with multiphoton absorption and a postexposure bake at 60°C for 10 min. (c) External diffusion of a gaseous monomer into the uncured resist in a closed chamber. (d) Flood UV exposure at 365 nm and subsequent hard-bake to stabilize the whole wafer.

Download Full Size | PDF

After being spin-coated on a 2 in. silicon wafer, the gel-like polymer is solidified on a hot plate for several hours; see Fig. 3(a). The resist exhibits high adhesion to silicon wafers without an adhesion promoter. The proper choice of temperature and time for this soft bake controls the solvent concentration and guarantees a tack-free surface.

Two-photon lithography following a programmed trajectory initiates cross-linking to form the waveguide core. Postexposure bake at 60°C for 10 min speeds up the cross-linking in the exposed volume; see Fig. 3(b). The writing speed varies from 0.1 to $1\mathrm{mm}/\mathrm{s}$, and the input laser intensity ranges from 5 to 40 mW, producing different doses for the cross-linking process.

Figure 3(c) depicts the external diffusion of a gaseous low-index monomer into the uncured cladding in a closed chamber at room temperature. The diffusion time varies from 96 to 168 h for diffusion depths in the range of 40–70 μm. Current research is going on to decrease the diffusion time and to increase the diffusion depth. As proven in Section 2, the refractive index of the core remains unaffected during this process.

The UV flood exposure at 365 nm crosslinks the diffused monomer and host oligomer by means of single-photon absorption. Finally, hard-baking the sample at 140°C for 10 min fully cures the cladding and stabilizes the embedded waveguides [Fig. 3(d)]. The incorporation of the guest monomer ($n_D=1.445$) into the host oligomer ($n_D=1.59$) matrix reduces the refractive index of the cladding which yields a high index contrast between the core and the cladding for light guiding.

This fabrication process possesses various advantages over other waveguide fabrication methods. It requires only one layer of a single material. Multilayers of waveguides can be inscribed in a single writing step without any stacking or alignment effort. Finally, this rapid prototyping technique no longer involves masks, contact, or any wet chemical process steps such as wet etching.

C. Modification of Voxel Size

The possibility to manipulate the waveguide feature size is a prerequisite to structure complex devices. Despite the fact that the height of the waveguide in traditional photolithography is limited by the thickness of the core layer, 3D laser lithography manages to control both waveguide width and height at the same time.

The volumetric pixel (voxel) is the focus volume where the laser intensity is sufficient for photopolymerization to occur via nonlinear multiphoton absorption. The ellipsoidal voxel has a high aspect ratio of at least $6:1$ for this material, which increases with increasing exposure dose. To achieve symmetrical waveguides, two typical spatial beam-shaping methods for femtosecond direct laser writing were used in literature, either adding a slit [20] or two cylindrical lenses [21] in front of the objective. For simplicity, we used multiple-sweep writing. In particular, we used the parallel writing of several in-contact lines. Figures 4(a) and 4(b) show the end-facet image of a waveguide using six parallel sweeps. The proximity effect, the effect where the photopolymerization in two adjacent sweeps link up to each other, smoothens the upper and lower border of the waveguide core laterally.

 

Fig. 4. (a) End-facet image of a waveguide. (b) Illustration of multiple sweeping. (c) Side-view SEM micrograph of single lines after developing the unexposed resist, and a closer view of the periodic height variation along the writing direction. (d) Relation between input laser intensity and voxel height in direct laser writing at different writing speeds of 100, 150, and $200\mathrm{\mu m}/\mathrm{s}$. The margin of error from the voxel height measurement is $\pm 0.2\mathrm{\mu m}$. Linear fitting shows the possibility to adjust voxel height by adapting laser intensity at different writing speeds. Higher writing speed results in a smaller core dimension.

Download Full Size | PDF

Although the width of the waveguide core can be controlled by the number of sweeps and the distance between them, the smallest height of the waveguide core is limited by the voxel height. Thanks to the 3D confinement of two-photon polymerization, the laser power and writing speed define not only the voxel width but also its height. Scanning electron microscope (SEM) micrographs help determine the voxel heights, as shown in Fig. 4(c). Nine single lines at the writing speed of $100\mathrm{\mu m}/\mathrm{s}$ and increasing laser powers from front to back were written through two supporting blocks. The subsequent development bath removed the unexposed resist to characterize the voxel heights. The first free-standing line from the front reveals the threshold laser power for two-photon polymerization of 18 mW at this writing speed. The further magnified view in Fig. 4(c) shows the periodic height modification along the writing path. The maximum roughness amounts to 0.4 μm with a period of 5 μm in the front line. This height variation reduces with increasing laser power, as shown in the lines toward the back. This effect is attributed to the combination of the oscillation of the driving piezo and the change of cross-linking degree corresponding to different laser power. The piezo oscillation, whose magnitude was measured to be 0.02 μm, leads to the modulation of scanning speed, varying the exposure doses along the writing direction. In the first line, the laser power is just above the threshold for cross-linking to occur; therefore, the laser intensity at the boundaries of the voxel is not sufficient for cross-linking. At the bumps, where the degree of cross-linking is higher, the resist is more resistant to the developer compared to neighboring regions. Once the laser power is well above the cross-linking threshold (in the back line), the impact of the piezo stage oscillation becomes dominant.

Figure 4(d) plots the measured voxel height according to the writing speeds of 100, 150, and $200\mathrm{\mu m}/\mathrm{s}$ and the laser intensity from 18 to 60 mW. The voxel height varies from 5 to 20 μm. The fitting reveals a linear dependence of laser power and voxel height. A systematic study on the scaling law of voxels in two-photon polymerization has been reported [22], where a nonlinear dependence of voxel width and height on laser intensity was shown instead. However, when closely investigating these relations, a trend to a linear dependence with high laser power and high repetition rate could be identified, which agrees well with our experimental results.

D. Free-Form 3D Waveguides and Arrays

The first prototypes of freeform waveguides made by our approach include single 3D waveguides, and waveguide arrays (Fig. 5). The trajectory of the 3D waveguide follows sinusoidal functions in both the $x–y$ and $x–z$ projection plane. Both top-view and end-facet-view pictures show a good contrast between the core and the cladding. The structures are up to 4 cm long after dicing.

 

Fig. 5. (a) Design of a 3D embedded waveguide (blue, solid) and its trajectory projections (red, dotted). (b) Microscope top-view image of a $7\times 1$ waveguide array. (c) Microscope end-facet pictures of a $4\times 1$ waveguide array with a pitch size of 25 μm.

Download Full Size | PDF

The fast prototyping fabrication permits various pitch sizes and trajectories of waveguides to be tested without time and expense for masks. Figure 5(c) shows the end-facet image of a $4\times 1$ waveguide array produced with a pitch size of 25 μm. As also seen in this image, a narrow shadow appears below each waveguide core which was not observed before the flood exposure of the cladding at 365 nm UV light. The reason and impact of these shadows will be discussed in the next section.

The first generation prototypes indicate excellent thermal, mechanical, and chemical stability. Baking samples up to 200°C did not affect its near-field pattern and transmission loss. Subsequent analysis of waveguide positions revealed neither bends nor shifts. Storing samples in a laboratory environment without any protection for more than one year did not degrade its optical performance. Furthermore, sequential chemical baths with acetone and PGMEA did not detach the resist from the silicon wafer.

4. OPTICAL CHARACTERIZATION

In order to analyze the waveguide performance, we determine its index profile, guiding modes, mode field diameter, numerical aperture, and transmission loss.

A. Refractive Index Profile

The refractive index profile of a waveguide determines its performance in the mode propagation including the number of modes and the mode field diameter.

As the waveguide structures are deeply buried in the resist layers, characterization methods for planar devices such as ellipsometry and prism coupling are not suitable here. Therefore, the refracted near-field (RNF) technique through measuring the change in the power of refracted rays has been used. RNF provides a high-resolution measurement with step sizes of 0.1 μm and an index precision of ${10}^{-5}$. Figure 6 plots the two-dimensional refractive index profile of the core and the cladding at 678 nm. The 3D graph shows a symmetrical core with the size of 7.4 μm. The refractive index of the core fluctuates in a small range of 0.003 but decreases more significantly at the boundaries to the cladding. On the one hand, the high total index step of 0.013 assures single-mode guiding at communication wavelengths. On the other hand, the gradient index at the outer part of the core increases the light confinement. Refractive index distribution of the cladding around the core is homogenous with deviations of only 0.001. This variation is due to the exponential distribution of monomer concentration along the diffusion depth.

 

Fig. 6. Area refractive index profile of a sample waveguide at the end facet measured with the refracted near-field method in (a) a 3D graph and (b) its cross section at $x=0$. The peak index contrast between the core and the cladding at 678 nm is 0.013.

Download Full Size | PDF

Waveguide cores behave as a lens for the UV light during the flood exposure due to their gradient index near the core border. This explains the focusing shadows below each waveguide core described in Fig. 5(c). This interpretation was again confirmed by the tilting of the shadows following flood exposure coming at different angles.

B. Near-Field Pattern

A Hamamatsu near-field pattern (NFP) measurement system was used to analyze the intensity distribution, the mode field diameter, and the symmetry of the guided modes within the fabricated waveguides. This system uses expansion optics with a high magnification ($500\times$) and a digital CCD camera to acquire high-resolution NFP images from the output facet of a waveguide. An appropriate single-mode fiber couples light to its input facet from a laser at transmission wavelengths.

Figure 7 presents the near-field patterns and their cross section profiles for transmission wavelengths of 850 nm [Figs. 7(a) and 7(b)] and 1550 nm [Figs. 7(c) and 7(d)]. The beam propagation method from BeamPROP was utilized to simulate and optimize the waveguide core dimensions. The actual cores in comparison with the near-field patterns reveal the light confinement of the guided single mode. All intensity profiles fit well with Gaussian distributions with a mode field diameter derived at $1/e^2$ intensity of 5.7 and 7.4 μm at 850 and 1550 nm, respectively. These cross section profiles show highly symmetrical mode fields with diameters comparable with single-mode optical fibers.

 

Fig. 7. Near-field pattern of the waveguides measured at transmission wavelengths. The near-field intensity profile in false color measured at 850 nm in comparison with the physical core, and its cross sections in $x$ and $y$ (a),(b) at 850 nm and (c),(d) at 1550 nm. The $1/e^2$ mode field diameters are both $5.7\pm 0.3\mathrm{\mu m}$ at 850 nm and both $7.4\pm 0.4\mathrm{\mu m}$ at 1550 nm. Black curve: measured intensity profile; red curve: Gaussian fitting.

Download Full Size | PDF

C. Far-Field Pattern

While a near-field pattern determines the number of guiding modes and their field diameters, a far-field pattern analyzes the beam divergence. The resulting numerical aperture defines the coupling efficiency of a waveguide to another waveguide, a fiber, or a laser source. A Hamamatsu far-field pattern system helps measure this far-field pattern by identifying the angular dependence of the intensity of the electromagnetic radiation at the end facets of the waveguides.

Figure 8(a) depicts the far-field pattern of a waveguide coupled from a Corning HI780 fiber at 850 nm and its cross section profiles. The full divergence beam amounts to 17°. The resulting numerical aperture amounts to $0.148\pm 0.01$ and is comparable to the numerical aperture of the single-mode fiber at 850 nm. This result will be helpful for coupling loss calculations as done in the next section.

 

Fig. 8. (a) Far-field pattern measured at 850 nm and its intensity profile cross sections. (b) Insertion loss measurement for cut-back method at 850 nm.

Download Full Size | PDF

D. Transmission Loss

Since polymer materials commonly have higher absorptivity in the spectral ranges relevant to optical communications compared to silicon, it is important to determine its transmission and coupling losses. We applied the cut-back method and the overlap integral method to determine these values.

The cut-back method exploits different insertion losses at various waveguide lengths to determine the transmission loss. In this experiment, two Corning HI780 single-mode fibers couple the 850 nm light in and out of the waveguide by butt coupling. Index matching oil is used to avoid Fresnel reflection losses and polishing. Insertion losses at 8, 13, and 18 mm lengths, plotted in Fig. 8(b), are linearly fitted. The slope of this line reveals a transmission loss of $0.37\mathrm{dB}/\mathrm{cm}$, while its zero-intersection shows the total coupling loss of 1.12 dB.

Alternatively, the coupling loss was derived using an overlap integral. Two main contributions of coupling loss, which are mode field mismatched and numerical aperture mismatched, were calculated as 0.07 and 0.42 dB, respectively, for each fiber–waveguide coupling. The total coupling loss for both input and output of the waveguide is $2\times (0.07+0.42)$ which equals 0.98 dB. This result is consistent with the cut-back method, since the coupling loss difference of 0.14 is attributed to the imperfect coupling.

There are some reasons to expect a lower transmission loss by further improvement. The surface roughness, a main contribution to attenuation apart from material absorption, was of the order of $\lambda /2$ for waveguides with a core height of 5 μm. It could be reduced significantly to the intrinsic oscillation of the translating piezo of $\lambda /10$ by increasing the material absorption at 390 nm wavelength. Furthermore, material absorptions at 850, 1310, and 1550 nm are 0.03, 0.35, and $0.88\mathrm{dB}/\mathrm{cm}$, respectively. Finally, single-photon absorption experiments of a similar photopolymer based on the same epoxy oligomer has been used to fabricate single-mode polymer waveguides with a transmission loss of $0.14\mathrm{dB}/\mathrm{cm}$ at 808 nm [23].

5. CONCLUSION

We have demonstrated a novel technique for three-dimensional waveguide fabrication into a newly developed photopolymer using two-photon lithography in combination with external diffusion of a gaseous monomer. A high-index contrast between the core and the cladding of 0.013 has been achieved. By adjusting the writing speed and laser intensity for the laser writing, it is possible to produce symmetrical square-shaped waveguides with adjustable core dimensions in the range of 5 to 20 μm. Fabricated waveguides exhibit high reliability and a transmission loss of $0.37\mathrm{dB}/\mathrm{cm}$ at 850 nm wavelength. The fabrication process requires only one layer of a single material. Multilayers of waveguides could be produced without any stacking or alignment effort. Transmission losses might be further reduced by reducing the waveguide’s roughness by improving the optical sensitivity of the photoresist. Therefore, this technique could pave the way for three-dimensional optical interconnects at the board level with high complexity and bandwidth density.

Current work is going on to demonstrate the applications of this technology such as fan-in/fan-out for multicore fibers, 3D splitters, and 3D optical routers. In view of commercial use of the process, we are studying the possibilities to reduce the diffusion time and to increase the diffusion depth.