1. Introduction

There has been a growing research interest in recent years in the area of PhCs [1], which are artificially engineered media that can be used to control the propagation of electromagnetic waves. Generally, PhCs consist of periodic structures that possess periodicity comparable with the wavelength that the PhCs are designed to modulate. If material and periodic pattern are properly selected, PhCs can be applied to many applications based on their unique properties, including photonic band gaps (PBG) [2, 3], self-collimation [4, 5], super prism [6], etc. Strictly speaking, PhCs need to possess periodicity in three dimensions to maximize their advantageous capabilities.

One important approach to realize low cost fabrication of 3D PhCs is based on 3D photolithography techniques. For example, with a highly collimated x-ray beam, a thick photoresist film can be “drilled” by three angled exposures to form the Yablonovite structure [7]. By properly arranging multiple laser beams, interferometric lithography has been used to generate many types of defect-free 3D periodic structures [8]. Arbitrary 3D polymer structures can be fabricated by multi-photon polymerization [9]. Although substantial progress has been made, these approaches still suffer from significant disadvantages. X-ray lithography relies on an expensive radiation source. Interferometric lithography lacks the flexibility to create engineered defects, and multi-photon polymerization is currently not suitable for parallel fabrication.

In our previous work [10], we developed a multilayer photolithography process that is based on the so-called “desensitization effect” of an image-reversal process. This method makes it possible to fabricate 3D PhCs with arbitrarily introduced defects. Unfortunately, multiple high temperature treatments result in reduced contrast in the final flood exposure due to the over desensitized resist in the lower layers.

In this paper, we extend the generality and manufacturability of its predecessor with a new multilayer lithography process based on absorption-enhanced exposure. In the following sections the development of the fabrication process is described first followed by demonstration with the fabrication of “woodpile” PhCs.

2. Fabrication process

In this work, we use multiple planar patterns of negative resist to construct the desired 3D structure in a layer-by-layer fashion. Figure 1 shows the lithography process. For each layer, the resist is first spin-coated and dried, which is followed by a 3D confined exposure and a post-exposure bake (PEB) to form a latent pattern by crosslinking the exposed resist. The same processing loop is repeated for multiple times such that the planar patterns in different layers, which are independent with each other but can be predetermined by the mask design, combine together to form a latent 3D PhC structure with arbitrarily placed defect. This latent 3D structure is revealed by the final development.

In the above fabrication processes, it is important to control two processing parameters, the spin thickness and the exposure penetration depth, to faithfully reconstruct the 3D photonic crystal structure. As an example, for the “wood pile” fabrication, the spin thickness defines the lattice constant in the vertical direction, and the exposure penetration depth determines the thickness of each “log”. These two parameters are actually associated with two fundamental problems for the multi-layer photolithography process, namely the 3D confined exposure and multilayer resist application. In this section we discuss our solution to these problems as well as the infiltration of resist template with a high-index precursor.

2.1. Three-dimensionally confined exposure

In the above fabrication process, it is critical to confine the exposure in the vertical direction. However, in conventional planar lithography the information on the mask is usually transferred to the resist layer as extruded 2D features. Straight, vertical sidewalls of high aspect ratio representing the faithful reproduction of the 2D information on the mask are desired. For this reason, transparency of the resist is very important for uniform exposure. Resist transparency, on the contrary, is not desired in the proposed fabrication process, as it causes re-exposure of the lower resist layers that have been previously patterned. In order to achieve surface-confined exposure, we modify the conventional exposure by enhancing the absorption of the resist.

 

Fig. 1. Fabrication process of 3D PhCs by multi-layer photolithography

Download Full Size | PDF

There are essentially two ways to achieve enhanced absorption. One is to add absorption component into the resist formula. This method has been widely employed for production of lift-off resist. Although it is difficult to find a commercially available resist that can provide enough absorption to confine the exposure to only a small depth close to the surface, it is possible to develop a customized resist, i.e. a resist with particular dye or dyes added that strongly absorbs at certain wavelength range giving the desired absorption character. Alternatively, knowing that many resists were abandoned due to their high absorption during the rapid shrinking of the exposure wavelength in the last decades, we can enhance absorption by using a shorter wavelength for exposure. For example, absorption can be greatly enhanced if an i-line resist is exposed with a deep-UV source. Compared with the first method, the latter is more easily demonstrated since it allows us to select from a wide range of commercial resists as well as standard exposure sources.

In this work, the absorption-enhanced exposure is studied by a specially designed experiment where the thickness of a series of suspended cantilevers is measured to reflect the effective exposure depth as a function of varied wavelengths and doses. In the experiment, we first spin-coat a 4µm resist film using a negative i-line resist NR7-1500 (Futurrex, Inc., 12 cork hill road, Franklin, NJ, 07416, USA). The same cantilever pattern is then scanned with varied doses to expose resist, employing both 254nm and 220nm deep UV sources. These cantilever patterns are anchored to the substrate with an additional exposure using 365nm source. After the post-exposure bake and development, we obtain a series of suspended cantilevers. Figure 2(a) shows the SEM micrograph of the cantilevers exposed with 254nm. Similar experiment was performed using 220nm source, as shown in Fig. 2(b). The results clearly demonstrated the capability of confining the exposure to the resist surface when the light source is mismatched with the resist. Moreover, varying the exposure dose for a given wavelength of light source can control the depth of the effective exposure. The thickness of the suspended cantilever structure is actually comprehensively determined by both the absorption and the efficiency at which photons can initiate photochemical reactions. Therefore, even though the absorption at 220nm is larger, the structure formed by 220nm is thicker for the same energy dose because fewer photons are required to generate equal amount of photoacids. In practice, we use the experimental data to calibrate the effective exposure penetration depth.

 

Fig. 2. Cantilever structures fabricated with absorption-enhanced exposure. The thickness of cantilever, which is measured using an environmental SEM system, reflects the effective exposure penetration depth. (a) SEM micrograph of cantilever structures that are formed using 254nm wavelength. The dose of exposure is increased from 0.9mJ/cm² (upper left) to 18.45mJ/cm² (lower right) with 0.45mJ/cm² increment; the thickness is increased accordingly from 329nm to 1860nm. (b) SEM micrograph of cantilever structures that are formed using 220nm wavelength. The dose of exposure is increased from 0.33mJ/cm² (upper left) to 6.6mJ/cm² (lower right) with 0.33mJ/cm² increment; the thickness is increased accordingly from 287nm to 1430nm. (c) Measured penetration depth as a function of varied doses. Circle dots are the result for 254nm exposure; Triangular dots are the result for 220nm exposure.

Download Full Size | PDF

2.2. Multiple resist applications

Having the capability of confining the exposure in a given distance from the resist surface, the next step is to control the resist application process such that a new layer of resist can be deposited with a controlled thickness on the top of previously applied resist films. In this work, we choose to spin-coat the resist because it is the most widely used method. With the presence of the previously applied resist films, multiple resist application differs from the conventional resist spin-coating by the additional solvent/resist interaction at the interface between new resist solution and the dried resist film.

The solvent/resist interaction strongly depends on the degree of crosslinking of the patterned resist film. Figure 3 shows typical results of the spin-thickness curves for spinning a new layer on top of well-crosslinked resist and on un-crosslinked resist. The spin-thickness of the well-crosslinked case increases linearly with number of layers. For a given spin condition, the thickness of each layer is the same except for the first layer, which is spun on the substrate surface. We therefore conclude that the solvent/resist interaction on a well crosslinked resist surface can be neglected. However, under the same spin condition, the spin-coat has very different behavior for the case of spinning onto un-crosslinked resist. The thickness of each resist application decreases first and then increases as more layers are applied, indicating the dependence of the spin-thickness on the thickness of previously applied resist.

This dependence, when spinning onto un-crosslinked resist film, can be explained by comparing the spin-coating process with or without the solution/resist interaction. In conventional spin-coating, the weight loss mechanisms consist of out-flow (convection) and evaporation [11], where out-flow causes the loss of both solid resist and solvent while evaporation only causes the loss of solvent. Usually, out-flow dominates the beginning of the spin process and evaporation dominates the remaining time. In the case of multiple resist application, although out-flow and evaporation are still the only weight loss processes if we consider all the resist films and solution as a whole system, they are further influenced by the solvent/resist interaction. The interaction may act as an enhancement for out-flow if the solvent re-dissolves the resist film and makes it “re-flow”. As a result, spin-thickness is reduced since more solid resist, from both resist solution and previously applied films, are lost at the beginning of the spin. The interaction may also serve as a new mechanism for solvent loss from the new resist solusion if the previously applied resist films “re-absorb” a substantial amount of solvent while remaining immovable. The out-flow-dominated phase is therefore terminated earlier, leaving more solid resist on the substrate and increasing spin-thickness. Therefore whether out-flow or evaporation dominates for a given spin condition depends on the thickness of the previously applied resist film. Thin films are more easily re-flowed and therefore behave as out-flow dominated while thick films can re-absorb more solvent and therefore behave as evaporation dominated.

 

Fig. 3. Typical spin-thickness curves for different conditions. Blue lines with triangular dots are the spin result on un-crosslinked resist; pink lines with square dots are the spin result on crosslinked resist; green lines with circular dots are the spin result on patterned PhC structure with controlled airflow. In the first case, resist thickness is measured each time after a new resist layer is applied and soft baked. In the second case, every resist layer is soft baked, measured, flood exposed and post-exposure baked before the new resist layer is applied. In the third case, resist is processed exactly using the proposed 3D lithography method. The airflow rates used upon spinning are 0, 3.2m/s, 2.1m/s, 1.6m/s, 1.2m/s and 0 respectively. Solid lines represent the total resist thickness; broken lines represent the thickness result of each spin. Except the first layer in the third case, which is spun at a reduced speed to obtain desired thickness, all the resist layers are spun at 3000rpm with 500ramp for 52 seconds.

Download Full Size | PDF

For the fabrication of a photonic crystal, new resist is applied on the top of a resist film that has an interconnected network of crosslinked resist filled with un-crosslinked resist. Therefore the spin process behaves intermediately relative to the cases of well-crosslinked resist and un-crosslinked resist. In practice, we use enhanced evaporation, which terminates the out-flow-dominated phase earlier and reduces the loss of solid resist, to compensate for the “re-flow” effect caused by the solvent/resist interaction. This is experimentally realized by increasing the airflow rate above the sample upon spinning. As more resist films are applied, the airflow rate is reduced because the re-flow effect is partially compensated by the re-absorption effect. The spin-thickness result with controlled airflow is also shown in Fig. 3.

2.3. Infiltration of resist template

The above lithography process results in 3D PhC structures in photoresist. In many cases, the photoresist-air structure does not possess enough index contrast to open a complete photonic bandgap. Therefore other materials with higher indices are usually employed to infiltrate the resist sample [12]. After removing the resist template, an inversed 3D PhC structure with the desired index contrast can be obtained. In this work, we infiltrate the resist template with zirconia using a sol-gel process. Precursor consisting of 1 to 1 ratio zirconium acetate and methanol is thoroughly mixed with a magnetic stirrer. Then the precursor is applied to the resist template. In order to reduce the stress produced upon the solidification process, the sample is dried in a sealed chamber with slowly decreased solvent (methanol) vapor pressure. The dried sample is finally heated to remove the resist template by using a programmable furnace Barnstead/Thermolyne 47900 (BarnsteadInternational, 2555 Kerper Boulevard, Dubuque, Iowa USA 52001-1478). The heating begins at the room temperature and increased to 450°C with a rate of 2°C/min. The temperature is then stabilized for 1 hour to completely remove the resist. The temperature is decreased back to room temperature with a rate of 5°C/min.

3. Experimental demonstration

We choose a commercial negative resist NR7-1500PY for the experiment due to its strong absorption and high development selectivity between crosslinked and un-crosslinked resist. At room temperature, the resist has a viscosity of approximately 35 cSt depending on the shelf time of the resist and the ambient conditions. For unexposed resist, its absorption coefficients are 0.494µm^-1 and 2.442µm^-1 at 365nm and 254nm respectively. All the resist layers are spin-coated at 3000rpm and 500rpm/s ramp with varied airflow according to the calibration result shown in Fig. 3. The resist is then soft baked by a hotplate at 135°C for 1min and exposed by an ABM mask aligner (ABM, Inc., 6280 San Ignacio Ave Ste M, San Jose, CA USA 95119) with a dose of 6mJ/cm² using 254nm deep UV source. The post exposure bake is accomplished at 120°C for 1min using a hotplate. Finally, the resist is developed using RD6 negative developer to reveal the latent structures. Figure 4 shows the SEM images of the 3D structures fabricated using the proposed method. Exposed in a contact mode, the trapezoidal cross sections of resist rods are formed mainly due to the non-uniformly light intensity distribution in the lateral direction, which is caused by diffraction. This effect can be reduced if edge beads are properly removed thus reducing the air gap between resist and mask or can be compensated if a reversed trapezoidal aerial image can be formed using a projection system.

 

Fig. 4. SEM micrograph of multilayer “woodpile” structures that are fabricated using the proposed 3D lithography method. (a) Six-layer “woodpile” structure. (b) Eight-layer “woodpile” structure.

Download Full Size | PDF

 

Fig. 5. SEM micrographs of infiltrated Zirconia “woodpile” structures. (a) A 4-layer 100µm by 100µm woodpile structure. (b) A 4-layer woodpile structure with arbitrary defects of “UD” letters embedded in the second layer from the top. (c) Sidewall view of an 8-layer woodpile structure.

Download Full Size | PDF

We finally infiltrated the resist template using the sol-gel process discussed above. Figure 5 shows the SEM micrographs of the inversed zirconia “woodpile” structures. Since it is difficult to control the thickness of infiltrated zirconia film, the PhC structure is usually embedded in the zirconia. To view the PhC structures, we peeled the solidified film off from the substrate surface. Therefore, the micrographs are actually observed from the “bottom” of the PhC structure.

4. Summary

In this paper, we demonstrated a 3D multilayer lithography method for the fabrication of 3D photonic crystals. Modified from the planar lithography, this approach allows arbitrary 3D features as well as batch fabrication. Following the introduction of this method, we discussed its two fundamental derivations from the conventional planar lithography, namely 3D confined exposure and multiple resist application. The first is experimentally realized by using a mismatched resist/light source system to enhance the absorption upon exposure, which results in controlled exposure penetration depth as a function of both wavelength and dosage. Different mechanisms that influence the spin thickness upon multiple resist applications were discussed. We proposed and demonstrated an environmentally controlled spin technique to achieve controlled spin thickness over multiple layers. We fabricated “woodpile” structure up to eight layers using the 3D lithography method. We finally showed that the resist structure could be further infiltrated with zirconia using a sol-gel process.