Abstract: In this paper, we proposed an efficient and high-precision process for fabricating large-area microlens arrays using thermal reflow combined with ICP etching. When the temperature rises above the glass transition temperature, the polymer cylinder will reflow into a smooth hemisphere due to the surface tension effect. The dimensional differences generated after reflow can be corrected using etching selectivity in the following ICP etching process, which transfers the microstructure on the photoresist to the substrate. The volume variation before and after reflow, as well as the effect of etching selectivity using process parameters, such as RF power and gas flow, were explored. Due to the surface tension effect and the simultaneous molding of all microlens units, machining a 3.84 × 3.84 mm2 silicon microlens array required only 3 min of reflow and 15 min of ICP etching with an extremely low average surface roughness Sa of 1.2 nm.
1. Introduction
Traditional optical lenses have the disadvantages of being large in size and heavy in weight, which do not satisfy the development trend of miniaturization and lightweighting for optical systems. Microlens arrays (MLAs) have emerged as an alternative to large aperture optical lenses and are composed of numerous tiny micron-sized lens units arranged in a specific layout. MLAs can provide excellent optical functions, including illumination, collimation, focusing, imaging, and light redistribution, coupled with significantly reduced mass and volume compared to conventional lenses, enabling great application potential across multiple fields, such as imaging systems, optical communication, and sensors.
The fabrication processes of microlens arrays include ultra-precision machining technologies, laser direct writing, grayscale lithography, nanoimprint lithography, and thermal reflow processes. Ultra-precision machining technologies, such as single point diamond turning (SPDT), are capable of fabricating microstructures with complex morphology and excellent precision. However, continuous cutting for an excessive duration will affect machine tool accuracy and significantly damages the diamond tool, resulting in poor uniformity of the processed microstructures. Yan and Makaida studied a tool-servo driven segmented turning method to fabricate silicon concave microlens arrays. The intermittent contact between the tool and the workpiece can effectively limit tool wear, achieving a surface roughness of 5 nm, but the machining area needs to be further improved. Laser direct writing technology utilizes optical systems to focus a laser beam onto the surface of the workpiece, thus melting and vaporizing the material to fabricate micro/nano structures. Hua et al. used a femtosecond laser with a wavelength of λ = 343 nm and a pulse duration of tp = 280 fs to directly fabricate a convex MLA on silicon with a surface roughness of below 3 nm. Femtosecond laser technology works directly on the target material without additional masks, but it is not suitable for machining large-area microstructures due to the low processing efficiency caused by its inherent characteristic of removing material point by point. The laser used by Hua had a material removal rate of 120 µm3/s, and it took 1 h to ablate the MLA with a footprint of 100 × 100 µm2 , which is clearly not applicable to large-area microlens arrays. Deng et al. attempted to process large areas, but the SEM images showed that the uniformity of array units was not satisfactory.
Grayscale lithography requires only one-time exposure and development to obtain 3D microstructures on photoresist (PR) using grayscale masks. However, the difficulty of producing high-precision masks greatly limits its promotion and application. Nanoimprint technology transfers the microstructure from the mold to the polymer through mechanical hard contact and is classified into hot embossing lithography (HEL) and UV-NIL, according to the polymer molding principle used, which are shaped by high temperature and UV light, respectively. This technique is characterized by process simplicity, high efficiency, and good reproducibility, thus enabling its use in mass production. Notably, it requires ultra-precision machining of the opposite target microstructure on a mold first, and as mentioned above, the ultra-precision machining technology is still insufficient for machining large areas. Once, we successfully machined a 6 × 6 microlens array on a silicon substrate using a hot embossing process; however cutting a 600 × 600 µm2 highprecision mold already approached our machine tool limit and the surface roughness of the microlens unit was only 17.7 nm, which would affect the light throughput quality of the lens.
2. Materials and Methods
2.1. Materials
The 4-inch optical monocrystalline silicon wafers used as the substrate material were customized from Meixin Electronics (Tianjin, China), with N-type doping, a crystal orientation of <111>, and a thickness of 200 ± 15 µm. The silicon wafers were diced into 1-inch samples using a laser ultra-precision processing system (DelphiLaser, UP-D, Suzhou, China). All the samples were ultrasonically cleaned with acetone, isopropyl alcohol, and anhydrous ethanol sequentially for 3 min and were baked at 100 'C for 10 min beforethe experiment. Photoresists, such as AZ5214 and S1813, and developer NMD (TMAH2.38%) were sourced from Resemi (Suzhou, China). The NMP solution obtained fromShanghai Aladdin (Shanghai, China) was used to remove any residual photoresist on thewafer surface.
The fabrication process for large-area silicon MLAs in this work is depicted in Figure 1First, the photoresist, including AZ5214 or $1813, was spin-coated on the silicon wafer(step i). The thickness of the film depended on the spin speed and the viscosity of thephotoresist. This was followed by prebaking on a hot plate at 100 'C for 60-90 s to ensurethe adhesion of the PR layer to the substrate. Next, the exposure process (step ii) wasperformed using an MA/BA8 lithography machine (SUSS MicroTec, Garching, Germany)and the prescribed intensity of UV light was 23.45 mW/cm-2. Since the thickness ofthe photoresist film was mainly below 3 um, an exposure time of7 s was sufficient. Wechose soft contact with a photomask as the exposure mode. Although not as effective asnon-contact exposures, such as laser direct writing, displacement Talbot lithography(DTL) can completely avoid the issue of wafer warpage. The lower contact pressureof the soft contact can maximize the uniformity of the lithography microstructures whileensuring full contact between the photomask and the photoresist layer on the 1-inchsilicon substrate. Subsequently, the wafer was immersed in NMD solution for 45 s fordevelopment (step ii), thereby removing the photoresist illuminated by the UV light instep ii. After exposure and development, the initial cylinder array was processed on thesubstrate, but the actual structure was similar to a circular truncated cone due to theimperfect photolithography process, as shown in Figure 2a.
Figure 1. Illustration of thermal reflow and the ICP etching process.
Figure 2. Simp lified model before and after reflow. (a) The circular truncated cone before reflow and(b) the spherical structure after reflow.
Finally, the target MLA was etched onto silicon through ICP etching (step v), whichwas performed using the Oxford etchingsystem (Oxford Instruments, Plasma lab system100 ICP 180, Yatton, UK). The etcher was equipped with two RF sources, both with afrequency of 13.56 Mhz, one of which acted as an ICP generator and was connected to aspiral coil wound outside the chamber to generate an inductively coupled electric fieldThe other RF source was connected to an electrode below the sample plate inside thechamber, called the CCP generator. During ICP etching, selected gases enteringthe etching chamber generate high density plasma via glow discharge under the effectof the electric field, and the plasma density is affected by the power of the coil (ICPpower). Then, the ionized plasma is accelerated by the bias voltage generated by theCCP generator to bombard the wafer downward, thereby removing the surface materiaphysically and chemically. We chose SF6, CFs, and O, to be injected simultaneouslyas process gases, where SF6 and Oz were mainly used for etching silicon and the photoresistrespectively. The addition ofC Fs increased the anisotropy of the etching process, whichincidentally slowed down the etching rate. Sometimes the microstructure after reflow maydiffer from the desired size, fortunately, this error can be corrected by adjusting the etchingselectivity (the ratio of the etching rate of silicon to photoresist) to control the finalizedstructure transferred to silicon. lf the reflow size is higher than ideal, a selectivity of lessthan 1 can be applied to reduce the height. Conversely, if there is an undersized error, aselectivity greater than 1 is required, In order to obtain accurate microlens dimensions onsilicon, the influence of RF power and gas flow on etching selectivity under constant ICPpower and chamber pressure was investigated. The selectivity S, is calculated using thefollowing formula (illustrated in Figure 3).
Fig 3. Schematic diagram of the selectivity calculat.
3. Results and Discussion
3.1. Results and Analysis of Thermal Reflow
After lithography and the thermal reflow process, photoresist microlenses formed on the silicon substrate. To clarify the effect of the size before reflow on the lens shape, group experiments with different bottom diameters d1 were designed at spin-coating speeds of 1000 rpm and 1300 rpm. The obtained results are listed in Table 1. The thickness of the PR film was roughly 3 µm at 1000 rpm and 2.6 µm at 1300 rpm. For the reflow experiment, we customized another photomask with multiple 5 × 10 cylinder arrays. The cylinder units of these arrays were of different diameters, but all had a pitch of 100 µm. Through a pre-experiment on reflow time, it was observed that all the photoresist cylinders could reflow into complete spherical structures within 2 min. After 3 min, the reflow tended to stabilize and the microlens shape was basically unchanged. So, we uniformly set the reflow time to 3 min to avoid time interference.
The section profiles before and after reflow were extracted for comparison using the stylus profilometer, as shown in Figure 4. Combined with d1 and d3 in Table 1, it can be seen that the bottom diameter of the photoresist before and after reflow remained almost unchanged, which is due to the fact that the contact surface between the photoresist and substrate was already solidified when the photoresist temperature had not risen above the glass transition temperature, resulting in the reflow being essentially unaffected by the contact angle. This means that controlling the bottom diameter of the photoresist cylinder before reflow can determine the bottom diameter of the microlens after reflow. The cylinder layout was replicated from the mask to the photoresist using lithography. So, when the mask is designed and fabricated according to the target microlens array before the experiments start, this indicates that the bottom aperture of the lens has been defined. From the experimental results of groups 1–4 and 5–8 in Table 1, the volume reduction percentage k gradually reduced with an increase in the truncated cone diameter at the samethickness. This was attributed to the larger exposed surface area resulting from increaseddiameters, which exacerbated the volatilization of the photoresist solvent during heatingThe variable k makes it difficult to control the lens height h2, which requires numerousexperiments to achieve the desired size. The prolonged time required for thermal reflowposes an unavoidable problem. Fortunately, ICP etching pmvides a solution by controllingthe etching selectivity ratio, which we have presented in the next section.
Fig4
4. Conclusions
In conclusion, thermal reflow processes combined with ICP etching proved to be an efficient fabrication process for machining large-scale microlens arrays. The cylinder array after lithography was reflowed into smooth spherical microstructures using reflow processes driven by surface tension, which were subsequently delivered to the silicon substrate by ICP etching. We discovered that the bottom diameter of the microlens remained invariant because the contact surface of photoresist and substrate had already cured under the influence of high temperature before the reflow began. This characteristic makes it convenient for controlling the microlens dimensions. Although the photoresist volume reduction percentage varied with the thickness of the cylinder, resulting in a deviation of the size after reflow from the designed size, this can be rectified by adjusting the etching selectivity. The selectivity increased with O2 and SF6 flow, with an opposite contribution of C4F8 and RF power. A 128 × 128 silicon microlens array with good uniformity and a low surface roughness of 1.2 nm was fabricated in a short period of time, illustrating the excellence of the thermal reflow process for mass production. Our future work includes fabricating larger-scale microlens arrays and constructing optical platforms for measuring the optical performance of MLAs.
下一篇： 磨粒磨削中GaN晶体的损伤演化和去除行为
