Precision glass molding is a more cost efficient process for the large volume manufacturing of highly complex optical surfaces than direct manufacturing. Glassy carbon (GC) molds are used for precision glass molding, because they can be operated at temperatures up to 2000°C. Used today mainly for manufacturing aspheric lenses, we consider here material technology for diffractive optical element (DOE). For diffractive optics the surface structuring is in the micrometer range and a surface roughness Ra lower than 20 nm is required. We introduce a reactive ion etching process with a titanium hard mask. Fused silica (FS) molds with identical optical functionality were fabricated for comparison. All molds were used for precision glass molding of a low Tg glass L-BAL42. We will compare GC and FS as mold materials in terms of quality and robustness. Optical performance measurements of the molded glass DOEs are shown and are in good agreement with the theoretical predictions. The results confirm that precision glass molding based on GC molds is a very promising technology to economically fabricate small structures in glass for DOEs.
© 2016 Optical Society of America
Today precision glass molding is used for fabrication for highly complex refractive optics, like asphers or freeform elements [1,2]. The cost to machine such a surface shape are very high for direct manufacturing (e.g. single point diamond turning), which is why precision glass molding is used. In the current applications smooth surfaces are manufactured. To fabricate diffractive optics, surfaces have to be micro-structured at very high precision. Important here are the minimum feature size and the aspect ratio of the structure, which enables larger diffraction angles. Sharp profiles are required to obtain high quality diffractive optical elements (DOEs) [3,4]. Profile roundings can cause significant reduction of the efficiency. It is very challenging when not impossible to fabricate molds with such specifications by mechanical machining.
No replication process for glass DOEs has been introduced to industry so far. Mainly, plastic DOEs fabricated by plastic injection molding are used offering low cost mass production . Much less optical plastics are available than optical glass materials. They have a lower transmission coefficient and cannot be used for UV light, where the polymer degrades rapidly due to absorption. High power light sources e.g. LEDs for illumination optics or lasers for manufacturing have to be used with glass optics, because polymers cannot withstand the thermal environment. Also plastic injection molded elements suffer often from straylight and haze. Which is why glass DOEs are preferred for high quality DOEs.
Glass DOEs today are fabricated by dry etching of fused silica (FS) wafers , which offers good transmission over a large spectral range. Precision glass molding of DOEs is a high potential alternative to direct etching in quartz, especially when the direct FS fabrication costs become very high e.g. for multilevel DOEs or ultra-high resolution DOEs written with e-beam lithography. Few attempts have been committed to use precision glass molding for DOEs. A micro Fresnel lens was replicated with silicon molds . Beamsplitting elements were fabricated with FS molds  and glassy carbon (GC) molds  and sine wave gratings with silicon (Si) molds . Also subwavelength relief surfaces were fabricated with precision glass molding: phase plates with SiC mold  and GC molds ; and antireflection structures with SiC molds on plane surfaces  and on curved surfaces . However, little or no effort has been spent to evaluate the optical performance of such elements. Mostly the replicated surface relief was examined for its structure and morphology and compared to the master. In our study, the question under discussion is how good the optical performance of molded DOEs can be and by what it is influenced.
One main limitation of precision glass molding can be seen in the availability of a high resistant, chemically stable mold materials, which can be micro-structured with the necessary quality and feature sizes. Si, SiC and FS have been successfully used as molds materials, but need an anti-adhesion coating, because the glasses used for molding are chemically very reactive in the glass transition temperature range. Not only are the costs higher and a coating is more prone to breaking, but also a feature rounding will be introduced in the order of the anti-adhesion layer thickness. One of the most promising materials is GC, since it does not require an anti-adhesion coating. Sasaki et al. molded successfully a Borofloat glass from Schott at 655°C with a GC mold used 120 times . GC can be operated up to temperatures of 2000°C, which makes it possible to mold FS at ca. 1400°C . One problem with GC is the availability of high quality wafer materials that are defect free. Recently, such materials become available and we demonstrate here the next step in mold fabrication with GC with wafer sizes of 4 inch, a size very practical for wafer scale fabrication.
Precision glass molding of DOEs was demonstrated with GC molds [8,11], but the surface roughness was too high with more than 20 nm and the available surface areas were very limited. Optical applications require much lower roughness values. Youn et al. introduced a dry etching process that accomplishes the requirement of low roughness . They used the high quality GC molds for replication of microfluidic channels. No DOE was fabricated or tested. In our study, we seek to adapt this approach for DOEs with optical glass materials with the potential to use any kind of glass for molding and increase the active surface to wafer level.
A reliable process for the structuring of GC at the micro-scale has to be set up. We employ standard mask lithography, which was designed with binary DOEs, to find the best plasma etching conditions. As starting point we use the results from Youn et al., who used an Au hard mask for plasma etching of GC. When using Au in dry etching systems, the Au contaminates the machine and thus the machine can only be operated for limited processes. Furthermore, Au is not compatible with semiconductor fabrication. This is why we investigated alternative etching masks to replace Au. Below we will discuss different dry etching materials and introduce Ti as a good choise as a hard mask material. This process is than used to fabricate high quality GC molds, which will be used for precision glass molding. The quality of the replication will be analyzed by careful surface inspection. Thereafter, the molded DOEs will be optically characterized in terms of efficiency and uniformity of a beamsplitting array. For comparison to standard FS technology, we fabricated the same DOE design with FS. We will show that the precision glass molding can give a good optical performance and show the connection to the quality of the mold fabrication and replication quality.
2. Micro-structuring of GC
The surface structuring needs to fulfill the geometrical requirements for diffractive optical structures, which includes small feature size, very high sidewall angles (80-90°) and a high surface quality. The elements designed on the photolithography mask have critical dimensions of 2 μm and active areas with diameters of 7.5 mm and 17 mm. The binary design elements act as beamsplitters. We note that the feature sizes are relatively large for DOEs, but we used these designs as a starting point in order to establish a working process. Later work is focused on improving the resolution with e-beam lithography.
Polished 4” GC wafers from Nisshinbo with an excellent roughness Ra (rms) lower than 2 nm were used as substrate. GC wafers exhibit often surface defects like small holes. After careful inspection no surface defect could be found on the mentioned 4 inch wafers.
The process flow that we developed for the micro-structuring of GC is shown in Fig. 1. After a cleaning in Piranha the GC wafers are sputtered with a 250 nm thick Ti layer, which acts as a mask layer for the Reactive Ion etching (RIE). Then the wafer is spin-coated with a photoresist AZ ECI 3007 giving a thickness of ca. 1 μm. The exposure of the resist is done with the mask aligner Süss MJB4 in hard contact. Vacuum contact was not available, but would be preferred for an enhanced resolution. The opening of the Ti hard mask is done in a STS Multiplex. The process parameters for the inductively coupled plasma (ICP) RIE are: electrode frequency of 13.56 MHz, coil power of 800 W, platen power of 150 W, chamber pressure of 3 mTorr and gas flow rates of 10 sccm for Cl2 and BCl3. The etching of the GC is done in an ICP-RIE system (SPTS Advanced Plasma System). The process parameters are: electrode frequency of 13.56 MHz, coil power of 950 W, platen power of 100 W, chamber pressure of 37.5 mTorr and gas flow rates of 40 sccm for O2 and 10 sccm for SF6. The stripping of the Ti is done in a hydrofluoric acid bath at room temperature for 5 min. A final cleaning step is carried out in Piranha to remove residual photoresist.
The process results are summarized in Table 1. We included the results for a process where the photoresist is used as the etching mask for comparison. Although this approach is offering a shorter process flow, it suffers from a very limited etch selectivity between the GC and the resist: the resist is etched 5 times faster than the GC. This process would require very thick resist layers and will result in flat etch walls. Thus, a hard mask with the highest possible selectivity has to be used. Ti offers potentially a high selectivity. We realized a selectivity of 1:0.2 and demonstrate wall verticality between 80° and 85°. The final results of the micro-structured GC molds show very good quality. Figure 3(a)-3(b) illustrates these results by presenting SEM images of a fabricated GC molds with a depth of 690 nm.
3. Precision glass molding
To verify the usefulness and versatility of GC for molding we used the micro-structured GC wafers to mold a low Tg glass L-BAL42, which has a glass transition temperature of Tg = 506°C. Details of the precision glass molding process can be found for instance in [7,8]. Typical parameters are: a force of 0.8 kN, a molding temperature higher than the glass transition temperature at 550°C and a molding time of 180 s. The cycling time is between 15 and 25min. A microphotograph of a molded DOE is shown in Fig. 2. The glass preform has a diameter of 21 mm and the GC stamp had the form of a hexagon slightly visible in Fig. 2 by marks around the rim of the preform.
A scanning electron microscope (SEM) was used to perform an in depth quality analysis at the structure level. SEM images of the glass element G2 visible in left upper area of Fig. 2 are shown in Fig. 3. To allow closed up comparison, the GC mold and the replicated glass are given at the corresponding position. It can be seen that the features are replicated at good precision. The edges of the glass are slightly rounder than for the master. This means that the glass did not completely fill in the microstructures during the molding step. We found that from the molding parameters the temperature has the strongest influence. An increase in the temperature will decrease the viscosity of the glass, which becomes more liquid and flows easier into the microstructures. A higher pressure can improve to the same extend the molding result too, since it is working against the surface tension of the glass. A minimum time should be used in order for the glass to flow into the structures and limit the amount of time for one molding run. For longer times, we found no significant improvement.
For all molding experiments the GC molds could be separated from the glass without effort. No chemical binding was observed. This was expected owing to the excellent anti-adhesion properties of GC surfaces. It is thus not necessary to use additional coatings. To compare our approach to state-of-the art technology with coated molds, we fabricated FSmolds. E-beam lithography with RIE was used. In order to use FS masters for precision glass molding anti-adhesion coatings are required. The FS masters were sputtered with 15 nm thick Platinum-Iridium (PtIr) layer, the typical choice for such a case. SEM images of two FS masters after molding are given in Fig. 4. Figure 4(a) has the same design as the GC mold from Fig. 3, but shows no corner roundings due to the higher resolution of the employed e-beam lithography step for mold fabrication. Figure 4(b) depicts a mold were a severe amount of L-BAL42 is broken and sticks to the FS mold. In the higher magnification images it can be seen that a part of the PtIr coating is missing at the edges. It is thus more difficult to separate the FS mold and glass. The wear of the FS molds is very high. Only a few cycles are possible, usually less than 5 cycles. A thicker PtIr anti-adhesion coating would increase the lifetime of the mold, but the quality of the micro-structured mold would degrade due to the edge rounding.
In comparison, we found the GC more robust, offer a much longer mold lifetime and do not require a coating, which would increase the cost and degrade the quality.
4. Durability of GC molds
One of the main failure mechanisms of glass molding is the different thermal expansion behavior of glass and mold material during the cooling stage. This phenomena can be described by the difference in the coefficient of thermal expansion (CTE) between the mold material and the glass. To be more specific let us consider the linear CTE for the temperature range where it is known (100 – 300 °C) of L-BAL42, which is 8.8 10−6/K. Close to the glass transition temperature this parameter becomes highly nonlinear and reliable measurements are not available. Within the given temperature range where data are available (100 – 300 °C) GC has a 4 times lower CTE of only 2.2 10−6/K. Although data at molding temperature are not available, it can be safely stated that the glass will shrink more than the GC mold during the cooling stage of the precision glass molding process. The volume difference translates into a lateral shrinkage in the order of micrometers and causes stress between mold and glass that can result in extreme cases in the rapture and /or destruction of micro features. Smaller features, which are more fragile, are more prone to be altered or break. We found that at borders between microstructured areas and flat areas on the same wafer this phenomenon is more likely. Most probably this is caused by a stress built-up between unstructured areas, where the glass is sliding over the surface, and structured areas, where the glass is fixed by the mold microfeatures and cannot move laterally.
The difference in lateral shrinkage of glass and mold depends on the temperature range, for which the glass and the mold stay in contact. Thus, a higher molding temperature might on the one hand increase the replication accuracy, but on the other hand cause stress high enough to break features in glass and/or mold. It is also desirable to separate mold and glass early during the cooling stage. For experiments with the machine set-up described in [7,8], the glass and mold were separated manually after a temperature of around 100°C was reached and could be taken out of the machine. A better practice is to implement a controlled removing procedure.
The molding machine was therefore modified so that the GC mold could be fixed to the upper machine arm. Thus, it is possible to separate mold and glass at a much higher temperature than before. This set-up was used for lifetime tests. For most of our experiments we used the same mold between 1 and 10 times. This was done mostly to see the influence of individual molding parameters. In order to analyze the durability of the GC molds, we conducted 50 cycles with the same mold. The molding temperature was 555°C and the separation temperature was 450°C.
Optical measurements of the 6x6 beamsplitter showed that the performance quality degrades mainly for the first 3 replications. The Uniformity error UE, the efficiency of the 0th order and the efficiency of the design orders increased. The performance of replicas 3 to 50 stay constant. SEM images Fig. 5(a) showed that the mold was slightly damaged. Replica 1 shows no denegation depicted in Fig. 5(b). Whereas replica 2 shows the corresponding broken part of the mold shown in Fig. 5(c). We believe that the problem is caused by a lateral movement of the upper molding arm during demolding, causing the partial breakage of the GC mold features. After the first replications, the replicas did not show any alterations anymore. Replica 50 given in Fig. 5(d) shows the same features as replica 2. Even with the present problem, one can see that the molding was stable after the first few runs.
AFM measurements of the mold show that the roughness does not significantly change. The roughness Ra of the etched surface after mold fabrication was 2.3 nm and after 50 replications 2.6 nm. Also no change of the texture like holes could be seen. This shows that GC molds have the potential for multiple use of more than 50 replication cycles.
5. Optical performance
The quality of replication is determined by its final intended use and thus, the optical application. It is therefore necessary to test the performance of the optical function of the replicated elements. A typical diffraction pattern recorded with a CCD camera for illustrating the optical functionality is shown in Fig. 6(b). The diffraction pattern was projected on a white board. The DOE, of which SEM images are presented in Fig. 3, acts as an 11x11 beamsplitter. The used wavelength of 950 nm does not correspond to π phase step, which would correspond to a wavelength of 795 nm. Thus, we can see also the 0th order, which would be otherwise not visible in the image. The design of this DOE suppresses of all on-axis orders, which are horizontally and vertically aligned with the 0th order. We used the set-up depicted in Fig. 6(a) to qualitatively characterize the optical performance.
It is a goniometer set-up with a power meter. The collimated laser light is send through the DOE. The power meter measures the optical power in each diffraction order individually with very high dynamical range. It is mounted on an x-y stage and can scan all diffraction orders automatically. The power without the DOE is also measured and the power in the design orders is compared to this number. It is desired that the amount of light and thus the efficiency of the design orders is as high as possible. The uniformity error UE is a parameter specific for beamsplitting elements, where a uniform power distribution over all design orders is desired and the fabricated element is tested against this criteria. The UE is defined as the difference between the maximal power Pmax and the weakest diffraction order power Pmin as UE = (Pmax - Pmin)/(Pmax + Pmin). The diffraction pattern of our samples consists of spot arrays, a typical beamsplitting element that can be qualified with values for the efficiency of the design orders, the 0th order efficiency and the uniformity. These give a clear and significant judgement about the quality of the fabricated DOE.
The results of the GC molded DOE are summarized in Table 2. The test wavelength for the DOE molded with GC was λ = 750 nm, where the ideal wavelength corresponding to the depth would be 795 nm. This will cause deviation and a careful analysis is needed because of the relation between wavelength and ideal etch depth. Jahns et al. discussed numerically tolerances of etch depths and transition points for Dammann gratings based on the Fraunhofer approximation . For binary DOEs, an error in etch depth causes a wavelength to depth mismatch and thus an increase of the intensity in the 0th order. Jahns et al. showed also that the etch depth error will not influence the uniformity error, but uniformly decrease the power in all design orders. In our case, because of an etch depth error, the efficiency is decreased by ca. 2% and is in good agreement with the theoretical efficiency. The Uniformity error increase from 5.2% to 11.9% cannot be caused by the wavelength to depth mismatch. The design was calculated with Fraunhofer approximation, since rigorous calculation for this amount of diffraction orders would be too heavy. This means the UE of the actual DOE might be higher than the predicted value of 5.2%. For comparison we fabricated a FS DOE with e-beam lithography. The element showed a uniformity error of 8%, which can be seen as the best possible performance and lowest uniformity error possible in fabrication.
Two sources of error for the worsened performance of the molded DOE should be considered: the quality of the mold and the accuracy of the replication. The quality of the mold is already lower than the FS DOE, which means the FS DOE is closer to the design than the molded DOE (compare Fig. 3(a) for GC mold and 4(a) for FS mold). The GC mold is fabricated with standard mask photolithography. The corner rounding is introduced by diffraction effects during the resist exposure. The resolution of the FS molded DOEs is much higher due to the more accurate e-beam lithography. Jahns et al. showed that corner roundings and over- or underexposure can considerable increase the uniformity of beamsplitters. If the glass does not completely fill the mold during molding, the final glass DOE will have different phase step heights. We believe that both the mold quality and the replication accuracy caused the increased UE of 11.9%.
Although FS as a mold material can be micro-structured with a higher quality, the lifetime of the FS molds is too limited to be used for mass replication, which asks for process cycles of more than 1000. The resolution of the mold fabrication needs to be increased. This is why the use of higher resolution lithography techniques (e.g. e-beam lithography) for the micro-structuring of GC molds needs to be considered. High resolution GC molds would offer very good optical quality and high resistance during the glass molding.
GC stamps on wafer level were fabricated. Beamsplitting diffractive optical elements were used as test elements. We molded L-BAL42 at temperatures in the range of 550-565°C. The GC molds could be easily separated from the glass after molding and have the potential to be used for more than 50 units. The molding parameters such as temperature, pressure and time need careful optimization in order to reach an accurate replication and a low mold wear. Optical performance measurements of the molded DOEs are in good agreement with the theory, which qualifies GC as an excellent mold material. For the first time, we could reach efficiencies of GC molded DOE in the range of the theoretical predictions. The employed mask photolithography of the mold causes a resolution limitation, but can be overcome in the future by using higher resolution lithography like electron beam lithography. This would also offer the potential for smaller feature sizes and thus smaller periods for the mold. However, the limits of the molding given by the possible replicable aspect-ratios and feature sizes have still to be determined.
Commission for Technology and Innovation CTI (12824.1 PFNM-NM); Federal Ministry of Education and Research BMBF (FKZ 02PJ1234).
The authors thank the technical support from the Center of Micro-nanotechnology (CMI) at EPFL.
References and links
1. M. Hünten, F. Klocke, and O. Dambon, “Precision glass molding – An integrative approach for the production of high precision micro optics,” Proc. SPIE 7591, 75910X (2010). [CrossRef]
2. L. Mertus and A. Symmons, “Implications of Diamond-Turned vs. Diamond Ground Mold Fabrication Techniques on Precision-Molded Optics,” Proc. SPIE 8489, 848902 (2012). [CrossRef]
3. H. P. Herzig, Micro-Optics: Elements, Systems and Applications (Taylor & Francis, 1997).
4. B. C. Kress and P. Meyrueis, Applied Digital Optics, from Micro-Optics to Nanophotonics (John Wiley & Sons, 2009).
5. H. N. Hansen, R. J. Hocken, and G. Tosello, “Replication of micro and nano surface geometries,” CIRP Ann. – Manuf. Techn. 60(2), 695–714 (2011). [CrossRef]
7. A. Y. Yi, Y. Chen, F. Klocke, G. Pongs, A. Demmer, D. Grewell, and A. Benatar, “A high volume precision compression molding process of glass diffractive optics by use of a micromachined fused silica wafer mold and low Tg optical glass,” J. Micromech. Microeng. 16(10), 2000–2005 (2006). [CrossRef]
8. Y. Chen, A. Y. Yi, F. Klocke, G. Pongs, and A. Demmer, “Manufacturing of glass diffractive optics by use of molding process,” ASPE 2006, 395–398 (2006).
9. Y. Chen, L. Li, and A. Y. Yi, “Fabrication of precision 3D microstructures by use of a combination of ultraprecision diamond turning and reactive ion etching process,” J. Micromech. Microeng. 17(5), 883–890 (2007). [CrossRef]
10. T. Mori, N. Yamashita, H. Kasa, K. Fukumi, K. Kintaka, and J. Nishii, “Period sub-wavelength structures with large phase retardation fabrication by glass nanoimprint,” Jpn. J. Appl. Phys. 48, 06FH20 (2009).
11. T. Mori, K. Hasegawa, T. Hatano, H. Kasa, K. Kintaka, and J. Nishii, “Surface-relief gratings with high spatial frequency fabricated using direct glass imprinting process,” Opt. Lett. 33(5), 428–430 (2008). [CrossRef] [PubMed]
12. K. Yamada, M. Umetani, T. Tamura, Y. Tanaka, H. Kasa, and J. Nishii, “Antireflective structure imprinted on the surface of optical glass by SiC mold,” Appl. Surf. Sci. 255(7), 4267–4270 (2009). [CrossRef]
13. T. Tamura, M. Umetani, K. Yamada, Y. Tanaka, K. Kintaka, H. Kasa, and J. Nishii, “Fabrication of Antireflective Subwavelength Structure on Spherical Glass Surface Using Imprinting Process,” Appl. Phys. Express 3(11), 112501 (2010). [CrossRef]
14. T. Sasaki, M. Takahashi, R. Maeda, T. Tanaka, T. Maeno, and Z. Yang, “Microstructures formed on a low fluorescent glass using glassy carbon molding,” Conf. Proc. IEEE Eng. Med. Biol. Soc. 7, 7126–7128 (2005). [PubMed]
15. S.-W. Youn, A. Ueno, M. Takahashi, and R. Maeda, “A process of glassy carbon etching without the micro masking effect for the fabrication of a mold with a high-quality surface,” J. Micromech. Microeng. 19(12), 125010 (2009). [CrossRef]
16. J. Jahns, M. M. Downs, M. E. Prise, N. Streibi, and S. J. Walker, “Dammann gratings for laser beam shaping,” Opt. Eng. 28(12), 1267–1275 (1989). [CrossRef]