Use of SU-8 Negative Photoresist for Optical Mask Manufacturing

Use of SU-8 Negative Photoresist for Optical Mask Manufacturing Alexei L. Bogdanov * MAX-Lab, University of Lund, SE-221 00, Lund, Sweden ABSTRACT The requirements for better control, linearity, and uniformity of critical dimension (CD) on photomasks in fabrication of 180 and 150 nm generation devices result in increasing demand for thinner, more etching durable, and more sensitive e- beam resists. Novolac based resists with chemical amplification have been a choice for their sensitivity and stability during etching. However, difficult CD control due to the acid catalyzer diffusion and quite narrow post exposure bake (PEB) process window are some of the major drawbacks of these resists. SU-8 is recently introduced to the market negative photoresist. High sensitivity, fairly good adhesion properties, and relatively simple processing of SU-8 make it a good substitution for novolac based chemically amplified negative e-beam resists in optical mask manufacturing. The replacement of traditional chemically amplified resists by SU-8 can increase the process latitude and reduce resist costs. Among the obvious drawbacks of SU-8 are the use of solvent-based developer and demand of oxygen plasma for resist removal. In this paper the use of SU-8 for optical mask manufacturing is reported. All steps of resist film preparation, exposure and development are paid a share of attention. Possibilities to use reactive ion etching (RIE) with oxygen in order to increase resist mask contrast are discussed. Special exposure strategy (pattern outlining) was employed to further improve the edge definition. The resist PEB temperature and time were studied to estimate their weight in overall CD control performance. Specially designed test patterns with 0.25 µm design rule could be firmly transferred into a chromium layer both by wet etching and ion milling. Influence of exposure dose variation on the pattern CD change was studied. Keywords: Negative e-beam resist, photomask, e-beam lithography, SU-8 1. INTRODUCTION Photomasks with ever smaller and better controlled linewidth are required for deep ultra-violet (DUV) lithography exploiting phase-shifting and optical proximity correction features. The reticles are known to consume a significant part of CD error budget in 0.25 µm DUV lithography. 1 DUV lithography resolution enhancements like phase shifting and OPC require even more reduction in the mask critical-dimension error. 2 E-beam resist is of the key components in the maskmaking. The quality and accuracy of placement of the resist mask are decisive for the final reticle CD error performance, regardless of the etching method in use for the pattern transfer. Positive resists have long been used for their high contrast and resolution, non-swelling development, and possibility to quench pattern dimensions on the development stage. On the other hand negative e-beam resists were always considered as low contrast and poor resolution ones, though with somewhat higher sensitivity. Development of chemically amplified resists (CAR) with increased efficiency of crosslinking or polarity change due to the catalytic reactions during PEB made the negative resists more attractive for submicrometer resolution lithography. However, the nature of the image formation in CARs implies certain mobility of the catalyst molecules produced by the exposure. A considerable diffusion rate of the photogenerated acid can degrade the latent image in the course of PEB. 3 Therefore, for high-sensitivity CARs parameters of the PEB process such as the * Correspondence E-mail: Alexei.Bogdanov@maxlab.lu.se; Web: http://www.maxlab.lu.se/beamlines/bld811; Phone: +4646 2228824; Fax: +46462224710

uniformity of heating, the temperature, duration of the PEB, and the timing of the lithographic process as a whole are crucial for the pattern edge definition and the CD control. The diffusion rate of the catalyst in the resist is mainly defined by two parameters: (1) molecular weight of the catalyst molecules (i.e. their size) and (2) viscosity of the base polymer, which is dramatically decreasing upon the heating of the polymer above its glassing transition point T g. To address the second issue epoxy functional groups have been added to low-molecular phenolic polymers to increase their ability to form the crosslinks. Photoresist SU-8 was first formulated at IBM as early as in 1982 4 and its first application in thick film photolithography was reported in 1995. 5 The resist isbased on an epoxy resin with high functionality (Fig. 1). Its main lithographic Figure 1. SU-8 (glycidyl-ether-bisphenol-a novolac) polymer structure (from Ref. 3) application was so far deep optical and x-ray 6, 7 exposures to produce high aspect ratio structures for MEMS and MOEMS applications where the resist is used in very thick films (5-500 µm). As it can be seen from its formula the polymer has quite low molecular weight (~7000) and thus when non-crosslinked can easily be dissolved by a number of solvents (e.g. propylene-glycol-methyl ether (PGME), gamma-butyrol-acetone (GBL), and methyl iso-butyl ketone). Each molecule of SU-8 has ~16 epoxy functional side groups providing a very dense three dimensional network of crosslinks when the resin is cured. A triarylium-sulfonium salt based photoinitiator is converted into an acid upon the exposure. During the PEB the acid molecules react with the epoxy side groups producing radicals attached to the backbone of SU-8 molecule. Upon a crosslinking act between two such radicals the acid molecule regenerates and can induce further polymerisation. T g of unexposed SU-8 is approximately 50 o C. Such a low T g value would have prohibited the use of PEB temperatures any higher than the room temperature in order to keep the linewidth undisturbed by the acid diffusion. T g value of the polymer begins to grow rapidly with the increase of the number of crosslinks thus dramatically decreasing the acid diffusion rate. For fully crosslinked SU-8 T g exceeds 200 o C. In this way the polymerisation process is contained in the areas of the resist where the initial acid concentration was exceeding certain threshold value. Therefore, SU-8 resist possesses a threshold exposure behaviour, which in other words means a high contrast and probably high resolution. The idea of the present work was to utilise SU-8 for mask making in order to replace more expensive Shipley s SAL-601 and SAL-605 negative e-beam resists. Developing the SU-8 process we tried to use the conditions similar to those for SAL resists. The main goal was to establish all the process parameters and to investigate practical resolution limits of mask making with SU-8.

2. RESIST PROCESSING 2.1 Resist spinning Both antireflection oxide coated and plain chromium mask blanks were tested. Prior to spinning the blanks were washed in deionized water and gently etched with O 2 plasma (Plasmatherm Batchtop RIE machine, 250 mtorr, 50 W, 15 s) to remove possible organic contaminants. We used SU-8 5 supplied by the Microlithography Chemical Co. with a solid content of 52% diluted in two parts of GBL. Spin curve for the resist is shown in Fig. 2. The spinning was performed in two steps. On the Figure 2. Spin curve of SU-8 5 diluted by 2 parts of GBL (1:2). first step the rotation speed was set to 250-300 rpm with total step time of 5 s. About 1.5 ml of resist was spilled in the beginning of the first step to coat a 4 x 4 mask. On the second step the spinner was quickly ramped up to 3000-4000 rpm and remained at that speed for about 2 minutes. Resist thickness we used was in the range of 140-160 nm 2.2 Softbake The soft bake was done at 90 o C both on a hotplate and in an oven. At this temperature the resist did not appear to be sensitive to the softbake time. However, the time had to be long enough to dry out the solvent. The drying rate was roughly estimated by observation of the film color change. Typical times of the softbake were: 3-5 min on the hotplate and 15-20 min in the oven. 2.3 Exposure dose and current Resist was exposed in JEOL JBX 5DII machine at the Swedish Nanometer Laboratory at Chalmers. The beam energy was 50 kev. Sensitivity curve for the resist made with flood exposure is shown in Fig. 3. Low (6 MHz) bandwidth of the JEOL s Figure 3. Contrast curve of SU-8 exposed with 50 kev electrons; PEB: 10 @90 o C hotplate; Development: 5 in GBL @23 o C.

scanner did not allow us to use currents higher than 100 pa for 25 nm distance between exposed pixels the maximum distance we used for resolution limit tests. For less demanding jobs writing currents of up to 5 na were used. 2.4 Post-exposure bake To be able to tell what temperature range could be used for PEB we conducted a simple experiment. 8 Undiluted SU-8 5 was spun on a silicone wafer to obtain 20-25 µm thick films. The film was then softbaked on a hotplate at 90 o C for about an hour to remove the solvent. After that resist samples without being exposed were postbaked for 10 min each on the hotplate at different temperatures. Eventually, the samples were etched in pure GBL and the time of complete dissolution was recorded. The plot of the cleaning time vs. PEB temperature is shown in Fig. 4. The onset of the beginning of the cleaning Figure 4. Cleaning time vs. PEB temperature for unexposed 25 µm thick SU-8. time rise was observed at approximately 133 o C. This experiment shows that the temperature of thermally initiated crosslinking lays above 130 o C and that it will be quite safe to undertake PEB at the temperatures below that point. Although PEB with elevated temperatures might give higher resist sensitivities for low resolution jobs, increased diffusion rate of the acid may result in poorer linewidth reproduction unacceptable for fine line lithography. Our PEB routine was the following: Exposed masks were placed into 90 o C oven on a pair of thin supporting NiCr wires to avoid direct contact with the hot parts of the oven and to ensure that the temperature of the plate would rise gradually. The smooth PEB temperature ramp provides suitable conditions for the mentioned in Section 1 mechanism of acid containing in crosslinked polymer matrix. In the case of PEB on the hotplate another technique was employed to achieve the smooth temperature ramping. The mask was put onto a massive metal spacer kept at room temperature before PEB. The spacer with the mask on it was then placed on the 90 o C hotplate. The monitoring of the hotplate temperature showed that the heat equilibrium conditions were settling in about 45 s after PEB start. The masks were kept on the hotplate for 10 minutes after the temperature had settled. Mask cooling after PEB did not seem to require any special precautions. 2.5 Development Both GBL and PGME were used as developers. Although for thick SU-8 films (thicker than 5 µm) used in photo- or x-ray lithography GBL produced better results, 8 for thin (< 200 nm) films the performance of both solvents appeared to be the same. All development work was done at 23 o C. Sufficient development times were 3-5 min, though development for 10-15

min did not produce any significant change in the resist masks appearance. After development the plates were rinsed in a fresh developer and dried with N 2 flush. 2.6 Plasma processing To achieve the best resolution the edge profiles of the resist masks were enhanced by reactive ion etching (RIE) process. The process is illustrated in Fig.5(a, b). Partially crosslinked resist remains after development near the exposed pattern edges forming some kind of a skirt of sparse polymer. In the case of dense pattern like in Fig. 5(a) these skirts overlap totally covering the spaces between the line features. If no exposure correction was used the edge offset resulting from the skirt formation could be as much as 0.2 µm. To get rid of the skirts RIE process using a mixture of Ar and O 2 (4:6) was developed. Ar was added to provide mechanical milling action to remove Cr oxide forming in O 2 plasma in open mask areas. Total gas pressure was 150 mtorr and RF power 50 W. For uncorrected exposure with high e-beam current the best results were obtained when resist films were thinned by 25-30% of their initial thickness. However, the amount of etching strongly depended on the e-beam sharpness and in the case of exposing with narrow low-current e-beam and a special writing strategy (see Section 3) only a very slight plasma ashing of the masks prior to the Cr etching was necessary. a 5 4 3 2 1 d Ar + b RIE c Cr etch Figure 5. Process of mask making with negative resist: a) exposed and developed resist film; b) thinned resist mask after RIE; c) mask after Cr wet etching. Another RIE step for resist removal is required; d) Ar + ion milling. Resist is removed in the process. RIE step is omitted. Legend: 1 - Glass substrate; 2-100 nm Cr layer; 3 - ideal resist profile; 4 - crosslinked resist; 5 - partly crosslinked resist. 2.7 Chromium etching Standard Balzers #4 yellow Cr etcher was diluted to prolong Cr cleaning time to 4-4.5 minutes. This was done to make the detection of the etching endpoint more accurate and thus to minimize etching undercut. In the course of the process the etcher was well stirred to ensure the homogeneity of etching. After the etching masks were rinsed with deionized water and an oxygen RIE resist stripping was finally performed.

2.8 Ion milling The Ar + ion milling was applied in order to eliminate undercut of Cr features resulting from the isotropic wet etching process (Fig. 5c). Oxford Instruments ion milling machine was used. The process parameters list is presented in Table 1. Table 1. Ion milling process parameters Background pressure Process pressure Energy of Ar + ions <1. 10-6 mbar 2. 10-4 mbar 500 V Ion current density 0.25 ma/cm 2 Ion beam angle of incidence at the mask 86 o The masks were rotated under slightly inclined ion beam to minimize building up of the fences produced by Cr redeposition onto the resist walls. The rate of Cr milling in our conditions was measured to be 60 Å/min. For SU-8 the figure was almost the same. As a result SU-8 masks with initial thickness of 100-110 nm were completely removed in the process of milling thus eliminating the need in resist striping. The main disadvantage of the method besides of its higher cost is the rounding of the Cr mask edges (Figures 5d and 9a) as the resist mask shape is partially transferred into Cr during the milling. 3. EXPOSURE STRATEGY 3.1 Enhancement of SU-8 resist edges The quality of resist mask edges is of crucial importance for CD control. Due to the proximity effect in e-beam exposure absorbed in the resist dose is slightly smaller near the boundaries of the exposed area. For negative resist that would result in reduced resist mask thickness and increased resist porosity at the mask edges. If wet etching of Cr is used the resulting Cr pattern has quite rough edges and CD reproducibility is poor. The problem may be solved by applying a proximity effect correction algorithm (PCA) that would iteratively calculate the exposure dose profile to produce an optimal distribution. However, even the most sophisticated and quick of the PCAs take tremendous amount of processing time, specially when the patterns to be corrected are irregular. Another drawback of using the automatic correction is a huge amount of exposure data produced by the PCAs, which has to be stored and than fed through the data-path of a lithography system. Nevertheless, with a moderate proximity effect and using an e-beam system with a sharp Gaussian electron beam with the beam diameter maximum allowable CD error, quite a trivial method of resist edge enhancement can be used. The mask features are first exposed with the dose Q A corresponding to the flood exposure sensitivity of the resist on a given substrate. That results in a slight underexposure at the edges and in small insulated features (Fig. 6). Then the edges are again exposed (outlined) using a single beam path exposure mode. The dose of additional line exposure Q L is high enough to produce a narrow line of the exposed resist even if drawn separated. As a result of such exposure the very edges of the resist mask will be well overexposed and thus the density of the crosslinks as well as the T g of the polymer at the edges (darker resist areas in Fig. 6) will be higher and the rate of the catalyst molecules diffusion lower than in the rest of the polymer. The situation may be regarded as an erection of fences around the exposed areas to prevent escape of the acid catalyst to unexposed parts of the resist during the post-exposure bake.

Resist mask profile without enhancement Additional exposure Dose / Resist thickness Q L Q A Standard dose profile H 0 25-50 nm Enhanced resist profile Distance x 0 x 1 x 2 Figure 6. Negative resist edge enhancement by outlining. H 0 is nominal resist thickness and x 0,..x 2 are positions of exposed area boundaries. 3.2 Test mask design The test mask design is shown in Figure 7. Apart of the wedge-like features the mask pattern is an assembly of elementary squares with the size D the same size as the step of the grid the squares are placed on. Therefore, D represents characteristic dimension or design rule of the mask. In our experiments with SU-8 we used test masks with D = 0.05, 0.1, 0.25, 0.5, and 1 µm. The features on the mask are designed so that any failures in the lithographic process such as inappropriate exposure 20. D Figure 7. Test mask design. D is characteristic dimension of the mask

dose, poor resolution of the tool, uneven exposure due to the proximity effect, erroneous etching parameters, and others will be revealed observing resulting masks. The wedge structures in the center of the mask and below can be used for prompt estimation of ultimate resolution of the lithography. The optimal process conditions can be checked by comparing the lengths of the vertical bright and horizontal dark wedges. The closer are the lengths the better is the process balance. For more accurate estimation checking should be done on the masks with different values of D. 4. RESULTS Two sets of samples were produced. The firs set (Figures 8 and 9) was exposed with 50 kev electrons with doses ranging from 1 µc/cm 2 to 8 µc/cm 2. Moderate edge enhancement exposures were applied. This allowed to find exposure conditions at which test masks with D = 0.25 µm could have been reliably reproduced in 100 nm thick resist (Fig. 8) with quite a low dose (< 2 µc/cm 2 ) *. Unfortunately, resist masks were too porous to be used in wet etching. Instead Ar + milling was applied as described in the Section 2. During the milling resist was almost completely removed as it is shown in Fig. 5d. The resulting Cr structures are presented in Figure 9. Obviously, masks with D 0.25 µm came out well. Very small size (~100nm) isolated dots can be seen in the top-right of Fig. 9b., though as a whole, the mask with D = 0.1 µm did not look well already after the lithography step. The micrograph of 45 o tilted mask in Figure 9a gives some impression of the shape of Cr mask profile. The side walls seem to be almost vertical and the tops of the metal lines are rounded. The second set of masks was produced with strong outlining (Section 3, Fig. 6), RIE profile enhancement, and wet etching of Cr (Section 2, Fig. 5 a, b, c). The results are presented in Figures 10 and 11. The exposure doses between 3 and 3.5µC/cm 2 with approximately 0.1 nc/cm outlining dose were found to be optimal. As with the ion milling the structures with D 0.25 µm were of sufficient quality. The maximum edge roughness was estimated as 30-40 nm. Structures with D = 0.25 µm exposed to different doses varying ±20% from the optimum were examined in order to estimate the process stability against exposure dose variation. It was found that the linewidth changed for approximately 70 nm when exposure dose changed from Q 0-25% to Q 0 +21% (Fig. 11). It was also noticed that overexposure in excess of ~ 30% did not result in a dramatic increase of linewidth but rather produced a harder "skirt" more difficult to remove by RIE. 5. CONCLUSION SU-8 resist was tested for optical mask making. The process conditions were similar to those used with other chemically amplified negative e-beam resists. Reliable reproduction of structures with minimum feature size of 0.25 mm was possible using both Ar + ion milling and wet Cr etching. E-beam exposure was organized in such a way that the edges of the resist mask were exposed to much higher dose than the rest of the resist. Such exposure profile helps to contain acid catalyst molecules in the exposed areas increasing the edge contrast. A RIE process was introduced to remove partially crosslinked resist from the areas adjacent to the resist edges. The linewidth stability against the exposure dose variation was shown to be very good. The resist processing appears to be quite easy and the process window is wide. So if one can make it up with the need to use a solvent developer and plasma striping, SU-8 can be a negative resist of choice for practical mask making with e-beam. * Compare with 7-12 µc/cm 2 for 50 kv - the dose we use in mask-making with SAL-601.

a D = 0.05 µm a D = 0.25 µm b D = 0.1 µm b D = 0.1 µm c D =0.25 µm c D =0.25 µm d D = 0.5 µm d D = 0.5 µm e D = 1 µm Figure 9. Cr masks produced by ion milling of the structures shown in Fig.8. e D = 1 µm Figure 8. SEM photos of SU-8 structures on Cr surface. D is the design rule for the test masks. QA = 1.6 µc/cm2, QL = 0.008 nc/cm, I = 100 pa

a D = 0.1 µm a Q0 25% b D = 0.25 µm b Q0 10% c D =0.5 µm c Q0 +10% d D = 1 µm d Q0 +21% Figure 10. Cr structures produced by wet etching with RIE pre-shaping of the resist. * Figure 11. Fragment of wet etched Cr mask with D = 0.25 µm. Exposed with various doses. Q0 = (3.2µC/cm2, 0.1 nc/cm)*. The difference between the linewidths in (a) and (d) is ~70 nm The first of the two doses is the area exposure dose and the second is the line exposure dose used for the edge outlining

ACKNOWLEDGEMENTS The work was conducted as a part of TFR (Swedish National Council for Engineering Sciences) contract. Author is deeply grateful to the staff of the Swedish Nanometer Laboratory at Chalmers University of Technology in Göteborg for having opportunity to use the lab equipment. SEM time for inspection was kindly provided by the Nanometer Laboratory at the Department of Solid State Physics of Lund University. REFERENCES 1. Kuijten, J. P.; Duray, F.; Kinderen T. D., "Analysis of Reticle Contributions to CD Uniformity for 0.25 mm DUV Lithography," Proc. SPIE-Int. Soc. Eng. 1998, 3334, 620-628. 2. Wong, A. F.; Ferguson, R.A.; Liebmann, L. W.; Mansfield, S. M.; Molless, A. F.; Neisser, M. O., "Lithographic Effects of Mask Critical Dimension Error," Proc. SPIE-Int. Soc. Eng. 1998, 3334, 106-116. 3. J. M. Shaw, J. D. Gelorme, N. C. LaBianca, W. E. Conley, and S. J. Holmes, Negative photoresists for optical lithography, IBM Journal of Research and Development, 41(1997), pp. 81-94 4. H. Ito and C. G. Willson, Proceedings of the SPE Regional Technical Conference on Photopolymers, Society of Plastics Engineers, November 1982, p. 331. 5. N. LaBianca, and J. Gelorme, High aspect ratio resist for thick film applications, in Proc. SPIE vol. 2438, SPIE, (1995), pp.. 846-852 6. N. C. LaBianca and J. D. Gelorme, "High-Aspect-Ratio Resist for Thick Film Application," SPE Conference, Society of Plastics Engineers, November 1994. 7. A. L. Bogdanov and S. S. Peredkov, Use of SU-8 photoresist for very high aspect ratio x-ray lithography, Proceedings of Micro-, Nano-engineering 99, Rome (1999)