Highly spatially resolved chemical metrology on latent resist images

6 Patterning photoresist with extreme control over dose and placement is the first crucial step in 7 semiconductor manufacturing. However, how can the activation of modern complex resist 8 components be accurately measured at sufficient spatial resolution? No exposed nanometre-scale 9 resist pattern is sufficiently sturdy to unaltered withstand inspection by intense photon or electron 10 beams, not even after processing and development. 11 This paper presents experimental proof that infrared atomic force microscopy (IR-AFM) is sufficiently 12 sensitive and gentle to chemically record vulnerable yet valuable lithographic patterns in a 13 chemically amplified resist after exposure prior to development. Accordingly, IR-AFM metrology 14 provides long-sought insights into changes in the chemical and spatial distribution per component in 15 a latent resist image, both directly after exposure and during processing. With these to-be-gained 16 understandings, a disruptive acceleration of resist design and processing is expected.


18
To support the global trend of further shrinkage of transistors, satisfactory pattern fidelity in the 19 photoresist at the nanometre scale is needed (1) (2). After lithographic patterning, resist processing 20 is the next crucial step in semiconductor manufacturing. By processing, the vulnerable latent pattern 21 is translated, via a more sturdy form in the developed resist, into the target wafer by e.g. etching or 22 ion implantation (3). Resist processing requires extremely sophisticated know-how and control to 23 optimally and reproducibly transfer (feature size) or suppress (variation of feature size), especially 24 the high-spatial frequencies from the latent pattern via processed resist into the substrate. Since the 25 economic value of lithography can only be harvested after defect-free pattern transfer into silicon, 26 tight control over resist processing is of key importance. Recently, metrology by AFM has been 27 demonstrated for 16 nm half-pitch lines on processed and developed resist (4) and 26 nm posts (5). 28 Ideally, similar metrology would be made available to quantify the pattern development per step, 29 i.e. after postexposure bake, develop and rinse, thus providing insight into which step is the most 30 critical bottleneck (6). However, currently, no commercial method exists that can resolve control 31 parameters such as critical dimension (CD) or line edge roughness (LER) in latent or partially 32 processed resist patterns. This paper introduces a metrology method that provides insight into 33 changes in photoresist during processing, which hitherto were not observable at the nanometre 34 scale, without inflicting damage to or shrinkage of the exposed resist. 35 The semiconductor industry has developed a very sophisticated infrastructure and way of working to 36 establish high-volume manufacturing of semiconductor circuits with deep submicron dimensions, as 37 reported by, e.g., (7), (8) (9) (10). These sources all emphasize that LER control is critical for 38 manufacturing, as excessive LER correlates with yield loss (11). Such will amplify for sub-10 nm 39 patterning, as the industry foresees the need for control over the stochastics of the lithographic 40 (EUV) image as well as the granular components of photoresist (12 identified and quantified the impact of (blur by, e.g., shot noise in) the exposure tool image on the Z-51 factor (32) (33) (34) (35) (36) (37) (38), which led to the introduction of k 4 as the latest resist metric 52 (39). Following the proposed protocol, it is now possible to separate the contribution to the 53 nonuniformity by shot noise in the exposed image on the one hand from the stochastic variations in 54 the resist and processing on the other hand. Nevertheless, accurate metrology for resolution and 55 uniformity is required for a reliable evaluation of both the Z-and k 4 -factors. 56 As outlined in a recent LER metrology review paper by Cutler et al., there is additional unique 57 information on a photoresist's qualities disclosed in the power spectral density (PSD) of all spatial 58 frequencies that contribute to the total LER (40). Thus, if a line pattern is properly imaged, measured 59 and analysed, the obtained metrics for the LER PSD can be used to accelerate the development of 60 both high-performance photoresists and resist processing and metrology equipment, recipes and 61 protocols. However, at the moment, LER can only be measured after resist processing. This renders 62 it impossible to identify the LER contribution by a specific resist component or per processing step. 63 For the same reason, one cannot investigate the impact of exposure tool parameters on these 64 individual contributions. Hence, process development and control would benefit tremendously from 65 a metrology tool that can resolve (some of) these blur factors during processing and before 66 development. This paper shows the first experimental results for such a disruptive metrology 67 method by deploying infrared atomic force microscopy (IR-AFM) (41). In IR-AFM, an IR beam is tuned 68 to a resonance specific to an activated resist component, while a scanning probe records the locally 69 induced force field in noncontact mode. This unlocks the practical evaluation of the two geometrical 70 factors R and L in Gallatin's R-L-S resist metric in latent images, thereby enabling the tracking of a 71 resist's performance between exposure and development. We believe that in the near future, IR-72 AFM will record chemical maps of the latent image per resist component after (during) each 73 processing step that directly reflect or even resolve these stochastics to resolve where patterning 74 resolution and uniformity are lost most. 75

76
Topographical versus chemical contrast using AFM versus IR-AFM imaging 77 To perform metrology on latent patterns in cured photoresist, images were recorded by regular AFM 78 as well as with infrared-atomic force microscopy (IR-AFM) (42). To serve metrology, images shall 79 have both a sufficient signal-to-noise (S/N) and spatial resolution (21). Figure 1 shows AFM and IR-80 AFM images of a pattern of a 500 nm wide line at 2000 nm pitch as written by electron beam 81 lithography. The 200 nm thick chemically amplified resist layer has been cured by a post-expose 82 bake but has not been developed, yielding a relatively flat sample that is easy to scan by AFM and IR-83 AFM. As is to be expected from undeveloped resist, the S/N in the topography image by regular AFM 84 is too low to extract a clear waveform, let alone to sustain metrology (see, e.g., the left panel of 85 Figure 1). In addition, the observed pattern is hard to interpret. Resist shrinkage is inherently 86 nonlocal in effect and nonlinear, and the resulting patterns are not straightforwardly linked to the 87 actual lateral distribution of chemical changes in the resist. In IR-AFM, to enhance the chemical 88 contrast in the sample, a narrow-band IR beam is tuned to an IR absorption resonance that is specific 89 to exposed (and cured) resist only, while a nanometre-sized tip scans the sample with a spatial 90 resolution of at least 10-20 nm, or better (42) (43) (44) (45) (46). The IR-AFM image (centre panel of 91 Figure 1) shows a remarkable increase in S/N that has a direct link to the exposed pattern when 92 compared to regular AFM. As a result, edge detection in the IR-AFM image is straightforward (right 93 panel of Figure 1) but nearly impossible from the regular AFM image. 94 95 96 104 by AFM for a practical range of peak force settings (28). In our application of IR-AFM, a noncontact 145 mode has been selected that is known to be even gentler on the sample surface than the peak-force 146 tapping mode. 147 In summary, IR-AFM metrology on latent resist patterns has the potential to provide pivotal 148 information to develop and optimise resist composition and processing, as it may resolve the 149 development of resolution and roughness without altering the resist by charging, contamination, 150 shrinkage or heat (41). 151

152
A chemically amplified resist was activated by electron beam lithography and then cured by a post-153 expose bake. To find infrared fingerprints, two sets of experiments were conducted. First, the 154 presence of any IR fingerprint was established by FTIR. Then, using the strongest line in the IR 155 fingerprint, the pattern resist was imaged by IR-AFM at sufficiently high spatial resolution for 156 metrology of pitch, line width and edge roughness. The raw data were purified by rejection of 157 statistically significant outliers in edge positions. By the purification of the raw IR-AFM images, the 158 artifacts in their power spectral density (PSD) of both LER and LWR have disappeared, and the 159 corresponding LER/LWR metrology now complies with SEMI Standard P47-0307 (51). 160 Sample preparation by electron beam lithography and post-exposure bake 161 A 4" silicon XX flat wafer was precleaned in an acetone bath for 5 minutes and, after rinsing in 162 isopropyl alcohol, spun dry at 2000 rpm and dried further by a 10-minute bake on a hot plate at 163 150°C. An HDMS primer was applied to the wafer shortly prior to spinning a layer of 200 nm thick 164 NEB 22A (Sumitomo Chemical Co.) chemically amplified e-beam resist at 6000 rpm (52). To remove 165 the solvent, the wafer was baked at 110 °C for 2 minutes. In a Raith EBPG 5200, 4 mm wide squares 166 and 500 nm lines at 2 micron spacing were exposed at areal doses of 40 and 80 µC cm -2 . The EBPG 167 raster pixel size was set to 120 nm, and a spot size of 160 nm was selected, warranting continuous 168 exposure of the patterns by the minor overfill of each pixel. To cure the resist, the wafer was baked 169 at 105 °C for 2 minutes. The wafers with the cured resist were stored at room temperature in a 170 closed box until either large-area analysis by FTIR or high-resolution imaging by IR-AFM. Figure 2  171 shows a cross-sectional schematic of the line pattern in the resist. 172

176
Scout for chemical fingerprint of resist activation  Figure 3 shows FTIR results recorded in transmission mode for cases II and III. The 185 difference between the spectra before and after curing exposure is relatively small because the 186 resist layer is rather thin (150 nm) but significant. The most prominent differences occur at ~985, 187 1030, 1175 and 1220 cm -1 , with a typical width of tens of cm -1 . The presence of the IR peaks in the 188 large exposed and cured areas provided experimental evidence that chemical changes in thin layers 189 of photoresist can indeed be noticed in IR spectroscopy, albeit not yet at high spatial resolution, as is 190 required for metrology on undeveloped resist images. 191 IR-AFM to resolve chemical contrast at the nanometre scale 192 The sample, as prepared by e-beam lithography, was examined by IR-AFM. A commercially available 193 system, Vista One by Molecular Vista with a Block Engineering QCL tuneable mid-IR laser, was used 194 with an NCH-Au cantilever with a tip radius of 20 nm. For topography imaging, the second cantilever 195 mode was selected with an approximately 1 nm free air amplitude and a setpoint of 75%. The 196 written pattern could be recognized from the topography images (figure 1) due to the well-known 197 effect of resist shrinkage (55). This pattern allowed us to locate positions to capture IR spectra from 198 exposed and unexposed areas of the same sample. For the IR-AFM operational conditions, we 199 applied standard settings as suggested by the manufacturer (42). For chemical contrast imaging by IR 200 excitation, the difference frequency between the first and second modes was used. Frequency 201 mixing then generates a signal in the first cantilever mode. The IR laser power was set to 20%, the 202 scan speed was 0.25 lines s -1 and the images were 512x512 pixels at a field of view of 10x10 microns. 203 Figure 3 shows how large area FTIR spectra compare to single-pixel IR-AFM spectra. Generally, there 204 is good agreement between the IR-AFM and FTIR spectra with reproduction of the same absorption 205 peaks in both methods. At several frequencies, there are clear IR absorption differences between 206 exposed and unexposed resists. Although IR-AFM is mainly surface sensitive, it appears that there is 207 a contribution of the SiO 2 absorption peak present in the data. For imaging in IR-AFM, we selected 208 the wavelength of 984 cm -1 (indicated by the black line in Fig. 3), as the corresponding signal 209 enhancement is not influenced by the SiO 2 absorption peaks at approximately 1100-1200 cm -1 and 210 therefore provides the best contrast. 211 212 Figure 3 Measured infrared spectra of the (un)coated and exposed-and-cured NEB 22 resist on a Si wafer. Left

215
It is appropriate to have a short look at the comparison between topography and IR-AFM operation. 216 Although topography imaging (left panel of Figure 1) reveals hints of the printed pattern, it has poor 217 contrast, and the AFM image is not directly/easily related to the exposed pattern (55). The latter is 218 attributed to nonlocal changes in topography by resist shrinkage and/or local resist stiffness 219 changes, which fundamentally complicate accurate edge placement metrology from these data. 220 Instead, the IR-AFM image is directly related to the written pattern through the induced chemical 221 changes. Figure 4 shows the distribution of AFM and IR-AFM signal strength after classification of 222 each pixel as either "exposed" or "nonexposed" resist based on the detected edges (right panel of 223 Figure 1). Clearly, the topography contrast is only a minor fraction of the distribution widths, while 224 chemical contrast by IR-AFM is a significant fraction of the distribution widths. This confirms that IR-225 AFM provides a reliable signal that, in addition to delivering straightforward-to-interpret images of 226 latent resist patterns, supports quantitative analysis, e.g., LER metrology. 227 228 Figure 4 Bottom: histograms of pixel values for exposed (green) and unexposed (red) areas. The 0,16 nm average height 229 difference is just a fraction of the distribution width (σ AFM = 0,23 nm). The 21,7 µV difference in IR-AFM signal strength 230 between exposed and unexposed areas is approximately twice the distribution width (σ IR-AFM = 11,7 µV).

231
Waveform extraction, edge detection and data purification 232 To extract the line edge and line width roughness from the IR-AFM images, line edge positions need 233 to be evaluated over a length of at least 2 micrometres (21). To this end, after image processing to 234 purify the IR-AFM images from instrument artifacts, the average waveform was determined by 235 fitting a raised-cosine function to each edge in the average line profile. Figure 5  of Figure 6) showed unphysical features, e.g., the LWR at lower spatial frequencies was less than the 247 corresponding LER. Analysis of the extracted edge profiles yielded the insight that ~5% of isolated 248 edge positions have been shifted by more than 2σ from neighbouring lines. Such shifts can be 249 considered unphysical; hence, we purified the edge profiles by replacing these outliers by the 250 average edge position of the neighbouring lines (right panel of Figure 6)

267
The purified edge positions displayed in Figure 1 were analysed for LER and LWR in three 268 independent ways, yielding a highly similar outcome, as reported in Table 1 and Table 2. First, the 269 RMS value from the average edge position was calculated. Second, the PSD of the edge positions has 270 been integrated after (25). Third, the PSDs of the LER and LWR from both areal doses were fitted 271 after (40); see Table 2. For the latter, the obtained IR-AFM LER as calculated by Eq. (1) in (40) differs 272 by a factor of 2 from the values reported in Table 1. This difference is attributed to the arbitrary 273 choice of evaluating the PSD(f) with or without negative lateral frequencies, which obviously 274 changes the energy per frequency. 275