Stochastic defects continue to draw attention in the area of EUV lithography. It is now widely recognized that stochastic issues not only come from photon shot noise due to low (absorbed) EUV photon density, but also the resist material and process factors [1-4].
It stands to reason that resist absorption of EUV light, which is depth-dependent, will also have an important bearing on the occurrence of stochastic defects. EUV resists have initially been mainly chemically amplified, with a typical absorption coefficient of 5/um [5], meaning that exp(-5) ~ 0.67% represents the fraction of light that is transmitted by a 1 um thick layer of the resist. For a layer only 20 nm thick, on the other hand, 90% of the light is transmitted, meaning that only 10% of the light is absorbed. For a 40 nm thick resist layer, it would be interesting to compare the absorption in the top half vs. the bottom half (Figure 1).
Figure 1. EUV photon absorption in the top half (left) vs. the bottom half (right) of a chemically amplified resist. The threshold to print here is taken to be 24 photons absorbed per 2 nm x 2 nm pixel. The assumed dose (averaged over the displayed 80 nm x 80 nm area) is 60 mJ/cm2. The oval outline is for reference to visually assist observing the stochastic absorption profile.
The bottom half receives less light (90%) than the top because some light has already been absorbed, and it also absorbs a small fraction (10%) itself. Consequently, it is more likely for some areas in the bottom half of the resist layer to fall below the nominal threshold photon absorption density to print. This leads to the higher probability of stochastic defects forming from underexposure in the lower half of the resist.
Given that lower resist absorption aggravates stochastic effects, we would expect the higher absorption of metal oxide resists [5] to be much better. From Figure 2, we see something significantly different.
Figure 2. EUV photon absorption in the top half (left) vs. the bottom half (right) of a metal oxide resist. The threshold to print here is taken to be 24 photons absorbed per 2 nm x 2 nm pixel. The assumed dose (averaged over the displayed 80 nm x 80 nm area) is 60 mJ/cm2. The oval outline is for reference to visually assist observing the stochastic absorption profile.
The lower half of the resist layer receives significantly less light to begin with, so the absorbed photons ultimately define a smaller region at the bottom of the resist. Since the metal oxide resist is a negative-tone resist, the photon absorption determines where the resist remains after development. This means toppling of resist features (or missing resist features) from a narrower bottom than top (‘undercutting’) can be a stochastic defect specific to negative-tone resists, particularly metal oxide resists.
DUV photoresists would never have this problem in common use, even with a lower dose and lower absorption coefficient on the order of 1/um (Figure 3). It’s because the resist is thicker and the pixels are effectively larger.
Figure 3. ArF photon absorption in the top half (left) vs. the bottom half (right) of a chemically amplified resist. The threshold to print here is taken to be 1200 photons absorbed per 7 nm x 7 nm pixel. The assumed dose (averaged over the displayed 280 nm x 280 nm area) is 30 mJ/cm2. The oval outline is for reference to visually assist observing the absorption profile.
A fairer EUV vs ArF comparison would use realistically difficult scenarios. Figure 4 compares the photons per pixel absorbed in the bottom half of the resist for 40 nm square pitch EUV (0.5 nm x 0.5 nm pixel) with 40 nm metal oxide resist thickness vs. 80 nm square pitch ArF (1 nm x 1 nm pixel) with 100 nm chemically amplified resist thickness, using a negative tone hole pattern. The EUV image assumes an ideal binary mask with quadrupole illumination, while the ArF image assumes a 6% attenuated phase shift mask with cross dipole illumination.
Figure 4. EUV (left) vs. ArF (right) photon absorption in the bottom half of the resist. The EUV resist thickness is 40 nm, while the ArF resist thickness is 100 nm. The pixel size is 0.5 nm x 0.5 nm for the EUV case, and 1 nm x 1 nm for the ArF case. The threshold to print here is taken to be 1.6 photons/pixel for EUV and 3.3 photons/pixel for ArF, to target the half-pitch. The assumed dose (averaged over the pitch) is 60 mJ/cm2.
At the finer pixel size, the roughness of the edge becomes more apparent for both wavelengths. However, the EUV case has lots of spots in the background where the exposure is sub-threshold, which will lead to potential resist removal during development, whereas the ArF case is free from such spots. The reason for this, in fact, has to do with the higher contrast of phase-shift masks compared to binary masks. The bright background has closer intensity to the central dark spot in the binary case, or less contrast, giving more opportunity for the background noise variation to reach levels comparable to the dark spot.
Acid generation (in chemically amplified resists) and electron release (following EUV exposure) lead to smoothing effects, which are simulated here using 4x Gaussian smoothing (sigma=2 pixels). This leads to an effective resist blur of 2 times the pixel size, i.e., 2 nm for the EUV case and 4 nm for the ArF case.
Figure 5. EUV (left) vs. ArF (right) latent image in the bottom half of the resist, after 4x Gaussian smoothing (sigma=2 pixels). The EUV resist thickness is 40 nm, while the ArF resist thickness is 100 nm. The pixel size is 0.5 nm x 0.5 nm for the EUV case, and 1 nm x 1 nm for the ArF case. The threshold to print here is taken to target the half-pitch. The assumed dose (averaged over the pitch) is 60 mJ/cm2.
With smoothing, the general spottiness of the image is removed, leaving residual edge roughness, for both ArF and EUV cases (Figure 5). However, since the EUV case had more random background counts, the edge looks relatively rougher, and there is a tendency for defects occurring at and near the edge. These can also impact the effective edge placement.
To conclude, the higher absorption coefficient is not helping to avoid stochastic defects which are occurring near the bottom of the resist layer, so the higher dose to compensate lower absorption is still necessary.
References
[1] https://www.jstage.jst.go.jp/article/photopolymer/31/5/31_651/_pdf
[2]http://ww.lithoguru.com/scientist/litho_papers/2019_Metrics%20for%20stochastic%20scaling%20in%20EUV.pdf
[3] https://www.spiedigitallibrary.org/journals/journal-of-micro-nanopatterning-materials-and-metrology/volume-20/issue-01/014603/Contribution-of-EUV-resist-counting-statistics-to-stochastic-printing-failures/10.1117/1.JMM.20.1.014603.full?SSO=1
[4]http://euvlsymposium.lbl.gov/pdf/2014/6b1e6ae745cd40aba5940af61c0c908e.pdf
[5] http://euvlsymposium.lbl.gov/pdf/2015/Posters/P-RE-06_Fallica.pdf
This article was first published in LinkedIn Pulse: EUV Resist Absorption Impact on Stochastic Defects
Also read:
Pattern Shifts Induced by Dipole-Illuminated EUV Masks
Share this post via:
TSMC Unveils the World’s Most Advanced Logic Technology at IEDM