Lithography is critical to the continuous increase in device performance and cost reduction that drives the semiconductor industry. With an imaging wavelength of 13.5nm (∼92eV), extreme UV lithography (EUVL) is widely accepted as the next candidate for extending that paradigm to 22nm half-pitch (hp) technology and below. While several challenges remain before it can be generally adopted, EUVL has recently moved to the development stage. Dutch semiconductor manufacturer ASML shipped the first NXE:3100 EUV lithography system to a customer site in late 2010, is planning to deliver five additional NXE:3100s in 2011, and has received orders for nine of the systems for delivery in 2012.1
One primary concern for scaling the technology with current materials arises from the exposure mechanism. In EUVL, secondary electrons are ejected when high-energy radiation is absorbed by the photoresist. Photoacid generators are activated by these secondary electrons to provide an image. Consequently, the photoacid is not produced at the location where the primary EUV photon is absorbed but rather at some distance away that is determined by both the length over which the secondary electron(s) travel and the reactive properties of the generator. This so-called secondary electron blur (SEB) is estimated to be as large as 7nm and thus may limit the ultimate resolution of EUVL to ≥16nm hp.2,3
SEB has, however, proven difficult to quantify experimentally. The key challenge is to separate SEB from other components in chemically amplified resists (CARs) that give rise to the total blur. We have used 22 and 24nm hp EUV imaging data with varied PEB temperatures as our input to two resist models.4 In such models, all blur contributions that are not explicitly described in the image formation appear as acid diffusion. At high temperatures, where acid diffusion during the post-exposure bake (PEB) step overwhelms SEB, acid diffusion is expected to be the main contributor to blur. As this is a thermally activated process, it results in a negative slope for an Arrhenius plot. If, at low-temperature PEB, acid diffusion is reduced to the extent where it is about equal to nonthermally activated blur contributors (such as SEB), the Arrhenius behavior will flatten out and deviate from linearity. Our experimental data set allows for such Arrhenius analysis in an attempt to quantify SEB.
Figure 1. Arrhenius plot for rate constant of acid diffusion (k) as a function of post-exposure bake (PEB) temperature (RT). The points are extracted-fit parameters from the continuum resist model. The solid line is a best linear fit to these data points. ln: Natural logarithm.
Arrhenius analysis of the extracted model fit to the experimental data demonstrates a perfectly linear behavior with negative slope (see Figure 1) as would be expected for a thermally activated process. In the low-temperature regime of the plot (right-hand side) there is no evidence of the curve flattening as described above. While the data from Figure 1 was obtained with a PROLITH continuum model fit, similar results were obtained by fitting the PROLITH stochastic resist model to the data, where critical dimension as well as linewidth roughness (LWR) data were used as model input.
Figure 2. Acid-diffusion length as a function of PEB temperature as extracted from continuum (red) and stochastic (blue) resist models for both 22 and 24nm hp experimental data.
Acid-diffusion lengths that correspond to the rate constants of Figure 1 are plotted as a function of PEB temperature in Figure 2. The quantitative agreement between both modeling approaches is excellent, especially at low temperatures. The discrepancy at the highest PEB temperature is most likely caused by uncertainty in the experimental data set under these conditions. More important, both models result in an acid-diffusion length ∼4nm at the lowest PEB temperature. As this is a convolved blur of all blur contributions, SEB should be significantly lower. Our stochastic model allows explicit optimization of SEB to the experimental data. From a best fit, it is estimated that the maximum probability for acid generation occurs about 2.4nm from the photon absorption site. However, this blur length is so small compared to the dimensions that have been used as experimental input for the resist model generation that we believe it should be regarded as an upper limit.
We have applied the calibrated stochastic resist model to predict the EUVL-imaging capabilities that will be used to scale to 16nm hp. We show that exposure latitude and LWR are only moderately affected (see Figure 3) going toward these dimensions, as long as the acid diffusion can be controlled to <5nm.
Figure 3. Extrapolation of the model towards 16nm hp at 0.32NA (numerical aperture) demonstrates that both exposure latitude (EL) and linewidth roughness (LWR) are manageable at these dimensions.
In conclusion, this work shows that EUVL using a chemically amplified resist may be scaled beyond the 22nm hp node and should not be limited by SEB as dimensions move to 16nm hp (assuming that orthogonal failure modes such as pattern collapse and defects can also be addressed by this material set). To further refine the quantification of SEB, an exercise similar to the one described here using a higher-resolution EUV system than is currently available would be required. We plan to demonstrate the predicted sub-22nm hp patterning performance using the NXE:3100 that is currently being installed at IMEC.
Todd Younkin, Michael Leeson
Carlos Fonseca, Joshua Hooge
Tokyo Electron America
Tokyo Electron Kyushu
John Biafore, Mark Smith
2. T. Kozawa, T. Tagawa, Radiation chemistry in chemically amplified resists, Jpn J. Appl. Phys
. 49, pp. 030001, 2010. doi:10.1143/JJAP.49.030001
3. T. Kozawa, S. Tagawa, Theoretical study on difference between image quality formed in low- and high-activation-energy chemically amplified resists, Appl. Phys. Express 1, pp. 107001, 2008.
4. R. Gronheid, T. R. Younkin, M. J. Leeson, C. Fonseca, J. S. Hooge, K. Nafus, J. J. Biafore, M. D. Smith, EUV secondary electron blur at the 22nm half pitch node, Proc. SPIE
7969, pp. 796904, 2011. doi:10.1117/12.881427