(573e) Mesoscale Kinetic Monte Carlo Simulations of Molecular Glass Photoresists

The shrinking of feature size in microelectronics is the primary driving force for smaller, faster, more powerful, more energy efficient computers and electronics. The critical enabling factor for the reduction in feature size is advances in photolithography, the process wherein well-defined areas are patterned onto a semiconductor substrate. Current state-of-the-art semiconductor manufacturing prints features as small as 45 nm with a 3ó deviation in line edge (line edge roughness) of around 5-6 nm. While might appear to be a small value for line edge roughness (LER), it is significantly larger than the desired 8% of the line width. Since LER does not scale with feature size, this value is completely unacceptable for future generations of semiconductors with features of 32 nm and below. To obtain acceptable values of LER, the fundamental causes of its formation during processing must be understood and mitigation strategies developed. Since it is difficult to experimentally probe resist dynamics at this length scale, modeling provides a way to investigate the root cause of LER formation.

Deterministic modeling has long been used to describe resist physics and simulate resist behavior. This has allowed for much faster process development and better prediction of the effect of exposure tool and process changes on lithographic process performance. Unfortunately, as feature sizes shrink, the important length scales rapidly approach that of individual resist molecules, and the behavior of individual molecules is decidedly non-deterministic. For example, the random-walk diffusion and reaction of single photoacids may have a large effect on the final resist profile, especially in terms of line edge roughness (LER). Since the current dominant length scale for LER is in the range of 3-10 nm, the most efficient model would be grained small enough to capture effects of this length scale, but not too significantly smaller such that the computational burden is prohibitive (e.g. atomistic modeling). As a result, our model is a mesoscale model that has a grain size of 1 nm, corresponding roughly to that of an individual molecular glass resist. Our current model is a full kinetic Monte Carlo simulation of all pertinent resist physics, including PAG dispersion, photoacid generation kinetics, acid and base diffusion, acid catalyzed deprotection reactions, and acid/base neutralization reactions. This allows for investigation of a large number of different resist formulations, properties, and conditions (e.g. low and high activation energy resists, diffusion and reaction limited processing conditions, high acid/low base diffusion coefficients, low acid/high base diffusion coefficients, and fast and slow acid/base neutralization). These parameters are all being systematically investigated to determine the possible underlying statistical causes for LER and thus also provide solutions or mitigation strategies.