Analysis of mitigating factors for line edge roughness generated during electron beam lithography using machine learning

Abstract

Machine learning has been reported to be useful for the analysis of the trade-off relationships among major properties of chemically amplified extreme ultraviolet resists. The resist materials and processes used in photomask production using electron beam lithography seem similar. However, they involve distinct processes and factors. As one of the critical issues in resist pattern formation, line edge roughness (LER) was investigated using machine learning based on six variables, namely, sensitivity, half-pitch, exposure pattern width, beam blur, and the concentrations of photoacid generator and photodecomposable quenchers in terms of a chemical gradient (an indicator of LER). The relationship between the chemical gradient and these six variables was well formulated using the 5th degree polynomials of these six variables. The coefficients of feature values indicated that the process blur is a relatively more important factor than the beam blur in the 1.95–3.70 nm range in 11–16 nm half-pitch patterning.