Empirical Analysis of Electron Beam Lithography Optimization Models from a Pragmatic Perspective

Abstract

Electron Beam (EB) lithography is a process of focussing electron beams on silicon wafers to design different integrated circuits (ICs). It uses an electron gun, a blanking electrode, multiple electron lenses, a deflection electrode, and control circuits for each of these components. But the lithography process causes critical dimension overshoots, which reduces quality of the underlying ICs. This is caused due to increase in beam currents, frequent electron flashes, and reducing re-exposure of chip areas. Thus, to overcome these issues, researchers have proposed a wide variety of optimization models, each of which vary in terms of their qualitative & quantitative performance. These models also vary in terms of their internal operating characteristics, which causes ambiguity in identification of optimum models for application-specific use cases. To reduce this ambiguity, a discussion about application-specific nuances, functional advantages, deployment-specific limitations, and contextual future research scopes is discussed in this text. Based on this discussion, it was observed that bioinspired models outperform linear modelling techniques, which makes them highly useful for real-time deployments. These models aim at stochastically evaluation of optimum electron beam configurations, which improves wafer’s quality & speed of imprinting when compared with other models. To further facilitate selection of these models, this text compares them in terms of their accuracy, throughput, critical dimensions, deployment cost & computational complexity metrics. Based on this discussion, researchers will be able to identify optimum models for their performance-specific use cases. This text also proposes evaluation of a novel EB Lithography Optimization Metric (EBLOM), which combines multiple performance parameters for estimation of true model performance under real-time scenarios. Based on this metric, researchers will be able to identify models that can perform optimally with higher performance under performance-specific constraints.