Comparison between the process windows calculated with full and simplified resist models

Abstract

While numerical simulation is generally regarded as indispensable for wavefront engineering tasks such as OPC decoration and phase-shift mask design, full resist models are rarely used for this purpose. By "full resist models", we mean models derived from a physical, mechanistic description of the chemical response of the photoresist to exposure and the subsequent PEB and develop processes. More often, simplified models such as an aerial image threshold model or the Lumped Parameter Model (LPM) are used because these models are much faster and make optimization of optical extension technology more tractable. Simplified resist models represent a compromise between computational speed and simulation accuracy. The purpose of this study is to quantify the differences between the process windows calculated with simplified and full resist models.Our approach is first to fit the parameters in the simplified models to match results obtained with a full resist model, and then to compare the predictions of the simplified resist models with those obtained with the full model. We take two approaches to model tuning: mathematical derivation of relationships between the models, and least-squares fitting of FE matrix data for isolated and dense lines.

Selection approach of critical patterns for calibrating the physical resist model based on spectrum coverage

“…Different imaging characteristics are observed due to the varying pitch-to-CD ratio values. Dense and isolated line/space patterns are frequently used for resist modeling 27 , 28 , 30 – 32 and calibration 33 – 38 due to the significant differences in imaging characteristics between them. Consequently, the EXP-method 1 selects the patterns with the smallest and largest pitch-to-CD ratios as the critical patterns.…”

“…modeling 27,28,[30][31][32] and calibration [33][34][35][36][37][38] due to the significant differences in imaging characteristics between them. Consequently, the EXP-method 1 selects the patterns with the smallest and largest pitch-to-CD ratios as the critical patterns.…”

Lumped parameter model for chemically amplified resists

“…Although the 3D-LPM model works well to match a full resist model for certain resists [1] it has poor predictability for other resists. Generally it has been noticed by these authors that the LPM model works well for resists optimized for logic applications but does not have acceptable predictability for resists optimized for contact hole or isolated trench applications.…”

“…The chosen resist model greatly determines both the speed, accuracy, and predictability of any lithography simulation. Currently the available resist models can be classified into four groups [1]: aerial image only, empirical lumped parameter, full physical continuum, and meso/molecular scale models. Although the accuracy varies even within each class of model it is generally assumed the full physical continuum models are more accurate but slower than the aerial image and empirical lumped parameter models.…”