Automatic detection of photoresist residual layer in lithography using a neural classification approach

Abstract

Photolithography is a fundamental process in the semiconductor industry and it is considered as the key element towards extreme nanoscale integration. In this technique, a polymer photo sensitive mask with the desired patterns is created on the substrate to be etched. Roughly speaking, the areas to be etched are not covered with polymer. Thus, no residual layer should remain on these areas in order to insure an optimal transfer of the patterns on the substrate. In this paper, we propose a nondestructive method based on a classification approach achieved by artificial neural network for automatic residual layer detection from an ellipsometric signature. Only the case of regular defect, i.e. homogenous residual layer, will be considered. The limitation of the method will be discussed. Then, an experimental result on a 400 nm period grating manufactured with nanoimprint lithography is analyzed with our method.

Introduction

Miniaturization has always been one of the main topics in the semiconductor industry since the 1960s. For this reason, a lot of funds have been invested for the development of appropriate manufacturing techniques that can continuously meet the ITRS (International Technology Roadmap for Semiconductors) expectations. Among these techniques, photolithography is a basic process and is still extensively used to achieve a very large scale integration density. In photolithography, light is used in order to reproduce desired patterns on a semiconductor material. The first step of this technique consists of coating a thin photoresist film on the substrate. The desirable patterns are then transferred onto the film by exposing a mask with a photon beam. The film is hence structured and acts as a mask for the substrate etching. The presence of a residual layer in the regions that should be etched is considered as a mask defect and must be controlled in preference inline before the final step of the substrate etching. Considering nanoimprint lithography [1], patterns etched on a mold surface are pressed into a polymer coated on a substrate. During the imprint, the polymer is heated above its glass transition temperature. Then, the polymer is cooled and the mold is removed. A thin residual layer is voluntary left in the compressed area in order to avoid any destructive contact between the mold and the substrate. This residual layer is then removed using appropriate etching processes and should be controlled before transferring the patterns on the substrate. For instance, such layer requires an over-etching that can affect the critical dimension of the structure transferred to the substrate. To fulfill these needs, rapid and reliable control techniques have to be considered. Standard techniques mainly used for dimensional metrology such as atomic force microscopy (AFM) and scanning electron microscopy (SEM) cannot satisfy these requirements. Indeed, they are time consuming, destructive and not adapted for inline metrology. Optical emission spectroscopy can be employed [2] in order to identify the chemical elemental composition of the sample. However, this technique does not provide quantitative information about the existing elements. Moreover, it is not adapted for structures with high line/space ratio because the optical emission corresponding to the residual layer is very weak. Alternative optical methods have emerged in the last decade as an alternative solution for critical dimension control. Scatterometry [3], [4], [5], [6] is one of the most promising techniques that can achieve this goal, being in the same time non destructive. It is based on the analysis of light scattered from a periodic structure in order to extract the corresponding geometrical profile. This technique requires the measurement of an optical signature diffracted from the sample that can be measured using different experimental set ups [7], [8], [9], [10], [11], [12], [13], [14]. The geometry is then deduced by solving an inverse problem using optimization or regression algorithms [4], [6], [7]. In general, some hypotheses have to be stated, in particular the geometrical model of the sample as well as the knowledge of the optical refractive index of each medium. In general, the user should have a priori information about the structure in order to set out suitable hypotheses. In our previous work [16], we have developed a neural network classifier in order to identify the geometrical profile of a diffraction grating before the characterization step. We have demonstrated [16] the feasibility of a neural classifier designed to distinguish between two geometrical profiles in a simple case. Other similar work [6] has been achieved using library search method and a multilayer profile model previously generated with a special engine. In this paper, the concept of classification will be proposed for another issue concerning residual layer identification in optical lithography. The idea is to measure the optical signature of the sample, and then to use a neural classifier in order to detect whether a residual layer is present or not.

In recent years, artificial neural networks ANNs have become very popular because of their performances of optimization and classification [7], [15], [16], [17], [18]. In classification task, they notably present the ability to estimate posterior probabilities without any prior assumptions about the statistical distribution of the data. Moreover, ANNs provide instantaneous treatment that has made them the only potential method suitable for inline characterization. To meet the rapidity requirement, the optical signature is measured using a phase modulated spectroscopic ellipsometer. This set up undergoes ongoing development and is widely employed in semiconductor industry for optical and dimensional metrology.

This paper is organized as follows. The architecture of the neural network classifier, dedicated for the analysis of an ellipsometric signature, is presented in the next session. Then, simulated results are summarized showing how to determine the minimal residual thickness that can be detected by the classifier. In this work, we will consider the case of a periodic grating with a homogenous residual layer over the structure, i.e. the case of a regular defect. The final section deals with experimental application of the method using a 400 nm pitch grating manufactured with nanoimprint lithography. The choice of imprinted grating is supported by the fact that homogenous residual layer can be obtained.