Placement of Sub-Resolution Assist Features Based on a Genetic Algorithm

Wei-Ming Kan,Chun-Sheng Wu,Hung-Fei Kuo,Chao-Yi Huang,Kao-Tun Chen,Yu-Ru Huang,Yu-Shin Lin

doi:10.1109/access.2019.2926102

Abstract

Resolution enhancement techniques compatible with an ArF (193 nm) immersion optical lithography system may constitute an effective means of minimizing the size of technology nodes of the dynamic random access memory. This paper investigated one such technique, namely mask optimization (MO), and applied sub-resolution assist features (SRAFs) in the MO to improve the aerial image quality of a target pattern that had undergone optical proximity correction (OPC). This paper first developed an optical model based on the Hopkins model to create an interference map, which was then used to create a cut-level map. The cut-level map was instrumental in predicting potential SRAF sites and randomly generating SRAFs that would serve as the initial population for a genetic algorithm (GA). Chromosomes were defined as a section map and encoded genes were used to define SRAF and target pattern. Using a GA to identify SRAF geometric measurements and placement was revealed to increase the process window and improve the image performance of the target pattern. This paper used 1D and 2D line/space (L/S) images as the baseline to test the convergence of the proposed method. 2D images were also used to test improvements in aerial image performance. The results indicated that the 1D L/S pattern converged at the 100th iteration. Furthermore, the depths of the focus of the 2D L/S array and 2D contact hole patterns were successfully increased by 113 and 21 nm, respectively. The proposed SRAF method, which integrated the GA and interference map, was able to ensure the diversity of potential SRAF solutions. Moreover, it was able to restrict the SRAF solutions to rectangular structures through the application of mask rules, thereby reducing the cost and improving the feasibility of photomasks.

Highlights

Dynamic random access memory (DRAM) is extensively used in components for smart televisions, self-driving cars, internet of things applications, and high-end servers, rendering the continual minimization of DRAM technology nodes a critical task [1], [2]
PROJECTION LITHOGRAPH MODEL The mask optimization (MO) process consists of two parts: the first part involves the optical proximity correction (OPC) for the pre-OPC target pattern, and the second part involves the addition of sub-resolution assist features (SRAFs) in the post-OPC target formed upon completion of the OPC for the target pattern [23], [24]
Because an overly complicated photomask design increases the cost of the photomask, this study introduces a set of mask rules to eliminate potential SRAFs that fail to meet its conditions

Summary

INTRODUCTION

Dynamic random access memory (DRAM) is extensively used in components for smart televisions, self-driving cars, internet of things applications, and high-end servers, rendering the continual minimization of DRAM technology nodes a critical task [1], [2]. The optimized target pattern in this design was primarily measured by the main focus and main exposure dose, meaning that it could not guarantee an increase in the process window To address these shortcomings, Xu et al proposed applying a genetic algorithm (GA) to both positive and negative SRAF placement rules [21], [22]. This study integrated the intensity map with the mask rules on the focal plane to determine SRAF sites and employed a GA to identify the optimal image quality and maximize the process window This approach differed from the conventional method and prevented the development of overly complicated SRAF structures, reducing the manufacturing cost of the photomask.

PROJECTION LITHOGRAPH MODEL

EXPERIMENTAL RESULTS

CONCLUSION