Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

A Method to Select Patterns for Etch Model Calibration

IP.com Disclosure Number: IPCOM000249087D
Publication Date: 2017-Feb-03
Document File: 5 page(s) / 204K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a flow to select optimal structures for the etch model calibration that guarantees a stable and accurate model. The process comprises clustering patterns, etching parameter space, and then clustering etch parameters and selecting parameters. This novel method provides an unambiguous, objective, and reproducible way to select a minimum set of design patterns for etch model calibration.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 43% of the total text.

Title A Method to Select Patterns for Etch Model Calibration Abstract Disclosed is a flow to select optimal structures for the etch model calibration that guarantees a stable and accurate model. The process comprises clustering patterns, etching parameter space, and then clustering etch parameters and selecting parameters. This novel method provides an unambiguous, objective, and reproducible way to select a minimum set of design patterns for etch model calibration. Problem The Optical Proximity Correction (OPC) technique is applied in semiconductor manufacturing to enable a robust patterning process. Very accurate optical and resist models are used to simulate a printed photoresist profile and correct for proximity effects by adjusting litho mask layout in an iterative fashion. However, every new technology node imposes more stringent patterning accuracy requirements than the previous one did. To fulfill these requirements, accurately predicting photoresist image is not enough. Starting 22fdx tech node adequate etch effects compensation is absolutely necessary to enable reliable patterning and high yield levels; therefore, etch process models, capable of predicting etch effects, are needed. An etch model usually starts from a litho contour, to compute at every litho contour point the litho-etch bias that can be strongly dependent on the local geometry. Figure 1 illustrates strong variations in the litho-etch bias among patterns with different dimensions, shapes, and proximities. Figure 1: Strong dependency of the litho-etch bias on the pattern dimensions and surrounding

Etch models are typically constructed from a set of arbitrary basis functions or kernels adjusted to fit at best silicon data. Consequently, etch modelling quality is strongly dependent upon the calibration data set (collection of the patterns used to build the model == sampling plan). An incorrectly selected sampling plan may result in both under-fitting and over-fitting of silicon data, or even model instability. Currently, no objective method or tool is available to rigorously select structures for etch model calibration (i.e., selecting structures that cover at best the various litho-etch bias responses). Common practices to define structures are purely empirical and subjective: based on know-how, experience, and trial and errors. Due to these flaws, existing etch models cannot achieve accurate simulations, are prone to overfit, and are not reliable to extrapolate (i.e. predict contours of structure different from those used during the calibration). Solution/Novel Contribution The novel solution is a flow to select optimal structures for the etch model calibration that guarantees a stable and accurate model. The method is not limited to the etch model, but also applies to predicting the shape post-fill or post-Chemical Mechanical Planarization (CMP). Method/Process Clustering The novel method consists of the following independent clustering steps. (Figure 2)

1. Split an i...