Browse Prior Art Database

A Method for Predicting 3-D Lithographic Hotspots from Layout Data Disclosure Number: IPCOM000239403D
Publication Date: 2014-Nov-05
Document File: 4 page(s) / 84K

Publishing Venue

The Prior Art Database


Disclosed is a method for predicting three-dimensional (3D) lithography hotspots by applying linear regression on 3D simulation data to train a model, followed by full chip analysis using the trained model.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 52% of the total text.

Page 01 of 4

A Method for Predicting 3-D Lithographic Hotspots from Layout Data

Resist scumming and top loss are becoming major issues in lithography. These problems are not easily detectable with conventional two-dimensional (2D) simulation methods. Resist scrumming and top loss problems are discoverable by three-dimensional (3D) litho simulation, using accurate models, but this comes at a prohibitively high runtime cost.

The novel solution is a machine learning technique for developing a 2D shapes-based model trained on 3D simulation data, capable of predicting resist scumming or top loss. The model is trained using 3D lithographic simulation data, or hardware data. Then it can be applied on full-chip layouts at fast runtimes, since it consists of a 2D shapes- based analysis.

The solution is comprised of the following methods:

 Gathering model training data using 3D lithographic simulations

 Training the model with resist data from different process corners

 Training the model with wafer data

 Developing a set of features for the model

 Developing the model using linear regression to training data (obtained from above)

 Applying the model on full-chip layouts at fast runtimes using the same feature set generation, followed by prediction

 Re-training the model when process/Optical Proximity Correction (OPC)/retargeting changes

The core process for implementing the methods follows:

1. Identify set of training patterns selected from known failures and non-failing clips. Training can be done one-time or repeated with newly found hotspots.

2. Extract feature vector (key step) based on design shapes or retargeted shapes

3. Train a machine learning model; Linear Regression on feature set

Figure 1: Machine Learning-Based 3-D Hotspot (HS) Recognition


Page 02 of 4

Feature extraction is a key step in the process. The method uses lithography insight to derive features. Most 3D defects occur due to poor contrast; lower intensity leads to scumming and higher intensity leads to top loss.

The novel method derives features based on concentric circles (Ci ). (Figure 2) Assume all features w...