Moon-gyu Jeong

Improving model prediction accuracy for ILT with aggressive SRAFs

For semiconductor IC manufacturing at sub-30nm and beyond, aggressive SRAFs are necessary to ensure sufficient process window and yield. Models used for full chip Inverse Lithography Technology (ILT) or OPC with aggressive SRAFs must predict both CDs and sidelobes accurately. Empirical models are traditionally designed to fit SEMmeasured CDs, but may not extrapolate accurately enough for patterns not included in their calibration. This is particularly important when using aggressive SRAFs, because adjusting an empirical parameter to improve fit to CDSEM measurements of calibration patterns may worsen the models ability to predict sidelobes reliably. Proper choice of the physical phenomena to include in the model can improve its ability to predict sidelobes as well as CDs of critical patterns on real design layouts. In the work presented here, we examine the effects of modeling certain chemical processes in resist. We compare how a model used for ILT fits SEM CD measurements and predicts sidelobes for patterns with aggressive SRAFs, with and without these physically-based modeling features. In addition to statistics from fits to the calibration data, the comparison includes hot-spot checks performed with independent OPC verification software, and SEM measurements of on-chip CD variation using masks created with ILT.

Pixel-based learning method for an optimized photomask in optical lithography

Abstract. Circuit design is driven to the physical limit, and thus patterns on a wafer suffer from serious distortion due to the optical proximity effect. Advanced computational methods have been recommended for photomask optimization to solve this problem. However, this entails extremely high computational costs leading to problems including lengthy run time and complex set-up processes. This study proposes a pixel-based learning method for an optimized photomask that can be used as an optimized mask predictor. Optimized masks are prepared by a commercial tool, and the feature vectors and target label values are extracted. Feature vectors are composed of partial signals that are also used in simulation and observed at the center of the pixels. The target label values are determined by the existence of mask polygons at the pixel locations. A single-hidden-layer artificial neural network (ANN) is trained to learn the optimized masks. A stochastic gradient method is adopted for training to handle about 2 million samples. The masks that are predicted by an ANN show averaged edge placement error of 1.3 nm, exceeding that of an optimized mask by 1.0 nm, and averaged process variation band of 4.8 nm, which is lower than that of the optimized mask by 0.1 nm.

Prediction of biases for optical proximity correction through partial coherent identification

Abstract. Most approaches to model-based optical proximity correction (OPC) use an iterative algorithm to determine the optimum mask. Each iteration requires at least one simulation, which is the most time-consuming part of model-based OPC. As the layout becomes more complicated and the process conditions are driven to the physical limit, the required number of iterations increases dramatically. To overcome this problem, we propose a method to predict the OPC bias of layout segments with a single-hidden-layer neural network. The segments are characterized by length and based on intensities at the corresponding control points, and these features are used as input to the network, which is trained with an extreme learning machine. We obtain a best-error root mean square of 1.29 nm from training and test experiments for layout clips sampled from a random contact layer of a logic device. In addition, we reduced the iterations by 27.0% by initializing the biases in the trained network before performing the main iterations of the OPC algorithm.

Comparison of OPC models with and without 3D-mask effect

OPC models with and without thick mask effect (3D-mask effect) are compared in their prediction capabilities of actual 2D patterns. We give some examples in which thin-mask models fail to compensate the 3D-mask effect. The models without 3D-mask effect show good model residual error, but fail to predict some critical CD tendencies. Rigorous simulation predicts the observed CD tendencies, which confirms that the discrepancy really comes from 3D-mask effect.