High performance lithographic hotspot detection using hierarchically refined machine learning

Abstract

Under real and continuously improving manufacturing conditions, lithography hotspot detection faces several key challenges. First, real hotspots become less but harder to fix at post-layout stages; second, false alarm rate must be kept low to avoid excessive and expensive post-processing hotspot removal; third, full chip physical verification and optimization require fast turn-around time. To address these issues, we propose a high performance lithographic hotspot detection flow with ultra-fast speed and high fidelity. It consists of a novel set of hotspot signature definitions and a hierarchically refined detection flow with powerful machine learning kernels, ANN (artificial neural network) and SVM (support vector machine). We have implemented our algorithm with industry-strength engine under real manufacturing conditions in 45nm process, and showed that it significantly outperforms previous state-of-the-art algorithms in hotspot detection false alarm rate (2.4X to 2300X reduction) and simulation run-time (5X to 237X reduction), meanwhile archiving similar or slightly better hotspot detection accuracies. Such high performance lithographic hotspot detection under real manufacturing conditions is especially suitable for guiding lithography friendly physical design.

Logic synthesis for manufacturability considering regularity and lithography printability

“…In order to define CHSci, it is necessary to evaluate the lithography hot spots. Lithography hot spots are patterns in the layout which are more susceptible to suffer a large variation during lithography [9][10][11][12].…”

“…This work presents a novel yield model for ICs, which considers lithography printability problems [9][10][11][12] as a source of yield loss. Moreover, a technology remapping approach considering this yield model as cost function is proposed and implemented, with good results presented.…”

DRC-based hotspot detection considering edge tolerance and incomplete specification

“…They create a dual graph to represent a given layout and then filter out over-weighted edges and faces according to a user-specified threshold value. Later, Ding et al in [3], Ding et al in [4], Wuu et al in [5], and Yu et al in [6] propose variant machine learning frameworks based on artificial neural networks or support vector machines. They extract hotspot features to train their learning machines.…”

A Fuzzy-Matching Model With Grid Reduction for Lithography Hotspot Detection

“…2) Machine learning techniques such as artificial neural network (ANN) [4], [6], [20] and support vector machine (SVM) [3], [4], [12]- [14] may be better in prediction accuracy, (that is, an unseen hotspot can be recognized as a hotspot). Nevertheless, this type of method normally suffers from the problem of false alarm, (that is, many nonhotspots are identified as hotspots).…”