Research On Modeling Method In Design And Manufacture Of Nano - Sized Integrated Circuits

From the birth of the first Integrated Circuit(IC) to the achievement that integration of billions of devices on a single chip, the integrated circuit industry has created a myth in human being’s technological history in just a few decades.However,during the60years, the IC industry is still developing according to the Moore’s Law.The continuous scaling down of critical dimension and the highly integration of IC pushes the design methodology of the IC to update itself.The design methology of the IC has experienced the revolution of device based generation to interconnect based generation,and now it’s entering the third generation of Design For Manufacturing(DFM) and Design For Yield(DFY).As the techonology node of IC scales down to nanometers, the yield of IC is suffering from the the more and more severe process variation, leading to higher and higher expense of IC design and manufacturing. Modeling not only reduces the process ramp up but also save extra cost of research and manufacturing. Thus, as one of the most important methods, modeling is widely used in IC designs and manufacturing.The dissertation focuses on discussing and studying two problems. One is the lithographic hotspot clustering problem caused by the continuous shrinking of critical dimension, the other is the model order reduction(MOR) of large scale integrated circuit caused by the highly integration of IC systems. The solutions to these two problems are given with the modeling method.In the first part of the dissertation, an improved tangent space(ITS) based distance metric classification system is proposed to deal with the lithographic hotspot classification problem. Based on the clustering analysis method in data mining technique, firstly, an improved tangent space based distance metric is proposed to describe the similarity distance between lithographic hotspot.Then a tree based incremental density clustering algorithm is proposed to complete the clustering process.Compared with the traditional clustering method,our improved tangent space based distance metric classification system can achieve better accuracy,which is more reliable for hotspot classification and more suitable for industry applications. In the second part of dissertation, we propose a multi-integral based time domain model order reduction method to deal with the large scale integrated circuit. We derive the state variables relation by integrating the time-domain equation of the original interconnect circuit with multi-step integration formulas. The projection matrix is then generated by the second order Arnoldi method, and is used to transform the large-scale original-systems into reduced-order systems. The proposed method can guarantee the value matching of the state vector for the original interconnect circuits and the reduced-order models. The accuracy of the reduced-order models can thus be guaranteed. The MOR procedure is numerically stable and the reduced-order models also preserve the passivity. Compared with the existing frequency domain MOR methods such as PRIMA, the proposed method can achieve higher accuracy with lower computational cost. Compared with the existing wavelet-collocation based MOR method, the proposed method exhibits lower computational cost with comparable accuracy. Compared with the existing time domain single-step integration based method, the proposed method can achieve higher accuracy with comparable computational complexity.A complete set of numerical experiments are given in this dissertation to demonstrate the effectiveness and validity of these proposed algorithms.