Font Size: a A A

Study Of Electromagnetic Modeling And Parameters Extraction Of On-chip Interconnects In High Speed Integrated Circuits

Posted on:2003-04-24Degree:DoctorType:Dissertation
Country:ChinaCandidate:X Z ZhongFull Text:PDF
GTID:1118360065951233Subject:Microelectronics and Solid State Electronics
Abstract/Summary:PDF Full Text Request
With the tremendous improvement of semiconductor technique, today's integrated circuits (1C) technology has entered into deep submicrometer (DSM) region. Now that a single chip is an entire system (the concept of system-on-a-chip, SOC), on-chip interconnect is now one of the most challenging areas of 1C processing. The parasitic effects of interconnect is the bottleneck of the entire circuit system performance. The traditional methods of parasitic parameter extraction in EDA, which based on the concept of lumped component, have lost their accuracy. Electromagnetic modeling and parameter extraction play an important role in 1C design and it is a foundation of the works later.The subject of dissertation is electromagnetic modeling and parameters extraction of long interconnects in SOC. As analyzing the electronic properties, methods studying are particularly emphasized at the same time. Accuracy and speed are two aspects to be concerned and the later is given more attention for the sake of practical using in EDA/CAD. This dissertation has made the following research:For the structures comprising discontinuities and three-dimensional specialities, which largely present in SOC, the planar gap and the tridimensional vertical bend of shielded coplanar waveguide (CPW), are analyzed respectively as two typical examples. The finite-difference time-domain (FDTD) method are used for electromagnetic simulation using its highly accuracy. Structure electronic properties are investigated and equivalent circuit parameters are obtained at the first time.For the lossy multiple conductors configuration, which is essential unit of interconnects, its rapid and accuracy extraction are studied and achieved. Two and three coplanar lines are studied as examples and their equivalent circuit parameters are obtained. An nonuniform mesh partitioning technique is developed by modification. It solves the difficulty of grid producing by an easy and flexible way. Simulation results show that when combine it with the compact FDTD method and the signal curvefit prediction approach, the computation magnitude is tremendously cut down compared to that when only the basic FDTD method is used. So simulation efficiency is greatly enhanced and the analyzing difficulty in dealing with multiple conductors is solved to some extend.Meanwhile, owing to the fact that there is huge hardness in accurate extracting of attenuation constant, one of the most new modified compact FDTD methods is applied to analyzing lossy multiple conductors to give accurate results. It is combined with the nonuniform mesh partitioning technique to reduce the computing time.Aiming at the fact that numerical simulations are very time-consuming and are used very limitedly in alternative CAD, Artificial neural network (ANN) is introduced to electromagnetic modeling to solve the problem. For the planar gap andthe tridimensional vertical bend of shielded CPW, FDTD simulations are carried out to produce training and testing samples and error-back propagation algorithm is used to train the multilayer perceptron neural networks (MLPNNs). Rapid and accuracy CAD models of these structures are successfully obtained for the first time. These developed MLPNN models can give the same reliable results as that of the electromagnetic simulations, but their computing time are negligible. Since this approach can efficiently remedy the time-consuming problem of single numerical method, it is very potential in the accurate and efficient interconnect modeling of circuit EDA/CAD.Fortran and Matlab programs are written for all the above approaches. Some primary work are done in soft realization.
Keywords/Search Tags:On-chip interconnect, electromagnetic modeling, parameter extraction, Finite-Difference Time-Domain (FDTD) method, Artificial neural network (ANN)
PDF Full Text Request
Related items