A FEM Punch-Through IGBT Model Using an Efficient Parameter Extraction Method Rui Chibante Armando Araújo, Adriano Carvalho Instituto Superior de Engenharia do Porto (ISEP) Faculdade de Engenharia da Universidade do Porto (FEUP) Rua Dr. António Bernardino de Almeida, 431 Rua Dr. Roberto Frias, s/n 4200-072 Porto 4200-465 Porto Portugal Portugal rmc@isep.ipp.pt {asa, asc}@fe.up.pt Abstract - A Finite Element physics-based punch-through IGBT model is presented. The model’s core is based on solving the Ambipolar Diffusion Equation (ADE) trough a variational formulation, resulting in a system of ODEs. The approach enables an easy implementation into a standard circuit simulator SPICE by means of an electrical analogy with the resulting system of ODEs, solved as a set of current controlled RC nets that describes charge carrier distribution in low-doped zone. The issue of parameter extraction for physics-based IGBT models is also addressed. An optimisation-based algorithm enabling an efficient parameter extraction method for IGBT model is discussed. Model is validated comparing experimental and simulated results. I. INTRODUCTION Modeling charge carrier distribution, in low-doped zones, shown in all bipolar power semiconductor devices, is known as the most important issue for accurate description of dynamic behavior of these devices. Some accurate and complex models have been proposed but due to cumbersome implementation, models like those have been incorporated in powerful, but also very expensive ones, simulation programs (like SABER). Generalized use of physics based models by power circuit designers requires accurate device models running on inexpensive simulators, like those of standard SPICE family. So, in recent years several SPICE models have been reported in literature, with an interesting trade-off between accuracy and computation time. Besides Hefner’s reference model [1], the ADE Fourier based solution [2, 3] have shown interesting results. Although they are adequate for most circuit simulations conditions they present some inaccuracies in predicting dynamic carrier distribution. In Hefner’s model, calculation of redistribution current assumes a linear carrier distribution over n - region which causes some problems at describing switching behavior at high blocking voltages [4]. In Fourier solution, series truncation can give rise to oscillations in carrier concentration which can affect voltage drop estimation [5, 6]. A new approach that overcomes the above limitations has been proposed recently [7, 8]. The core of this new model is based on Finite Element solution of ADE. The approach allows at solving ADE using a variational formulation and Finite Element Method (FEM) solution with one-dimensional simplex elements. The main advantage of the approach is an easy implementation of physics based models into general circuit simulators through electrical analogy with the resulting system of ODEs, solved as a set of variable RC nets, describing hole/electron behaviour in low doped zone. The success of this method has already been proved for PIN diodes, BJTs [9, 10] and non-punch-through (NPT) IGBTs [7, 8] and the aim of this paper is to extend the practice of this approach to punch-through (PT) structures. Using standard models for other zones of the device (emitter, junctions, space charge and MOS) and with the knowledge of hole/electron time/space distribution and instantaneous base width, device’s currents and voltages are accurately determined. With this hybrid approach it is possible to describe device dynamic behaviour with high accuracy while maintaining low execution times. Development and implementation into the standard circuit simulator SPICE of this new physics-based PT-IGBT model is presented in section 2. The main drawback of physics based models is that they require extraction of numerous parameters. Model developers have recognized that an extraction scheme is crucial in order to design power circuits easily through simulation. Different papers in literature have addressed this issue for past decade. Usually, proposed extraction methods are based on extrapolation of datasheet information and equations relating some parameters [2, 11, 12]. With this method they are not expected excellent results. Alternatively, some extraction procedures based on experimental measurements [13, 14] can give accurate results but they are very complex and require so precise measurements that they are not useful for needs of simulation [11, 13, 15]. As pointed out by Kraus et al. [16] probably the best approach is to combine previous methods to get an initial satisfying guess and then use parameter optimisation to extract optimum parameter set. In recent publications parameter optimisation has been used successfully but with some drawbacks. In [17] parameters are extracted using an optimisation procedure by means of a relaxation algorithm. The algorithm is fast but requires a satisfying initial guess for model parameters and still some quite large errors are observed. In [18] a direct search technique is used to extract some parameters of the Fourier model. However, optimisation is neither based on current/voltage waveforms nor switching parameters. It is oriented to produce accurate power loss estimates by analyzing instantaneous power dissipation but it could happen that inaccuracies in voltage and current waveforms are mutually compensated. In this paper an extraction procedure based on an 679 0-7803-9252-3/05/$20.00 ©2005 IEEE