Elsevier

Solid-State Electronics

Volume 171, September 2020, 107839
Solid-State Electronics

A modified finite difference model to the reverse recovery of silicon PIN diodes

https://doi.org/10.1016/j.sse.2020.107839Get rights and content

Highlights

  • It is found that due to approximating the second-order partial space differentiation by a finite difference (FD) form with a time-dependent constant space step, the FD method has a large error in describing the current-voltage (IV) reverse recovery of a silicon PIN diode.

  • In the un-depleted N region in a silicon PIN diode, the total carrier number can be obtained either by directly integrating the carrier density from the ambipolar diffusion equation (ADE) or by integrating the continuity equation when simulating the reverse recovery of a PIN diode. Their difference points out the accuracy of a numerical algorithm.

  • A modified FD (MFD) method is proposed to use a time-independent constant space discretization step to approximate the second-order partial space differentiation by a FD form. The ADE is solved by using a single-step back-Euler method. Then a new grid is set up based on new boundary coordinates and a cubic spline interpolation is used to transfer the carrier density distribution from the old grid to the new one.

  • Through the comparison of the data with two sets of SRH life time parameters (slow-set and fast-set), it is found that in some cases the current boundary condition at the boundary of P+N or NN+ junction may fail in the usual ADE-based computation.

Abstract

In this paper, Silvaco Atlas TCAD device-circuit mixed simulation and MATLAB programming are used to compute the reverse recovery processes of silicon PIN diodes. The latter is based on solving the ambipolar diffusion equation (ADE) with the moving boundaries. The results of the ADE-based Fourier expansion (FE) and finite difference (FD) method are first compared with that from the Atlas simulation. It is found that the result from the FE method agrees very well with that from the Atlas simulation, while the result from the FD method is much worse. The reason is attributed to approximating the second-order partial space differentiation by a FD form with a time-dependent constant space step in the FD method. One clear phenomenon is that the voltage in a FD simulation shows a very steeper drop followed by a very steeper rise. To solve this problem, we propose a modified finite difference (MFD) method in which the space discretization step is fixed when solving the ADE by a single-step back-Euler method and the new coordinates of two moving boundaries of the un-depleted N region are iterated through the zero-value of the boundary carrier density, current and voltage requirement. Then a new grid is set up based on the new boundary coordinates and a cubic spline interpolation is used to transfer p(x, t) from the old grid to the new one. The result from our MFD method agrees very well with those from Atlas and FE simulation. In addition two sets (a slow set and a fast one) of carrier concentration dependent Shockley-Read-Hall recombination life time parameters are used to study the validity of the usual boundary conditions at two boundaries of the un-depleted N region in the ambipolar diffusion approximation. Our results in some cases the boundary conditions may fail.

Introduction

The PIN diodes are widely used either as independent power devices or a part in insulated gate bipolar transistors [1], [2], [3], [4]. The main technical goal of fast PIN diode is to perform fast and soft reverse recovery [1], [2]. As PIN diodes usually work in high injection levels, so the ambipolar diffusion equation (ADE) with the moving boundaries needs to solve in a reverse recovery simulation [3], [4], [5], [6]. To meet this goal, different approaches have been proposed such as analytical approximation [4], finite difference (FD) [5], [6], finite element (FEM) [7], Laplace transform [8], Fourier expansion (FE) [9], [10], and numerical methods [11], [12]. The FD and FEM methods are often implemented in SPICE simulators, while the FE methods are usually implemented in MATLAB Simulink. Different approximations have been made in each of these methods, so even with the same input parameters, the simulated reverse recovery current and voltage (IV) profiles may be different for different simulation methods. However, many papers show the excellent agreements between their ADE-based simulation results and the reference results, which either from the experimental results or the results based on the mixed circuit and technology computer aided design (TCAD) simulations. If the reference results are experimental results, the agreement may be achieved by curve fitting and the parameter adjustment. If the reference results are TCAD simulation results with a set of input parameter values, the curve fitting may extract a different set of parameter values for an ADE-based simulation. Even the excellent agreement may be true for one PIN device structure, but is not guaranteed to be true for other PIN device structures. The fair comparison is to have a same set of parameter values for both TCAD-based and ADE-based simulations. Currently few papers have performed such comparisons among different ADE-based methods.

For FE methods, the implementation of MATLAB Simulink has both advantages and disadvantages. There are at least two advantages: there are many reliable and easy-to-use MATLAB functions or algorithms; and a system is easily built by connecting different Simulink blocks. One disadvantage is that in most cases the final diagram of the Simulink blocks is too big and has complex signal flows. As only a limited number of Fourier expansion coefficients are taken, the Gibbs’ phenomenon may happen. Usually a triangular window function is applied to Fourier coefficients to reduce the spatial oscillations in the carrier density [10]. However, it may change the total carrier number and cause the loss of the total carrier number. The same question may also take place in other methods such as FD and FEM methods. So it is worthwhile to examine whether an ADE-based algorithm can keep the carrier conservation.

During a reverse recovery, it is observed that the carrier densities decrease in both P+N and NN+ junction sides [9]. In a usual FE method a Fourier expansion form of the carrier density is assumed in the central N region. Often a small number of Fourier expansion coefficients in this method are taken, the depletion region is observed to appear only in the P+N junction side during a reverse recovery. Our Silvaco Atlas TCAD simulation shows that a small depletion region may appear in the NN+ junction side during a reverse recovery. One may wonder if the Fourier series is truncated to keep more expansion coefficients, this phenomenon is still observable in an FE simulation. In addition, increasing the number of Fourier expansion coefficients in FE method, the edges of the carrier density in the un-depleted N region become steeper, the usual feedback control of boundaries of the un-depleted N region may have problems.

For numerical methods, Goebel et al.. adopted the first-order forward Euler method to solve the ADE [11], [12]. As the forward Euler method is not as stable as the backward Euler (BE) method for the stiff differential equations, the computation time step should take a very small value. The advantage is that the FE does not need to solve complex nonlinear equations as the BE does. Yamashita et al. compute the evolution of carriers in the N region of a PIN diode using the Green’s function of the basic diffusion equation [13]. Their computation is shown very fast and the resultant current–voltage (IV) profiles are shown to agree very well with those from TCAD simulations. However, in their model an electron and a hole have the same carrier density and move together, so the same electron and hole fluxes are obtained in either side of the un-depleted N region. It is totally different from the ambipolar diffusion during a reverse recovery process, in which holes mainly move toward to P+ region and electrons move to the N+ region at a reverse bias. For the FD method, Buiatti et al. discretize the second-order partial space differentiation by a FD form with a time-dependent constant spatial step in the ADE, then convert the resultant equations to a circuit form and get the reverse recovery IV profiles through a fast SPICE simulation [5], [6]. In mathematics, a second-order partial space differentiation should be discretized by a FD form with a fixed constant space step. Thus, the FD implementation by Buiatti et al. may give some errors.

Another important problem is the validity of the boundary conditions at two boundaries of the un-depleted N region from the ambipolar diffusion approximation (ADA), in which some further approximations should be made to compute the current components such as electron, hole and displacement current in the depletion layer of P+N or NN+ junction. The reason is that in the ADA the electron, hole continuity equation and Poisson equation are not solved self-consistently, we cannot determine all current components including electron, hole and displacement current when the boundary carrier density is approximated as zero for either reverse-biased junction.

In this paper, we compare FE and FD methods through MATLAB programming. We use a first-order BE method to solve both the ADE and the total carrier-current density equation. In order to examine whether both carrier density and current density follow the ADE and the equation of carrier conservation, the focus is given to solving the nonlinear equations of carrier density and boundary current density iteratively using three different approaches. The values of ∂p/∂x at two boundaries of the un-depleted N region determined from the boundary electron and hole current component in the ADA and the FD approximation based on the carrier density profile may not equal. We compare them in detail in this paper. We take two sets (a slow set and a fast one) of the carrier concentration dependent Shockley-Read-Hall (SRH) recombination life time to study their effect on the validity of the boundary conditions at two boundaries of the un-depleted N region in the ADA. The result from the Silvaco Atlas TCAD simulation is taken as the reference.

Section snippets

Ambipolar diffusion equation

Fig. 1 shows a schematic diagram of a PIN diode and the corresponding doping profile. The N base has a width of W, extending from the P+N junction (x = 0) to the junction NN+ (x = W). Under high injection, the carrier distribution p(x,t) is governed by the following ADE [9], [10]:p(x,t)t=Da2p(x,t)x2-p(x,t)τhl.

Here Da = 2DnDp/(Dn + Dp) is the abmbipolar diffusion coefficient and τhl is the high-level injection life time. The notation xl is used to describe the thickness of the depletion

Results and discussion

Fig. 3a presents the reverse recovery current–voltage (IV) curves computed using Silvaco Atlas TCAD, FE and FD method and the slow-set of consrh lifetime parameters τn0 and τp0. The initial current is set to 50 A. The value of M is set to M = 25 in the FE and FD computation. In Fig. 3a we denote IV curves as FE25 and FD25 and Atlas. For FE25 and FD25, the x2-dependent Vd1 model (10a) is used. The feedback control approach (29), (37) are used for the FE and FD method, respectively. It is found

Conclusions

In summary, we have used Silvaco Atlas TCAD device-circuit mixed simulation and ADE-based FE and FD methods to simulate the reverse recovery processes of silicon PIN diodes. We use MATLAB programming solve the ADE with the moving boundaries. We find that the result from the ADE-based FE method agrees very well with that from the Atlas simulation, while the result from the ADE-based FD method is much different. The reason is attributed to approximating the second-order partial space

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Manhong Zhang received the M.S. degree in condensed matter physics from University of Science and Technology of China, and Ph.D. degree in condensed matter physics from Institute of Physics, Chinese Academy of sciences, in 1997. He received another Ph.D. degree in solid electronics from University of Texas at Austin, Texas, USA, in 2008. From 2009 to 2013 he was with Institute of Microelectronics, Chinese Academy of Sciences. During the period his research was focused on memory devices. Since

References (16)

  • P. Xue et al.

    Comprehensive physics-based compact model for fast PIN diode using MATLAB and Simulink

    Solid State Electron

    (2016)
  • Onozawa Y, Takahashi K, Nakano H, Nemoto M, Otsuki M, Ikawa O, et al. Development of the 1200V FZ-diode with soft...
  • Hille F, Bassler M, Schulze H, Falck E, Felsl HP, Schieber A, et al. 1200V Emcon4 freewheeling diode – a soft...
  • S. Bellone et al.

    An analytical model of the forward I-V characteristic of 4H-SiC PIN diodes valid for a wide range of temperature and current

    IEEE Trans Power Electron

    (2011)
  • S. Bellone et al.

    An analytical model of the switching behavior of 4H-SiC p+-n-n+ diodes from arbitrary injection conditions

    IEEE Trans Power Electron

    (2012)
  • G.M. Buiatti et al.

    Physics-based PiN diodes SPICE model for power-circuit Simulation

    IEEE Trans Ind Appl

    (2007)
  • Buiatti G M. Modeling and Simulation of Power PiN Diodes within SPICE, [Ph.D. thesis], the Dipartimento di Elettronica,...
  • R. Chibante et al.

    Finite-element modeling and optimization-based parameter extraction algorithm for NPT-IGBTS

    IEEE Trans Power Electron

    (2009)
There are more references available in the full text version of this article.

Cited by (5)

Manhong Zhang received the M.S. degree in condensed matter physics from University of Science and Technology of China, and Ph.D. degree in condensed matter physics from Institute of Physics, Chinese Academy of sciences, in 1997. He received another Ph.D. degree in solid electronics from University of Texas at Austin, Texas, USA, in 2008. From 2009 to 2013 he was with Institute of Microelectronics, Chinese Academy of Sciences. During the period his research was focused on memory devices. Since 2014, he has been a Professor in the School of Electrical and Electronics Engineering, North China Electric Power University, China. His research interests include simulation of power devices and CMOS analog circuit design.

View full text