Large-signal Modeling of MESFETs and HEMTs

Optotek Ltd.
Kanata, Ontario, Canada

Accurate device models are essential for the efficient use of both linear and nonlinear simulation programs. Conventional procedures for extracting MESFET and high electron mobility transistor (HEMT) large-signal models are complex, time consuming and require the use of expensive test equipment and software.

LASIMO™ large- and small-signal modeling software, first introduced in 1995, facilitates the development of accurate large-signal models by simplifying procedures for the extraction of MESFET and HEMT large-signal models using a PC.1,2 The program provides designers with a set of proven transistor models and offers flexibility in optimizing different parameters for these models to match actual transistors by fitting the measured and modeled bias dependence of the device characteristics. Three steps are involved. First, transistor S parameters are sampled at gate and drain voltage points in the operating current-voltage characteristics. Next, small-signal transistor models are extracted at each sampling point. Finally, the nonlinear drain current Ids, transconductance Gm, output conductance Gds, gate-source capacitance Cgs and gate-drain capacitance Cgd, are optimized at each sampling point to agree with the supported nonlinear large-signal model.

LASIMO supports nine DC large-signal models and four capacitance models, including the Advanced Curtice, Statz-Raytheon and TriQuint Own Model. A DC large-signal model defines the set of nonlinear equations describing the behavior of the drain-source current, transconductance and output conductance, which LASIMO optimizes to match the corresponding measured, bias-dependent small-signal parameters. Similarly, a capacitance large-signal model defines the nonlinear equations describing the behavior of the gate-source capacitance and gate-drain capacitance, which the program optimizes to match the corresponding measured, bias-dependent small-signal parameters.

For measurements, the program requires only an automatic vector network analyzer and two programmable power supplies. The Vds and Vgs bias sweeps are selected and the program then automatically extracts the data files, sorts the associated voltage and current dependencies, and stores the RF and DC data in a compact database readable by the program.

Small-signal Model Extraction
The characteristics of the bias-dependent S-parameter files are determined automatically. This step extracts the bias dependencies of Gm, Gds, Cgs and Cgd prior to large-signal model optimization. The extraction of each parameter is performed within an independent, user-specified frequency range. Parasitic resistances and inductances, and other bias-independent intrinsic elements (source-drain capacitance, Cds, internal resistance, Ri and time constant t) also can be extracted via a special mode, which combines optimization and extraction.

Large-signal Model Optimization
Two types of Newton optimizers and two types of random optimizers are provided for solving and fitting the measured and modeled data. The nonlinear transconductance and output conductance expressions for any chosen supported model are obtained by deriving the drain-source current with respect to the gate
and drain voltages, and then fitting these results to the measured data. The measured drain current is used to de-embed parasitic effects and obtain the intrinsic voltages. Bracketing algorithms are included to vary the search step and arrive at a minimum error function. Optimization can be performed for the Ids, Gm, Gds set and the Cgs, Cgd set separately or in combination. In addition, weighting functions are defined separately for each of those parameters. Figure 1 shows the LASIMO optimization screen.

Output Types
LASIMO supports an extensive range of test and graph outputs. Measured and
modeled S parameters can be displayed in Smith, polar and x-y charts, and in text. Ids-V as well as Gm-V, Gds-V, Cgs-V and Cgd-V characteristic curves also can be displayed in x-y charts or in text form. Figure 2 shows a typical Gm vs. Vds data display. The small- and large-signal errors at each bias and frequency point also are presented as tabular data. Vector network analyzer measured S-parameter data also can be viewed graphically and in text form.

LASIMO exports the large-signal model parameters to a SPICE netlist. A novel output generates bias-dependent small-signal MESFET and HEMT models corresponding to the various large-signal models, which benefits users of the MMICAD™ small-signal linear simulator program. The model accepts the large-signal parameters as data in addition to Vds and Vgs, and then computes the S parameters at user-defined bias points. These data can be used for the simulation of small-signal circuits such as amplifiers as a function of operating bias.

User-defined Model Version
Applications requiring the utmost accuracy call for a modification of existing models or the development of new models. To address this situation, LASIMO's capabilities have been extended to allow new models to be incorporated and edited by the user. In addition to simplifying the extraction of key transistor parameters, LASIMO Version 2 allows for the creation of models ideally fitted to actual devices.3 Provision has been made for five DC and five capacitance user-defined large-signal models.

The new user-defined models are implemented as dynamic link libraries (DLL), which are created and added outside LASIMO. The program links to the DLL at run time. A DLL is provided for each of the five DC and five capacitance large-signal models. Hence, whenever the program needs to obtain data from a given user-defined model, it links dynamically to the appropriate DLL. (Multiple DLLs allow the user to conduct trials on model variations or create different models without the contents of the DLL.)

Each DLL is generated from a set of project files written in C language and loaded into a C compiler. Each project is preconfigured to be loaded into the Visual C++ environment.3 However, LASIMO users are not required to know the details of programming the DLLs. Each project that generates a DLL is configured as a set of files with fixed presets and is provided with a default model, which is intended to be edited and compiled to create the user-defined model. The source code for the project files is provided with the program. (The fixed presets for the project group of files enable them to be compiled into DLLs in the format accepted by LASIMO 2.0.) Presently, the default DC model is the Curtice model and the default capacitance model is the Basic Semi-junction model. Only a single function that computes the nonlinear parameters in one designated project file needs to be edited by the user. Only elementary level C programming is required.

The function used to create a user-defined model receives a set of arguments from the program and the function returns certain nonlinear parameters. The user is able to define and access up to 13 large-signal parameter coefficients to be optimized for each DC model and up to 15 large-signal parameters for each capacitance model. These DC and capacitance large-signal parameters can be selected to suit various nonlinear simulators.

When LASIMO 2.0 accesses a DC user-defined large-signal model, a maximum of 13 DC large-signal model parameters are passed to the function in the DLL. (These parameters are the coefficients of the equations describing the behavior of the drain current Ids, transconductance Gm and output conductance Gds, as well as the current bias values for the gate and drain voltages Vds and Vgs.) The DLL returns the nonlinear computed values of the drain current Ids, transconductance Gm and output conductance Gds. Local variables or functions can be added by the user as needed in order to arrive at these three values.

When the program accesses a capacitance user-defined large-signal model, a maximum of 15 capacitance large-signal model parameters are passed to the function in the DLL. (These parameters are the coefficients of the equations describing the behavior of the gate-source capacitance Cgs, drain-source capacitance Cgd, as well as the current bias values for the gate and drain voltages Vds and Vgs.) The DLL returns the computed values of the gate-source capacitance Cgs and the drain-source capacitance Cgd.

Besides allowing users to create their own large-signal user-defined models in DLLs, LASIMO 2.0, through its capability of linking to the model, permits the designer to define the number of large-signal parameters for the model and to assign name labels for both the model and the large-signal parameters. In addition, the designer can verify the computed values returned by the model, verifying the validity of the model DLL. These facilities allow for the streamlined integration of external user-defined large-signal models into the program and, in terms of subsequent operations, renders them indistinguishable from the built-in models.

The new user-interface features that appear in the LASIMO file form are shown in Figure 3 . The file form also lists the supported built-in models. In subsequent use, a dialog box prompts for the name label of the selected DC user-defined model, as well as the name labels of the large-signal parameters. A similar menu is provided for the capacitance models.

The test model dialog box shown in Figure 4 allows the user to link dynamically to the DLL and is used to set the values of the large-signal parameters and the bias values, and send them to the DLL selected by the user-defined model. The computed values are returned in addition to the status of the DLL. The Analyze Model selection key initiates the process and the returned values can be inspected to verify the validity of the DLL. A similar test menu is provided for the capacitance models.

In subsequent operations, the user-defined model version operates in the same manner as the built-in large-signal model version. Once a model is coded and a DLL is generated, the number of DC or capacitance parameters in the file form are defined. The operations for extracting large-signal models proceed identically to LASIMO Version 1.0.

LASIMO 2.0 requires a 486 or Pentium™ PC with Windows™ 3.1 or above, and a minimum of 8 MB of random access memory. A National Instruments IEEE-488 card is required for data acquisition. The manual and on-line help system include easy-to-follow tutorials that guide the user in the handling of data acquisition, and small and large-signal model extraction. An appendix to the manual documents all the small- and large-signal model extraction algorithms.

Prices for the user-defined model version of LASIMO start at $3900. Delivery is from stock. Additional information is available from the company's Web site at

LASIMO Version 1.0 was developed with contributory financial assistance from the National Research Council of Canada (Grant IRAP RDP 22713U; team leader S.F. Dindo). Alpha Industries (C.J. Wei, Y.A. Tkachenko and D. Bartle) provided assistance with the evaluation of the user-defined model version of LASIMO.

1. H. Willings, C. Rauscher and P. deSantis, "A Technique for Predicting Large-signal Performance of a GaAs MESFET," IEEE Transactions on Microwave Theory and Techniques, Vol. MTT-26, 1978, pp. 1017–1023.

2. M. Miller, M. Golio, B. Beckwith, E. Arnold, D. Halchin, S. Ageno and S. Dorn, "Choosing an Optimum Large-signal Model for GaAs MESFETs and HEMTs," IEEE MTT-S Digest, 1990, pp. 1279–1282.

3. C.J. Wei, Y.A. Tkachenko, D. Bartle, S.F. Dindo and D.I. Kennedy, "An Accurate Dispersive, Self-heating Large-signal Model of GaAs MESFETs and its Parameter Extraction," to be published in an upcoming issue of Microwave Journal.

Optotek Ltd.,
Kanata, Ontario, Canada
(613) 591-0336 or (800) 361-2911.