Analysis and comparison of four TCAD simulation methods for RF device design

The most common purpose of applying computer-aided design technology (TCAD) to the design field is prediction. Simulation can enable engineers to understand the design. In some cases, simulation can be used to deal with things that cannot be measured in the laboratory. Simulation is increasingly used to improve manufacturing yield. When the prediction is reliable, it has a great advantage in the technology development process. Although the predicted trends and quantities are sometimes different from the actual measured data, in most cases, TCAD can still provide the most accurate prediction based on the existing data.

Problems in the radio frequency ( RF ) domain can be divided into two categories: small signal and large signal. TCAD-based small signal predictions are often difficult because of the complexity of replicating device characteristics into the simulator. In large signal predictions, distortions may occur due to operating limitations and device imperfections. In those instances where we can accurately simulate a small signal solution, we can probably also simulate a large signal solution.

TCAD Principles

Let's look at a practical problem shown in Figure 1, which occurs with both small-signal and large-signal amplifiers. In the small-signal case, the amplifier is usually a class A amplifier, we assume a conjugate match, and the design is formalized. In the large-signal case, the amplifier can be class AB or even class B, in which case the best performance is usually achieved by optimizing the input and output matching networks, input bias, and harmonic termination. If these parameters are considered together with the intrinsic device design parameters, a large research space will be obtained. For multi-stage amplifiers, which often use different technologies, this parameter increase may occur.

To understand the performance of different device designs, we should not compare all device designs at the same matching, harmonic termination, and biasing, but must compare to those values ​​that produce the best performance of the device design. By comparing the optimum performance conditions for each device in the amplifier, the selection of the best performing device design can be made.

Figure 1: RF amplifier circuit.

This makes a clear case for full large-signal simulation of the circuits and devices under consideration. If engineers wish to evaluate new devices for which no measured data is available at the time, a TCAD solution must be employed. Large-signal simulation of TCAD data is accomplished in four ways:

1. Use mixed-mode transient simulation in TCAD;

2. Use integrated harmonic balance directly in TCAD tools for large signal simulation;

3. Special tools integrate composite simulation results with circuit design;

4. Extract large-signal compact models from TCAD data and use these models to understand large-signal characterization parameters.

It is important to investigate the possibility of using the most basic approach, mixed mode simulation. In mixed mode, transient circuit simulation is done directly in the TCAD software. Typically, circuit designers do not use transient simulation but harmonic balance to solve these large signal simulation problems. The challenge is to accurately describe the RF circuit in the workbench and then perform TCAD simulations using simple single (or dual) frequency carrier ( CW ) input power sweeps. Transient simulations must be performed in steady state and may take several cycles. This simulation can be more numerically challenging when the device is in compression. The simulated time series must be long enough to describe the lowest frequency required. For two-tone or multi-tone problems, very long transient simulations may be required. Therefore, running power sweeps can be very time consuming. In addition, most designs in mixed mode workbench are very poor because it lacks key terms such as complex impedance. The main problems with this approach for real circuits are speed and convergence.

The second method is to implement harmonic balance in TCAD simulation tools. The harmonic balance method is more correctly called KCL-HB or Kirchhoff's Current Law Harmonic Balance and is used in Agilent's Advanced Design System ( ADS ), Cadence's Spectre-RF, and other circuit simulators commonly used in RF and analog design. Harmonic balance is a nonlinear frequency domain technique used to determine the quasi-periodic steady-state solution of a system with a wide frequency variation content. This method uses the following equation:

This equation describes the relationship between the linear and nonlinear circuit currents. The parameters in brackets are the linear part and the rest are the nonlinear part. _{I s} is the supply current, Y is the linear circuit admittance matrix, V is the internal node voltage vector, Ω is the angular frequency matrix on the diagonal, Q is the charge vector in the frequency domain, and I _G is the nonlinear circuit current in the frequency domain. This solution begins to converge when the linear and nonlinear circuits reach a balance.

Implementation using TCAD requires a lot of development work. Although there is a lot of research in this area and programs from universities are available, there has not been a reliable tool available on the market. Harmonic balance is a method used for large signal RF problems, usually performed in circuit simulation tools. Harmonic balance is a nonlinear frequency domain steady state simulation.

Linear circuit components are modeled in the frequency domain only, nonlinear components are modeled in the time domain and converted to the frequency domain at each step. The algorithm generally limits the number of harmonics processed to 7-11. The memory requirement to reach the 11th order is 4-8GB, not including the memory required for device simulation. Iterative solutions that require less memory can be used. Due to limited resources, these content requirements lead to the limit on the number of harmonics, and the analysis of multi-stage amplifiers cannot currently be adopted by this method. The sweep may take several hours, and the time required for actual devices may be longer.

The third approach was investigated by Loechelt in 2000 and is called Computational Load Pull ( CLP ). In this approach, simulations (or measurements) of large signal transients are used to characterize intrinsic devices and tools are used to bring it all together for circuit evaluation. This approach has several advantages. Once a data set is constructed to characterize the intrinsic devices, it can be used in multiple circuit simulations. Of course, this approach also has disadvantages. Since RF Workbench is built inside the CLP tool, it can only be used for designs that are executed in this tool.

The issues with these approaches so far have been speed, RF bench functionality, performance, and setup time, as summarized in Table 1.

Table 1: Comparison of four methods for large signal simulation of TCAD data.

The fourth approach is to extract compact models from TCAD simulation data. The main advantage of this approach is that the simulation-based model uses the same procedures, extraction methods, and can use the same design as the measurement-based model. This allows the use of very powerful RF circuit simulation capabilities that have been developed and the original RF design. The disadvantages are the time required to run TCAD, the time required to extract the model, and the limitations on the compact models used. This is an important limitation because the TCAD simulation may contain physical characteristics that cannot be reflected in the compact model. There are two remedies for this deficiency, one is to create a user-defined version of the model with better physical characteristics, and the other is to use a table-based model. In order for this method to be practical, automatic extraction must be created to achieve rapid extraction of a large number of device models.

Since we know from Figure 1 that the best performance occurs in uncertain source and load matching, simulation must be performed at the entire source and load level to search for the highest performance point. Assuming that there are 60 source states and 60 load states that must be searched alternately, it may take about 300 power sweeps to determine the highest performance point.

Large Signal TCAD Simulation Example

TCAD simulations were performed for the device using Synopsys tools. The model was extracted using an automated method from those simulated data and the forward, reverse Gummel, I/Vs and CV characteristics were compared as shown in Figure 2.

Figure 2: Comparison of forward and reverse Gummel, I/V, and CV characteristics, with TCAD data in blue and model data in red.

The TCAD data is shown in blue and the model data is shown in red. The good match between the two shows that the model accurately reflects the original TCAD data. Figure 3 shows a comparison of the S-characteristic parameters. The good match again shows that the model accurately reflects the TCAD data.

Figure 3: Comparison of S-parameter features with TCAD data in blue and model data in red.

The model is used in a circuit similar to that shown in Figure 1. An algorithm that repeatedly sweeps the source and load planes is used to select the best performing source and load match. The resulting load plane efficiency is compared between measured data for similarly designed devices in Figure 4, with the black line being the reference measured data and the red line being the simulated data using the model.

Figure 4: Efficiency contours.

A comparison of the power sweep at the maximum efficiency point with measured data for a similarly designed test device is shown in Figure 5.

Figure 5: Maximum efficiency power sweep.

The power sweep plot shows excellent predictions of efficiency, output power, and gain. In addition, the comparison also shows the measurement results of linear error vector magnitude (EVM), adjacent channel power (ACP), and alternate channel power ( ALT ). These measurements show that the gain and phase relationships are well simulated. Accurate predictions of linear characteristics, EVM , ACP, and ALT are very important for today's wireless communication device designs .