Section 2: Comparison Between Measured Data and Factory-Provided Models

Measured vs. Simulated Data - MJL3281A

f_T vs. I_C

Below are the MJL3281A measured data of f_T vs. I_C and the corresponding data predicted by the SPICE model. The measured data are shown in green and the simulated data are shown in red. The solid lines are for V_CE = 10 V and the dashed lines are for V_CE = 5 V.

At an I_C of 100 mA (a typical bias current for the device), the simulated f_T is 3.45 MHz, while the measured f_T is about 21 MHz. For small-signal analysis, the simulated bandwidth of an emitter follower will therefore be low by about a factor of 6. Also, as V_CE is reduced from 10 V to 5 V, the simulation shows a greater reduction of f_T than the measurement does. Clearly there is much room for improvement in the ability of the models to accurately predict f_T vs. current.

C_ob and C_ib vs. Reverse Voltage

Below in Figure 3 are the plots of simulated and measured data of C_ob and C_ib vs. reverse voltage for the MJL3281A. Measured data are again shown in green and simulated data are shown in red. The solid lines represent C_ib while the dotted lines represent C_ob. The simulated data were not obtained directly from the simulator. Instead, they were obtained by substituting the model parameter values into the junction capacitance formulas given in Massobrio and Antognetti. The capacitance formulas are described in detail later in the section on parameter extraction, together with the model parameters they use. However, it's worth noting that one important parameter of the formulas is the extrapolated C_ib value at V_CB = 0. This parameter has a significant effect on the simulated f_T vs. collector current in the low-current region. Therefore, if the simulated values of C_ib vs. V_CB deviate significantly from the measured values, there will be a corresponding error in the simulated values of f_T vs. I_C in the low-current region. This turns out to be the root cause of the discrepancy between measured and simulated f_T for the MJL3281A in the low-current region.

It can be seen that the simulated data for C_ib are much larger than the measured data. This explains the discrepancy between the measured and simulated low-current values of f_T. It is also seen that the simulated data for C_ob are lower than the measured data. However, the effect of C_ob on f_T is minimal, so the error in C_ib dominates the f_T error.

β vs. Collector Current

Below in Figure 4 are plots of β vs. collector current comparing the measured datasheet values with the simulated values obtained from the OnSemi model with V_CE = 5V. The green trace is measured data and the red trace is simulated data.

It is seen that the simulated β vs. collector current shows a pronounced reduction with increasing collector current that's not present in the actual data. One excellent feature of this device is its nearly constant β from low currents all the way up to 7 A. As with the data for f_T vs. collector current, simulations will show performance worse than actual devices.

Measured vs. Simulated Data - MJL1302A

f_T vs. I_C

Below in Figure 5 are the MJL1302A measured data of f_T vs. I_C and the corresponding data predicted by the SPICE model. As for the MJL3281A, the measured data are shown in green and the simulated data are shown in red. The solid lines are for V_CE = 10 V and the dashed lines are for V_CE = 5 V.

Correlation of measured and simulated data is much better than with the MJL3281A, but could still be improved.

C_ob and C_ib vs. Reverse Voltage

Below in Figure 6 are the plots of simulated and measured data of C_ob and C_ib vs. reverse voltage for the MJL1302A. Measured data are again shown in green and simulated data are shown in red. The solid lines represent C_ib while the dotted lines represent C_ob. It is seen that the correlation between measured and simulated capacitances is nearly perfect. As mentioned in the discussion of the corresponding data for the MJL3281A, having accurate simulated C_ib vs. reverse voltage is essential to getting accurate data of f_T vs. current in the low-current region. But despite the almost perfect matching of measured and simulated C_ib vs. reverse voltage, the error in f_T at low currents is still not negligible. In the parameter extraction section, it will be shown that fine-tuning of the capacitance parameters of the model can result in significant improvements in the accuracy of f_T vs. current in the low-current region without sacrificing accuracy in the simulated data of C_ib vs. reverse voltage.

β vs. Collector Current

Figure 7 shows a plot comparing the datasheet and simulated β values vs. collector current obtained from the OnSemi model with V_CE = 5V. The green trace is measured data and the red trace is simulated data.

It's clear that the current gain obtained from simulation will be much poorer than actual device performance. Simulated current gain is only a little more than half of the actual current gain at low currents, and shows a gradual reduction with increasing collector current that doesn't occur with the actual devices. At 10 A, the simulated current gain is only a little more than one-third the measured value. Much improvement in the simulated behavior is possible.

Summary of Simulated vs. Measured Data

Checking the simulated data against the measured data for both device types shows that in most cases, the simulated data will be pessimistic compared to actual performance. In some cases, the differences are quite pronounced. The largest discrepancy is in the simulated data of f_T vs. collector current for the MJL3281A in the low-current region. At the typical quiescent current value of 100 mA, the simulated f_T is low by a factor of six. In Section 3, a procedure which matches the simulated data as closely as possible with the measured data will be demonstrated. This procedure will then be used to create new models. Lastly, the data for the newly-created models will be compared to the measured data using graphs similar to Figures 2 through 7. It will be seen that the simulated data for the new models are much closer to the measured data than that of the original models.