This article focuses on measurement techniques to accurately evaluate high performance GaN transistors and evaluates high bandwidth differential probes for use with non-ground-referenced waveforms.
The increase in switching speed offered by GaN transistors requires good measurement technology, as well as good techniques to capture important details of high-speed waveforms. This article focuses on how to leverage the measurement equipment for the user’s requirement and measurement techniques to accurately evaluate high performance GaN transistors. The article also evaluates high bandwidth differential probes for use with non-ground-referenced waveforms.
In order to demonstrate measurement techniques and requirements for the various types of GaN power devices, we will look at the following EPC eGaN FETs: a (i) high speed, 10 MHz switching frequency, 65 V eGaN FET EPC8009 (Q1 and Q2 in figure 1) based half-bridge board, and a (ii) lower speed, 500 kHz switching frequency, EPC9080 half-bridge demonstration board which uses the 100 V eGaN FET EPC2045 as the top switch (Q1) and 100 V EPC2022 as the bottom switch (Q2). Both boards are configured to operate as buck converters as shown in Figure 1.
Impact of bandwidth on measurement
The highest bandwidth available from the combined scope and probe system is given by :
where BW-3dB, BW-3dB,Scope, and BW-3dB,Probe are the bandwidths (in Hz) corresponding to the system, scope, and probe, respectively. For this article, a 2 GHz oscilloscope (Tektronix MSO 5204) was used. The passive probe (Tektronix TPP1000) offers a maximum bandwidth of 1 GHz. The lower value between the scope and probe bandwidths has the greater effect on system bandwidth, which in this case is 1 GHz.
When evaluating the layout of PCB designs, typical measurements include rise and fall times, peak overshoot, undershoot and the expected switch node rising edge ringing frequency, which can be estimated by using the ringing frequency equation:
In Equation 2, Lloop is the high-frequency loop inductance comprising of the high frequency decoupling capacitors, the eGaN FETs (Q1 and Q2), and the PCB connections of the components. Co2 = Coss + Cpar, includes Coss, which is the output capacitance of the bottom side FET Q2 at the Q2 blocking voltage and Cpar is the parasitic and probe capacitance at the switch node. Lloop is estimated to be approximately 200-300 pH for the demo boards considered in this article . Coss is 30 pF for the EPC8009 in the test voltage range , and Cpar is roughly about 10 pF for this demo board. This translates to a ringing frequency fr1~ 1.6 GHz. For the larger capacitance EPC2045 and EPC2022 based design, the ringing frequency is estimated to be fr2~ 0.44 GHz.
It is clear from  that the highest system bandwidth available to us is below the ringing frequency of the EPC8009 based design. We will now observe how choice of system bandwidth affects the capture of the switch node waveform for higher speed GaN transistors, such as the EPC8009, and relatively slower switching GaN transistors, such as the EPC2045 and EPC2022.
The measurement system is acting like a low-pass filter, attenuating the high frequency content. This is demonstrated in Figure 2 (top). It is observed in Figure 2 that the rise times of the captured waveforms vary significantly. This can be attributed to the relationship between the system bandwidth and the rise time according to the following Equation :
The fastest rise time captured in Figure 2 (left) is roughly 0.4 ns, which corresponds to a system bandwidth of ~1 GHz. Using the same probe and oscilloscope with the 500 MHz bandwidth digital filter, the measured rise time is 0.8 ns. Clearly the rise time of the measured signal is limited by the system bandwidth. Since the measured rise time is equal to the calculated system rise time, the input signal is faster than the measurement system rise time. Therefore, the input signal rise time is likely much less than 0.4 ns.
The measured ringing frequency (fr1) is 1.176 GHz for the EPC8009 based design, made with the highest bandwidth 1 GHz probe. The lower bandwidth cases shown in Figure 2 (top) further degrade the accuracy of the ringing frequency measurement. When looking at the peak voltage overshoot, it is also clear that lower bandwidth measurements underestimate the peak voltage across the switching devices. For timing dependent dead time measurements, the system bandwidth is also important. Shown in Figure 2 (top), the dead times are visible for the bandwidths of 500 MHz and 1 GHz, although not very precise for measurement. At lower bandwidths, the dead time periods are almost non-existent. Table 1 shows the system bandwidths impact on the ability to make critical measurements for the highest speed, EPC8009-based board.
Figure 2 Effect of probe/system bandwidth on captured waveform (EPC8009-based board, top, and EPC9080, bottom)
The other test case is shown with the EPC9080 demo board, which has a much lower ringing frequency and switching speed from lower on-resistance and higher capacitance eGaN FETs . The corresponding waveforms are shown in Figure 2 (bottom). The ringing frequency (fr2) of 438 MHz and its amplitude measured with the 1 GHz (blue) probe is valid, since f r2 is below the -3dB frequency of the system. The 1 GHz (blue) and 500 MHz (green) waveforms capture all the details accurately. However, for system bandwidths of 350 MHz (orange) and 250 MHz (brown), f r2 is above the system bandwidth. As a result, it captures the ringing waveform shape, but the attenuation on the ringing is evident, thus underestimating the overshoot. The rise time measured by the different system bandwidths are roughly ~ 3 ns. The lowest bandwidth we have used is 250 MHz, which corresponds to a rise time of 1.6 ns according to (2) and for all of the cases the rise time can be captured accurately. This discussion is summarized in Table 2.
[Continue reading on EDN US: Measurement techniques]
Suvankar Biswas is a senior applications engineer, David Reusch is executive director of applications engineering, and Michael de Rooij is vice president of applications, all at Efficient Power Conversion.