Group Actions
Engagement
  • Author Author: Jan Cumps
  • Date Created: Date Created
  • Views 2838 views
  • Likes 7 likes
  • Comments 14 comments
Related
Recommended

High Voltage GaN FETs - Part 3: Probing the LMG3410 Half-Bridge

Jan Cumps

The series will use high-voltage FETs in half-bridge configuration.

This time I look at the bridge's output signal.

image

I'm particularly interested in the overshoot, because controlling that is one of the claimed advantages of GaN technology.

 

The high and low side of the bridge are controlled by an opposite signal. Both signals have a dead time inserted when one signal goes low and the other switches high.

GaN semiconductors don't need a lot of dead time. The setup here is configured for a 50 ns band.

 

image

*1: I was sloppy when capturing this scope window. The falling edge of the input signal should be where the orange line is. Check comments below for more info.

Thank you jc2048 for spotting this.

 

 

The scope capture above shows the input (low-power) signals.

The white line is the input signal, with 22% duty cycle and a frequency of 200 kHz.

The yellow line is the driver signal for the high side FET, the blue signal for the low side one. Check the schematic below to get an idea of where the 3 signals occur.

(excuses for the mismatch in scope horizontal scale setting between yellow and blue. I only saw this after I disconnected my setup. They have the same amplitude.)

 

image

 

I'm going to probe the switcher's output signal. I'll zoom in on the overshoot positions, so you can see that the signal is behaving decently in the dead time.

 

image

This trace is a full high part of the output at 200 kHz, 22% duty cycle.

It gives an impression of the switching artifacts just before the edges.

 

A detailed capture of the rising edge overshoot looks like this:

image

The input signal is 60 V. The overshoot is approximately -8 V:

image

 

On the falling edge, there's less switching residue:

image

(the time scale is extended to the limits of what I can show on my scope screen. Check the full signal above to get an idea of how steep (50 ns) this ramp-down and overshoot is.)

 

image

 

The evaluation board that I'm using has reasonable filtering at the output. Let's check how the DC signal after that flter looks like.

image

The load I'm using is approx. 12 W (12 V, 1 A). The capture below shows only the AC (noise) component of the output.

I get close to 650 mV of ripple peak-to-peak.

image

 

Not surprisingly, the largest artifacts appear at the edges of the PWM driver signals. This is a reasonable DC signal.

Further filtering will smoothen it more - the board I'm using is to show how the GaN FETs perform, so it doesn't focus on output filtering.

 

image

 

 

 

 

Part 1: Several 100 Volts
Part 2: Test Setup with LMG3410 Half-Bridge
Part 3: Probing the LMG3410 Half-Bridge
Parents

  • Good stuff, Jan. Thanks for posting this.

     

    The high side FET anticipates the input signal and turns off 50nS before it arrives? Something tells me that might not be right. Line drawn in wrong place?

     

    I take it the waveforms are with the filter on the output. Are they with a load of 1A?

     

    What does the falling edge look like at a load current of 0.5A?

  • in reply to jc2048

    I expect that if I would re-take the capture, and not exaggerate the saved input signal,

    the white signal would look like the (drawn by me in Paint image ) red signal here:

    image

  • in reply to Jan Cumps

    Ah. I see. I was hoping you might say you'd discovered a way to send information back to the past and we could clean up on the stock markets.

     

    The reason I was asking about different loads was because I was curious about the fall in voltage when the top FET turns off. It looks very even and controlled. If there were no capacitance at the switching node and the switch was perfect, the coil would take the voltage down very rapidly in order to try and keep the current flowing and would hit whatever is clamping the negative excursion [before the low FET turns on] much quicker.

     

    So, possible theories.

     

    One is that we just see the capacitance of the FETs being discharged by the coil [and it's merely a coincidence that the period for the load current you've chosen is about the same as the dead period]. If I've got it right, moving a capacitor 60V in 40nS by discharging it at 1A indicates a capacitance of 666pF [perhaps 630 by the time we take off 15pF for your scope probe and a bit more for the board and a little bit more for the interwinding capacitance in the coils]. That's does seems like a sensible sort of value - silicon power FETs can have drain capacitance of up to several nF - but I don't know enough about GaNs to be very sure about it.

     

    Two is that there is some control of the FET drive to slow the turn off [and again it's a coincidence that it more or less matches the deadtime]. Not sure that would be linear - need to think about that one a bit more.

     

    The other thought I had is that if the control side is fast enough, it might be that the controller is actively controlling the descent to fit the interval. I don't quite believe that - I'd be impressed if it could. [The advantage, if you could pull it off, would be the clean handover to the bottom FET and the consequent reduction in emissions.] That you see some overshoot speaks against that one.

     

    So the test was to see what happens with half the load. If the capacitance theory is right, it will only fall 30V in the 50nS and then go down more quickly when the bottom FET joins in and helps. If it were being actively controlled, it would still take the 40nS to come down. Not sure about the other case.

     

    Of course it could be I've got this all wrong and you'd see something else entirely [which would also be interesting].

     

    Nice use of Paint there.

  • in reply to jc2048

    I'll try the test with half the load, . I was going to use the programmable DC load of the roadtest for that image .

    I'll have to go to the electric shop to buy a second bulb with the same specs as the one I'm using now and put it in series.

     

    The drama I have with my current lab setup is that I'm working right on the border of what my scope is capable of (Rigol DS1052E). Everything looks like 40-50 ns it seems image.

  • in reply to Jan Cumps

    I've just looked at the spec for your scope. It's very good value for money - I couldn't quite believe the price. But 50 MHz bandwidth is on the wrong side of marginal for this. [If anyone from element14 is reading this, we need a roadtest for a 1GHz scope with an active probe, please.]

     

    Traditionally, that would equate to a risetime spec of 0.35/fT = 0.35/50M = 7nS. Whether that applies to a sampled digital scope I don't know. It may do. The 1GSps is shared with the 100MHz version and gives enough oversampling that the traces should match an analogue scope fairly well if the sample timing is accurate and the signal processing is done nicely.

     

    It's not hopeless, but we need to be very skeptical about what you're seeing with the faster edges - that top GaN is turning on much faster than we think it is. And the shape of the waveform as it goes up is probably as much do with how a passive probe that matches a 50MHz scope behaves when it is hit by a 60V step with a rise time of a nanosecond or two (or maybe less) as what is going on in the circuit.

     

    The depiction of the falling edge is closer to the reality of what's happening in the circuit, so looking at different loads would still be interesting if you wanted to and had the time.

  • in reply to jc2048

    Jon Clift wrote:

     

    I've just looked at the spec for your scope. It's very good value for money - I couldn't quite believe the price....

    I've purchased it second hand, it costed me the money I got for selling my old CRT scope + I believe 10€.

  • in reply to jc2048

      wrote:

     

    I've just looked at the spec for your scope. It's very good value for money - I couldn't quite believe the price. But 50 MHz bandwidth is on the wrong side of marginal for this. [If anyone from element14 is reading this, we need a roadtest for a 1GHz scope with an active probe, please.]

     

    Traditionally, that would equate to a risetime spec of 0.35/fT = 0.35/50M = 7nS. Whether that applies to a sampled digital scope I don't know. It may do. The 1GSps is shared with the 100MHz version and gives enough oversampling that the traces should match an analogue scope fairly well if the sample timing is accurate and the signal processing is done nicely.

     

    It's not hopeless, but we need to be very skeptical about what you're seeing with the faster edges - that top GaN is turning on much faster than we think it is. And the shape of the waveform as it goes up is probably as much do with how a passive probe that matches a 50MHz scope behaves when it is hit by a 60V step with a rise time of a nanosecond or two (or maybe less) as what is going on in the circuit.

     

    The depiction of the falling edge is closer to the reality of what's happening in the circuit, so looking at different loads would still be interesting if you wanted to and had the time.

     

    I have done the exercise with a 100 MHz scope and different loads:.

     

    Load is 1 A:

    image

     

    Load is 100 mA:

     

    image

     

    I've used proper probing techniques this time, with the pigtail spring on the probe.

Comment
  • in reply to jc2048

      wrote:

     

    I've just looked at the spec for your scope. It's very good value for money - I couldn't quite believe the price. But 50 MHz bandwidth is on the wrong side of marginal for this. [If anyone from element14 is reading this, we need a roadtest for a 1GHz scope with an active probe, please.]

     

    Traditionally, that would equate to a risetime spec of 0.35/fT = 0.35/50M = 7nS. Whether that applies to a sampled digital scope I don't know. It may do. The 1GSps is shared with the 100MHz version and gives enough oversampling that the traces should match an analogue scope fairly well if the sample timing is accurate and the signal processing is done nicely.

     

    It's not hopeless, but we need to be very skeptical about what you're seeing with the faster edges - that top GaN is turning on much faster than we think it is. And the shape of the waveform as it goes up is probably as much do with how a passive probe that matches a 50MHz scope behaves when it is hit by a 60V step with a rise time of a nanosecond or two (or maybe less) as what is going on in the circuit.

     

    The depiction of the falling edge is closer to the reality of what's happening in the circuit, so looking at different loads would still be interesting if you wanted to and had the time.

     

    I have done the exercise with a 100 MHz scope and different loads:.

     

    Load is 1 A:

    image

     

    Load is 100 mA:

     

    image

     

    I've used proper probing techniques this time, with the pigtail spring on the probe.

Children