2

A part we're using in a half-bridge LLC soft-switching DC-DC converter is getting hard to find (STW20NM60, 600 V, 20 A, 0.25 Ω, 192 W), and I need to find a replacement. What parameters do I need to care about?

The first thing to check is the voltage rating. The voltage supplying the converter is nominally 400 V, but can temporarily rise to 460 V or so. 600 V parts should be plenty, right?

According to Selection of MOSFETs in Switch Mode DC-DC Converters, it's probably dissipating around 10 W, so the >140 W power ratings of every TO-247 component I've found seem perfectly adequate.

So how about the current rating? I'm not sure how to calculate this. The AN2492 example circuit is for a 400 W supply, and they use a 14 A FET. I think the actual peak current through them is more like 3 A, though. Why such a big margin?

What other parameters are important? I think a "fast diode" version would be best for reliability and efficiency:

This revolutionary Power MOSFET associates a new vertical structure to the company's strip layout and associates all advantages of reduced on-resistance and fast switching with an intrinsic fast-recovery body diode. It is therefore strongly recommended for bridge topologies, in ZVS phase-shift converters.

endolith
  • 28,494
  • 23
  • 117
  • 181

1 Answers1

1

The gate characteristics are important.

The total gate charge will dictate the size of the gate resistor you'll need, and subsequently the power dissipation due to drive. It will also affect the switching speed of the MOSFET, which means if you choose a replacement with different gate characteristics, you'll also likely have to play with the gate resistor values in order to achieve similar switching characteristics.

The gate threshold voltage will also play a role in the switching speed of the device. Make sure your replacement is in the same ballpark as the original.

Other parasitics (\$C_{iss}\$, \$C_{oss}\$) are important in soft-switched topologies, but for an ordinary half-bridge shouldn't be too crucial.

600V should be fine for a 460V half-bridge, since the FETs only see Vin worst-case.

Big-current MOSFETs tend to have low \$R_{DS(on)}\$ values, and are often chosen to minimize conduction losses. There may be high peak currents during abnormals (transformer short, etc.) which may make the part appear to be over-rated at first glance.

Calculating the peak current without knowing things like the transformer inductance, switching frequency, etc. can be hard - it might be easier to just stick a current probe in there and measure it (or measure across any current-sense resistor you may find).

Fast body diodes may improve robustness (assuming there aren't discrete diodes on the PCB in parallel with the body diodes) and wouldn't hurt in my estimation, so long as their current rating is sufficient. Again, this is more important for soft-switching topologies.

Adam Lawrence
  • 32,921
  • 3
  • 58
  • 110
  • I don't have a current probe, but there is a current-sense resistor for overcurrent protection. What's the worst-case scenario for current testing? – endolith Sep 08 '11 at 16:01
  • 1
    Use a differential probe if possible, to avoid inadvertently earthing the primary. It's much safer than floating a scope. You should be able to use limited bandwidth (20 MHz) and still get a meaningful signal. – Adam Lawrence Sep 08 '11 at 17:18
  • Well... now that I have a diff probe, this looks pretty ugly. At idle, according to the 0.12 ohm current sensing resistor, there are spikes reaching +9 A and -21 A during the rising edge of the low side gate driver, and smaller current spikes during the falling edge. Shoot-through current? If the scope graphs in app notes are to be believed, these spikes should not be here. https://imgur.com/a/mEIag – endolith Sep 09 '11 at 20:50
  • I am skeptical of those spikes. There could be a lot of current freewheeling around during the bridge dead-times which could be picked up on your probe. I'd make sure that BWL is on and try to establish the maximum 'ramp' (non-spike) current across the resistor. – Adam Lawrence Sep 09 '11 at 21:24
  • Regular bandwidth limiting doesn't get rid of them, but scope's digital filter does. Then the hump during high load reaches 8 A, but there are still 12 A spikes in overload. https://imgur.com/a/kjq3h – endolith Sep 09 '11 at 22:00
  • 12A doesn't sound unreasonable considering they used a 14A part. I would go with a part rated 15-20A steady-state. – Adam Lawrence Sep 10 '11 at 12:32
  • I was looking at the Safe Operating Area chart, and it seems like the 20 A continuous rating is determined by the RDSon and the power dissipation limit? This would be the relevant value for the ZVS hump when there's no voltage across the FET? And there are also pulse ratings in the SOA chart, when there's D-S voltage, which would be more relevant to the spikes during transitions? – endolith Sep 10 '11 at 13:46
  • 1
    The SOA is generally based on power dissipation, yes - it's a function of how well the channel is established (how 'on' the FET is), the drain current and the current characteristic (DC, pulsed, etc.). The drain-source voltage influences how 'on' the FET is for a given gate-source voltage (see http://electronics.stackexchange.com/questions/18885/mosfet-as-a-switch-when-is-it-in-saturation/18903#18903) but ultimately, it's a thermal situation - if the part doesn't overheat with the existing cooling, it's most likely fine. – Adam Lawrence Sep 12 '11 at 15:48