Choosing an SiC FET part

[davekni]

So far my only SiC FET projects have been my two class-E HFSSTCs.  I've been searching a bit recently for good SiC FETs to purchase for H-Bridge use.  Found this part that looks interesting to me, IMZA65R015M2H:
    https://www.infineon.com/dgdl/Infineon-IMZA65R015M2H-DataSheet-v01_00-EN.pdf?fileId=8ac78c8c8d2fe47b018dd63bcd9652eb
    Cost of 10 parts from Mouser or directly from Infineon is $170.80 (+ $4.99 shipping if ordering from Infineon).
This is a 650V part.  Plenty of margin for 240VAC line use.

What I like in specifications:
    395A peak current, higher rating than many IGBTs.  75A continuous at 100C case temperature.
    Specified for 0-18Vgs drive.  (Gate threshold is typically 4.5V, higher than most SiC FETs.)
    79nC typical total gate charge at 0-18Vgs.
    4-pin TO247 package for Kelvin source connection.
    Reasonable Vgs max limits, transients to -10/+25Vgs, DC to -7/+23Vgs.

The one specification that is less than ideal in my thinking:
    Diode peak current of 395A is good for only 250ns.
For phase-shift or pulse-skip use, gate needs to be turned on during free-wheeling times.  Would reduce power dissipation for any FET, but is more necessary with this part.

Any thoughts about this part?  Other good or better options?

For anyone who has purchased SiC FETs on EBay, how do you avoid counterfeit Chinese parts?  (Relabeled or from an inferior quality fab.)  Or is the SiC market too new to have counterfeit parts yet?

Would appreciate feedback.

[Anders Mikkelsen]

My recommendations would depend a bit on the use case, but the proposed part is not a bad one. My experience with Infineon parts has been quite good. One thing to note is that their parts tend to have a bit higher Rdson temperature coefficient so a fair comparison with other devices should account for your expected junction temperature and current (both of which affect conduction loss)

Having a high pulse current rating in the datasheet is nice, but I wouldn't place a very high weight on this number as it might include some unspecified reliability qualifier that's not consistent between manufacturers.. Conduction loss as a function of current and temperature, along with transient thermal impedance, is a better way to gauge the robustness of the parts, but the Idm might come from these specifications in the first place, I didn't look too deep into it.

I see you chose a part with a kelvin source, which has a pretty major impact on Eoff, I would definitely stick with that package if you look at other parts as well.

Body diode conduction is not recommended for significant conduction angles both due to the high conduction loss, impacts on reverse recovery charge, and also some reliability concerns. For reasons of stacking fault growth (BPD degradation) the body diode is usually only rated at 1000h operation at the rated current. This is not usually a concern with common drive schemes, as long as you keep dead-time moderate. Diode-emulation mode to have DCM in synchronous converters is not recommended. Some bricks do have antiparallel SiC schottky diodes and this would be less of a concern with those. Some Toshiba parts also have a lower Vf body diode built into the structure. One last note on this is that SiC schottky diodes are not actually pure schottkies, they also have minority carrier conduction at higher currents, so using them with very high crest factor current waveforms is not recommended for the same reasons. For phase shift operation, body diode conduction should only happen during dead-time, and for pulse skip operation you need to override the switching states rather than disabling the gate drivers. This also gives you the freedom of distributing the conduction losses by alternately using the upper devices to short out the tank during pulse-skipping, or by inverting the control signals if you want a faster envelope decay rate.

I didn't have major issues with counterfeit parts, but I only purchased Wolfspeed parts second-hand and these have an uncommon package overmolding shape so it's easy to tell if they are genuine. There are also more of them on the surplus market since they have been marketing these parts for longer. I particularly like the older 900 and 1000 V devices as they tend to have a better Rdson(T) vs Qg FOM. They are not as area-efficient as newer devices so they are expensive when bought new, but they can sometimes be found cheap. The only issue I had with second-hand parts was buying some older generation parts on AliExpress that seemed to have been desoldered from boards, but they were genuine.

On voltage rating, I would also consider higher voltage parts, as they tend to have a better FOM, the Rdson per die area doesn't rise as quickly as Vdsmax up to 1200 V, so if you can operate at a higher bus voltage you get more bang for your buck. The main reason for this is that the blocking region forms a small fraction of the Rdson for 650 V parts, so Rdson doesn't rise as drastically with an increase in the blocking region thickness. This also depends on the device generation and technology, Wolfspeed 900 V parts are particularly good in this respect.

Paralleling devices is also a valid strategy to get the target Rdson, and this makes the Rth_cs, allowing you to get away with Al2O3 insulators instead of AlN for example, also making paste selection less critical.

If you're interested, I have a lot of C3M0030090K, and I could sell them for 5 dollars a piece or 120 per tube of 30. Plus shipping from Europe, which looks to be around 25 dollars for up to half a kilo. These can be used with doubled mains for example, and I measured the onset of avalanche to happen around 1250 V. Device breakdown voltages usually come from terrestrial neutron flux FIT rate targets (under one per 1e9 device hours typically) and not from the onset of uncontrolled avalanche. The avalanche voltage rises with junction temperature as well. Max Vgs comes from the gate oxide lifetime targets, usually 1e7 hours, and there is some conduction loss benefit to going higher if you can accept lower lifetime. Negative Vgs limits are much lower than for Si MOSFETs, as the short blocking region of these parts leads to the blocking region E-field extending into the gate oxide, you can go outside the limits as long as drain voltage is not present.

[davekni]

Thank you for your detailed reply including bits of device physics!

One thing to note is that their parts tend to have a bit higher Rdson temperature coefficient so a fair comparison with other devices should account for your expected junction temperature and current (both of which affect conduction loss)

Yes, I did notice the bit higher temperature coefficient.  Based my net evaluation on rated continuous current at 100C case temperature.  75A is higher than most SiC FETs, higher than any I'd found around 79nC gate charge.

This also gives you the freedom of distributing the conduction losses by alternately using the upper devices to short out the tank during pulse-skipping

Yes, I used this alternation in my short-lived phase-shift QCW using IGBTs.  (Even with fast IGBTs, data sheet calculations suggested power would be too high for QCW pulse width.  Thought it might be close enough to try.  I had extra IGBTs.  Repaired H-Bridge and run from buck converter now.)

On voltage rating, I would also consider higher voltage parts, as they tend to have a better FOM, the Rdson per die area doesn't rise as quickly as Vdsmax up to 1200 V, so if you can operate at a higher bus voltage you get more bang for your buck.

Yes, I know that SiC is more efficient at higher voltage.  But I don't use line voltage doublers or diode bridge directly to bulk caps to avoid low power factor.  Most of my use is for rectified 120V or 240V line or after a PFC stage.  Could make PFC generate higher voltage.  But I have plenty of 450V caps and am more familiar with working at that voltage and below.  So far only my class-E and ZVS (Mazzilli) designs needed higher voltage switches.

If you're interested, I have a lot of C3M0030090K, and I could sell them for 5 dollars a piece or 120 per tube of 30. Plus shipping from Europe, which looks to be around 25 dollars for up to half a kilo.

Although not a part I would choose normally, $145 per tube is a good deal.  I'll send you a message about details.

Negative Vgs limits are much lower than for Si MOSFETs, as the short blocking region of these parts leads to the blocking region E-field extending into the gate oxide, you can go outside the limits as long as drain voltage is not present.

Thank you for that explanation.  Do you have any estimate of how tolerated Vgs varies with Vds?  Would there be any allowable Vgs improvement at 400Vds instead of rated 900Vds?  Of course life time is a big help too.  Most projects run for a few hours total ever.