Mystery CREE power module investigation

[Anders Mikkelsen]

I recently got a good deal on some Cree branded power modules on eBay, provided without much info. I gambled on these being MOSFET halfbridge modules, since Cree makes SiC MOSFETs mainly, but there was some risk that they were diodes. My hope was that they would be useful for a large QCW coil. I have managed to more or less conclude what they are, and thought I'd document the info and process here in case anybody else comes across some of these parts. I know that at least one other Tesla coiler has some in his possession.



Firstly starting with a look at the module itself, it is marked with Cree HT-3233-R GPN 400-00409-01 and a serial number. The part number gives no relevant hits on google aside from other auctions for the same parts. The box it came in has no more information on it either, aside from a date stamp from 2017, suggesting it was produced during or after this year.

Cree changed their name to Wolfspeed since these were made, and from their catalog the HM series of parts look very similar. Ratings are in the 1200 - 1700 V range, with current ratings in the range of 380 - 760 amperes. None of the parts match the part number on these, and the gate drive connectors are also different, so no dice. There was one more auction on eBay for the same part at a higher price, which I ignored initially, but revisting it I noticed a pictured sticker with some more info.



1700 V and 2.4 mohm, that sounds promising. Also a mention of a Google car division, explaining the GPN code (Google part number). This does not match any of the present parts from Wolfspeed, but it's close in ratings to the CAB500M17HM3 which has a full datasheet, this is a good start. Digging a bit further into old Wolfspeed presentations, I found similar parts described as HT-3000 series, which is at least a partial match for the part number. Searching for HT-3000, I found that the package was originally developed by APEI, who were later bought by Cree. These use an AlSiC metal matrix composite base plate and a Silicon Nitride insulator internally, for excellent thermal performance and lower thermal cycling stress compared to the usual Al2O3 over Copper. DC loop inductance is only 5.5 nH. All of this is pointing in the direction of these being very very interesting modules.

I managed to find some more data on the original APEI website via archive.org, where they have the HT-3231-R provided with a datasheet, found at https://web.archive.org/web/20160410104953/http://www.apei.net/Files/Download?name=HT-3231&v=635713839740000000 . The specified Rdson is much higher for this module, but at least this gives us a mechanical drawing matching the gate connectors and pinout for these. I found many further papers and press releases mentioning these parts, but nothing in the range of 1700 V and 2.4 mohm.

I contacted Wolfspeed, but apart from confirming the Vdsmax and Rdson, they couldn't provide further info as these were for an old customer project, and that they were likely 2nd generation parts. The previously mentioned CAB500M17HM3 is a 3rd generation part, so this opened up the question of how applicable the datasheet is to these. Knowing the device generation is especially critical, as the gate voltage is not the same for second and third generation parts. By this point I had received the modules from eBay, so I decided to do some measurements to try to figure it out. Gate capacitance measures 72 nF, which is higher than the CAB500, possibly suggesting older generation parts, and the Cgs/Rds ratio looked closer to second generation parts I compared with.

At this point, I came across a very interesting reference in a book on archive.org, containing details on a project by a company called Makani, who experimented with wind generation mounted on tethered kites. The idea was to have a 4800 V tether for power transfer, using motor inverters stacked in series. More info can be found at https://archive.org/details/theenergykite

Available Modules
Unfortunately there are a limited number of SiC modules on the market and even fewer 1700V
SiC modules. Those available for sample or purchase at this time are listed here along with
known useful parameters [internal ref].
A promising module is the APEI / Wolfspeed module fitted with Wolfspeed Gen3 1700V die.
APEI and Wolfspeed are currently investigating a 1700V module with 10kV baseplate isolation
(a change that would greatly simplify motor and controller chassis grounding). The above sheet
has extrapolations to Gen3 HT-3000 modules with and without 10kV baseplate isolation.

Unfortunately the mentioned table is not included, but I decided to dig a bit deeper still. Makani was apparently bought by Google in 2015, so this could have some connection to these parts since one of the labels had a Google logo printed. Going back to the label I noticed -MKANI- as part of the order code, so this is getting interesting. It seems likely that the note from the book refers exactly to these modules, suggesting they are 3rd generation, so some deeper investigation is needed.

I noticed that the CAB500 has Rdson specified as 2.16 mohm in the datasheet, while being listed as a 2.5 mohm part. I had assumed that 2.4 mohm was the real Rdson on these, but this prompted me to actually measure it. This part measured 2.12 mohm, which is suspiciously close to the CAB500 datasheet value. Recalculating the Cg/Rdson ratio put it between the G2 and G3 devices, so no clear conclusion there. Having another look at the datasheet, I noticed that the specified Rdson does not include the package resistance, which is 0.23 mohm, meaning that the acutual MOSFETs are 1.89 mohm rather than the measured 2.12. Now the Cg/Rdson ratio matches G3 parts perfectly. So we actually have a more powerful device than the CAB500.



Wolfspeed dont't make a wide range of different dies for these parts, and looking at their catalog it was clear that the CAB500 used eight 20 mohm dice, while this part has nine per position in parallel. The catalog also confirmed that this can't be a 2nd generation part, as the lowest Rdson available in a 2nd generation 1700 V die is 45 mohm, while pictures I found of the inside of this package show that the required number of dice to get 1.89 mohm would not fit.

The last piece of the puzzle is the baseplate isolation rating. While measuring the dimensions of these to draw a busbar structure, I noticed that they were significantly larger than any of the datasheets showed. The plastic part of the package has more overhang and multiple creepage slots, suggesting that Makani did go for the 10 kV baseplate isolation in the end.

So to summarize these are essentially CAB500M17HM3 but with nine CPM3-1700-R020E dice instead of eight per position, gate drive connections and pinout as described in the HT-3231 datasheet including the RTD temperature sensor, and increased base plate isolation up to 10 kV. That's a lot more than I had hoped for. I don't plan to go easy on these modules in terms of pushing voltage and current, and it should be obvious that a QCW type coil made with these bricks will be a serious endeavour, with an aim of breaking at least a few world records if I get it together.

[davekni]

Impressive find!  A few years ago at work I designed with 1200V 400A CREE SiC FET half-bridge bricks, but that project didn't survive even to prototype stage.

Looking forward to seeing your eventual QCW coil!  Will need a massive bank of bulk caps to feed it.

[Anders Mikkelsen]

Not bad at all for 60 dollars a module, even considering shipping and import to Europe. I still have a bunch of 30 mohm 900 V parts (well over a thousand) from another eBay find to play with as well, so I should probably consider selling some to the community as I have no chance of using all these parts. At least it makes it very low threshold for me to push the limits without worrying about losing some devices.

I feel like SiC doesn't get the attention it deserves in the *SSTC community, as the merits for this application are hard to beat. The general perception seems to be that they are expensive and hard to use, with unclear benefits. On the first point, I will gladly sell transistors to anybody intereste for a price that can't be argued with. On the drive aspect, galvanically isolated drivers and suitable DC/DCs are widely available and not expensive, and I plan to publish some designs for inspiration, in addition to the phase shift induction heater I already documented here which has a reference implementation of cheap and robust gate drive for SiC. On the third point, I feel the parts are at a disadvantage because the common approach with IGBTs is to push them until they explode and back off a bit, while the designers experienced enough to jump into SiC also have a mindset of designing with huge margins, partly due to the cost of new devices. But as mentioned there are tons of surplus deals, and prices are also coming down strongly with production scaling. At the moment there are over thirty companies offering parts on the open market so prices are rapidly coming down.

To quantify the merits, if you scale a SiC transistor to have similar conduction losses to an IGBT, switching losses will be reduced by roughly 85 % in a CW application. For very low duty cycle applications like traditional DRSSTCs, the quadratic conduction losses with current diminish the advantages somewhat, but for applications like buck converters, induction heaters and CW/QCW coils the difference is night and day. It's possible to make a 30+ kW buck with a single TO-247 transistor for example, or to get 25 kW from a single full bridge of TO-247s parts at 400 kHz in induction heating applications.

I would be curious to know more about your project with the bricks, if you are in a position to share information.

[Anders Mikkelsen]

I was able to confirm that these used the CPM3-1700-R020E dies, but eight of them. I was mislead by the gate capacitance and Rdson, these seem to be early generation dies with lower Rdson and higher Cgs compared to the nominal values.



This means that the datasheet for the CAB500M17HM3 should be applicable for these in full. It also seems that the Silicon Nitride insulator is the standard thickness despite the increased isolation rating, so the thermals specifications from the datasheet should also apply. The only remaining piece of the puzzle is the RTD behavior. It seems like the nominal resistance at room temperature is around 530 ohm. Temperature coefficient is likely around the standard 400 ppm/K but this is to be confirmed.

[rikkitikkitavi]

SiC switcher is taking over much of IGBT territory at larger power conversions but the very largest such as grid-line converters (HVDC lines at hundreds of MW) but they are coming in there too probably. Stack em high, as they say att ABB.

They will likely not be used for small off-line converters any time soon due to costs, more complex driving and small benefits from reduced losses.

But yes, for HV experimentation they are a gem, considering the drive issue.

Anders, it would be very interesting looking into some ideas of topologies. As you write, surplus is increasing and there is a lot of support circuitry around nowadays from commercial suppliers at reasonable prices.

[Anders Mikkelsen]

They will likely not be used for small off-line converters any time soon due to costs, more complex driving and small benefits from reduced losses.

But yes, for HV experimentation they are a gem, considering the drive issue.

Anders, it would be very interesting looking into some ideas of topologies. As you write, surplus is increasing and there is a lot of support circuitry around nowadays from commercial suppliers at reasonable prices.

It depends on how you define small, they are less common in applications below a kilowatt, most commonly seen in auxiliary offline flyback converters operating off rectified 400/480 V mains. In the few-to-tens of kilowatt range they are becoming more common in EV chargers for example, both on- and off-board, particularly in bidiriectional ones where hard turn-on is unavoidable precluding superjunction MOSFETs. They are becoming more common in solar inverters as well, as far as I know. Server and computing power supplies are also adopting SiC in the PFC, where the hard-switching loss benefits make them a good option. Gate drive is more challenging than with regular Si due to the asymmetrical voltage requirement, but there are standard SiC MOSFETs now that are characterized and recommended for 0 V Vgs turnoff, many from Infineon for example.

For driving, the most interesting option in my opinion are galvanically isolated gate drivers. These are usually based on capacitive or inductive isolation, and contain an integrated output driver stage as well. There are half-bridge drivers available with integrated overlap protection and dead-time insertion making them very easy to use, and they have better common mode transient immunity than gate drive transformers and none of the duty-cycle limitations due to volt-second balance. Compared to older junction isolation parts like the IR2110, they are not sensitive to high-side source undershoot, so they are a lot more forgiving when it comes to fast switching and layout issues. They also provide a galvanic barrier between the switches and control electronics, so failures usually don't propagate to the control side when something goes wrong. I've been using the 2EDB9259Y as they are very cheap in volume, but other parts to check out are UCC21520 https://www.lcsc.com/product-detail/Isolators-Gate-Drivers_Texas-Instruments-UCC21520DWR_C601651.html and similar parts from Silabs, Analog, 2Pai, onsemi and ST.

[rikkitikkitavi]

Good tips there. Yes, they seem pretty rugged, the integrated drivers nowadays.

Compared to the hassle of DIY your own low-leakage-while-retaining-insulation GDT they simplify much.

I was thinking of < 2kW off line converters but as you write development of SiC Mosfet goes on.
Maybe SiC will take over again from IGBTs?

[Anders Mikkelsen]

One way to make lower lekage GDTs without increasing interwinding capacitance is to use nanocrystalline cores instead of ferrite. I have not seen this discussed here, so I thought I'd mention it. They have much higher saturation flux density, which allows using less turns and therefore having less leakage inductance. For a given coupling capacitance, the transmission line impedance of the winding can also be lowered, further reducing leakage inductance.

One big problem with GDTs is that they have significant interwinding capacitance compared to isolators. With the edge rates you get with SiC parts, but also Superjunction MOSFETs and even some modern IGBTs, the common mode current coupled into the gate drive chip outputs can be problematic. As an example, a twisted pair from a CAT5 cable has capacitance around 50 pF/m and inductance of 500 nH/m. A small toroid has a mean lenght per turn around 5 cm, and with 10 turns that gives 25 pF of coupling capacitance. SiC MOSFETs tend to give a dV/dt of around 50 V/ns with practical values of gate resistors, superjunction MOSFETs can exceed 100 V/ns due to the highly non-linear output capacitance, and IGBTs tend to be under 30 V/ns. This gives common mode coupling currents of 1.25, 2.5 and 0.75 A respectively, values that can cause output stage latchup and damage in many gate drive chips.

With isolated drivers, barrier capacitance tends to be in the 1 - 2 pF range, greatly reducing the coupled current. Parts are also available with characterized CMTI behavior of 100+ V/ns, with some above 400. There is still the issue of current coupled through the stray capacitance of the gate drive power supply, but DC/DC converters are available with less than 4 pF of isolation capacitance and CMTI rates of 200 V/ns, for example the Murata MGJ2 series.

I think we will see gradual SiC adoption in certain applications, but not across the board. For converters below a few hundred watts, I would not be surprised if GaN will completely dominate within a few years. SiC diodes are already used in most boost PFC converters as of today, but the transistors are less common on the consumer market outside of onboard EV chargers, EV traction inverters and solar inverters. I think we will see them more in applications of home energy storage as that becomes more common, possibly also in induction stoves for heating non-ferrous pans if the market pressure is big enough to push that application.

[rikkitikkitavi]

If there was a thumbs up button I would use it for your last reply, Anders.

Very informative regarding GDT.

[Anders Mikkelsen]

If there was a thumbs up button I would use it for your last reply, Anders.

Very informative regarding GDT.

There is the "applaud" button next to each post

I managed to find some info on the internal RTD from one of the original project engineers, closing the last open topic on these bricks. The resistance is 500 ohms at 0 degrees C, with a temperature coefficient of 3850 ppm/K. It appears to be a platinum type RTD, so the higher order coefficients could be assumed to be similar to common parts.

I did get some more information around the project, which I'm planning on sharing when I have a chance to compile it.