OSHW TO247 IGBT watercooled laminated half-bridge

[Arlo1]

3d board2.jpg

IN this picture the trace that comes from the boost transistors on the top right side and wraps around the last igbt on the right side then runs from right to left.

[marcos]

IN this picture the trace that comes from the boost transistors on the top right side and wraps around the last igbt on the right side then runs from right to left.

Ah, I see. In theory its awful, it builds up all kind o parasitics. On the other hand, this is a tesla model S inverter, its 300+kw. Look closely at the red board, you can clearly see individual gate resistors, 2 on each gate, and the long copper traces. Those are the gate traces for all those igbts.
https://electrek.files.wordpress.com/20 ... &strip=all

And here you can see that mechanical assembly and copper stackup
https://www.google.com/patents/US20120305283

I try to keep that long trace wide for low resistance unbalance (the individual gate resistors should make that negligible). And I try to keep the inductance at a minimum by making the current loop minimal using the 0.15mm pcb stackup.

[helno]

Really nice looking design. Can't wait to see how it performs after manufacturing.

How much do you estimate it will weigh? I am looking for a 400 volt capable esc but only need around 100 A for a few minutes and 30-60 A continuous.

[marcos]

It should be fairly light. Its hard to tell the weight until the assembly is ready, I'd guess 3kg, just to say a number.

You're asking for a 40kw inverter, that is no small feat, keep in mind that in a "few minutes" you might already reach steady state, so plenty of cooling should be considered.

[fechter]

Great project. I like the pictures of the Tesla inverter. You can bet they did their homework designing that one. I always like to copy proven features.

[helno]

plenty of cooling should be considered.

It would be for an aircraft. So cooling air should not be a problem. I would prefer to avoid liquid cooling to get rid of the pump as a failure point.

[HighHopes]

about your link to tesla invverter. i have seen other tear downs to the inverter board and i'm impressed they can get that much current through there PCB. it doesn't even look like very thick copper... 2oz? impossible... yet that's what it looks like

[marcos]

about your link to tesla invverter. i have seen other tear downs to the inverter board and i'm impressed they can get that much current through there PCB. it doesn't even look like very thick copper... 2oz? impossible... yet that's what it looks like

That red pcb does't carry much current, it only drives the gates. Its connected to the sources and gates, no connection to the drains.

Below the red pcb there are copper sheets welded to the TO247 leads. Check the first image of the patent link above, it makes it much more clear.
And also this one:
https://patentimages.storage.googleapis ... D00006.png

[HighHopes]

copper sheets, sure, but they aignt using copper sheets in this one

[marcos]

Yes, thats an old tesla roadster half bridge. Its around 10 years old by now. Its suprising they rated that drivetrain at 180kw.
I recall doing the math of copper thickness and pcb size for that board, and it made sense at some point.
The thing I never figured out is how to get the current out of that pcb. I don't know which kind of connection they used.
Anyway, I don't like that approach, pcbs are not the right thing for that power level.

[bj001]

what's with the nuts almost shorting the drains of the fets?
p channel fet on the top?
I'm missing something...

Sent from my SM-G920I using Tapatalk

[marcos]

Those are IGBTs, and I don't remember what are those nuts for. On the other side there are several dc link+ snubber caps.
Top layer carries dc+, bottom layer is dc-. And both layers carry phase in the middle. Its a laminated bus, good stuff.

[HighHopes]

glad we feel the same way. didn't tesla put all their patents up for open use? perhaps they don't mind if we copy...

i know of a battlebots builder that uses discretes sintered onto copper sheets. lab tests show impressive results. there's definitely something worth investigating here

or

just drop in a large module. lower overall inverter volume, more robust for same power level. more expensive though but its not a whole lot more

[marcos]

Now that I sorted out the controller for this setup I should get back to this. Microcontroller stuff is fun and all, but I already know how to build those things, and gate driver design is a new challenge for me. Not much free time though.

So next time I get a moment of peace on this computer I'll start copying the right side of this board to the left side so bottom and top gate driver have the exact same layout. I didn't test the power supply, but I'm on a winning streak with power supplies so I'll just send it to mfg and give it a try

The day that I'll be stuck without a couple of rogowski probes is coming... I'm very aware of that. At least having full control of the controller inner workings allows for some deeper probing and debugging using firmware tools.

[marcos]

or just drop in a large module. lower overall inverter volume, more robust for same power level. more expensive though but its not a whole lot more

Heard you and the module approach became stage I. Now stage II is getting the actual thing done

[marcos]

Quick update, its looking nice and symmetric

On the control side I was using 3.3v but now its 5v for better noise immunity, and placed a bigger regulator so it doesn't heat up that much.

It needs a connector on the right for the current sensor and a way to measure temperature.

Here is an overlay so its clearer whats going on

The green stuff is a tesla patent, by default it switches slow and safe -no overshoots- but under the right conditions you can make it switch off harder for increased efficiency (and higher overshoots).

[HighHopes]

what i remember is that tesla controls the gate voltage to manage switch time which influences voltage transients. if that's the one you're talking about, i don't know how they can patent this as its an idea been around for ages. it was called dynamic gate driver back in the day, almost 20 years ago. i've seen all sorts of different incarnations over the years, some work, most don't. good luck! glad to see you're still working on this inverter, came a long way.

[marcos]

Maybe the patent describes a higher level something. I recall they use a particular feedback (or was it feedforward) overshoot management loop, and they happened to describe how the actual hardware was laid out, and with a couple of pictures I may have figured it out. Or maybe not, we'll find out way after I test the basic stuff.

Someone posted something about being silly about not seeing the desat circuit, and his signature linked to a good thread about a diy gate driver that I didn't read before. Spent quite some time today reading the complete thread on the phone and something caught my attention. A dude found that his desat does not work at high temperature because this diode sinks all the desat blanking current as reverse leakage current.

In a sense, that would affect me too.

It looks like at 100°C half the current coming from the internal 500uA current suply is sunk in the diode, so the blanking time last twice the time if I'm reading this correctly.

In the case of the avago gate driver that has a 250uA current for the blanking time constant, since the diode sinks 250uA it gets stuck on that blanking time and you get no desat protection. My TI gate driver would just have the desat disabled for 4us instead of 2us, for example, and completely disabled at near 120°C.

This is not going to reach such high temperatures. 100°C or even 85°C in the gate driver pcb is way too much for my taste, there are electrolytic caps on board that should live a long, cool life. But hey, this came from seemingly nowhere and it could have hit me big time if I had a poorly chosen schottky diode.

So thanks to the dude who pointed me to that thread. Schottky reverse currents are something to keep in mind.

[HighHopes]

good point. you probably don't need that diode at all if have low voltage transients anyway. i built many high power gate drivers without it.

[marcos]

good point. you probably don't need that diode at all if have low voltage transients anyway. i built many high power gate drivers without it.

In that case I'll keep the footprint but change the part# for one that I can use with confidence at 100°C

[marcos]

Diode changed for one with less leakage.

The gate driver pcb looks ready to me, its $85 for a set of 5 boards in seeedstudio. A bit higher if I choose 2 Oz copper. But before that I need to continue with the copper sheets, maybe next week.

3d board3.jpg (70.3 KiB) Viewed 230 times

[SplinterOz]

Looking really good.

[SjwNz]

Someone posted something about being silly about not seeing the desat circuit, and his signature linked to a good thread about a diy gate driver that I didn't read before. Spent quite some time today reading the complete thread on the phone and something caught my attention.

That may of been me ? I posted about not seeing desat in your Sch on page 1, but then I clicked on the link and saw your did have desat so deleted the post
as I felt a bit silly. I think it was Futterama, he did some tests and was talking about the leakage current through the Zener/schottky at high temperatures on my thread.
With the gate drive I was using (ACPL-333J) the desat current was only 250uA, but the newer version of that gate driver chip (ACPL-337) is now using 1mA.

[marcos]

Yeah, it was you but with the post deleted I couldn't remember your name or the thread name

So thanks sjwnz, it was a good read. I don't think there is a better place on the internet to get involved with these subjects.

[HighHopes]

did u use seeeds to fab the board too? 1oz is good pricing, but 2oz is more than 10x?