Connector interface, solid state swtiching

[CarbonWerkes]

Hello

I have an application where I will need to employ 8-10 low-side mosfets for simple DC power switching- replacing some conventional mechanical switches in an automotive environment. The expected current is 20A at 12V max. Id like to know if anyone has any practical experience with physically interfacing mosfets (Dpak, maybe T0220) to the low side of the switched load. Specifically, I dont think I can implement a conventional through-hole type edge connector (Phoenix, Molex etc), as I run into trace width minimums at the interface to the mounting hole. Is it resonable to just mount a connection tab very near the Drain on each low side switch, perhaps even soldered to the FET tab in question, and then just tie the Source side of the Fets to some bus bar to ground? This seems plausible, but with all things automotive, vibration and temp swings tends to throw a monkey in the works...

Regards,
R

 

[itsmoked]

Often you can use both sides of the board. This allows a big trace to the two outer pins and a big one to the center pin on the opposite board side.

A common cheap design method is how they build multi hundred watt inverters. They put the TO-220s along the edges then sink them to a finned extrusion that holds the board. Drop a lid on you have a box. If you can have a fan all the better.

You should consider picking up 500W inverter and learning from its assembly.

Keith Cress
kcress -

http://www.flaminsystems.com

 

[VEBill]

Or visit a junk yard and purchase some used modules that perform similar functions.

I've seen many examples where the TO-220 tab was soldered directly to the PCB traces. But probably not quite as much current as 20A.

 

[MacGyverS2000]

Make sure to use 2 oz. or better PCBs if track width is a concern with standard 1 oz boards.


Dan - Owner

http://www.Hi-TecDesigns.com

 

[danonboost]

What will the duty cycle be? That are going to be driving the gate with 10V? 4? I ONLY use surface mount FETS; TO-220s are a pain in the butt. Like Keith mentioned, use 2oz copper if you can.

I'll pick out some great FETs for you if you tell more. You shouldn't have any heat issues running 20A @ 12V with a low RDsON.

dan

 

[itsmoked]

Hey Dan suggested 2oz. and it's an excellent idea I might add, actually a requirement.

CarbonWerks; As dan mentions 20A is not a power issue 98% of the time. 40+ amps? Yes, that becomes a problem.

Check my projects page and look at the 300W charger shots. You can see what I did for traces there. The general situation is that, yes, the traces get sub par next to the actual leads but the leads themselves actually heat sink the heat created at the pinched traces. Also since copper is dah bomb for heat conduction the heat is rapidly drawn away from the warmer narrow trace areas to the wider areas.

I much prefer the DPAks for layouts but you will rapidly have heat removal issues at about +15A if you do not have forced air. With forced air you can go much further with the DPAKs. Forget them without air flow as you can't get the heat much more than an inch away from the DPAK tab on 2oz material. This makes giant board area around a DPAk a rapid diminishing returns situation.

I have just been through this with the depicted charger. The entire thing was DPAK'd but the customer decided to put it somewhere warm and stagnant. Which would toast it. So I redesigned it to be TO-220'd with some serious heat sinking. This did the trick.

(Sorry for the 'scroll hell' - suggest you "open in new window" to avoid it if it bothers you.)

http://www.flaminsystems.com/page/page/2456297.htm

Keith Cress
kcress -

http://www.flaminsystems.com

 

[danonboost]

To add to what Keith said, DPAKs are great for pulling heat out. Don't think in terms of trace width on power supply design. Think planes. The wider, the shorter, the thicker, the better. Always. 90% of an efficient, cool, and generally awesome supply is the layout (okay and the parts you pick

).

 

[CarbonWerkes]

Hey all-
Thanks so much for the very detailed feedback. The real nightmare is, the device will be exposed to nasty vibration- 50-120hz at 2-5g. So Im not sure that a design for an inverter, or even some automotive-rated stuff like power amps, translates well- at least in terms of ability to survive in a high-vibration environment.

The switches are 20A because 95% of the common loads in automotive are < 20A. I can parallel a couple if required to manage a special case.

Due to space and vibration concerns, Dpak was the plan; D2Pak may be a little large. I thought the TO220 might be useful though, because it would allow me to extend the Source leg to a bus bar for Gnd. Im not sure how well I can get 20A out of a DPAK pin-3 solder pad. Maybe the longer-legged Dpak variant (forgot the designator) would help there.

With RdsOn in the 8-15 milliohm range, my parts' heat dissipation problem is not terrible as long as I keep out of the linear zone (not really a concern with a simple switch like this). Gate voltage is undefined, but likely 3.3V. If required, I can use an LT and get it to 5V or 12V. The frustration is mainly that the heat-transport mechanism on these DPAK/TO220 NChannels tends to be the Drain. If it were the Source, then boy would my life be easier- I could just mount the things right to a chassis-mounted (grounded) sink and kill two birds.

Even though with no thermal concerns, I still have this nagging connector-fatigue problem to consider. Maybe it is as simple as using a terminal block with some 300mil traces (2oz copper). And some careful part placement. Or maybe I have to hardwire internal jumpers from the FET tabs (or near the tabs) to the chassis/bulkhead connector. With solder, though, I have seen tons of fatigue failures evolve over time in automotive- and I exoect this project would be worse because of the higher-than-typical expected G/vibration loading. Ugh.

You all think Im OK with just a terminal block? Maybe those screw-downs are pretty tolerant of vibration, and I dont have to worry about solder fractures. Any thoughts on this or any aspect of the problem(s) are apprecaited.

Regards,
R

 

[danonboost]

IPAK? Anyway, DPAKs can go well over 20A. I'd drive them with a lot more than 3.3V.

I ran across this on a quick search...

http://www.fairchildsemi.com/ds/FD/FDD8870.pdf

Gate charge is pretty high; still not sure what you're driving it with.

dan

 

[MacGyverS2000]

For vibration, do not count on the solder to hold things down. For your connectors, specify a type that includes mounts on the side (and if wide enough, a mount or two in the middle). You may also need to do a resonance analysis on the entire package to determine where the resonant frequency of the system is... if it ends up in that 50-120Hz range, you're in for a fight.

Dan - Owner

http://www.Hi-TecDesigns.com

 

[CarbonWerkes]

Dan/Dan

Ipak may be right. I know Dpaks are fine at 20A, Im just not 100% I can get 20A from a solderpad on the source side- that is a small foot on a Dpak. Thats why I was thinking a longer leg extended to a bus bar I can machine. I guess I can design for both, and switch from Dpak to the legged version with a bar if there are problems.

On vibration- you are saying a PCB locking mount then? As with a typical DB9 for example? I saw a few terminal blocks in the catalogs that seemed to have some holddowns- that might help. Board resonance is well controlled due to the PCB mounting strategy. My concern is any internal cabling (well, wiring, with 14-16AWG stranded wire). I suspect that, like with a stringed instrument, those wires are going to want to flap around a bunch- even if not at resonance. Maybe just some nylon ties would be enough. If I can do a terminal block approach, and therefore not use solder for board-to-wire connections, I suspect that fatigue cracks, and stress on the PCB mount holes, will be greatly reduced. But, that is only what I have read. So you guys think Im OK getting 20A from a trace to the pin of a terminal block? Id rather not have to do a 4-6 layer board just to get enough copper interface for that. Maybe a block with a wider pin spacing, but Im still limited to that pin diameter-meets-2mil-copper interface...

Regards,
R

 

[MacGyverS2000]

For 2oz copper, an external trace of 370 mils will have a power loss of about 0.6W over 2". Split that over a top and bottom plane and the individual trace width drops to 140 mils. If you're willing to let the trace temps rise 30C over ambient, tracks width drops to 70 mils. See a trend? It's not that unreasonable to get a working design...

Dan - Owner

http://www.Hi-TecDesigns.com

 

[danonboost]

Dan,

What are you using to calculate the loss over a given trace? (...hoping you have a link)

dan

 

[CarbonWerkes]

Apparently that just works out- since my calcs for current on a via/post indicate a 60mil throughhole with 2oz copper would get me to 20A. So a bit of a pinch from the 170 into the post, but I can live with that loss.

Danonboost, a popular calculator is at

http://circuitcalculator.com/wordpress/2006/01/31/pcb-trace-width-calculator/

Or just search by 'circuit width calculator' online-

Best,
R