PWM pushing 50 amps

[nicoverduin]

Hi,
I am doing a project where I need to handle 50 Amps through mosfets using Arduino PWM. I am using the PSMNR90-30BL which is an N-Channel Logic Mosfet capable of handling 90A with a RDSon of 0.001 Ohm. As these are surfacemounted Mosfets, I am planning to give the cooling through PCB on both top and bottom layer using stitching on 4Oz copper on both sides. The traces to and from the mosfet are at least 10mm wide on both sides. These trace will be connected to each other with vias with 100mil spacing.

But the circuit first:

The headers are for configuring the circuit to give each Mosfet its own PWM input or all Mosfets on PWM Cell1.

Some questions:

• Will this work?
• What problems do you see and suggestions to improve it

[CrossRoads]

Not all of us can open the link you included.

Modify your post, click Additional Options below, Browse to your locally stored .jpg, and Attach it.

[MarkT]

Never try switching really high currents with a MOSFET directly driven from a logic
chip. You risk destroying the Arduino.

With high power on the output of a MOSFET there is a lot of kick-back through the
drain-gate capacitance when it switches, capable of driving the full output voltage
onto the gate if it isn't driven by a low-impedance driver. This issue gets worse with
higher drain currents and higher drain voltages. I don't know what voltage you are
dealing with, nor can I see if you are planning 50A per package or 50 shared between
4 packages...

The correct way to handle high power with a MOSFET is to drive it with a proper
MOSFET driver than can hold its own against the gate-drain capacitance - these
chips can put somewhere between 0.1 and 10A directly into the gate depending on
which one. I've used the MIC4422 before which is available surface mount and DIP,
and is pretty pokey. Remember lots of ceramic decoupling on the MOSFET driver,
essential...

If you drive MOSFET gates in parallel you need to add individual gate resistors or
you risk inter-device oscillation (10 ohms would be fine).

Once you move to using a MOSFET driver you don't need logic-level devices, so the
choice of MOSFETs is much larger. You will need a 12V supply for the driver though.

[ Oh the other, important, point - a driver chip will switch the MOSFET nice and fast
so switching losses are reduced - at 50A switching losses from a sluggish switchover
can totally dominate heat dissipation. The MOSFET may be 0.001 ohm when its properly
on, but if it takes 10us to switch it'll be dissipating high power for a non-trivial fraction
of the time ]

[CrossRoads]

That part has pretty high Input Capacitance, 14850pF

PSMNR90-30BL,118 | Digi-Key Electronics

You might want to consider a MOSFET driver that is specifically designed to drive that so it switches from on to off & ff to on very rapidly and avoids dissipating power in the middle region where I^2*R heat build up could burn it out.

[system]

Will this work?

In principle, yes.

What problems do you see?

1. Availability of 4oz PCB, but you may have someone who will plate up to that;

2. Stability of 4oz PCB if you can obtain it (making it flat an keeping it flat could be an issue on FR4);

3. Switching losses with a simplistic gate drive; and

4. The ease of causing fire or explosion when handling currents this big! Especially when commissioning the gate drive PWM functions ]:D.

suggestions to improve it?

a. Consider single-sided aluminium substrate PCB material for the MOSFET stage, like http://www.bergquistcompany.com/thermal_substrates/t-clad-product-overview.htm. I can connect you with a UK PCB manufacturer who specialises in this, but you may find one in NL;

b. If you have 12V available in your system, consider an active gate drive circuit, either using discrete transistors or an integrated bottom-switch driver. I'm using the UCC27533 on my current project, (see http://www.farnell.com/datasheets/1712835.pdf); and

c. Consider a high-speed over-current protection circuit, with a low-value sense resistor between source and Gnd and a comparator to sense an excessive voltage drop. I'm not so familiar with the MEGA family, but on the XMEGA I configure either in-chip fault protection functions, or events (faster versions of interrupts), to shut off the gate drive if the source current exceeds a safe limit.

[MarkT]

The OP said they were building up to 4oz with copper foil on top of the PCB (well
that's the only interpretation I can make of

I am planning to give the cooling through PCB on both top and bottom layer using stitching on 4Oz copper on both sides

And a good point about explosions - always wear eye-protection when handling power
electronics, devices fail by exploding violently.

[ BTW a single device will be enough if its something like this:
http://uk.farnell.com/ixys-semiconductor/ixfn140n20p/mosfet-n-sot-227b/dp/1427319
]

[nicoverduin]

I will have to go through all of the comments, but as for the PCB 4Oz.. EuroCircuits can deliver it on 2mm PCB.
That is the easy part

[nicoverduin]

MarkT:
The OP said they were building up to 4oz with copper foil on top of the PCB (well
that's the only interpretation I can make of

I am planning to give the cooling through PCB on both top and bottom layer using stitching on 4Oz copper on both sides

And a good point about explosions - always wear eye-protection when handling power
electronics, devices fail by exploding violently.

The PCB will have 4Oz on both top and bottom layer. Each Mosfet is intended to supply around 12.5Amps each.

[MarkT]

2oz will be fine then.

[nicoverduin]

MarkT:
[ BTW a single device will be enough if its something like this:
http://uk.farnell.com/ixys-semiconductor/ixfn140n20p/mosfet-n-sot-227b/dp/1427319
]

I agree the device can handle it. However Even with clean switching it will dissipate I^2 * R = 50^2 * 0.018 = 45Watts which is a lot of heat to get rid off. And a decent cooling blok is quite expensive. Now they have an experimental solution with 4 x mosfet generating about the same amount of heat. The cooling blok costs about a 100 euro's. So any alternative is welcome

[system]

as for the PCB 4Oz.. EuroCircuits can deliver it on 2mm PCB

Ah, so you are increasing the through-pcb thermal resistance to decrease the accross-pcb thermal resistance. The net benefit is not certain, but chances are you will make a net gain, because if the PCB is large enough, you will be conducting heat better to a larger area, then through the PCB. I'm assuming here that you have a heatsink on the back of the PCB, not just free air. Thermal management is one of the trickier parts of power electronics... XD

I agree the device can handle it. However Even with clean switching it will dissipate I^2 * R = 50^2 * 0.018 = 45Watts which is a lot of heat to get rid off. And a decent cooling blok is quite expensive. Now they have an experimental solution with 4 x mosfet generating about the same amount of heat. The cooling blok costs about a 100 euro's. So any alternative is welcome

Hmm, so you need everything possible to reduce waste heat. Driving the gate at 12V, rather than 5V will reduce your on-resistance. What type is the cooling blok/heatsink, and does the heat go through the PCB to the blok?

[nicoverduin]

Billysugger:
a. Consider single-sided aluminium substrate PCB material for the MOSFET stage, like http://www.bergquistcompany.com/thermal_substrates/t-clad-product-overview.htm. I can connect you with a UK PCB manufacturer who specialises in this, but you may find one in NL;

My supplier has this stuff, but it can only deliver PCB's with 1Oz copper. To handle this much current I need thicker copper.

Billysugger:
b. If you have 12V available in your system, consider an active gate drive circuit, either using discrete transistors or an integrated bottom-switch driver. I'm using the UCC27533 on my current project, (see http://www.farnell.com/datasheets/1712835.pdf); and

I have 12V available. Interesting chip.

Billysugger:
c. Consider a high-speed over-current protection circuit, with a low-value sense resistor between source and Gnd and a comparator to sense an excessive voltage drop. I'm not so familiar with the MEGA family, but on the XMEGA I configure either in-chip fault protection functions, or events (faster versions of interrupts), to shut off the gate drive if the source current exceeds a safe limit.

There are already a number of safety checks. Do you think the ACS758 running at 150Khz sampling is unsufficient? Basically I am using PWM (Arduino about 600 Hz). On average I expect the circuit to only need about 20-25A. But peak current will reach 50A max.

[system]

My supplier has this stuff, but it can only deliver PCB's with 1Oz copper. To handle this much current I need thicker copper.

Er, why? The current-carrying capacity of a track is dependent of its cross-sectional area (you can make the track wider) and its allowable temperature rise. Its temperature rise is dependent on how well it can dissipate heat, and on Aluminium, a track can get rid of heat much better than it can on FR4. Aluminium power plates routinely handle well in excess of 50A with standard 1oz copper.

Do you think the ACS758 running at 150Khz sampling is unsufficient?

Oh no, that's perfectly sufficient!

Basically I am using PWM (Arduino about 600 Hz)

Then you're in the grey-area, and I'd use active switching at 12V with a series gate resistor (begin with 10 ohms) to limit dV/dt. If you have any wound components in your load, it could be quite noisy at 600Hz. With active gate drive, if you have sufficient PWM resolution, kicking the frequency up to 20kHz would cure that. But if sound noise is not an issue, it may be more efficient at 600Hz.

[nicoverduin]

Okay after drowning in all the great comments of which I am very grateful btw, I made a new piece of the schematic (just one mosfet but the principle is the same) leading to this:


Did I pay attention?

I still have to check that aluminum thing (my lack of experience in this). But I also need the thicker copper (I think to handle the current throug the traces) withoug creating 10 lane highways on the PCB

[CrossRoads]

Did I pay attention?

Not fully. I repeat:

Not all of us can open the link (the original one) you included.

Modify your post, click Additional Options below, Browse to your locally stored .jpg, and Attach it.

[nicoverduin]

Mea Culpa..... Done

[CrossRoads]

UCC27533 spec shows 1uF decoupling cap, your schematic has 0.1uF.

Not sure this part will be sufficient. The specs all refer to driving 1800pF load - the MOSFET you listed originally is 10X that, 18450pF.

ucc27533.pdf

That will translate directly into slower rise & fall times and greater heating.

[nicoverduin]

So this one would be better : PSMN1R0-30YLC about 7000 pf.

[system]

Did I pay attention?

Yes, very good, 9/10. Now connect pin4 to Gnd also (In+ must be high AND In= low for the driver to switch on the mosfet).

UCC27533 spec shows 1uF decoupling cap, your schematic has 0.1uF.

Ah, me too. I've put 100nF on my application too, but I'm okay with it for reasons which (I hope) will become clear.

The specs all refer to driving 1800pF load - the MOSFET you listed originally is 10X that, 18450pF.

Really? Not 14850pf? Anyway, the parameter which determines switching loss is gate-source charge, rather than input capacitance.

That will translate directly into slower rise & fall times and greater heating.

Hmm, well without any series gate resistance I'd partly agree. But the rise/fall time is going to be limited by the series gate resistance (that's its purpose) not the drive capabilities of the driver. This part will drive a couple of amps into the gate, given the chance, so with a 10 ohm series resistor and a 12V supply, you won't get anything like that, whatever the gate capacitance.

The reason for limiting the switching speed is that as the fet turns on, and the drain voltage falls, gate-drain capacitance can couple high dV/dt rates back into the gate causing oscillation, often seen as multiple switching edges on the drain if you look for them. The switching loss caused by multiple transitions can be far worse than the increased switching loss due to slightly extending the transition, (not to mention the EMC implications). There is a balance to be struck, and when you have the first board up and running, the gate resistor value can be fine-tuned by trial and error. There are several effects superimposed (by increasing the resistance, dV/dt reduction decreases oscillation, lower driver coupling increases freedom to oscillate, and increased resistance damps oscillation), so it's not something that is straightforward to calculate up front.

If two engineers with plenty of experience, but perhaps in different areas, disagree, then it's time to get this circuit onto copper and see where the smoke comes out.