12V 50A PCB Help

[Ditiris]

I'm the only EE at my company and we need a board that takes 12VDC (battery from an alternator), so more like 14V, does some basic filtering and can switch on loads, and provides multiple outputs with current limits of 1A, 5A, 10A, 25A, and 50A. I need to monitor all of these currents and have short circuit protection.
I'm looking at NMOSFETs with a gate driver from a microcontroller. These are in Art of Electronics and there are copious notes here so I feel okay about the schematic side (I'm sure I'll be proven wrong when I post schematics later).
There are also some filters and current sense circuits in AoE - but I'll take whatever advice I can get.

Where I'm really struggling is PCB knowledge for transitioning these currents. I used some PCB calculators and for 50A at 55C, 30C rise, with top 2oz copper it computes ~17mm wide traces. But the NMOSFETs are 6x5mm... The shunt resistors are even smaller at 3.4mm. How do I transition between planes and these devices, and from these devices to some connector at the PCB edge?

Thanks in advance!

[Psi]

Do you need to design your system out of discreet mosfets/logic for cost? or can you use high side driver ICs ?
High side driver ICs would make this entire project very easy. (no shunt resistors, no gate voltage boost, automotive rated)
They are around $3.39 each for the high current ones.

eg, VN7003 is one I use on some automotive stuff. It would be great for your 10A, 25A, and 50A outputs.
https://www.st.com/en/automotive-analog-and-power/vn7003alh.html
You'd probably want something a little smaller for the 1A and 5A outputs. They make them in all sizes.

VN7003 is
- Designed for 12V systems, so rated for 24V continuous and 40 V max,
- This chip is rated for switching up to 135 A but that is unrealistic without like water cooling.
 I use it for 10A continuous with no heatsink and I'm sure it would handle much more.
- Like a lot of high current high side drivers the chip has its own built current sense output you can directly read with the MCU adc. So this means you dont need any extra current sense circuit, just read the current and impliment overcurrent shutoff in code.
- It has feature for reverse battery protection and can withstand some amount of transients and back EMF.
- They handle internally producing a higher voltage needed to turn the n-fet on, if you use discreet mosfets you must supply your own isolated or boosted voltage to turn the gate on. Which is annoying and costs money.

And there's something cool with paralleling mosfets/drivers.
Since paralleling a mosfet means around half the current flows through each you don't simply 'share the heat around', the total heat is 50% less using 2 instead of 1, and that reduced heat is also shared between the two chips, which is really nice.

eg, 50A through one 3.5mR high side driver would be 8.75 watts and you'd need a heatsink.
But if you use two chips you share that 50A, so 25A flows through each but the Rds is still 3.5mR.
So 25A * 25A * 3.5mR =  2.18W in each chip.  Which is easy doable with a large copper pcb fill as the heatsink.


As far as PCB layout copper goes. I use 3oz 6-layer to give me enough copper to handle switching up to 150A without having to use external heatsinks. I think I could get away with 3oz 4-layer, just have not tried it yet. Maybe even 2oz. My design doesn't need to be on for more than a few minutes though. So you'd have to do testing if you need your system to run continuously.
You're switching less current than me but you have multiple outputs. Maybe start out with 4 layer 2oz or 3oz and see how thermals go. Using thicker copper REALLY helps with thermals, but it comes at a high cost. So if you're making something at volume, or you have the time, then designing a external heatsink into the system will save money vs trying to use PCB copper to avoid needing one.
 
When going with thick copper you often want to only use it on the inner layers because the thicker you go the min track/gap spacing gets bigger and bigger to the point where you just cant use QFNs or find pitch parts at all. So your outer layers may not be able to use thick copper depending on what other parts you want to put on them. This makes it even more critical to do lots of via stitching to get the heat away from the thinner top layer into the thicker inner layers so it can spread out faster.

From memory 3oz is about as thick as you can go on top/parts layer before you start to get issues where you simply cannot use some SMT parts because the pitch is too fine.

[Ditiris]

Hi Psi, thanks for your reply. I typically do signal processing on FPGAs/GPUs/CPUs so I am way out of my wheelhouse here. Fun, but also scary as it needs to work.

Do you need to design the thing out of discreet mosfets/logic for cost, or can you use high side driver ICs.

No restrictions for the type of devices. This board needs to be solid as it's obviously a bad failure point for the system.

Probably pretty important info I left out in the previous post - this board has to be low power and the microcontroller will be asleep most of the time. Like, power everything on after a few months kind of thing (I think one year is the goal). So the quiescent current is an issue. The other issue is that this board also has to be in an environment that's 55C and must be passively cooled. I can definitely get the mechanical guys to give me a heat sink to the case and then attach that to a wall and get a good conduction path, but there is no airflow and the wall would also be 55C. So I was pretty much sorting NMOSFETs by R_DS_ON. I was already thinking of doubling the MOSFETs for redundancy - although it didn't dawn on me that yeah, 1/2 power is 1/4 heat rise, cause you know, equations.

So this means you dont need any extra current sense circuit, just read the current and impliment overcurrent shutoff in code.

Would this be fast enough to prevent voltages from dropping on the other outputs? I don't care about updating the current sense values super fast for the diagnostics but I was worried it wouldn't be fast enough to prevent drops in output voltages given the potentially large number of currents I'd be sensing (downside of doubling MOSFETs).

It has feature for reverse battery protection and can withstand some amount of transients and back EMF.

Oof, I missed this. In this system, the battery input is about the only thing they could hook up wrong. I definitely need to add reverse input protection from the battery. They can't hook up the loads wrong.
Got any reverse battery protection tips for the input? (I've never done this either)

So you'd have to do testing if you need your system to run continuously.

I think the worst-case runtime is about two days, so yeah, pretty continuous. I like testing.

As far as PCB layout copper goes. I use 3oz 6-layer to give me enough copper to handle switching up to 150A without having to use heatsinks.

What does your stackup look like? Would 4oz be better? I'm betting there's a cost tradeoff in here somewhere...

Make sure to put lots of via stitching around the drivers to get that heat spread out to all the layers.

Can you be more specific here - power-sized vias or something else?
Is this how you transition - do you just make giant traces with lots of vias and then neck down to the shunt/output/whatever?
Or are you stitching near inputs/outputs and running multiple traces in parallel through the layers to get that 1/4 reduction from i^2? I just realized we're fighting the resistance of the copper here, aren't we - it's all just about temperature rise?
If we get a board model back can the mechanical guys run this and accurately see what the temperatures would be?
Hopefully we can get a layout guy for this part but I want to have some basis of knowledge here.

Thanks again for your input!

[ejeffrey]

it computes ~17mm wide traces. But the NMOSFETs are 6x5mm... The shunt resistors are even smaller at 3.4mm. How do I transition between planes and these devices, and from these devices to some connector at the PCB edge?

Don't sweat the transitions.  Those PCB trace calculators are for "long" traces.  Having them attach to a smaller device pad is no problem.  The resistance of the necked down area is small and the heat generated will be easily conducted into the wide trace.

[Psi]

So this means you dont need any extra current sense circuit, just read the current and impliment overcurrent shutoff in code.

Would this be fast enough to prevent voltages from dropping on the other outputs? I don't care about updating the current sense values super fast for the diagnostics but I was worried it wouldn't be fast enough to prevent drops in output voltages given the potentially large number of currents I'd be sensing (downside of doubling MOSFETs).

If you want to make it faster you could do it in hardware using a comparator and logic gate on the high side driver to shut it off instantly without MCU processing time.
Or you could use MCU interrupt lines and it should respond within a small number of clock cycles. Only testing will tell you if these approaches are fast enough

Make sure to put lots of via stitching around the drivers to get that heat spread out to all the layers.

Can you be more specific here - power-sized vias or something else?
Is this how you transition - do you just make giant traces with lots of vias and then neck down to the shunt/output/whatever?
Or are you stitching near inputs/outputs and running multiple traces in parallel through the layers to get that 1/4 reduction from i^2? I just realized we're fighting the resistance of the copper here, aren't we - it's all just about temperature rise?
If we get a board model back can the mechanical guys run this and accurately see what the temperatures would be?
Hopefully we can get a layout guy for this part but I want to have some basis of knowledge here.

I can show you what i do, but im not saying that my design is perfect.

My 150A pcb is 6 layer all are 3oz. More copper would be better but cost is higher so I'm sticking with 3oz. (I should check cost of other layer thickness combinations and see if I can cut cost but have not done so yet).
The thermal vias are 0.3mm,  there is lots of them around the chips. (These are mosfets, not high side drivers, the VN7003 I used in a different product which only needs 10A)
My outputs are press fit M6 terminals (LFPE0506) so i use solid ground fill everywhere to carry current between input M6 to fet to output M6.
Here's one bank of fets (footprint for 6 in parallel but i only populate 4) Fets are TPH1R204
The other 5 layers are pretty much the same as the top layer, large copper fill joining vias on each side of the fets to the in/out terminal with some signal tracks in the gap under the mosfet.

It has feature for reverse battery protection and can withstand some amount of transients and back EMF.

Oof, I missed this. In this system, the battery input is about the only thing they could hook up wrong. I definitely need to add reverse input protection from the battery. They can't hook up the loads wrong.
Got any reverse battery protection tips for the input? (I've never done this either)

I solved reverse battery protection at 150A by simply having a large master relay to connect power to the board. The relay will only turn on if power is around the right way. Plus it's nice to have a proper mechanical disconnect. The relay never turns on/off with 150A flowing so the relay contacts don't get abused.  You can do it with mosfets as well of course.

[Ditiris]

Thanks, this is super-cool.

Or you could use MCU interrupt lines and it should respond within a small number of clock cycles.

Oh, yeah, good idea - that should be fast enough. If I have enough pins for this I think that would work. I'll have to look at how much the hardware costs, if it's less than a few dollars I'll probably go the hardware route.

LFPE0506

Beefy boys.

I solved reverse battery protection at 150A by simply having a large master relay to connect power to the board. The relay will only turn on if power is around the right way. Plus it's nice to have a proper mechanical disconnect. The relay never turns on/off with 150A flowing so the relay contacts don't get abused.  You can do it with mosfets as well of course.

Both of these will draw power all the time, right? Even if my loads are off? The real problem here is heat though... I'm not sure of any good solutions here... I might not be able to provide this feature.

[Siwastaja]

Be concerned about aspect ratio of connections. If your MOSFET leg is 1mm wide, don't worry about using a 1x1mm square to further connect it. Power dissipated on that square is tiny, it heatsinks from all corners. But if you connect 100 of such squares in series (100mm x 1mm track), then those trace width calculators model the situation better.

By far #1 advice on high-power layout is minimization of distances; miniaturization of design. The situation where the designer tries to satisfy track width calculators and as a result has to push things further away to make room for the wide traces is worst possible.

[Psi]

I solved reverse battery protection at 150A by simply having a large master relay to connect power to the board. The relay will only turn on if power is around the right way. Plus it's nice to have a proper mechanical disconnect. The relay never turns on/off with 150A flowing so the relay contacts don't get abused.  You can do it with mosfets as well of course.

Both of these will draw power all the time, right? Even if my loads are off? The real problem here is heat though... I'm not sure of any good solutions here... I might not be able to provide this feature.

yeah, for your application the constant current consumption of a relay being on may not work.

You can use mosfets, but you need two of them, or two banks of them for higher current, because you want to avoid the body diode issue. 
You could also look at solid state relays. See what options you have their for active current draw.

Or you could use MCU interrupt lines and it should respond within a small number of clock cycles.

Oh, yeah, good idea - that should be fast enough. If I have enough pins for this I think that would work. I'll have to look at how much the hardware costs, if it's less than a few dollars I'll probably go the hardware route.

You'll want to filter that signal a bit, so it only triggers an interrupt when current is high enough for long enough. Needs to reject current noise and random bursts of high current or inrush etc..

Some MCUs have a built in comparator with resistor ladder option on one of the inputs, so you can set the current limit in software while still getting a hardware based comparator.
Some MCUs even have a GPIO you can connect to the comp output pin, so you can trigger external things in hardware.
But you usually only get 1 or 2 comparator in the MCU.

I use the MCU hardware comp as a last resort/failsafe.
If current reaches a value above anything expected (including inrush), it triggers the comp and shuts everything down in hardware by clamping the photocoupler LEDs that are controlling the high current fets to ground.
The ADC is also running and monitoring the currents of everything and applying some "smarts" to things. eg. applying a X time-to-trip delay at Y amps to allow inrush for a short time but not continuously.

This means i get rapid response to a dead short circuit or anything catastrophic that is at risk of decorating the room with mosfets, but I also get a more normal slow-blow type response otherwise.

[max_torque]

Unless you are making a lot of these products and need to get the cost of each one down to the absolute miniumum (seems unlikely) then you'd have to be mad to not use one of the many, fully automotive rated high side driver ICs.  These do all the protection for you, are robust against surges and load dumps, and very easy to interface to a basic microcontroller, and critically, include all the complex gate drive, current mirroring, over-temp, over-current, under-voltage, over-voltage, loss of load and even reverse polarity protection in some cases!

Plenty to chose from, but i'd suggest infineon as you first stop, eg:  https://www.infineon.com/cms/en/product/power/smart-power-switches/high-side-switches/


Read and understand a few datasheets, and it should pretty quickly become clear that to do all the things those relatively cheap little power switches do, get anywhere near the level or reliability and robustness is going to cost you a huge amount of money in time and effort and certification and testing!

Not for nothing will every single automotive ecu you'll ever find include them as a matter of course!



For a robust reverse polarity protection for normal voltages (rather than surge) then a basic P channel mosfet arrangement can be used pretty simply. To get total ratings up into the 50 or 100A you'll need some parallel devices, but these days you can just about get a low enough RDSon for say 30V P channel fets to make it work with very little complexity (ie avoiding all the driving requirements of N channel devices in such a layout)

[max_torque]

This device

https://www.infineon.com/cms/en/product/power/smart-power-switches/high-side-switches/power-profet-plus-12v-automotive-smart-high-side-switch/bts50005-1lua/

I have used a lot, and you can have it sat their at 50A, jam a screwdriver across the output pins, and it pretty much just turns off, no damage, and no 'fast' controller intervention required

[Psi]

yep, if using a good high side driver you don't have to worry about damage from short circuits.

You can still implement your own soft current limits in code if you need, but the speed is no longer an issue, at least not from the perspective of protecting the driver itself. You may still want to disconnect the load quickly to prevent damage to the load itself, even if it's drawing current that's well within what the high side driver can handle.

For the VN7003 that I use the on-chip current limit doesn't kick in until 135A, but I do a software limit at 20A since the load normally draws 10A.

[mikeselectricstuff]

I've looked at a lot of these devices since they obsoleted my previous favorite IR3315.
What's particularly annoying is how few of them have a resistor to set the trip current. I also need 24V operation, which narrows the field further.
I ended up doing a descrete solution based on an ACS711 hall-effect current sensor, which has an evercurrent fault output, to turn off a P-channel MOSFET - still not adjustable but was about right for what I needed. The hall effect sensing means minimal IR losses in a shunt, and it will trip fast enough to protect a downstream SO-8 dual MOSFET from dead shorts at 24V.

[Ditiris]

Unless you are making a lot of these products and need to get the cost of each one down to the absolute miniumum (seems unlikely) then you'd have to be mad to not use one of the many, fully automotive rated high side driver ICs.  These do all the protection for you, are robust against surges and load dumps, and very easy to interface to a basic microcontroller, and critically, include all the complex gate drive, current mirroring, over-temp, over-current, under-voltage, over-voltage, loss of load and even reverse polarity protection in some cases!

Wish I'd seen this sooner, looks like what I'm attempting to build discretely...
I'm not going to be building a lot of these units, but I do need separate monitoring capability for 20-40 loads per power unit. They're reasonably price - I don't think it matters.
The only downside I see is the standby current seems high. That might be an issue, I'll have to do the calculations. Well, and JLCPCB doesn't have the parts I want...

I have used a lot, and you can have it sat their at 50A, jam a screwdriver across the output pins, and it pretty much just turns off, no damage, and no 'fast' controller intervention required

This may or may not have been my exact plan for short circuit testing...

[Psi]

Yep, high side driver ICs are awesome, use them unless you have a good reason not to.

In my case trying to switch 150A was bit much for them and I needed so many channels the cost reduction using mosfets was significant even for lower volumes.
But I'm always on the look out for newer cheaper high side drivers that I could switch to instead of mosfets.

[Ditiris]

After reviewing a few datasheets the only drawbacks I really see (for my very specific application) are

• these are substantially more expensive for low Rds(on) and still don't have Rds(on) as low as some FETs. I guess the lower Rds(on) variants just haven't made it in yet.
• Bigger issue for 0.65mohm guys is - Stock: 0. Lead time: 52 weeks

Maybe the discrete work has not been in vain.

[mikeselectricstuff]

What would be nice for maximum flexibility is a chip which combines with an external MOSFET to make an e-fuse - not sure if this exists, but haven't found one last time i looked :
Hall sensor for current sensing with minimal drop ( isolated so could go high or low side)
Resistor to set trip current from 5A upwards
Optional Resistor to set time/current curve/delay if require
Optional Resistor to set auto-reset time, or manual reset
MOSFET driver with charge pump to do high-side switching with N-channel
Configurable high or low-side switching
True open-drain outputs for status
Internal regulator for supply up to 48V

[Psi]

After reviewing a few datasheets the only drawbacks I really see (for my very specific application) are

• these are substantially more expensive for low Rds(on) and still don't have Rds(on) as low as some FETs. I guess the lower Rds(on) variants just haven't made it in yet.
• Bigger issue for 0.65mohm guys is - Stock: 0. Lead time: 52 weeks

Maybe the discrete work has not been in vain.

You shouldn't need that 0.65mR driver for 50A.  Two 3.5mR ones in parallel should be enough.
But yes, if you go the discrete mosfet path you have more options for low Rds and that makes heat management easier

[Ditiris]

if you go the discrete mosfet path you have more options for low Rds and that makes heat management easier

I need to have the MEs do some preliminary thermal analysis. The 20C/W numbers sound very reasonable to me if I keep Rds low enough.