Interest in a "Power Board" devkit for homebuilt controller?

rhitee05

10 kW
Joined
Apr 7, 2009
Messages
841
Location
Upstate SC
I've been tossing around ideas in my head for a homebuilt controller for a while now. I'm an EE, it's what we do for fun. I finally got around to choosing some components and drawing the schematic. My plan is to design a board that will contain the FETs, drivers, and misc support circuitry (voltage regulators, current shunt, etc.). Instead of integrating a uP onto the board, I'll provide headers and connect to a ready-made logic board to serve as the brains. Since C code isn't my strong suit, I plan to use an FPGA board (probably these: http://www.knjn.com/FPGA-RS232.html), but a PIC/AVR/etc would work just as well.

There seems to be a fair amount of interest around here in tinkering with controllers and some home-brew, but I know from experience that not a lot of people have the skills needed to design the FET side of things (I've blown up my share). Since I'm already planning on making it a modular design, I was wondering if there was interest around here in something like this? I'm not an expert in controller design, but I have PCB design experience and have done enough work with FETs and power circuitry to be confident I can design something that will work. I won't attempt nor claim to make an optimal design, just one that works pretty well without destroying itself.

If there is some interest, perhaps we can work something out where I'll provide a few of these dev kits on a limited basis (bare board, populated, whatever). If not, perhaps at least we can have some discussion on the finer points of controller design for the benefit of all? I know there are others around with more controller experience and I. I'm interested in comments and suggestions from the community, and I'm happy to make the schematic and CAD files available when complete. I have no interest in making profit or going into production, just experimenting and perhaps contributing to the community.

Design Goals:
For my application, the design specs are 10s LiPo (42V hot) and roughly 75A peak phase (motor) current. The parts I've chosen will all be good to 60V, and that can be raised to 75V by making a couple of substitutions (FETs and properly rated caps). By making a slight change to the regulator, that could be upped again to 100V. I've done some calculations for power dissipation, and I think 100V/100A is probably doable with a bit more aggressive cooling, at least for bursts. Calculations are only an estimate until the real thing can be tested.

FETs, Driver, Caps
I've chosen to use IRF7749L2 FETs for my design. They're 60V FETs in a surface-mount can similar to those used in the RC ESCs, but a better part in a larger package. They have an impressive 1.1mOhm Rdson (less than 2 mOhms hot) and the can package gives them low inductance and good thermal properties. I'm not under tight size constraints, so I can also spread the FETs out to give them more copper area. I'm also planning on providing airflow. The board will be laid out for 24 FETs, although I think 12 will be sufficient for my needs. IR makes similar FETs for this package in 40, 60, 75, and 100V versions. The 100V version (IR7769L2) has a 3.5mOhm Rdson which compares well to the popular 4110s.

The driver will be the Linear Tech LTC4444. It has a strong gate drive and I like the built-in shoot-thru protection. One driver for each half-bridge and separate resistors to each gate.

By leaving space between each set of FETs, it's also possible to provide more distributed caps closer to where they're needed. It'll be a proper bypass array with 3-4 different values spread across decades, smaller caps very close to the FETs, etc. I think a well-designed (and well-placed) set of caps can perform well without needing 1000's of uF, but its an easy thing to leave a few unpopulated slots just in case more is needed.

Voltage Regulators
It'll add a couple bucks to the cost, but I decided to go with a switch-mode supply for this design to cut down the heat. A linear regulator would be something like 1/3 of the total dissipation under nominal design conditions! I like the LM5010, which can source up to 1A from supplies as high as 75V. It also only requires a couple of external parts, so it's not much more cost. The high output current means lets me use the 15V rail to power a fan, light or some other external gadgets. If there's interest in a 100V version, I can put in provision for a pre-regulator. I'll use a secondary linear regulator to provide 3.3V-5V as a logic supply to the control board.

Measurement Circuitry
The board will include a shunt and associated circuitry for measuring battery current and a circuit for measuring battery voltage. I've chosen pars fast enough to regulate current on a pulse-by-pulse basis. I'm also planning to include the phase voltage circuits necessary for sensorless operation, but I haven't decided yet if I want to measure each directly or just provide comparators (suggestions?). Since I plan to use an FPGA for control, the board will include one or two ADCs for these signals. I can provide separate analog headers for a uP or something with internal ADCs. I'm also going to include at least one digital temp sensor thermally coupled to the FETs so thermal protection can be implemented. All of this will probably reside on an SPI bus.

Since I'm planning for forced-air cooling, the board will also incorporate a couple of low-side FETs set up to drive fans off the 15V rail. The intent (eventually) is to provide PWM control to the fan(s) based on the temp sensor(s). It could also be used to drive a light or some other accessory.

Thanks to anyone who actually reads this through. Let me know what you think!
 
sounds like your board will be very adaptable. it just may suit some of the extemists. i'm interested in seeing more.

i would like to see the provision for a pre-regulator to allow for higher input voltages. being able to adapt just by populating a few parts on the board would be a nice option. the extremists like to run very high voltages. for some even 100V isn't enough.

speaking of the extremists. for them the surface mount chips may not be enough. heck TO247's on a massive copper heatsink with forced air cooling would not be enough. personally though i would prefer to see the bigger packages like TO220 or TO247 because they are easier to replace if i happen to blow a couple.

rick
 
Eric,

Sounds good. I think it was in the 12kW motor thread that new controller designs were discussed and I said it would be worth splitting it into modules.

The thing to do is to draw up an interface spec between the modules. You might also want to think about splitting your part into smaller parts. Eg, the FET bridge itself with the minimum drivers as one module, and the accessory features as another module.

Nick
 
rkosiorek said:
i would like to see the provision for a pre-regulator to allow for higher input voltages. being able to adapt just by populating a few parts on the board would be a nice option. the extremists like to run very high voltages. for some even 100V isn't enough.

I don't think anything above 100V is practical using these surface-mount parts. IR does make 150V and 250V FETs in these packages, but the power density starts to become large. You'd also probably want 5-6 in parallel, at which point gate drive and switching power become an issue. You could probably get there if you wanted to go with liquid cooling, immerse the whole board in water or oil and use a radiator.

rkosiorek said:
speaking of the extremists. for them the surface mount chips may not be enough. heck TO247's on a massive copper heatsink with forced air cooling would not be enough. personally though i would prefer to see the bigger packages like TO220 or TO247 because they are easier to replace if i happen to blow a couple.

The calculations I've done so far indicate 100V/100A is probably doable with a 24-FET board and forced-air cooling. These are just calculations, so everything is subject to change upon collision with the real world. One favorable feature of these can packages is they are actually quite efficient at transmitting heat to both top and bottom surfaces, so for a high dissipation case like that you could actually heatsink both top and bottom and get a substantial increase in capacity.

Even around here, the >10kW market is pretty small! :) The design issues become exponentially more difficult as power rises. I don't need those kind of power levels for my purposes, so I'm just not really interested in spending the time needed for a good really high-power design.
 
Tiberius said:
The thing to do is to draw up an interface spec between the modules. You might also want to think about splitting your part into smaller parts. Eg, the FET bridge itself with the minimum drivers as one module, and the accessory features as another module.

It's a good idea for maximum flexibility. That could probably be a whole thread in itself - a discussion leading to a standardized open interface spec.

If I'm doing this just for myself, cost is king and that (usually) means that one board is cheaper than two. If there's some additional interest, volume makes the costs go down pretty fast. There are also some board shops that will let you perforate boards to separate later, so that's another option. I'm sure you could make a stack of boards in a smaller form factor than one larger board.
 
I'm certainly interested. You won't find much support for surface mount FETs here, but I will build a few and see how they perform. I have nothing against SMD power stages.

Feature set sounds great for a start. I would be interested in a 100v version, with an easy way to change low voltage cutoff and amp control.
 
johnrobholmes said:
I'm certainly interested. You won't find much support for surface mount FETs here, but I will build a few and see how they perform. I have nothing against SMD power stages.

I understand that, and I've read a number of past threads where it's been discussed. I understand LFP has made a hobby out of blowing them up! :D

These FETs are a newer part with much better specs than the ones shown on something like the Turnigy HV110 ESC (IR6648 I believe). For one thing, the Rdson is about 1/4 as much, so 2x of these devices are already better than 6x of the others. I think these are much more suitable for something in the mid-high-power range.

johnrobholmes said:
Feature set sounds great for a start. I would be interested in a 100v version, with an easy way to change low voltage cutoff and amp control.

For now, my only plan is to design a "power board" which would be mated to a separate logic board for control (BYO). My intent is to implement a controller in VHDL which will run on a commercial FPGA dev board. I would have no problems making the code available when complete (not anytime soon). I do intend to make the full complement of signals available, so the sky's the limit on functionality for any custom controller.
 
If you're trying to save money, doing it multiple times is always more $$. A few versions with SMD, then eventually going to a package that can be cooled. ;)

Remember, even if you only make 1/4th the heat (just an example), its still going to fail if the only places for that heat to go is either into the pcb traces, or out the back of an epoxy case with a thermally useless C/w.


That said, its super good to have more folks working on controler builds! I am super happy to have a smart guy like yourself helping to develop this stuff. The controllers are the weak link right now, and the market has a big gap in it. The more folks trying to fill that gap the better.
 
The last SMD board I got has 1/8" thick buss bars that connect from the PCB traces to the heatsink. There are ways to sink SM fets in all directions :wink:
 
johnrobholmes said:
The last SMD board I got has 1/8" thick buss bars that connect from the PCB traces to the heatsink. There are ways to sink SM fets in all directions :wink:


Pics! Pics!
 
Do the best you can! I want to see if they've got something that makes sense, or if they went the route of taking more board footprint and assembley labor to make an SMD work than it would have taken to use a coolable fet package. All the SMD spec sheets look so damn tempting on the spec sheets. It not until you're trying to cool something through copper foil epoxied to fiberglass that the painful reality hits. :(


johnrobholmes said:
I will see what I can do. Everything is glued together, so you can't see much :lol:
 
LFP, you should take some time to read up on these DirectFET packages from IR. I think they hold a lot of promise. A big problem with SMT, like you said before, is the lack of any thermal path to the top. The DirectFET uses a metal can with the die bonded directly to the top. Pads on the bottom of the die solder directly to the PCB, so you have good conductivity on both sides. The specs for the L8 package that I'm using give an Rth of 0.5 C/W to the PCB and 1.2 C/W to the can. That makes it quite reasonable to heatsink the top in addition to the sinking into the copper underneath. You'll still dissipate more heat with a bunch of TO-247s bonded to a huge hunk of metal, but it's a lot better than something like a D2Pak where all the heat is dumped into the board.
 
johnrobholmes said:
The last SMD board I got has 1/8" thick buss bars that connect from the PCB traces to the heatsink. There are ways to sink SM fets in all directions :wink:

That's an excellent idea. Not just the heatsinking, but it's nice to give the PCB some help with all that current, too.
 
No wonder that controller is so expensive, that's one very custom PCB! 3mm thick copper layer, wow! Those connectors are also very nice.

I like the concept of using a fuse as the current sense element. Dual-purpose and also probably more precise than other simple options - I'll have to check into that. Always nice to have some inspiration.
 
If I'm remembering correctly, he ran into no way to solve the exact problems I had warned him that would come up using SMT parts in a steady-state high current application, even with the super fancy pressed-lug board, and had to go to an out board through-hole power stage clamped to a sink (and running 1 giant SCR I think.)
 
regmeister said:
Wheras the board I just posted is a power board & obviously the shunt. The PWM was seperate.

I don't know how he incorporates the shunt in the new design, but you can see the terminals sticking our either side.
That looks like a fuse, to me. 250A; I have 100A versions of the same thing from powerchair SLA packs.
 
rhitee05 said:
No wonder that controller is so expensive, that's one very custom PCB! 3mm thick copper layer, wow! Those connectors are also very nice.

I like the concept of using a fuse as the current sense element. Dual-purpose and also probably more precise than other simple options - I'll have to check into that. Always nice to have some inspiration.

Yes, that PCB was expensive, however it worked very well for up to 200A continuous. I smoked a couple of those, one through excessive dV/dt and another one by too high impulse currents.

The problem is, it is quite difficult to desolder the broken Mosfets from the board as the 3mm copper plane sucks all the heat. I used one 80W iron to heat the internal copper and a second 40W iron to remove and replace the Mosfets.

I successfully tried IGBTs with my new P4 controller just because they withstand much higher voltage and for what they can do they are not that expensive. I wanted a scalable power solution for 100kW and beyond.

Yes, I used a Fuse for measuring the current on the first P4. That worked, however the resistance of the fuse (copper) varies with temperature, so your current measurement is not that precise. However, you will see more current as things heat up and thus will limit earlier ...

As for protection, this kind of fuse is much too slow. Things will blow up (Mosfets, battery cells) before this fuse will trip. You would have to use a faster semiconductor fuse ...

So for the new P4 I discarded the fuse altogether and measure the current with a manganin-shunt (stable over temperature).

Instead of the fuse I added a relay that switches the main contactor. One of the P4's micro-processors continually checks the status of the power transistors and if it detects a failure, main contactor will shut off within a couple of milliseconds.

I consider this safety feature very important if you have a larger power-system (especially on a motorbike).

Charles :)
 
Chucky said:
The problem is, it is quite difficult to desolder the broken Mosfets from the board as the 3mm copper plane sucks all the heat. I used one 80W iron to heat the internal copper and a second 40W iron to remove and replace the Mosfets.

Yes, I would imagine that's quite difficult, the copper plane would be an excellent heat spreader. It doesn't do much to get the heat out of the board, however. Was there a heat sink clamped to the back or something along those lines?

In the layout I'm working on so far, each group of 4 FETs is arranged in a square about 1" across. This is allows a heat sink to be attached to the thermally conductive top surface of the packages. There are a wide variety of small square pin-fin sinks made for BGA packages that are perfectly suited to this and which provide quite reasonable C/W figures (particularly with some airflow).

Chucky said:
Yes, I used a Fuse for measuring the current on the first P4. That worked, however the resistance of the fuse (copper) varies with temperature, so your current measurement is not that precise. However, you will see more current as things heat up and thus will limit earlier ...

As for protection, this kind of fuse is much too slow. Things will blow up (Mosfets, battery cells) before this fuse will trip. You would have to use a faster semiconductor fuse ...

Absolutely true, a fuse will not protect the FETs from overload. Most fast-blow fuses I've seen take seconds to blow even at 200% rated current, so a lot is smoking by that point. They are useful to limit damage (if, for example, a FET fails shorted). Pulse-by-pulse control, as you said, is the only way to guarantee protection.

I think having some temperature coefficient in the shunt could be considered useful. It will make the measurement somewhat less accurate but will serve to limit current as temperature rises. A useful characteristic so long as you don't require instrument-grade measurement accuracy.
 
After more thought, I think the trick for cooling is to make use of a heat spreader on each group of FETs then a large sink on top.

Each of these little FETs is about 1/4"x1/4" on top, so that's 1/16 sq inch surface area. The thermal resistance of interface materials is proportional to surface area. For example, one of the sinks I was looking at had an interface material that was .04 C-in^2/W. So, for 1/16 in^2, that would be 0.64 C/W. Maybe 1 C/W in the non-ideal real world. Not huge, but significant.

Instead, place a heat spreader on top of each group. Say, a 1" square piece of 1/8" copper or aluminum sheet. If you used copper I think you could solder it directly to the cans for very good conductivity. The thermal conductivity of copper is very high, so this becomes very close to an isothermal surface. Now place the interface material on top (electrically isolating) - for 1 in^2 the thermal resistance is now ~0.04 C/W, pretty much negligible. A large extrusion heat sink goes on top of all 6 groups, or perhaps tight against the side of an aluminum enclosure. If you want to get really crazy, make it another copper plate with some pipes soldered to the back - water cooling!

IR has a very nice app note (AN-1059) in which they give a very thorough discussion of thermal modeling and rating for these packages. Using their model, I used SPICE to simulate the thermals of this configuration using a few rough assumptions. Using a 0.5 C/W sink the model predicts almost 9W per FET for delta-T of 100C above ambient. The chips are rated for 175C, so that's conservative.

If my loss calculations are anywhere near accurate, 9W per FET is well past the 100V/100A mark. This seems to show promise for those wanting more serious power. The model shows quite a steep increase in power handling as the sink resistance drops.

To be continued...
 
I've been doing some reading on gate drive circuits and found some interesting things. I'm hoping some of the real MOSFET experts will come by and weigh in!

Some documents I found very interesting and helpful:
Very thorough discussion of gate circuits: http://focus.ti.com/lit/ml/slup169/slup169.pdf
Specific analysis of the dv/dt-induced turn-on problem: http://www.irf.com/technical-info/whitepaper/syncbuckturnon.pdf
Parallel Gate drive and oscillation: http://www.microsemi.com/micnotes/APT0402.pdf
Includes a very complete discussion of power losses: http://www.analogzone.com/pwrt1208.pdf

The basic gate drive circuit (IC driver with series resistor) is simple and effective, but the only tuning adjustment you have is the resistor value. If you need to slow turn-on, you can increase the resistor, but then turn-off is also slower. As far as I know, there's no reason not to have turn-off be as fast as possible.

Solution: Asymmetric drive circuit (see below)
Asymmetric Driver.JPG

This behaves as usual during turn-on, but during turn-off the diode bypasses one of the resistors allowing you to adjust the two independently. Omit R3 altogether for turn-off as fast as the driver IC will permit. Faster turn-off reduces the switching loss and also helps to prevent shoot-through.

When driving multiple MOSFETs in parallel, it's advisable to have individual series resistors on each gate rather than a single resistor driving all gates together. This helps to isolate the gates and prevent high-frequency ringing between gates. Higher values of R provide more damping to eliminate oscillation, but they also slow the switching as previously mentioned. A solution to this is adding a small amount of inductance in series with each gate, perhaps in the form of a ferrite bead. This provides more damping at the higher frequencies where oscillation occurs (MHz, apparently) but has minimal effect on the gate drive itself at much lower frequency.

Another problem is preventing the dv/dt-induced turn-on for the low-side MOSFETs. When the high-side FETs turn on, the drain voltage of the low-side changes rapidly. This causes current to flow through the "Miller" Cgd capacitance and can cause the low-side to turn on in extreme cases (causing shoot-thru). Slowing the high-side turn on will prevent this, but at the cost of higher switching loss. Other solutions involve biasing the low-side gate to a negative voltage or providing a very low impedence "clamp" to ground.

I like the clamp idea. A simple implementation is to connect a small-signal FET from the power FET's gate to the source (ground). This FET is turned on after the power FET is turned off and provides a very low-impedence path to sink the Miller current to ground and keep the power FET from turning on. Apparently some high-power IGBT drives implement this internally, but none of the conventional driver ICs we use do.

With all this in mind, here's the gate drive circuit I plan to implement (for low-side drive):
Clamped Driver.JPG

It uses the asymmetric drive shown above with a bypass diode. R6 sets the turn-off impedence (common to all gates). The turn-on resistance is R6+R7/N. Instead of a ferrite bead, I placed a section of meander line on the PCB to add some extra inductance (best guess around 50 nH). This trace is directly over the return plane, so the loop area is still small for EMI reasons. Each gate has it's own clamping MOSFET. I found a nice dual-package SMT unit which has on-resistance of about 50 mohm and can handle pulses of up to 15A. This will provide a very low-impedence path and should prevent spurious turn-on. Control is important, though - if the clamp is turned on too soon, the gate charge will cause a huge current spike and probably blow it up. I set up the op-amp as a comparator, with a threshold of around 1V. Once the power gate is below that threshold, the clamp gate begins charging through a resistor, which should provide another 200 ns or so of delay. It has to turn on fairly quickly, though, so the clamp is in place before the high-side FET starts turning on. The clamp will divert the gate current for a short period of time at turn-on, but the series resistors limit this to a value the clamping FET can handle. The bypass diode on the gate ensures turn-off happens very rapidly. The other diode D1 is used to clamp the input voltage below the op-amp supply to avoid saturating it or blowing it up.

I did a SPICE simulation of this circuit and it seems to work quite well. The clamp sinks a pulse of current as the high-side turns on but keeps the low-side gate less than 200mV above ground. This would allow the high-side gates to be driven as fast as possible with no worries about Miller turn-on.

The high-side drive will have diodes for asymmetric drive, but omit the extra clamping circuitry.
 
Eric,
I like the idea of a homemade 3-phase power stage as the RC ESC's seem inadequate (already blew a K-Force 120A turning a 30in. prop). I'd like to put together a controller that is capable of driving some of the larger rc outrunners. I'm starting to experiment with the microchip dsPICDEM MCLV development board and the large Turnigy HXT. The board is limited to 48V and 15A but is enough to spin the HXT under no load. Depending on the software and jumper settings you can do both sensored and sensorless. So far I have had some success with sensorless operation. I just have to be careful not to put too much load on the motor or I will blow the FET's (6-FQB55N10's). I already did this once... I'd also like to fit the motor with hall sensors and try sensored operation as well.

http://www.youtube.com/watch?v=T6GeCL_rnDc

The development board makes it easy to access all of the PWM signals and as such I would like to drive a more capable power stage board(100V/100A), perhaps using a gate drive optocoupler (e.g. HCPL-3180). The DirectFET parts you mentioned are attractive yet seem hard to work with for the do-it-yourselfer. Anyway, a modular power board that includes the fets, gate drivers,caps, current shunts, and an abundance of heat dissipation is what's really needed. I know, easier said than done...

I'll continue to post any progress I make and will monitor this thread for continued inspiration.

Martin
 

Attachments

  • IMG_0429small.jpg
    IMG_0429small.jpg
    101 KB · Views: 1,862
  • mclv_small.jpg
    mclv_small.jpg
    203.9 KB · Views: 1,862
very cool Mauimart,

If we can create a brain board with power supply out of the a devkit reference design that would be great!

Good luck with your designs Rhitee05!

That would be cool if we could come up with a controller interface standard between the board you come up with
and the board that I come up with!?!?

What do you think Rhittee05?

In the 6 fet thread, I was thinking a ribbon cable with connectors between the two boards.

It carry these signals:

Phase A, B, and C
V+, V-
Shunt +, -
Gates:
V1H
V1L
V2H
V2L
V3H
V3L

What other signals do we need?
 
Back
Top