Choosing a DC/DC converter topology for 500W, 40A max

[AnasMalas]

Hello! Here's something im trying to get into: power converters. I wrote a review paper for my bachelor's "research project" course talking about GaN switches (HEMTs) and device miniaturization, and now for my capstone project, I want to make a power converter to learn a bit. I spent a couple of days reading about the topic, went through every app note and tool I could find from Ti, and I think the following would be a pretty nice project

Project: A fully integrated, non-isolated DC/DC step up/down converter for battery powered robotics applications (motor drive), with controllable Vout for speed control, made using soft-switched GaN switches to get high power power in a small and light package relative to Si, possibly with an integrated "on-off" Si H-bridge at the output to allow reverse.

I want it to be rated 500W; 3 LiPo cells at 40A, 4 cells at 30A or 5 cells at 24A. So specifications would be:
Vin: TBD (min range: 10-23V)
Vout: TBD (min range: 8-24V)
Imax: 40A
Target power to volume ratio: 100-500 W/in

Design can be simplified by the use of a GaN switch with built in driver, for example, Ti LMG3422R030. It has a pretty high Rds(on) of 35 mohm, thus 28 W power dissipation with two in parallel under continuous current. Or I can use an external driver and something like EPC EPC2302 with an Rds(on) of 1.8 mohm (2.88 W cont power dissipation with only one switch, 5.9 W total power dissipation according to a Ti tool), but then many protections are gone and design is harder

Questions: I will have about 7 months to complete this project and all constraints are self imposed. Right now I cannot decide on the topology, I think a four-switch buck-boost topology would work, but it would need modications to allow soft switching, right? And soft switching may be harder to design, but it will give better effeciency thus allowing the converter to be smaller, correct? Am I being way too overly ambitious 

[jonpaul]

bonjour cher Monsieur

BRAVO pour votre effort.

Takes years of experience with SMPS to learn the topologies magnetics, components, and switch trade-offs

A 500 w as mentioned can be a single device switch or half bridges, also double Foward

GaN is not needed, consider SiC or FET.

Resonant mode, soft switching and class E are good starting points. Besides switch choice, and topologies,

Much of a SMPS design is layouts, magnetics, components and compliance eg EMI, safety and transients protection.

Suggest that you first research IEEE Power electronics, mfg app notes and textbooks.

Bon chance


Jon

PS 53 years as EE, my first transistors inverters in 1963, electronic ballast 1976, 12 kw HMI in 1980s, also laser, avionics and medical SMPS.

[AnasMalas]

Bonjour! Thanks for the advice

I think ive learned layout suffently, and compliance wouldnt be a requirment for me as I dont have a way to verify it, and I dont plan on selling this converter (for now). Magnetics will take me some time, but ill get there

Ive looked at many app notes, but I find that most help you make the converter, but those that help you decide on topology mention cases that I cannot relate my project to. Do you have a specific book or app note that you'd recommend? maybe tell me which company you like the app notes from, ive already went through everything I found from Ti as I like their documentation style. Im having a look at review articles for the topic now in the IEEE journals

Also, GaN will be needed for very fast switching with low losses. I plan on switching in the MHz range to use smaller inductors and capacitors. I also want to learn how to use these transistors, for when I may actually need them.

[T3sl4co1l]

You may want to build a scale model or a few, and use Si first, to get a feel for the switching loop.  Sounds like that would be worthwhile experience.

GaN must be laid out extremely tight, basically mandating multilayer PCBs.  And, the power dissipation rating of those chips is extremely small, well, not so extreme by themselves -- but the problem is getting it out.  They won't solder perfectly flat and planar, and some gap will need to be filled by thermal paste or gap pad.  And it's a small footprint, so any gap is a big deal.

IIRC, it's reasonable to get on the order of 5 or 10W from one of those, the larger eGaN chips anyway, but yeh, needs to be done very carefully.  Including careful clamping, so you don't bump off a corner of a chip putting it together, etc.

Also in terms of instantaneous power, as the active transistor volume is extremely tiny (very high power density), so they handle basically no duration short-circuit power (think fractional µs), nor can they handle any avalanche breakdown.  So your timing and control better be spot on, so that they never exceed ratings, even for an extremely short time, under any input and output conditions.  Per-cycle current limiting and current-mode control are mandatory.

Also consider multiphase for this current level.  It substantially reduces input and output ripple current, saving on capacitors.  Each inverter runs at lower current, making stray inductance proportionally less significant.  Current-mode control also makes paralleling phases fairly trivial.

For reference, here's a module of, somewhat different ratings, but about the same total power (organized as 3 channels for 2 outputs; 2 are tied internally for 2-phase interleave on the higher-current channel).  About 100mm square.  Just ordinary Si and diodes (not synchronous rectified), 140kHz, pretty easy.



The discrete control circuit is because I didn't find a controller with exactly the specs I was looking for (namely, CC/CV control), so I went and did it myself; you probably won't need to do that here, as a fixed current limit and variable output voltage will suffice, or maybe CC all the way, depending on how exactly you want to control the motor (and also what kind it is, but that's very simple if it's PMDC and unidirectional, and even BLDC behaves the same with an electric commutator circuit in place).

Tim

[AnasMalas]

Thanks for the great advice, Tim! Lots of great advice in there, I hadnt considered doing a Si first and then migrating to GaN. But yeah that's obvious as the process isnt all that different (same topology, looser requirements)

Quick question please: does this converter need to be bidirectional (because of PMDC motor drive)? and is bidirectional design more difficult

[T3sl4co1l]

Well, you tell me.

If it only needs to run forward and push against a mechanical resistance, then voltage and current will only ever be one way (1 quadrant operation) and a plain old buck converter (using a catch diode in the inverter) will do.
If it needs regenerative braking or whatever, that's bidirectional current, two quadrant operation -- needs a synchronous rectifier (or always-switching inverter), easy enough.
If it needs to be reversible, then voltage must reverse (two quadrant operation, along the other axis!), and that requires an H bridge -- at least a slow add-on one, or a reversing relay.
If both ways both ways (regenerative forward/reverse, full four quadrant operation), then the H bridge needs to be synchronous (always switching).  Or, I mean, you can combine both, but why use 6 transistors when 4 will do.

Or maybe you need one side of the load to remain grounded, so the "hot" side has to be driven bipolar, with +/- supplies and a half-bridge; or with a 3-level inverter (half-bridge, but add an active-driven GND state: more accurate braking around 0 RPM, less switching loss), or so on.

There are many ways to do it of course.  These particular options put equivalent voltage across the load, but note they differ in common-mode voltage -- so the choice has implications for EMI.

And the H-bridge, you might just treat as left and right inverters, leaving say the right one default-low, while left side does the switching.  Swap polarities by swapping left/right functions.  Simple enough.

And EMI isn't a big deal, anyway: simple enough to treat both legs of an H-bridge as independent, so attach a suitable LC filter (normal mode / single-ended) to each leg, and hang the load between those outputs.  Add extra CM filtering if needed.


To be perfectly clear: quadrants refer to the (V, I) space covered by the converter's output range.  The same terminology is also used in relation to TRIACs, which can be turned on for any combination of gate current (+/-) and applied voltage (MT1-2 +/-).  Same idea: two axes with two polarities each.

Tim

[jonpaul]

Hello again, GaN is very costly and less available than SiC (Silicon Carbide) or Silicon.

Besides FETs there are other types of switch semis that can work, IGBT, etc.

Modern FETs and drivers can handle a switch freq to 100 kHz...some MHz e.g.  soft switching,  resonant or class  E modes.

A single device properly driven, with soft switching can run a forward or Flyback 500W.

We used double forward in 1990 for 750W medical arc lamp ballast.

A half bridge is  easier to build and drive and more than sufficient at these powers.

The GaN shine at much higher switch freq and much smaller target volume than your goal.


The normal SMPS design approach is to choose   topology  first, then switch frequency, then mode , and LAST the choice of switch transistors types and materials.

You have it backwards if you start with a limitation to use only  GaN.


Finally the commercially available supplies are fine models and examples of design, eg

 Vicor (forwards with patented clamp),

https://vicorpower.com/

or HypeX Netherlands (for high end amps/powered speakers)

https://www.hypex.nl/p/technology/smps/

For research besides IEEE, check  conferences papers and seminars  eg  PESC, APEC, Nuremberg, app notes from the control IC and transistor manufactures.

Many fine textbooks were written about SMPS design.

Finally compliance is an issue for safety even in one off prototypes.

Also, a noisy EMI non compliant  SMPS is not a practical example.

An EMI shield and filter at mains and output should be easy to do.


Bon Chance!


Jon