Fixed Ratio DC/DC HV converters

[luky315]

I need to design a high voltage (800V in, 400Vout, 25Aout) DC/DC converter with the "highest possible" efficency. In my research I found the very efficient fixed ratio converters, which would be ok for my project, because it is only the preregulator for an older inverter design that I have designed years ago and that design should not be touched because it works and is certified and the higher input voltage is needed only in a few cases. But there are surprisingly little informations available on this class of converters. Does someone here have more informations or would regulated topologies (which one?) be a better approach?

[Wolfram]

Is isolation required? If not, then a simple SiC MOSFET + diode buck converter would be ideal here. A single phase one would do at this power level, and efficiency above 99 % is not extremely hard to achieve. Look at some of the example designs from Wolfspeed, they have some that are very close to your requirements.

Keep in mind that a project like this is dangerous and not trivial to implement if you don't have experience with high power switchmode design.

[jbb]

This could be a good job for a switched capacitor converter: needs little to no inductor.

The conventional hard switched type may work OK for you, or there are versions with modest added inductance to get a ZCS transition.  Sorry I don’t have any references handy. Note that you’ll need some careful startup to get things into operation; I suspect at least one switch will need to be rated at 1200 V and a precharge resistor may be required.

[ahbushnell]

10 kW switched capacitor.  Why not just a buck converter? 

[NiHaoMike]

The switched capacitor seems easy to design except the startup part. But there is a trick to precharging the output capacitor without having to switch HVDC in order to bypass the precharge resistor: have the precharge resistor in series with the converter capacitor. Then it would be AC that's getting switched.

[luky315]

@NiHaoMike sorry I don't understand the trick

[jbb]

Huh, I just happened to come across a resource on pre-charging.

https://youtu.be/05QprDgrP6E?si=rfvnB8YNja-zi6a0

[T3sl4co1l]

What is your budget and worst-case timeline?  You ask for ""highest possible"", but exactly this can take a very long time to optimize, a great many transistors in parallel, and large custom magnetics.  Do you, in fact, have a budget or size limitation that prohibits achieving truly unbridled highest efficiency?  Or is your actual requirement more like, there is some minimum efficiency you must meet, and given any other limitations, whether it is possible to achieve?  Finally -- is the efficiency constraint actually driven by another constraint such as power dissipation and surface area?  Like I said, the solution could end up quite large, in which case its power dissipation may be rather moot -- and then "highest possible" efficiency isn't actually required.

We also need to know all the usual power-related specs: input range, output range, source and load impedance, does it need to be fault-tolerant, how will it respond under transient conditions such as inductive surge or load short, how will it start up, is this from a battery, does it need precharge, is it precharging something else, does it need current limited output; what level of EMI filtering is required; what is the operating temperature range, is cooling water available if applicable, what contamination level (environmental dust, humidity, etc.) will it be exposed to, etc.

Also, I realize you quoted "highest efficiency", but is this quoting from somewhere else, a spec perhaps, an assignment?  If so, please show what you can.  Context matters; if this is merely a school assignment, say, then there is some expected amount of work, and at least the general form of it, or the approach, is "the correct answer".  If this is for a work project, then it matters what it connects to, how it functions within the system; requirements are driven by that, and whatever efficiency you get, is what you will get.

And other than that -- perhaps you used quotes for emphasis?  Sarcasm?  Do you not actually need it at all, but then why ask?  I'm inclined to take it at face value as you can see, but these are questions that are avoided by making a clear problem statement including all requirements.

And also not to sound like a nag, or some nefarious engineering genie looking for loopholes -- but tightly constrained engineering problems often do take the form of the proverbial genie problem, and it pays to think in terms of that.  The lesson is, understand that there are a great many constraints that apply to even very simple elements of a system.  Maybe, if you're having trouble with one of the parameters, consider others -- contemplate, strategize -- maybe there's something they missed, that does in fact constrain the other parameters more tightly, preventing an absolute statement; and now you can in effect "disprove" the given specification as impossible, and push back on the requirements and get something a bit more nuanced, and feasible.

Tim

[luky315]

Hello Tim,
at the moment I am simply looking for possible solutions for my problem, there are no hard requirements, the demand to have the "highest possible" efficency came from the project leader, therefore the quoting.
I am comparing different topologies and roughly estimating BOM and development costs.
The "load" is the B6 Bridge of an inverter with roughly 470µF DC-link capacitance. Input is a LiFePo4 battery pack "144S" (430V to 520V) and the peak power input is in the region of 25A with 430V input, with 520V its naturally a bit lower. It's not an automotive project, just the voltage levels are similar for obvious reasons.
Now the same inverter with all the control circuit and the motor should be used on a system with double the voltage (2 stacked Batteries which are also used for a different part of the system) - and a redesign of the power stage has to be avoided. So the idea was to use a preregulator to get down from the "800V" (1000V would be more correct) to an aceptable voltage for the inverter. I designed the inverter power stage hardware (not the software) so I know how to handle high voltage and high power on a PCB, but I have no experience with high voltage DC/DC converters.
So at the moment I am just looking which topologies are available and which ones would make sense in my application. I don't need bidirectional capabilities (no regeneration / braking is used)

[T3sl4co1l]

Well, a buck converter sounds fine for that, unless there's something about grounding of the new battery configuration.  Probably using 1700V SiC, because 1200V would be pushing it, but might still work out.

Synchronous rectification will be necessary for "maximum" efficiency, but certainly not just for conceptual development, and initial testing.

I do suggest developing a scale model, say at 12V 1A or the like, to get a feel for control and dynamics, then at 20A+ to get a feel for inductance and switching dynamics, then at 100, 200, 400, 800V maybe, to bring it all together.

Tim

[NiHaoMike]

@NiHaoMike sorry I don't understand the trick

The problem with the charge pump divider is that since it is fixed ratio, on initial startup, the peak currents would be very high as it charges up the output capacitors. (It cannot be soft started by ramping the duty cycle as a buck converter can.) The usual solution is to have a precharge resistor and bypass relay on the input or output, but a HVDC relay is really expensive. So put the resistor in series with the switching capacitor instead, then the bypass relay only has to switch AC which is much easier and lots of readily available relays can do that.

[luky315]

Ok but then you have the resistance of the relay always in series with the switching capacitor where you definitely don't want any unnecesary losses, I need to have a look at that...

[Wolfram]

You mention 1000 V, what's the actual absolute max voltage expected? This will determine if you can use 1200 V SiC, or if you have to use less performant and older 1700 V devices, independently of whether you do charge pump or buck conversion.

As a data point, I've made a 800 - 400 V 30 A buck converter inspired by one of the Wolfspeed reference designs, and got 99.17 % measured efficiency at half power, and around 99 % at full power, for a total parts cost of around 50 EUR, hence my suggestion of going this route.

[uer166]

+1 on buck. I don't think sync buck would make efficiency that much higher here's, an async stage diode would waste 1.2/400V = 0.3% which honestly would be similar to sync FET reverse recovery and misc losses. It's also quite simple controls-wise so a good chance of success if you worked with high power SMPS before.

[luky315]

Is digital power (TI C2000 or ST STM32G43x based) worth it or are "traditional" analog loops still the way to go with this requirements? I shouldn't have fast transient loads and a stable input voltage due to the big battery bank

[uer166]

I'd go with an STM32G4 since it provides most flexibility and it's something I'm family with.

For a buck though, you'll be able to use almost any generic PWM controller. Choose what you need in terms of operating point (DCM? CCM? DCM with QR?), and choose accordingly.

[uer166]

I should add that at your power level you might wanna do multiphase interleaved buck with a few stages, so instead of one 25A buck it's 2x12.5 or even more.

You can do this with off the shelf PWM controllers that support this as well, no DSP needed.

[T3sl4co1l]

I certainly wouldn't encourage a novice pursue from-scratch pure digital control.  If you aren't comfortable with the output going full-scale from a software error, don't do it.

Or, put another way: by the time you have adequate protection circuitry in place around the MCU to mitigate such failures, you've already implemented most of a proper analog/hardware control, and might as well ditch the MCU entirely.

Any buck or forward controller will do here.  Even TL494, if you won't be switching very fast.  You'll need a handful of support components to implement average current mode control.  Something like LM3481 maybe as well, for peak current mode control.  Or ye olde UC3842 for that matter.

The main thing with controllers is, they'll be made mostly for low-side drive, or very particular setups (e.g. internal bootstrap driver + high-side current sense), which won't be easy to adapt to your specific ratings.  The easiest solution, then, is to keep the controller's signals at ordinary levels (5V or thereabouts), and use transducers and isolators to couple the rest around.

What I would suggest, is one of these controllers, operating in average or peak current mode, with voltage dividers to sense input (if applicable) and output voltage, a Hall effect sensor for switch current, and an isolated gate driver for the high-side switch.  If common ground between input and output is not required, then low-side current sense is applicable.

You can still have an MCU for driving setpoints; this is the simplest and safest mixed-signal integration scheme.  The hardware keeps on controlling and regulating, even if the CPU locks up.  The setpoints don't need to update often, or at regular times.  A watchdog can monitor CPU function and disable the whole thing (or put it in a safe state in any case) if it stops responding.

Tim

[T3sl4co1l]

You mention 1000 V, what's the actual absolute max voltage expected? This will determine if you can use 1200 V SiC, or if you have to use less performant and older 1700 V devices, independently of whether you do charge pump or buck conversion.

As a data point, I've made a 800 - 400 V 30 A buck converter inspired by one of the Wolfspeed reference designs, and got 99.17 % measured efficiency at half power, and around 99 % at full power, for a total parts cost of around 50 EUR, hence my suggestion of going this route.

Very curious where you got the inductor from. That budget doesn't seem to leave any room for SiC MOSFETs, or bypass capacitors, or PCB, or enclosure...

Tim

[jbb]

I (re-)did the control platform for a 3 phase solar inverter using a C2000 chip. Overall, I was really pleased with it.

But I  can confirm that microprocessor control can work well (yay!) but software bugs will cause booms (boo!). Such surprises include but are not limited to:
- incorrect ADC driver and signal filter
- incorrect control algorithm
- incorrect PWM driver (what happens if you request 0% or 100% duty?)
- a printf() jamming the control loop; I was concerned when I inherited a platform and they told me “oh, don’t request the DC link voltage while it’s running or it’ll blow up”
- hitting a breakpoint on the debugger

For a buck converter, there’s a lot to be said for a HW controller. FYI: if using a peak current mode controller (they’re good!) and duty could go over 50% you’ll need ‘slope compensation’ which is described in some Texas Instruments app notes.

Microcontrollesrs become a compelling option once you get to an inverter, which needs more complex algorithms than DC-DC

[T3sl4co1l]

duty could go over 50% you’ll need ‘slope compensation’ which is described in some Texas Instruments app notes.

Just to add a note on this specifically:

It's not 50%. It's WIDELY repeated, appnote after appnote, that it's 50%.  But 50% has nothing to do with it.

It will be 50% here, or very close to it, as it happens.  But it's still not it.

All they really need to say is BCM.

It's current going continuous, so that there is some residual current in the inductor at the next turn-on, and thus persistent state between cycles, that does it.  Nothing more, nothing less, ever so simple, and always correct regardless of the actual voltage ratio you're running.

I have no idea how literally no authors since the 80s have noticed this, but so it is.  Case in point: appnotes are terrible.

You can operate perfectly free of subharmonic oscillation/chaos at 90, 95%, whatever, as long as it's DCM (so, that would be only for a very high voltage ratio).  Conversely, you'll be tripping over your shoelaces at low voltage ratio, or during startup/fault conditions where CCM is practically forced, and singing or hissing can be expected (and increased losses and ripple).

(Not calling you out in particular -- I'm sure one sentence suffices for that. The bulk of this is addressed to those authors who keep repeating it, and to everyone else, to see how incorrect information such as this can circulate and be repeated so easily.)

(I also love to talk about this, because the underlying quirk is that, peak current mode control is an analog implementation of the logistic map; subharmonic oscillation == period doubling, and as you go deeper into CCM, the period doubling doubles up and so on, and pretty quickly, duty cycle is jumping around fully chaotically, scattered by ripple, time delays and noise, hence the typical hissing sound deep in this mode. )

Tim

[Wolfram]

You mention 1000 V, what's the actual absolute max voltage expected? This will determine if you can use 1200 V SiC, or if you have to use less performant and older 1700 V devices, independently of whether you do charge pump or buck conversion.

As a data point, I've made a 800 - 400 V 30 A buck converter inspired by one of the Wolfspeed reference designs, and got 99.17 % measured efficiency at half power, and around 99 % at full power, for a total parts cost of around 50 EUR, hence my suggestion of going this route.

Very curious where you got the inductor from. That budget doesn't seem to leave any room for SiC MOSFETs, or bypass capacitors, or PCB, or enclosure...

Tim

That's the component budget, so not including the enclosure. The inductor would be a 74 OD by 35 mm height toroid if you go by the Wolfspeed reference designs, or an E65/32/27 which is pretty similar in terms of cross section and practical winding area. In 60 permeability Sendust that would be around 5 bucks in moderate quantity (100s). Wire and winding operations would add about 10, so that's 15 for the inductor. In single quantity it would be a bit more but not a lot if you find a good winding shop, 20 is not unrealistic. The SiC MOSFETs used to be around 6 a piece for a 75 mohm 1200 V class device, but the small quantity pricing went up a bit since then. The cheapest option at the moment is Genesic 75 mohm 1200 V, around 9 bucks each, and you need two of those . Pricing for larger quantities can be significantly lower through volume distributors. For the diodes, a pair of STPSC15H12WL would work for example, at 5 each.

On the control side you need an isolated gate driver, which can be had for around a dollar, and a SiC-specific floating gate drive supply like a Mornsun QA-series DC/DC (200 V/ns CMTI, 3.5 pF isolation capacitance) for around 4 bucks. Then you need a controller, I would be surprised if a current mode buck controller can't be found for less than 2 bucks. For the circuit board, this could be done on a 10x10 cm 4 layer 2oz board, which is currently 47 dollars for 10 boards at JLCPCB. Film caps would add another 10 dollars, so we're closer to 70 than 50 in the end, higher volumes are needed to approach the 50 mark.

[jbb]

Good point about the conduction mode, T3sl4co1l. You’re right that if you’re in Discontinuous Conduction Mode (DCM) then slope compensation is not required.