When 200 Watts plus is needed - to parallel or to not?

At lower voltages, there are a number of control chips that will provide this function - i.e. mutli-phase several outputs. But for higher voltages, for instance 300Vin to 300Vout at 500Watts there is a benefit of paralleling several smaller converters. One small converter is easier to design and can be made small and robust, so phasing several converters - or interleaving 2 converters has some benefit.

What are your thoughts? Do you use a load share chip? If so, which one and why?

Ed, other than squishing out the EMI footprint a bit, what advantages would there be to use multiple converters in parallel at such fairly low power levels?

As for design ease I don't find the effort very different between a 100W converter and a bigger one. Sometimes I have to spread out the FETs over multiple devices in order to sink the heat away but that's about it (so far).

Ed, I am with Joerg. Just build a bigger converter. The telco 48V guys parallel for redundancy and because they always have. The on board bucks are paralleled for bandwidth and thermal considerations (I think the last bit is true).

And also you might want to look at whether you will do a hard parallel or a soft parallel or somewhere in between.

Sorry Ed - just to be clear - do not parallel unless there is a good reason. If you do have good reason then consider if you need to hard parallel or soft parallel.

Joerg, I agree that paralleling can help in lowering the EMI (lower peak current switching for one), but it can also help with making the magnetics smaller for a lower profile design. It can also help spread the power semiconductors around for better thermal management. However, I would not have a problem with making a single 200 Watt converter - but what about 400 Watts? I guess this is a question of how much power can a single converter provide.

Hamish, I agree with you as well - but at what power level does it make sense to do? I am interested in the best technique to parallel (would you explain the 'hard' Vs 'soft'?).

When the voltage input and output are both somewhat high - like 300 VDC - I would feel even more comfortable in making a single 500 Watt converter. The current would be low and using a 2-Switch Forward, push-pull, or half-bridge would all help with the voltage stress. Would you agree?

Now let's say that I have a three phase system - and I have want to parallel a 500 Watt converter from each phase. I would need to use a transformer so that I could parallel the outputs - so this is one case where paralleling would be necessary. If the output has large step loads - which technique would be the best to use?

Ed, I don't remember the exact wattage of the biggest one I was ever involved in but it was somewhere around 3kW. Regular bridge design, just one converter, one ferrite core set. There was another converter with several hundred kW but I wasn't involved in the power side of that one. As Hamish said, there'd have to be a compelling reason such as a desire for redundancy, for graceful system degradation or for hot-swapping without power interruption to justify the extra cost of multiple converters in parallel.

As for magnetics it is often best to custom order those if you need an odd form factor. That's what we did on a recent design where we had, within reason, all the length we wanted but had very tight height and width requirements. Sometimes one just can't live with catalog cores. The core looked a bit strange, like a stretch limousine, but works fine.

Hello Ed,

I am reluctant to offer a solution for a design problem that has only been partially specified. You have not described several important system characteristics which could affect the requirements of the design solution. Some unanswered questions are:

What is the input voltage range over which the design must operate and how does the input voltage come up? Is the input source current limited and what is the source impedance over frequency (i.e., does it have a large output capacitance that can dump all its energy into a short)?

What is the regulation requirement for the output (can it just track the input or is tight regulation required)?

What is the current limit characteristic requirement for the output (continuous full current, foldback, a low duty cycle "burp mode" shut down and soft restart)? Will the load always be disabled until the output has ramped up and is stable? How much capacitance must the output charge?

What are the isolation requirements between input and output?

As an example of the impact to the optimum choice for the design that these unanswered question may have, consider that given the right requirements, the best choice may be a simple dc-dc converter with resonant transitions. This type of converter could be extremely simple, provide the very best utilization of the isolation transformer and be over 99 percent efficient at 300 volts, but its output would track its input and it would have to have a burp mode overload protection. Whether this approach may be feasible for your needs is unclear without a more complete description of the problem.

As far as your question about power stage granularity goes, it would depend on a myriad of factors, not all the same for each design situation. However, a good point at which to consider going to multiple stages is the power point at which additional switches have to be added anyway. Of course there is still additional cost associated with multiple drivers and perhaps control ICs. However, this may be offset somewhat by reduced power stage filter costs resulting from ripple reduction and higher effective ripple frequency, but that determination would require more information.

Ed, So the power level at which you parallel is down to a few factors.

1. If you need redundancy then parallel converters are good so long as one failing does not stop the entire show.

2. If you have limited power capability and need to go beyond this then paralleling is a good idea. 400VAC Multi-megawatt IGBT induction motor drives are often made up of power modules of maybe 250kW to 1MW because this is a sweet spot for IGBTs. That is the purchasing for the company becomes the main consideration.

Soft parallel solutions are where each converter manages the parallel connection for itself. That is each converter has a complete controller that allows it to operate with minimal information about what the other converters. Typically this means that each converter has a complete input and output filter of its own. There is no high speed comms between the converters and the switching of each converter is invisible to each other covnerter. This soft parallel is typical of the telco 48V supplies where each supply manages the current sharing via sensing its own current.

A hard parallel is where the control of each converters is managed be having high speed comms between the converters. This allows the filtering between the converters to be minimised often to the point of being non existent except for the stray inductances and capacitances. In the extreme case the switching instances of each converter are managed to minimise the current flow between the converters. Typically with reasonable high speed comms four 250kW IGBT blocks can be hard paralleled with minimal inductance between them. After that the timing precision required can be difficult to guarantee.

Best advice is to just build a bigger converter. Interacting control loops in a soft parallel are not always your friend as the output impedance of one converter ends up being the load impedance of the other.

As mentioned system reliability and ability to hot or cold swap units on site could be a major driver for paralleling power converters. If three phase power is the source of power directly rectifying the three phase line to get and input DC voltage, can in my experience, be done while getting very close to 0.9 power factor with little filtering. The rectified DC can then be used by one or more power converters and would have much less filtering than if a power converter was placed on each each of the three phases with their outputs in parallel.
While multiple supplies are touted to be more reliable it is actually the availability of output power that can be improved by paralleling power converters and that can have a marketing advantage or be a requirement. The actually failure rate would increase as you have more parts to fail.
As far as cost goes I believe at low production volumes multiple parallel supplies can give you flexibility to use in potentially more applications. That is a 50W converter can be used in 50W applications and two used in 100W applications etc.
If the product volume is low then using lower cost parts in larger volume may not cost as much more as many people assume; not including development cost which might tip the balance making lower power parallel supplies less expensive. And as said multiple supplies tends to spread the power distribution, which may or may not be an issue.
A lot to consider but I think the market is the key to your answer.

Joerg, thanks - great comments. I normally do design my own magnetics and I try to use off the shelf parts to make a custom solution - although a custom bobbin might be needed at times.
David, you are right, but I was trying to leave it open discussion to hear the various methods that might be proposed - and the pros/cons of each.
Hamish - great comments as well - and I do see the advantage when redundancy is required - even though, in the strict sense as Ernst points out - adding more parts can bring the failure rate up - if redundancy is the goal - then one failure doesn't bring you down so reliability can actually be improved. So, if an interleaved approach was made, the two converters would be 'hard parallel' and then if two of these were connected in parallel - that would be a 'soft parallel'? The output of one being a load of the next is an interesting subject.
Earnest, you have captured one of my common challenges - having a 3 phase input. I've actually been struggling between using an active PFC method on each phase and then paralleling the outputs. Or, as you describe, direct rectification stage with bulk cap to feed an isolated DC/DC converter stage. And then, using passive PFC to help the diodes commutate for lower emissions/harmonics. Actually I have an application where the requirement is to have a lagging power factor. The power output is 600Watts, so again, I'm back to a single converter or an interleaved topology. The load has large step-load changes, so that's a consideration.

The application often constrains methods to be used.

One of my experiences with somewhat low power conversion ( <200W ) for a military mobile Lion charger constrained the design to be under 4mm in height and also required very low EMI. The choice I made was a 4 section interleaved Sepic topology permitting the wide buck and boost needed as the Input Voltage could be either 3X or 0.33X the Output.
While a low cost of materials were not top priority to the design, the extremely low EMI produced by a multi-section Sepic eliminated the cost of an Input filter.
Reliability is very high despite additional components due to the much lower stresses.

The moral of the story is to let the application constraints define the best path for you.