Hardware Insights Forum

The way used by Hardware Secrets for calculating their current capacity isn't exactly correct and nor is merely adding their nominal ratings together...

The simplified circuit of a basic two-transistor forward converter (there is also a one-transistor version, which is a bit less efficient as it requires a primary-side snubber, but otherwise isn't much different) has the following schematic:

The components on the input are not shown, but not relevant to this topic. The two (small) fast recovery diodes to the left of the transformer return the energy from its inductive "kick-back" to the primary capacitors when the switchers turn off. The resistor and ceramic capacitor in series across the transformer secondary (and between the anodes of the two Schottky diodes) are a snubber (how those work is not relevant to this topic, but suffice to say that their purpose is to absorb voltage spikes). (In flyback, in case you're wondering, the secondary-side snubber is directly across the diode.) The diode whose anode is connected to ground is the "freewheeling" rectifier. The large coil between the cathodes and the first capacitor is the mag-amp, with the small ferrite coil between the two capacitors for additional filtering.

The thing about these designs is that the rectifier connected to the "top" of the transformer winding in the picture (I call it the pulse rectifier, as it conducts the pulses from the transformer) is loaded for the "on" time of the switcher, and the "freewheeling" rectifier loaded for the "off" time. This is not a problem with using the two diodes in a single component as they are on the same chip, and will stay at about the same temperature as each other. But if you use multiple parts with any wiring arrangement other than every part having one diode as the pulse rectifier and one for freewheeling, things get a lot more complicated, and nominal capacity will only be attainable at a specific duty cycle.

The current capacity of a diode is referred to as "maximum average forward current" which means the instantaneous current multiplied by the duty cycle. So based on this rule, an SBL2045CT could handle 20A at a 100% (continuous) duty, 30A at 67% (two thirds), 40A at 50% (half), 50A at 40% (two fifths), 60A at 33% (one third), or 80A at 25% (one quarter). In practice, though, the instantaneous forward voltage rises with higher instantaneous forward current. But there's no simple rule for that.

Here's my formula:

Instantaneous current per diode = output current / number of parallel diodes (has to be done separately for the pulse and freewheeling rectifiers)

Overall average current in a dual-diode part = ([I(A)]*[D(A)])+([I(B)]*[D(B)]) where I is the instantaneous current in the diode and D is its duty cycle, while (A) and (B) distinguish between the two individual diodes

I've heard a typical operating duty cycle is one third. Freewheeling occurs during the gap between switching pulses, so in that case the freewheeling rectifier would be on for two thirds of the cycle. One-quarter duty would have freewheeling for the other three quarters of the cycle, two-fifths duty would have freewheeling for the other three fifths, and I think you get the idea now.

Onto the examples:

With identical parts in all of them...
Example A works perfectly if the duty cycle is exactly 50%, but if it's 33%, the safe output is reduced by 25%!!!!! To optimise it for 33%, you would want to choose the rectifiers so that D2 has double the rating of D1.
Example B solves that problem, simply by swapping the connections.
Example C is the arrangement used for +5V in the Dell H305P-01. The optimal duty cycle is 25%. With a 33% duty cycle, the capacity is reduced by 10% (not the 25% you might expect if you didn't take into account that the change will also reduce dissipation in the freewheeling diodes).
Example D uses three parts and is optimised for 33% duty.

Try out the calculation for yourself...

A common ultrafast recovery rectifier known as BYQ28E-200 (10A) seems to be only rated at a 50% duty cycle, so I guess even with two of them in parallel, in half bridge, they'd be quite a bit less capable than the 20A schottkys out there rated at a 100% duty cycle (come to think of it, I don't see many schottkys out there rated at anything more than a 50% duty cycle)? Also, are you still of the opinion that a single schottky package is not derated even in forward topology (only when connected to another in the conditions you made mention of)?

Wester547 wrote:A common ultrafast recovery rectifier known as BYQ28E-200 (10A) seems to be only rated at a 50% duty cycle, so I guess even with two of them in parallel, in half bridge, they'd be quite a bit less capable than the 20A schottkys out there rated at a 100% duty cycle (come to think of it, I don't see many schottkys out there rated at anything more than a 50% duty cycle)?

My guess is that they used 20A pulses for that test, which does indeed average out to 10A with 50% duty.

c_hegge had a Rexpower with an FR1003G (10A 200V fast recovery) on +12V that survived 19.2A, by which point it would have been dissipating 25W or so - with a silicone pad, and average heatsinks and ventilation. I have to admit, it would be amusing to see how an STPR part would perform under the same kind of torture, but if it's any indication, the STPS1545CT on +5V in the Okia from the 2011 cheapo round-up burned as early as test 3 (there's no table for that test, but test 2 shows 10A from that rail), which, again, doesn't build my confidence in STMicroelectronics.

I don't need to post a response to your second question as the first post already has the answer.

That Okia unit look like it's true crap. I would not point blame to the rectifier used on the +5V rail. I'm sure it would be capable of more than its rating in a well designed power supply.

I know it's a crappier PSU, but I don't know any way "extra" current could flow through the secondary rectifiers, and the heatsinks actually aren't the worst I've seen (at least they have fins and a core that isn't paper-thin). And it would be a shock if it really was true, that those FR302s on +12V survived while a higher-rated Schottky on +5V died.

My guess about STMicroelectronics is that they aren't as conservative with their engineering as other semiconductor manufacturers.

The fan was directly wired to +12V in the "Rex" PSU, though; good cooling makes a big difference.

EDIT: Or maybe the thermal compound or pad Okia uses for the secondary rectifiers is significantly worse?

Wester547 wrote:The fan was directly wired to +12V in the "Rex" PSU

The same goes for the Okia. There are only two other explanations I can think of - that the Okia has a lower speed fan and/or worse ventilation.

The vast majority of controllers actually have 50% as the maximum duty cycle, and that's what I generally assume (unless the datasheet for the controller IC specifically states otherwise). As such, the theoretical maximum generally ends up being the rating of the rectifier.

c_hegge wrote:The vast majority of controllers actually have 50% as the maximum duty cycle, and that's what I generally assume (unless the datasheet for the controller IC specifically states otherwise).

Isn't that the absolute maximum, not just what you'll get if you put maximum load on the unit??? The units are designed to operate on a broad range of mains voltages (let's say 90~130VAC for the lower range, and 180~260VAC for the higher one - a difference of 1.44x between the lower and upper ends of each), so it stands to reason that the rectifiers have to allow for a wide variation in duty cycle. (In units without APFC, but the following applies even to APFC units) And there are several volts of ripple left on the primary capacitors, which again has to be controlled for, so the duty cycle can never stay exactly the same for any appreciable time.

Is there really no difference in terms of current limit, between both diodes fully conducting at a 50% duty cycle in a single schottky package and one diode freewheeling at a 33% duty cycle and the other diode conducting at a 66% duty cycle? I ask because I noticed that even at 230V, judging by c_hegge's reviews, the HP-D3057F3H isn't really efficient at all (it hovers between ~68% and ~75% efficiency at low and high loads respectively, possibly because the fan spins very slow at low loads, probably much slower than the H305P-01 fan controller has it spinning, but that should only be at low loads). In fact, it seems on par with the H305P-01 in terms of efficiency, and that linear regulates the +3.3V rail, which squanders lots of power. I get that the H305P-01 has a 6A bridge rectifier on a heatsink and the HP-D3057F3H has an 8A bridge without one, and that the HP-D3057F3H uses two 16A ultrafasts and only a 20A part for the +3.3V rail (the H305P-01 has effectively 30A of current on the +12V rail in a much more efficient manner and a 30A schottky for +3.3V, though I would contest it's still wasting more power by way of linear DC-DC conversion), but still... I guess it would have to do with the duty cycle of the secondary supervisor IC if that differs between either PSU? But then that limit is almost always 50%...

Also, the HP-D3057F3H does appear to have minimum load resistors for the +3.3V, -12V, and +5V rails as well, I guess those don't help efficiency. And the H305P-01 does have a better primary switcher, and the fan is much closer to the heatsink and main transformer in that one (the HP-D3057F3H uses its fan as an exhaust, the H305P-01 as an intake), but even then...