ATX PSUs: why dual bridge rectifiers instead of one?

[quicknick]

I'm repairing an LC-Power Arkangel3 850 W PSU which suffered a rather spectacular failure in the mains rectifier section. It uses two GBU1508 bridge rectifiers in parallel, mounted on opposite sides of the same heatsink.

There's strong indication that the failure was caused by a flashover between the + terminal of one of the bridges and the heatsink. Because of the asymmetric construction of the GBU package, the bridge that popped had its terminals significantly closer to the heatsink than the other. A large portion the terminal was completely vaporized, with clear evidence of arcing to the heatsink. I'm planning to modify the heatsink to provide adequate clearance.

This got me wondering about the reason for using two bridge rectifiers in parallel in the first place.

The obvious answer would seem to be current sharing, but bridge rectifiers don't necessarily share current particularly well: small differences in forward voltage, thermal conditions and even PCB resistance can cause one of them to carry more current than the other.

What puzzles me is that I see the same arrangement in considerably more expensive (and presumably better built) PSUs as well — for example, Corsair and Seasonic units I've worked on also use two bridge rectifiers in parallel.

So what is the actual engineering reason for using two bridges rather than one larger bridge?

Is it primarily:

a) increased current capability / thermal spreading,
b) higher surge-current capability,
c) lower conduction losses,
d) availability/cost,
e) something else?

And, assuming the thermal design is adequate, would replacing the two parallel 15 A bridges with a single high-quality bridge rated at roughly twice the current be a reasonable design choice?

I'm particularly interested in how this behaves in an active-PFC PSU, where the input current waveform is quite different from that of a simple bridge + bulk capacitor rectifier.

[asis]

Hi,

For clarity, I can provide a fragment of a schematic for a complex PSU for printing equipment with multiple output voltages: +24V; +30V; +41V; STB +5V; STP +5V; +12V;+17V.
The PSU has an active PFC and a rather complex control circuit.
The circuit design matches what you're describing.
In my opinion, the use of parallel bridges is entirely justified because the dynamic operating mode of the PFC varies significantly depending on the operating mode and load.
It also provides the ability to gradually switch on the line voltage to smooth out transients during startup.
You can see all this in the voltage waveforms for various operating modes.
-
In your case, resonant phenomena could well have occurred, causing an electrical breakdown.

[TimNJ]

Could be a 5-10% reduction in conduction loss with 50/50 sharing. if they are trying to hit an 80 Plus spec, could be a low effort solution to get the efficiency up. It's probably more about actual power loss (total watts to dissipate) than thermal spreading; there isn't that much additional surface area gained with the second bridge alone. The impact is less for a resistive/PFC load than for a capacitive load.

A generally better solution is to parallel diodes within the same package. (Short the AC pins - each diode bridge package now gives you two diodes instead of four.) Diodes on the same die will be very closely matched and the thermal impedance between paralleled diodes on the same die is lower than through the package, heatsink, and through the second package.

Surge current capability is probably not a driver as most diode bridges of this rating have high surge capability, 150-200A @ 8.3ms.

Size/space may have something to do with it. Similar diode bridge in a larger GBJ package would run cooler due to larger surface area in contact with the heatsink, if not strictly an efficiency concern, and more so a thermal concern.

[moffy]

Possibly a combination of load and surge current sharing, sharing losses across two packages and perhaps lower cost, one bridge might be marginal, the next larger bridge might be more expensive/inconvenient so two of the same becomes economical.

[coromonadalix]

a cheap way to have a more powerful rectifier bridge,  the problems are un-equal voltages drop between the diodes themselves, and the shared current is not equal

since they are not high precision / tolerances, the weaker ones wiill drop the overall efficiency   ...

[Psi]

It's definitely bad practice, but paralleling diodes does work to share power/heat.
However you can't assume perfect 50/50 sharing.

The Vdrop is dependent on current, so it does tend to self balance to some degree. As a diode draws more current or gets hotter the Vdrop gets worse which causes the other diode to take more of the load away from it.

If doing it I would plan on a possible 70/30 share, just to be safe.

[PGPG]

As a diode draws more current or gets hotter the Vdrop gets worse which causes the other diode to take more of the load away from it.

When a diode gets hotter its Vdrop gets lower and this diode takes more current making it more hotter and so on.

[Psi]

As a diode draws more current or gets hotter the Vdrop gets worse which causes the other diode to take more of the load away from it.

When a diode gets hotter its Vdrop gets lower and this diode takes more current making it more hotter and so on.

Sorry yes, I shouldn't have brought temperature into it.   
But as far as current goes, it self balances assuming temps are similar. Which they usually are if you have two diodes on the PCB in the same spot. The temp difference works in the bad direction to cause imbalance.

[quicknick]

Thanks everyone for the replies, this has been much more informative than I expected.

So it seems that the main drive behind this is the race for every % of efficiency, but wouldn't a Schottky bridge improve things? Anyway, I digress. I think there is enough here to make me want to actually test this myself. I have five GBU bridges on the way (plus the one that survived from the PSU, plus others recovered from scrapped units), so, if I get around to setting up a suitable low-voltage test, I'd like to compare pairs from the same and different batches and measure the current sharing, perhaps at different loads and temperatures. Can't promise this will happen reasonably soon, though.

[TimNJ]

Thanks everyone for the replies, this has been much more informative than I expected.

So it seems that the main drive behind this is the race for every % of efficiency, but wouldn't a Schottky bridge improve things? Anyway, I digress. I think there is enough here to make me want to actually test this myself. I have five GBU bridges on the way (plus the one that survived from the PSU, plus others recovered from scrapped units), so, if I get around to setting up a suitable low-voltage test, I'd like to compare pairs from the same and different batches and measure the current sharing, perhaps at different loads and temperatures. Can't promise this will happen reasonably soon, though.

There are no Schottky bridges in the 600-1000V range. PN Schottky runs out of steam at about 200V. SiC Schottky is the wrong tradeoff at 60Hz.

[asis]

Hi,

@quicknick
Unfortunately, I can't see your specific PSU schematic, but to avoid further problems during subsequent power-ups, I suggest checking the PFC MOSFET and other components, or better yet, disabling the PFC completely.
If you could index the components used, it would probably be easier to navigate and offer some advice.

[quicknick]

@asis

Thank you, but fortunately the power supply is almost repaired. I've cleaned up all the splattered metal, replaced the blown fuse and the damaged inrush NTC, and it now powers up normally with just a single bridge rectifier fitted as a temporary test.

I've ordered the replacement parts, which should arrive in a couple of days. I'll then fit a matched pair of bridges and remove (saw off) a small section of the heatsink so that it no longer overlaps the bridges' terminals. This will eliminate the clearance issue that appears to have caused the original failure.

I hadn't realised that it was possible to disable the PFC for diagnostic purposes, so I'll definitely keep that in mind when dealing with stubborn PSUs in the future. That could be a very useful troubleshooting technique.

The LC-Power 850 W uses a CM6500UNX PFC controller. The primary side is otherwise quite interesting; the platform is very similar to the Great Wall design used in some Corsair units, although there are some differences.