Hi, I have a general question about paralleling schottky diodes in an SMPS at the secondary rectifier section. I was wondering whether there are any disadvantages in doing so. For instance if you parallel BJTs, I know that the one that is slightly warmer will conduct better and cause thermal runaway (if you don't add an emitter resistor) due to more current going through it. I was wondering whether the same would happen for diodes, although in older AT / ATX power supplies they did that as a common practice on both the primary and secondary side and it seems to work very well without causing any problem. So I tried to Google a few legitimate sources such as books, IEEE, Researchgate and Sciencedirect instead of relying on logic and hear-say and could find zero articles or publications where it has been tested or documented. So I suspect that it should not be a major problem for diodes?
Maybe someone in here know where I can find references / evidence that this will not likely happen? I've done this before and have never had any problem with one of them getting warmer than the other, so I suspect the current division by the two should be close to 50 / 50. Anyone that can confirm this?
Maybe someone in here know where I can find references / evidence that this will not likely happen? I've done this before and have never had any problem with one of them getting warmer than the other, so I suspect the current division by the two should be close to 50 / 50. Anyone that can confirm this?
The problem is that two diodes are never exactly the same. So if one has just a little bit lower voltage drop than the other, that one will conduct the bulk of the current.
You can use a series resistor to migitate the effect but to be effective it must be large enough to equalize the voltage drops at the design current, which means you lose voltage and dissipation in those resistors.
If possible and necessary it is better to get a bigger diode.
Jan
Carel, are you limited to certain models of diode in your stock of parts, and want to parallel 2 diodes due to junction temperature or SOA?
If you use diodes from the same batch, and they are on the same thermal base, and you have suitable derating, then I can't see a concern. Including current sharing resistors imho would be a last resort - it would be easier to add another diode in parallel - especially as any low value resistor would itself have a large tolerance, and would really need in-situ testing.
Be mindful that adding diode parasitic capacitance may incur a distinct switching loss disadvantage, unless you are tuning for some form of zero voltage resonance operation.
If you use diodes from the same batch, and they are on the same thermal base, and you have suitable derating, then I can't see a concern. Including current sharing resistors imho would be a last resort - it would be easier to add another diode in parallel - especially as any low value resistor would itself have a large tolerance, and would really need in-situ testing.
Be mindful that adding diode parasitic capacitance may incur a distinct switching loss disadvantage, unless you are tuning for some form of zero voltage resonance operation.
@MOER - I didn't think there are many people out there that understands Afrikaans (the language & culture that my fellow South Africans would like to see getting killed off). Dit gaan goed dankie!
@trobbins, Yes, they are from the same batch. It's a DSEI60-06A. I am purely doing it because I want to go to about 80A (They are 60A each) and the two are mounted to the same giant heatsink. It is for an LLC Resonant Converter.
I have seen this almost in every AT / ATX SMPS out there that still uses shottky diodes and they usually ony derate them by about 10% each when placed in parallel. I have never seen a diode fail on any one of them. Usually only the MOSFETs / BJTs fail. I suspect that even without a resistor the imbalance would be no more than 10%, which is fine for me. But I have never seen official Manufacturers like ST, Fairchild, Texas Instruments or Infineon mention it anywhere in their Application Notes. Nor have I seen warning against it in textbooks, etc and I can't seem to find any resources that I can study to better understand how this balancing might be affected by temperature, current, etc.
@trobbins, Yes, they are from the same batch. It's a DSEI60-06A. I am purely doing it because I want to go to about 80A (They are 60A each) and the two are mounted to the same giant heatsink. It is for an LLC Resonant Converter.
I have seen this almost in every AT / ATX SMPS out there that still uses shottky diodes and they usually ony derate them by about 10% each when placed in parallel. I have never seen a diode fail on any one of them. Usually only the MOSFETs / BJTs fail. I suspect that even without a resistor the imbalance would be no more than 10%, which is fine for me. But I have never seen official Manufacturers like ST, Fairchild, Texas Instruments or Infineon mention it anywhere in their Application Notes. Nor have I seen warning against it in textbooks, etc and I can't seem to find any resources that I can study to better understand how this balancing might be affected by temperature, current, etc.
If you look at the data sheet a few things caught my eye.
1) The 60A rating is at 0.8 times the Vrrm, it's a 600V part.
A fair part of the heating is due to reverse recovery current at reverse bias voltages.
The simple RMS rating, ignoring this is 100A.
Lower reverse bias would raise the forward current rating.
2) Looking at the Vf vs If curves a diode at 60A and 25degC has the same Vf as one at 90A and 100degC.
So the heat sinking would have to be quite poor for the sharing to get seriously out of shape.How far apar can your Tj's get?
But OTOH I think most people parallel these things in search of lower losses rather than higher ratings?
1) The 60A rating is at 0.8 times the Vrrm, it's a 600V part.
A fair part of the heating is due to reverse recovery current at reverse bias voltages.
The simple RMS rating, ignoring this is 100A.
Lower reverse bias would raise the forward current rating.
2) Looking at the Vf vs If curves a diode at 60A and 25degC has the same Vf as one at 90A and 100degC.
So the heat sinking would have to be quite poor for the sharing to get seriously out of shape.How far apar can your Tj's get?
But OTOH I think most people parallel these things in search of lower losses rather than higher ratings?
Actually the power losses are larger with two // diodes. Because they are lower in temp than a single diode would be, the Vf is higher than a single would be. Since the current is the same, there's more power loss (I*V).
As to // itself: consider that no two diodes are the same in Vf. The data sheet for the DSEI60-06A doesn't even list a typical value, just a max value. A clear sign that it varies all over the place and the manu doesn't want to commit to any value.
The one with lower Vf will conduct more current than the other. It heats up more than the other, thus conducts even more, gets even hotter, etcetera. It can be partly migitated by a huge heatsink and getting them very close but you will never get it completely right.
A bad way to kludge a bad design practise.
Jan
Last edited:
I think you might have omitted a factor of approximately-two?
- in the first design (single diode), each diode carries I amperes
- in the second design (diodes in parallel), each diode carries I/2 amperes
Since the current goes down by 2X but the forward voltage goes up by less than 2X, . . . . what can we conclude?
Yes - losses are larger with two, but not by a significant amount. Overcooling a diode will cause the efficiency to drop, they conduct better at high temperatures - just like a BJT. MOSFETS does the opposite, they increase their resistance.
BUT I think it is very, very important to distinguish between theory and practice. The theory is that MOSFETs are immune to thermal runaway when running in parallel and BJTs are vulnerable. But the theory assumes that there is no microcontroller and that they are running with a constant load and at a constant supply. In practice, the increased resistance of the MOSFET will cause a current drop and consequently a voltage drop. The PWM controller will "detect" this and increase the duty cycle to maintain output voltage of the SMPS. Due to the increased I2R losses in the mosfet, you are likely to exceed the power dissipation value and it will almost certainly now also act like a BJT in the sense that it will suffer from thermal runaway (the only difference is that it is in this case not caused by the PN junction, but by the PWM controller increasing the duty cycle).
IXYS are one of those companies with the poorest datasheets available. They don't list all the specs and some you even have to calculate yourself.
I think paralleling the diodes is safe. The only alternative is to use synchronous rectification. The only problem is that it gets quite tricky because LLC resonant converters can operate up to 500kHz and the frequency are not constant. Costs will also increase when using synchronous rectification.
I've seen diodes in parallel in car audio amps, hydrogen fuel cell SMPSes, AT/ATX power supplies, etc. I will maybe do a load test and visually observe the heat with a flir thermal camera, but I think 80A from 2x60A (2 x 2x30A) diodes are totally fine. This will be the max load the supply will ever see.
BUT I think it is very, very important to distinguish between theory and practice. The theory is that MOSFETs are immune to thermal runaway when running in parallel and BJTs are vulnerable. But the theory assumes that there is no microcontroller and that they are running with a constant load and at a constant supply. In practice, the increased resistance of the MOSFET will cause a current drop and consequently a voltage drop. The PWM controller will "detect" this and increase the duty cycle to maintain output voltage of the SMPS. Due to the increased I2R losses in the mosfet, you are likely to exceed the power dissipation value and it will almost certainly now also act like a BJT in the sense that it will suffer from thermal runaway (the only difference is that it is in this case not caused by the PN junction, but by the PWM controller increasing the duty cycle).
IXYS are one of those companies with the poorest datasheets available. They don't list all the specs and some you even have to calculate yourself.
I think paralleling the diodes is safe. The only alternative is to use synchronous rectification. The only problem is that it gets quite tricky because LLC resonant converters can operate up to 500kHz and the frequency are not constant. Costs will also increase when using synchronous rectification.
I've seen diodes in parallel in car audio amps, hydrogen fuel cell SMPSes, AT/ATX power supplies, etc. I will maybe do a load test and visually observe the heat with a flir thermal camera, but I think 80A from 2x60A (2 x 2x30A) diodes are totally fine. This will be the max load the supply will ever see.
If you've got a total current of 20A, putting it through several diodes in parallel will be a lower Vf so a lower loss.
Yes they will run a bit cooler, but the delta from that is usually going to be smaller than the Vf drop due to lower current per device.
When it's a 3.3V rail, shaving a bit of voltage drop can be more worthwhile than when it's a 50V rail in an amplifier.
There are other issues too, like the resistance of tracks to a single device.
Anyway, regarding current sharing, as jan didden says, poor device matching will result in poor sharing. So I wouldn't expect the safe limit to be close to twice that of a single device.
You can make some estimates on the basis of a reasonable device-to-device variation and the V/I slopes in the data sheet, or you can look for a bigger diode.
It's possible that in the OP's case, the Vr is quite low and the single device would be OK anyway. A second device adds margin/derating.
Yes they will run a bit cooler, but the delta from that is usually going to be smaller than the Vf drop due to lower current per device.
When it's a 3.3V rail, shaving a bit of voltage drop can be more worthwhile than when it's a 50V rail in an amplifier.
There are other issues too, like the resistance of tracks to a single device.
Anyway, regarding current sharing, as jan didden says, poor device matching will result in poor sharing. So I wouldn't expect the safe limit to be close to twice that of a single device.
You can make some estimates on the basis of a reasonable device-to-device variation and the V/I slopes in the data sheet, or you can look for a bigger diode.
It's possible that in the OP's case, the Vr is quite low and the single device would be OK anyway. A second device adds margin/derating.
www.st.com/resource/en/application_note/dm00098381.pdf
This is for smaller devices but shows the principle?
This is for smaller devices but shows the principle?
I was figuring, the total current remains the same, but because the diode gets less hot, the Vf goes up. But I see what you mean, the lower current in itself lowers the Vf.
You think the net effect is lower Vf?
Most probably depends on the specific situation. With a lot of current, you are high on the (steep) Vf curve and the net effect might be higher Vf due to lower temp, no?
Jan
These are probably quite well matched in Vf due to being manufactured together (same die?).
Jan
I know some designs have paralleled more diodes than necessary just to ensure current is diverted between enough diodes so no diode gets too much. But beware the added diode parasitic capacitances!
Thnx for link. Very interesting. I was tought paralleling diodes is poor design, but evidently not always.
So spread in Vf is more important than thermal differences (within reason). And the lower reverse voltage rating the lower the spread in Vf, to the point that schottkys with Vreverse < 60V can safely be paralleled.
Obviously layout is critical since spread in path inductance will also unbalance current sharing.
Thnx for link. Very interesting. I was tought paralleling diodes is poor design, but evidently not always.
So spread in Vf is more important than thermal differences (within reason). And the lower reverse voltage rating the lower the spread in Vf, to the point that schottkys with Vreverse < 60V can safely be paralleled.
Obviously layout is critical since spread in path inductance will also unbalance current sharing.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.