ECS P6STMT with OST and G-LUXONS caps

[gonzo0815]

Re: ECS P6STMT with OST and G-LUXONS caps

By ading two mosfets in parallel, they resistance will be halved and therefore heat generation will be reduced. But to archive this, the Mosfets should have some resistors at the gate, to give each the same controllcurrent and voltage.
On some ECS boards with the 7500 vrm controler, there aren`t more than 1 phase. Insted of that, they let allmthe mosfets beeing one big phase....very bad desing imho. thgis includes the K7s5a, K7s6a and probably the k7VZA.

[Rainbow]

Re: ECS P6STMT with OST and G-LUXONS caps

Yes, boards with KA7500B/C (TL494 equivalent) have only single phase - either each MOSFET has two driving transistors that are connected to the same output of KA7500 (as on K7S5A) or there are only two driving transistors and the MOSFETs are simply connected in parallel to them (K7S6A).
IIRC, ECS used this only for Socket A boards. Socket 370 and Slot 1 boards use either MC34063 or those special CPU VRM controllers that other manufacturers use.

[trodas]

Re: ECS P6STMT with OST and G-LUXONS caps

Good point with the resistance being halved, gonzo0815 Why the hell I never think of that? And yep, the conductivity between the gates, collectors and emitors of the mosfet undubtely show that they will run in parallel - witch is different from most better designs, where each mosfet have own control on the gate pin, independent...

So, since no resistor is missing there - should I add there mosfets instead of just only one to archive the parallel operation witch promise low heat (but wait, did not the mosfet use / waste some power for itself? I mean - do not the wasted heat exceed the "saved" heat from the resistance of opened mosfet?) or not?

I think I should - at least they won't die if the Pannyes bridged with 100nF ceramics ever give up again...


PS. on the CPU vrm chip is written:
SC1189SW
0244
G31960-3

[davmax]

Re: ECS P6STMT with OST and G-LUXONS caps

Just in case you are not sure. MOSFET switch losses are in two categories 1. On resistance loss. 2. Switching loss as the ON and OFF action passes through a high dissipation area rapidly. This loss increases with switching frequency (more switching losses). Loss due to driving the gate can be neglected.

Your Mbd apparently does not use a very good design in the VRM. Yes parallel MOSFETs will reduce heat if both share the load current (lower loss for 1 & 2). The challenge is that they may not share the load current equally, that is why it is not a good design. To force devices to share current the normal practice is to insert a small resistance in series with each MOSFET. But this results in more loss and heat. So the most effective way of forcing the MOSFETs to share the load is for both to have an inductor in series instead of a resistor. This is efficient with low loss and heat. Your Mbd does not appear to provide an inductor for each MOSFET.

May be best to stick with original design. Keep it simple.

[gonzo0815]

Re: ECS P6STMT with OST and G-LUXONS caps

I agrre, for 14w i don`t think it should be any problem. The heat was due to bad caps, wich will lead to increased duty cycle and therefore increasing the heat on the MOSFET. With good caps this will be mutch lesser, even if those ECS desings aren`t that good from ECS.

[davmax]

Re: ECS P6STMT with OST and G-LUXONS caps

Hi Trodas,

Yesterday I thought that I would explore the actual loss differences if the load was shared by two MOSFETs.

I aborted yesterday's posting because of a mistake. Also it seems that it is better to attach a document with a table.

Please see attached.

[trodas]

Re: ECS P6STMT with OST and G-LUXONS caps

davmax - well, I almost neglected the switching losses, thanks for remonding me out. The gate this is, of course, neglectable. It is like arguing that the controling IO will heat more when it control 4 mosfets instead of two (halving the gate resistances = doubling the current to them...)

The design of the cpu Vcore power is crappy, no question about it. Only two phases and what is worser - they share the same inductor. Briliant... However the question is, how to improve it, optionaly make it generate less heat and last longer + become more robust when cap failure occur that it won't get destroyed again.

And keep it simple? Nope, mate. I want challenge. And I want robust safety. Therefore it is more challenging to try add these anoter mosfets to see, if it works and what happen to the teperature and load
Simple things aren't challenging enought to do them at all. Adding the mosfets could be a interesting experiment and definitively challenging thing


gonzo0815 - yep, 14W is laughable at best. However it still killed the mosfet. And yep, bad caps. But I fear that with fanless design, over few years and even with added ceramics - even these Pannyes might become bad caps and therefore making the Vcore switching stuff more robust by restoring the original design - that could be the safe approach, don't you think so?
Now just where to get one more Infineon 15N03L - or four replacements to this mosfet. Probably with better specs - faster switching time = less heat generated? Can anyone suggest someting, pls?


davmax - I would lie if I say I follow you there Some shortcuts (most of them) are unknown for me - I probably know well the czech equivalents, but that make me a bit lost...
However your calculation show, that the single mosfet will generate 10.4W of heat (6.4 + 4), while the two ones in parallel operation will generate only 6.9W ( (1.6 + 1.85) x 2 )
Is that about correct?
Then this is one more reason why add second mosfet then! And if you can do actuall measurments, then the VIA C3 1200Mhz Nehemiah run at 1.4Vcore voltage (bios don't allow me to change it...) and since VIA is saying that on load it should be only 14W (who can believe that low?!) - then then current should be 10A.
So now you have the exact values - just get me one more of these oldie mosfets, or suggest a good substitute that one can buy from Digi-key and I'm all for restoring the original idea about 2 mosfets in parallel, altrough I dubt they can share the load equaly by this simple way...

One of them - probably the upper one, where the heat go - will heat-up quickly more. More heat = less resistance in semiconductors. Less resistance = less generated heat by the current - but also more flow will go trough this more heated-up one now...
Dunno where it end, but somewhat it get balanced, because these aren't too far from themselves, so the temperature rise affect both... Still the upper ones will cary more load, I think...

[davmax]

Re: ECS P6STMT with OST and G-LUXONS caps

Hi Trodas, I have been waiting for someone to respond to my calculations. The calculations were just to see what losses can occur in a MOSFET. In actual fact the values are not directly related to any motherboard VRM because it is my understanding that in a VRM there is normally two MOSFETs per phase, one connecting the supply to the choke and switching OFF then the other connecting the choke to ground during the OFF period. This MOSFET connection is often called totem pole, both connected in series between supply and ground with the choke connected to their mid connection. This is the usual method and this means that the ground connected MOSFET does the job of a diode in the OFF period, therefore there is no diode loss in the supply side MOSFET, that leaves only switching loss and Ron loss.So less dissipation in a VRM MOSFET.

10A load in one 20 milliohm MOSFET = 2 watts loss + switching loss. Not much really.

As you are no doubt aware the risk in a totem pole method is that if any one of the two MOSFET fails short then the other connects into a short circuit when it switches ON. Why everybody changes both MOSFETs when failure occurs.

The 15N03L (should read 42N03L) seems a very adequate device for this system. I suspect that the currently available 15N03L devices are upgraded (42A) replacements for an original 15A device.
Did you manage to get some 15N03L devices?

[gonzo0815]

Re: ECS P6STMT with OST and G-LUXONS caps

On the older boards there are still shotky diodes used e.g. k7s6a, k7s5a and probably mutch more.
Anyway, this calculation seems to be a accurat enough to know the region.

[davmax]

Re: ECS P6STMT with OST and G-LUXONS caps

Thanks Gonzo0815. Yes the Schottky diode sounds right, it used to be the way to go. The challenge with switching losses is that they can be significant but impossible to quantify since they are dependant on circuit design, components and frequency. If we assume an 80% duty cycle for the MOSFET we get 1.6 watts of resistive loss. I would guess switching losses would not take the total beyond 4 watts, but a guess.

[trodas]

Re: ECS P6STMT with OST and G-LUXONS caps

Like I say, I'm not very familiar with the calculation around the switching loss. The resistance loss is, however, very simple thing Just if we know, for how much time it is opened...

10A load in one 20 milliohm MOSFET = 2 watts loss + switching loss. Not much really

And for two mosfets in parallel? 5A per each = 1W for both + switching loss?

But let's concentrate on other thing now. And that is getting my mobo fixed.
Big Pope said that he managed to get two 15N03L mosfets. That is enought (probably one will do...) to restore the mobo functionality, witch is great. However to prevent future disasters, I looking for 3 of them - to run two and two in parallel for each phase - as originally designed.

However they are old and so on - so what about THIS:
https://cdn.badcaps-static.com/pdfs/...3c183b454d.pdf
It is a D2Pak switching mosfet. In short - 30V, 64A and 12miliohm suxxka. Tell me, is this a usable replacement? I would bet it is, as the datasheed for the infineon claim 30V, 42A and 12.6miliohm.

I'm abit confused where the 20miliohm come to your calculation as well, as I'm a bit clueless into the other characteristic, but for me they looking comparable with the 15N03L infineon.
( )

And what is best - this one I can buy at Digi-key com for $3 a piece... Therefore I can buy 4 (or 5) of them and change the mosfets on the mobo completely to make it superior in term of robustness and reliability.
What do you think about that?

Yea, it is no simple approach. But it should give the best results and that is, for mission critical machine like my home server is, what does count.

Surely these 15N03L from Big Pope come handy - at least to test if the mobo kick in, when I solder the mosfet in place. I would say that this is very important to know, that the mobo is still working, before doing crazy things to it - hehe

[Rainbow]

Re: ECS P6STMT with OST and G-LUXONS caps

That MOSFET died because of bad caps, not because there are only 3 on the board. Even with 4, one would probably short too. The VRM does not work properly with bad caps and it has to end somehow - usually it ends with shorted MOSFET or damaged CPU.
Most PII/PIII boards have only two anyway, only some Tualatin boards have 3 or 4.

[davmax]

Re: ECS P6STMT with OST and G-LUXONS caps

Hi Trodas I have attached my reply. So that formatting is not lost.

[davmax]

Re: ECS P6STMT with OST and G-LUXONS caps

Trodas, now that I have a good idea of your VRM type better calculations and conclusions can be drawn. See attached

[gonzo0815]

Re: ECS P6STMT with OST and G-LUXONS caps

I agree to that equation, the loses aren't that high in those Mosfets. I belive, the failing caps has lead to an very high duty cycle and may be to other side efects, the controler chip forced the whole vrm, due to the high noise and fluktuating voltage at the output with bad caps. In the edn, the weaker or hotter mosfet had been died. So i think replace the caps, take two good mosfets and let it go back on duty.

[davmax]

Re: ECS P6STMT with OST and G-LUXONS caps

Trodas, The MOSFET you picked is fine , this one from Digikey is better.
https://cdn.badcaps-static.com/pdfs/...aad62025ae.pdf

[trodas]

Re: ECS P6STMT with OST and G-LUXONS caps

davmax - thanks for the explaination, mate. So, in conclusion the efforts making the mosfets run in parallel aren't worth the 14W VIA C3 scenario, probably not even running fanless. However the last mosfet you find rock

IRF3709ZS is not only rated at 87A, but mainly has a very small resistance - 6.3 miliohms and also the total gate charge 17nF is nice.

The one I found have twice the resistance (12 miliohms) as well, as it is "only 64A and has gate maximum charge 33nF (dunno why there is not the average one listed, it could be bellow 20nF, as the IRF3709ZS show 17nF typical and 26nF as max gate charge...

But mainly the resistance is FAR FAR lower, witch give - in fanless operation - enough adwantage to make me want it. There is only problem - Non Stock...

PS: I think is is good to note that the origninal Infineon 15N03L is rated to 42A, has 12.6 miliohms resistance and gate total charge 12.7 to max 15.9nF. So it have very low gate charge, but also higher resistance and average current only... Maybe that is why the gate charges are relatively high - on strong current ones?

[davmax]

Re: ECS P6STMT with OST and G-LUXONS caps

Trodas , Have you had any success searching for IRF3709ZS? AERI have the DPAK version in stock (smaller). I have requested a search for the right part from AERI. See what happens.

[trodas]

Re: ECS P6STMT with OST and G-LUXONS caps

Sucess!!!
A BIG thanks to Big Pope - package from him, containing two 15N03L make it today, thanks a lot! I replaced the mosfet and the mobo did not kick in - it allways shut off immediatelly. So I checked the bottom mosfet better and it seems to be conductive real well (0.4 ohms, while the wires used to measure has 0.2 ohms itself) between it's collector and emitor, so I replaced it (thanks a lot Big Pope for the second one!) and the mobo then kicked in w/o a glitch

Fantastic! Now just the cursed PSU and my server is kicking back again.

Anyway, after a quick look and re-setup into the bios the mosfets are hardly warm. The chipset is the hottest part - about 50°C quite fast. CPU is 42 (passive) and board sensor say 40°C, however it is close to the voltage regulator for chipset - another relatively hot thing.

Conclusion - the mosfets are hardly even loaded and replacing them with better ones might not be worth the hassles - Pannyes should not fail so miserably and when they does, it should - thanks to the ceramics now everywhere - be with some crash-warnings before, so I could replace them in time. I hope.

Still, having two of these IRF3709ZS is worth the bucks Digi-key asked me for it, however it is not yet worth the troubles with customs, so I better wait till my contact in Canada wake up...

I would definitively feel safer with them on hand

[Rainbow]

Re: ECS P6STMT with OST and G-LUXONS caps

I replaced the mosfet and the mobo did not kick in - it allways shut off immediatelly.

You were lucky with that. Even one remaining bad MOSFET is enough to get the smoke out again. That's why you should always test all MOSFETs. When in doubt, desolder the legs of all of them (leave them soldered by the tab only) and test.