Mobo MOSFET surgery advisable?

[Brians256]

I think I mentioned this before, but I wonder if it is possible to simply replace the MOSFETs with higher quality ones. I need some expert EE advice from some of you gurus out there.

Now, I've some some research to figure out what causes the heat in the DC-DC converter power MOSFETs. If I recall correctly, the main culprit depends upon the frequency at which the MOSFETs are switched on/off. I think Intersil or National has a good appnote on this.

At lower frequencies, the main factor in how much heat is generated is RDSon, which stands for the resistance from drain to source when the MOSFET is switched fully on. When the switching frequency is low, the transistor spends most of its time either fully on or fully off. So, the generated heat is mostly caused when current is flowing through it.

With higher power MOSFETs, you need to work harder to turn the MOSFET on and then off. Think of industrial strength valves. You don't ask grandma Hazel to turn off the water supply valve for Hoover dam, right? Well, it turns out that you need beefier MOSFET control circuitry to make sure that the MOSFETs switch on/off quickly so that the MOSFET spends most of the time fully on or fully off. If you don't, then the MOSFET starts acting like a high power heater element as it spends time time in the "almost on" state. Since I am sorta stuck with the MOSFET controller on the mobo, I need to pick a MOSFET that is as easy to turn on/off as possible while still being able to easily handle the power our CPU will need. This is one of the reasons that mobo mfgrs use multiple MOSFETs. It is easier to get three or four MOSFETs with lower power specs turning on/off quickly. Much cheaper.

At higher frequencies, other factors come into prominence. A MOSFET is turned on by current flowing in/out (depending upon whether it is a N-channel or P-channel MOSFET) of the gate. This gate acts like a capacitor storing up charge and then releasing it. However, and here is an important part, nothing comes for free. Moving those electrons in and out of the gate is lossy, because the gate has a charging "resistance". Every time you move the electrons in/out, you generate waste heat. Turning the gate on and off very rapidly means that you generate little bits of heat VERY often. It adds up.

The best thing to do is to lower the amount of electrons you need to shove in and then pull out, right? Well, that doesn't come for free either. A MOSFET with lower gate capacitance cannot handle as much power flowing through the Drain to Source (that is the main power path leading (in our case) to the CPU).

If I'm not mistaken, there are other effects that I (to be honest) can't remember. There is the Miller something or other and I think a couple of other effects.

Now, I'm sure that the transistors used by most motherboard manufacturers are not the best available, because that simply is not needed. But, can you get a transistor with a lower RDSon or lower gate capacitance without messing adversely with some other important parameter?

Also, the inductors (the little wire wrapped doughnuts) can get hot too, because you are dumping lots of current in and out of them. If you have higher value inductors, they tend not to get as hot. But, you can't do that without changing the filter capacitors, I think. The MOSFETs might get driven harder with higher value inductors too (I don't remember on this one).

Any thoughts on this one?

[bigben2k]

From what I know about MOSFETs, from experimenting with them on my PSU array, the driver doesn't necessarily have to be beefed up.

The heat mostly comes from having a large current pass through the MOSFET, which has a certain resistance, even when on.

If you think about it... 25 amps through a 1 ohm resistance creates a 25 volt drop. Obviously the MOSFETs have a lower resistance than that. The one I found has a resistance of 0.004 ohms which, at 40 amps, creates a voltage drop of 0.16 volts. It's that small voltage drop that heats up the MOSFET: P=V*I so the power dissipated by the MOSFET is 6.4 Watts! :eek:

Otherwise, you should be aware of the driver characteristic: what driving voltage does it supply, and how fast is it? Then you have to match that with the MOSFETs response time.

So your first step ought to be to identify the existing MOSFET.

I can help you search through Linear Technology's line of MOSFETs, for an equivalent.

[Brians256]

A lower RDSon value usually implies a higher gate capacitance, though (from what I remember). So, you might have a harder time driving the MOSFET from (for example) an Intersil DC-DC voltage controller chip.

[hydrogen18]

the problem isnt the resistance in the mosfets as far as i know, the fact that they get hot and then resistance goes thru the roof. Meaning shuldnt heatsinks take care of the problem

[bigben2k]

Originally posted by hydrogen18
the problem isnt the resistance in the mosfets as far as i know, the fact that they get hot and then resistance goes thru the roof. Meaning shuldnt heatsinks take care of the problem
I see what you mean, and you're right: as the temp increases, so does RDSon, and the power dissipation goes with it.

I'm guessing that a low RDSon to start with, would allow some operational temps to start lower.:shrug:

[Brians256]

I have no idea what the specs are on the MOSFETs for this cheapo MSI mobo on my work desktop (I didn't buy it!). They are labeled as BC240. Any ideas?

The DC-DC converter chip on my mobo at work is the Intersil HIP6302 (see PDF spec sheet (http://www.intersil.com/data/fn/fn4766.pdf) for details). It runs at about 275KHz, which means that the MOSFETs are being turned on and off (two operations!) 275,000 times a second.

In order to spend at least 90% of the time in a fully on or off state, you have to switch the MOSFET on and then off 20x faster than your PWM frequency (using PWM to generate DC is what the mobo DC-DC chip does). That works out to be about 1/6 of a microsecond (1.8e-7 to be more exact).

Anyway, there are a couple of datasheets out there talking about how to optimize the MOSFETs for these things. The Intersil PDF focuses mostly on RDSon, but other papers talk about the Miller effect, transconductance, and other parameters.

Also, the inductors get hot too. Anyone think I could change them out for higher efficiency?

[bigben2k]

You'll have to get the specs for the coil. ;) Then you can swap it for something that can handle a little more current.

I'll see what I can find about the "BC240".

[hydrogen18]

im planning on trying to watercool my mosfets, so i'll let u know how that goes. mine dont get very warm anyways b/c my mobo has 6 or so

[bigben2k]

Yeah, thanks for the reminder. I had to print up a lifesize pic of the A7N8X, so that I could get a look at the fit, and noticed that the MOSFETs aren't the familiar TO-220. I might have to resort to 2 MOSFET blocks, once I've figured out the electrical potentials.

The A7N8X has 4 MOSFETs. ;)

[murray13]

Just my two cents worth...

It's just not worth the time and effort to swap out components in this area of a MB. IMHO

The dc-dc converter is designed as a whole. Swap out one thing and you have to change another, and then another....etc. There are many more things that you have to design around than just Rdson, Gc etc. Trace length, Inductor resistance, ESR in the filter Capacitors. Change something and you really can't know (without a good spice simulation of the before and after) what affect it will have.

I'm not trying to shoot you down, just trying to convey the complexity of dc-dc conversion.

Oh one other thing, in the current multiple phase dc-dc converters used on MB's there are two FET's per phase. One of which is always on.

Even using the best components your still going to have to deal with 10 or 12 watts of heat being generated by the dc-dc converter. No matter what you use. Sounds like a new topic to discuss, 'The best way to cool MB FET's'.

Either way I couldn't find anything quickly about a BC240. Probably a end of life product that is discontinued. If you would like me to dig a little further just say so and I WILL find it.

[murray13]

Originally posted by bigben2k
....and noticed that the MOSFETs aren't the familiar TO-220....

Haven't taken a look at the board, but they are probably TO-252 or TO-263. Take a look here (http://www.irf.com/package/pkhexfet.html) and see if they match up. IRF calls them Dpak something.

[Brians256]

Thanks murray13! I appreciate someone answering me with enough knowledge to tell me what I'm up against.

I can definitely say that the MOSFETs are a DPAK of some sort. I've been reading up on the datasheet for the Intersil DC-DC converter control chip HIP6302 to learn some of what THEY think is important in selecting components.

What I was hoping to find was enough specs on the existing parts to find an alternative with similar but improved parameters. For example, with the improved processes available today, I should be able to get a MOSFET with a similar gate capacitance, switching frequency, and current capability with a lower RDson. Maybe I could even lower both RDSon and gate capacitance. After all, I can buy a couple of $5-10 MOSFET where a mobo manufacturer would want to stick with $1 or even cheaper MOSFETs. I've been reading all the marketing fluff coming out of IRF and similar manufacturers all proud of their trench FETs and so on. So, they must be making things better these days, right?

Also, you should be able to do the same thing with the inductors and capacitors. Manufacturers REALLY like to buy cheap stuff. I know that I should be able to buy a pair of low ESR caps that could lower the overall equivalent resistance while keeping the same filtering capacitance.

Keep in mind that I'm not talking Radio Shack stuff. I mean Digikey, Mouser or the like. Coilcraft makes some really nice stuff that you don't see on any motherboard. For example, using a MPP or High Flux core inductor, I would assume that waste heat would drop off because the overall core loss would decrease dramatically. Do you think that an $80 (retail) mobo is using something other than powdered iron core inductors?

Anyway, I was hoping to find out enough parameters on what is there so that I could make simple improvements. If that isn't possible (I doubt that!), I'll have to settle with heatsinks and (possibly) a low noise and low CFM fan on the VRM area.

[Since87]

I've been thinking about this issue as well, because I just bought an Abit IC-7 mobo and there is conjecture that these mobos may not be able to power the Intel Prescott processors when they come out.

The mobo has Phillips PHB108 (http://www.semiconductors.philips.com/acrobat/datasheets/PHP_PHB_PHD108NQ03LT-02.pdf) MOSFET's on it, which look to be state of the art. (I haven't found lower Rdson for a similar gate charge.) Perhaps by the time Prescotts have dropped into my price range, there will be better MOSFET's out there.

Short of better MOSFET's becoming available, I think the best solution is likely to be improving the cooling. I'd be inclined to just solder a strip of sheet copper to the tab of the MOSFET and stick it up into some airflow.

Brians256,

The reference design for the HIP6302 controller calls for using external gate driver IC's. Perhaps the gate drivers on your motherboard could be replaced with something beefier that would allow the use of MOSFET's with greater gate capacitance.

As far as the inductors are concerned, I'd guess they use Sendust/Kool-mu cores. It's fairly cheap and much better than powdered iron. I think the majority of the power dissipation in the inductors is likely to simply be the resistive losses in the copper wire. I wouldn't be surprised if the temperature of the inductors is mostly due to heat conducted from the MOSFET's. Rewinding the cores with heavier wire may be of some benefit, although you have to be careful that you don't increase the parasitic capacitance of the inductor. (You also have to ensure a beefed up inductor will still mount to the PCB acceptably.) I'd leave the inductors alone and focus on cooling/replacing the MOSFET's

[murray13]

Hey Since87, keep these in mind if you ever feel the need to do some mod'in. Found these (http://www.irf.com/product-info/datasheets/data/irlr3717.pdf) that could fit your requirements. They have one half the Rdson and about the same Qg.

I'd still want to cool them.

[SexyMF]

I have just completed a BE Electronics and computer engineering degree.

Brians256 :

You are clutching at straws here. Not to be rude but although your enthusiam isn't, your electronic knowledge and expertise are lacking.

The dc-dc converter is designed as a whole. Swap out one thing and you have to change another, and then another....etc. There are many more things that you have to design around than just Rdson, Gc etc. Trace length, Inductor resistance, ESR in the filter Capacitors. Change something and you really can't know (without a good spice simulation of the before and after) what affect it will have.

Oh so correct.

Rewinding the cores with heavier wire may be of some benefit,

STOP, you have no idea.

After all, I can buy a couple of $5-10 MOSFET where a mobo manufacturer would want to stick with $1 or even cheaper MOSFETs. I've been reading all the marketing fluff coming out of IRF and similar manufacturers all proud of their trench FETs and so on. So, they must be making things better these days, right?

Yeah, a MB manufacturer when buying by the millions can get a better price than you at Radioshack. And yes, they are making better semiconductor switching devices these days. You are out of your league.





Gosh, I was going quote some more but I'm tired.

If you want an efficient power converter then do not go past resonant power conversion. I suggest a Class-E resonant inverter with a resonant Class-E zero-volt-switching rectifier.


FYI. When designing with MOSFETs you must read the SOAR curves VERY carefully because your gate drive requirements change depending on how much current you intend to sink in what amount of time. Thus if you change your MOSFET the amount of current it can draw based on the [fixed] amount of gate drive given to it changes. For the better or worse? it depends. Also, the Miller effect comes in to play. When you change a voltage either side of Cds [gate-drain capacitance] the capacitance change dramatically. This [The Miller effect] manifests itself as increased gate drive requirements to achieve the same level of turn on. The Miller effect occurs due to increasing gate drive voltage (because your appling charge) and dropping drain voltage due to the transistion from blocking-state to on-state.

[Since87]

Originally posted by murray13
Hey Since87, keep these in mind if you ever feel the need to do some mod'in. Found these (http://www.irf.com/product-info/datasheets/data/irlr3717.pdf) that could fit your requirements. They have one half the Rdson and about the same Qg.

I'd still want to cool them.

Thanks murray13.

I will keep those parts in mind. The parts you found are spec'd for 20V max Vds, instead of the 25V of the Philips parts, but they may well work.

[Since87]

Originally posted by SexyMF
I have just completed a BE Electronics and computer engineering degree.

Originally posted by Since87

Rewinding the cores with heavier wire may be of some benefit,


STOP, you have no idea.


Oh callow one, please enlighten me. I've only been working as an EE for 20 years. Obviously your fresh education has made you aware of new realms of physics that weren't even dreamed of when I went to school.

Is the reduction in the winding resistance of the inductors going to send the Q of the inductors through the roof causing huge destructive oscillations?

Please explain the dire consequences of rewinding the inductors with heavier wire.

[Brians256]

Hmmm... sounds like SexyMF is quite skeptical of me being able to improve upon the mighty and arcane arts of graced upon us by the EEs at Asus, MSI, Abit, etc...

Not to put too fine of a point on it, but I disagree.

Besides, who cares if I don't have a blinding success? At least I'm learning something. He who is not willing to look foolish once in a while will never learn anything more difficult than eating with a spoon.

Now, back to our regularly scheduled program of sharing knowledge and being polite to each other.

[Brians256]

I now have a much better idea what the Miller effect is (it helps having a PhD EE (focus in microwave frequency analysis) to answer questions) and I know what Qg is (gate charge in coulombs for saturation although it appears that Qsw is more relevant for timing calcs). I'm not stupid, if someone will take the time to point out where I need to learn. AoE is certainly a good reference tome, for instance.

Also, some of the timing issues are less troublesome when you have what appears to be an adaptive PWM source and an adaptive MOSFET driver such as the Intersil 6302 and 6602. At least the timing is not hardcoded to a specific impedance seen by the MOSFET driver (oh, and YES I know that the impedance is a complex value containing resistance and reactance and has frequency specific components).

The HIP6602BCB chip on the motherboard I'm currently examining has the ability to drive any MOSFET up to 3000pF gate capacitance at a decent speed, so lowering RDSon while leaving input capacitance below 3000pF should work.

I haven't read up on SOAR curves yet.

[SexyMF]

Hmmm... sounds like SexyMF is quite skeptical of me being able to improve upon the mighty and arcane arts of graced upon us by the EEs at Asus, MSI, Abit, etc...

Yes I doubt that you could achieve any marked improvement over the rather smart people who design these multilayered motherboards. As Murray pointed out you are still going to be dissipating a fair amount of heat. Correct me if I'm wrong, but you objective here is to lower heat output by improving the efficiency of the dc-dc converter - so have you worked out what the temp difference would be achieved by your efforts?
.


I'm not stupid, if someone will take the time to point out where I need to learn.

Just ambitious


Oh callow one, please enlighten me. I've only been working as an EE for 20 years. Obviously your fresh education has made you aware of new realms of physics that weren't even dreamed of when I went to school.

Your just taking the piss.

[Brians256]

I'm going to lock this thread and start banning people if you don't start acting like adults.

Criticizing ideas because they are wrong is great. That's how we learn. Casting slurs and other name calling belongs back in Kindergarten.

SexyMF, stick with objective criticism please. Since87, please resist the temptation to flame him back. If we cannot sustain adult behaviour I will take action to preserve the SNR of this board.