kingpincooling.com

Go Back   kingpincooling.com > Overclocking & Modding > Hardware Modding

Hardware Modding Soldering, deciphering datasheets, tricks and tips.

Reply
 
Thread Tools Search this Thread Display Modes
  #1  
Old 04-07-2014, 09:40 PM
Khenglish Khenglish is offline
Ambient
 
Join Date: Mar 2012
Posts: 28
Default CPU voltage regulation MOSFET swap

So here's my motherboard's CPU voltage regulation schematic:



Now that looks like a decent 3-phase setup, but unfortunately those nice FETs are not the ones that made it to production. Instead all FETs are much weaker mdu1516 FETs, and half of the high-side FETs were removed.

mdu1516: 14 mOhm resistance when Vgs = 4.5V
mdu2654: 4.5 mOhm resistance when Vgs = 4.5V
mdu2657: 11.3mOhm resistance when Vgs = 4.5V
mdu1512: 5.0 mOhm resistance when Vgs = 4.5V

Now these resistance values are rough estimates since this is a MOSFET and not a resistor, but since all that differs between these transistors is likely width/length ratios, it's safe to say that the production high side is 2.48 times as resistive as planned, and the low side is 3.11 times as resistive. If we take the worst case scenario of 14 mOhm resistance, then pulling 89W at 1.371V (what I need for 4.4ghz) is the following:

I = V / R
I = 89W / 1.371V
I = 65A total
I = 21.7A per phase

P = I^2 * R
P = 21.7A^2 * 14mOhm
P = 6.57W per phase
P = 19.7W total due to high-side power FETs alone!

This is a ridiculous amount of power draw just from power FETs. Yes it's a worst case scenario, but it shows that power losses in the voltage regulation is extremely high. Adding in the inductors could make this well over 20W. Using the planned high side power FETs is a much more reasonable 8W worst case scenario, thus I am looking into removing the current FETs and putting the planned FETs in. I have some questions though:

1. Does any other component need to be changed to accommodate the new FETs?

2. Currently turbo turns off when the CPU is drawing high amounts of power at normal room temps with the CPU in the mid 80s. It takes several minutes of high power draw before this occurs. It will turn back on after a few seconds, then off again a few seconds later with this cycle repeating indefinitely. This does not happen in a cold environment, or if a screw up die contact so that the CPU runs hotter and can hit TJmax. My thinking is that a motherboard component is triggering an overheat signal, cutting power draw by disabling turbo to save the motherboard. The only component close to the FETs with a heat sensor are the MOSFET gate drivers. By putting in bigger FETs, will I be putting more strain on the gate driver and exacerbate this issue?

3. To simplify purchasing can I buy matching high-side and low-side FETs? I was thinking of getting all mdu2654 or all mdu1512 FETs for both high side and low side.

4. Does anyone have an mdu2654 datasheet? I'm mainly interested in the gate capacitance. I found current ratings from a screenshot, but the picture lacked the capacitance numbers. I am concerned with overloading the gate driver.

5. The FETs are far enough away from the CPU socket that I can shield the socket from my heat gun when replacing the FETs. The same cannot be said for the inductors. Downsizing the inductors and adding caps to spread out the increased ripple current can also significantly reduce voltage regulation power draw, and improve output voltage stability in the process. My thinking is that this may melt the CPU socket, but it might not since the socket was likely put on in the first place using a oven hot enough to melt lead-free solder, so maybe the socket can take the heat. Is replacing the inductors a possibility, or is melting the socket too likely?

I only checked the MOSFET part numbers and not the gate driver parts. It is possible that they also cut back on the gate drivers, but for now I am assuming that they did not.

Oh and some system info:

Clevo P150EM with HEAVILY modified cooling (better than m18x cooling)
mediocre overclocker 3920xm. Power limits and current limits are set over 200.
4.2ghz in P95 tops out at 81C, 76W power draw. No throttling. 1.291V VID
4.3ghz in P95 tops out at 86C, 81W power draw. Turbo turns off for 3 seconds roughly every few minutes. 1.321V VID.
4.4ghz in P95 tops out at 88C, 89W power draw. Turbo turns off for 3 seconds roughly every 20 seconds. 1.371V VID.
4.5ghz in P95 throttles terribly and would probably hit 100C if it did not.

The voltage regulation cooling is tied to the CPU cooling. Reducing voltage regulation draw should also reduce CPU temps slightly, hopefully keeping the CPU temp under 90C at 4.4ghz. Note that laptops only have around 60% of the VCC and GND pins as a desktop and have no load-line compensation, thus vdroop is immense and the CPU never sees anywhere near the VID input.

I have run a 2920xm sandy bridge in the system in the past and pulled 105W for a short burst. I have pulled mid 90s on the ivy bridge. BIOS power limits do not seem to be an issue. Again this seems motherboard overheating based.

Currently I am thinking of just putting all MDU1512 chips in, filling the empty high-side FET spot as well. This will cut high-side power draw by 80%. Maybe I will put MDU1511 chips in for the low side, but I am concerned they will be too hard on the gate drivers. Another possiblity is maybe all 1512 low-side, and a 1512 paired with a 1516 for each phase high-side. I feel like I can mix FETs like that if they are in parallel in the same phase, but I am not sure.


If anyone has any input I would really appreciate it. Even if you aren't sure but have your own story of something you did I would really like to hear it. Or if you see something I could do to improve the CPU overclock please let me know that too. I also have the CPU VRM schematic if that would be helpful.

Last edited by Khenglish; 04-08-2014 at 11:36 AM.
Reply With Quote
  #2  
Old 04-08-2014, 03:59 PM
RRainbo's Avatar
RRainbo RRainbo is offline
Sub-Ambient
 
Join Date: Jun 2011
Location: CZ
Posts: 53
Default

Quote:
Originally Posted by Khenglish View Post


I = V / R
I = 89W / 1.371V
I = 65A total
I = 21.7A per phase

P = I^2 * R
P = 21.7A^2 * 14mOhm
P = 6.57W per phase
P = 19.7W total due to high-side power FETs alone!
Thats wrong calc. Highside mosfet is fully turn on only [(Vout/Vin)/3] of time! Conduction losses are not that important as switching losses in this case.

Quote:
Originally Posted by Khenglish View Post
1. Does any other component need to be changed to accommodate the new FETs?
Maybe bootstrap capacitor will need to be bigger if you will use both highside fets or fet with higher total Q -check datasheet there should be something about selecting that cap.

Quote:
Originally Posted by Khenglish View Post
2. Currently turbo turns off when the CPU is drawing high amounts of power at normal room temps with the CPU in the mid 80s. It takes several minutes of high power draw before this occurs. It will turn back on after a few seconds, then off again a few seconds later with this cycle repeating indefinitely. This does not happen in a cold environment, or if a screw up die contact so that the CPU runs hotter and can hit TJmax. My thinking is that a motherboard component is triggering an overheat signal, cutting power draw by disabling turbo to save the motherboard. The only component close to the FETs with a heat sensor are the MOSFET gate drivers. By putting in bigger FETs, will I be putting more strain on the gate driver and exacerbate this issue?
No idea how it work didnt read intel datasheet. VRM controllers have usuallly thermistors(external ntc or ptc) placed around inductors. no temp sensors in gate drivers.

Quote:
Originally Posted by Khenglish View Post
3. To simplify purchasing can I buy matching high-side and low-side FETs? I was thinking of getting all mdu2654 or all mdu1512 FETs for both high side and low side.
Yes you can buy matching mosfets but best is highside fast switching and lowside low Rdson.

Quote:
Originally Posted by Khenglish View Post
4. Does anyone have an mdu2654 datasheet? I'm mainly interested in the gate capacitance. I found current ratings from a screenshot, but the picture lacked the capacitance numbers. I am concerned with overloading the gate driver.
Cant help you with that sorry.

Quote:
Originally Posted by Khenglish View Post
5. The FETs are far enough away from the CPU socket that I can shield the socket from my heat gun when replacing the FETs. The same cannot be said for the inductors. Downsizing the inductors and adding caps to spread out the increased ripple current can also significantly reduce voltage regulation power draw, and improve output voltage stability in the process. My thinking is that this may melt the CPU socket, but it might not since the socket was likely put on in the first place using a oven hot enough to melt lead-free solder, so maybe the socket can take the heat. Is replacing the inductors a possibility, or is melting the socket too likely?
Use proper equipment for that work not heatgun. Use alufoil as shielding and control temp and socket should survive.

What do you mean with downsizing inductors? With changing inductors you will totally change your LC filtration and also current sensing of VRM controller.
Reply With Quote
  #3  
Old 04-08-2014, 05:28 PM
Khenglish Khenglish is offline
Ambient
 
Join Date: Mar 2012
Posts: 28
Default

Thank you for the info!

Could you go over MOSFET selection in more detail? I was thinking that the reason why low-side FETs are usually more powerful than high-side was because Vds for the high-side is Vin - VID, while low-side Vds is just VID. In my case Vin is 16.5V (I measured it). So for 1.3V for example high-side Vds is 15.2V, while low side Vds is 1.3V. With Vgs = 5V, the high-side FETs will be in saturation, while the low-side are still only well within the linear region. With equal sized FETs, the high-side can push around 3 times the current as the low side.

But on the other hand, I see high-side as "pull-up" and low-side as "pull-down", with the VRM sensing the output voltage vs the VID voltage and turning on whatever set of FETs are needed to move towards the VID voltage. My thinking was the pull-down is really just there for regulation purposes in case the voltage drifts too high, while it's the high-side that supplies the current for the CPU. Is this thinking incorrect? Why do you say that conduction losses for the high-side are not as important as switching losses? What frequency do these circuits usually switch at?

OK so you say switching out inductors for smaller ones is no good since that will affect the VRM. Good to know. Won't bother with that idea anymore.

Also the VRM circuit if it helps you determine anything:

Last edited by Khenglish; 04-08-2014 at 06:23 PM.
Reply With Quote
  #4  
Old 04-10-2014, 04:52 AM
RRainbo's Avatar
RRainbo RRainbo is offline
Sub-Ambient
 
Join Date: Jun 2011
Location: CZ
Posts: 53
Default

Quote:
Originally Posted by Khenglish View Post
Thank you for the info!

Could you go over MOSFET selection in more detail?
Dont think i understand it that good so i will point you to page 19 here. Its not exactly for cpu purpose but its same topology and math should work same and every thing is explained.

Quote:
Originally Posted by Khenglish View Post
Is this thinking incorrect?
It doesnt look you understand it right.
-read wiki
-you dont want to work in linear mode at all! Transistors are there for hard switching
-Vds on highside will be just Rdson*I so few mV
-lowside is there to close circuit so current can flow from inductor. Voltage drop on it will be again few mV

Quote:
Originally Posted by Khenglish View Post
Why do you say that conduction losses for the high-side are not as important as switching losses?
Lets just say you have one phase
Vin 16.5V
Vout 1.3V
I 10A
Rdson 10mOhm
Ton 50ns
Toff 50ns
f 300kHz

Time when this mosfet is on is 1.3/16.5 so its only turned on 7.8% of total time
so final figure for conducting losses is P =( I^2 * R)*(Vout/Vin)
P=0.078W

Switching loss
P =[(Vin*I)/2]*f*(Ton+Toff)
P=2.475W

Thats why i say switching loss are more important


Quote:
Originally Posted by Khenglish View Post
What frequency do these circuits usually switch at?
They work in hundreds kHz. Modern close to MHz.
Reply With Quote
  #5  
Old 04-10-2014, 01:40 PM
Khenglish Khenglish is offline
Ambient
 
Join Date: Mar 2012
Posts: 28
Default

Thanks again. OK so I think I get the basics now. So the high-side directly powers the CPU, but the low-side helps out too by forcing a current switch in the inductor, and the inductor powers the CPU.

I looked up my gate driver equivalent resistance on the datasheet. The ratings are .9 source typ, .7 sink typ, with 2 Ohm max for both.

I tried to look up the VRM frequency. My datasheet is not public, but all public onsemi VRM datasheets had identical resistance-frequency tables. When using my resistance from the motherboard datasheet my voltage regulation frequency is around 850KHz.

I do not understand your switching loss calculation. From my experience switching loss is the power spent on charging and discharging transistor gates, and from the dead short that briefly occurs when high-side and low-side FETs are on at the same time. I did some math on the former and it's insignificant in this case (.03W). The latter looks more difficult to calculate than the formula you used.

Last edited by Khenglish; 04-10-2014 at 05:06 PM.
Reply With Quote
  #6  
Old 04-11-2014, 11:01 AM
RRainbo's Avatar
RRainbo RRainbo is offline
Sub-Ambient
 
Join Date: Jun 2011
Location: CZ
Posts: 53
Default

Both mosfets are NEVER on at the same time. there is always some deadtime between.

Here i found nice animation

Most of the switching losses is when you opening mosfet and current starting to flow but Vds didnt drop yet and opposite for turn off



Here is appnote and Here is excel tool to help you with math(i couldnt open it so not sure if its work or not)
Reply With Quote
  #7  
Old 04-11-2014, 12:46 PM
Khenglish Khenglish is offline
Ambient
 
Join Date: Mar 2012
Posts: 28
Default

OK there's a password on the excel doc, but I used the formulas to make my own doc and what I'm getting is that for high-side using a bigger FET hurts, and adding an extra 1516 is a wash. Basically 2W of loss with 1 or 2 MDU1516. What makes the high-side inefficient is the 3Ohm resistance to charge and discharge the gate. This is far larger than the gate driver equivalent resistance. When adding a 2nd 1516 I can cut this number to 1.5Ohm while doubling the gate charge, so there isn't much increase in switching losses, but while conduction losses improve tremendously they weren't very high to begin with.

For the low-side there is 2.8W lost with the existing dual MDU1516 setup. There doesn't seem to be much negative by stepping up to bigger MDU1512 or even MDU1511 chips, while there are big conductive loss improvements.

So it seems that the thing to do is to replace the high-side FET with 2 weaker high-side FETs to cut in half the gate resistance but maintaining the same gate charge, and replace the low-side FETs with bigger FETs.

UPDATE:

And I just discovered that specs on TI FETs BLOW AWAY Magnachip FETs. Same Rds values at 1/3 the gate charge and 1/3 the gate resistance.

2 of these in parallel look really good for the high-side: http://www.ti.com/lit/ds/slps243e/slps243e.pdf
And 2 of these for the low-side: http://www.ti.com/lit/ds/symlink/csd17501q5a.pdf

Last edited by Khenglish; 04-11-2014 at 05:44 PM.
Reply With Quote
  #8  
Old 04-12-2014, 11:22 AM
RRainbo's Avatar
RRainbo RRainbo is offline
Sub-Ambient
 
Join Date: Jun 2011
Location: CZ
Posts: 53
Default

They look really good but have relatively high R junction-ambient so put heatsink on them if possible but not really necessary.
Reply With Quote
  #9  
Old 04-12-2014, 01:22 PM
Khenglish Khenglish is offline
Ambient
 
Join Date: Mar 2012
Posts: 28
Default

The CPU heatsink already covers the inductors and FETs so that should do it.

FETs ordered. Got some extras that I'll first solder onto an older laptop to make sure I don't screw up and need a new motherboard.

Last edited by Khenglish; 04-12-2014 at 05:29 PM.
Reply With Quote
  #10  
Old 04-20-2014, 12:56 AM
Khenglish Khenglish is offline
Ambient
 
Join Date: Mar 2012
Posts: 28
Default

I successfully swapped a high side FET on my spare laptop. I left the other 2 phases untouched. I'll swap out the low-side FETs on the same phase tomorrow. I probably won't get to the clevo until Wednesday due to a test on Monday.

Update: low-side FETs for 1 phase successfully swapped on old system. Soldering the chips to the clevo Wednesday.

Last edited by Khenglish; 04-20-2014 at 04:27 PM.
Reply With Quote
Reply


Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)
 
Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT -5. The time now is 04:13 AM.

Contact Us - www.kingpincooling.com - Archive - Top
Powered by vBulletin® Version 3.8.3 Copyright ©2000 - 2017, Jelsoft Enterprises Ltd.