[Worklog] Ivy Bridge IHS Soldering

[der8auer]

 

After changing the stock TIM on Ivy Bridge CPUs we saw some impressive improvement for 24/7 setups using air- and watercooling. I could also improve my CPU for wPrime on LN2 and hit the 2nd best score in wPrime32m (meanwhile dropped down to position 3).

 

Source:

Thread

http://forum.hwbot.org/showthread.php?t=47539

 

Unfortunately not everybody had the same improvements and also my CPU died few days later on air cooling. Even though I'm not sure whether the TIM change caused the malfunction.

 

After the disappointing results of the TIM change I thought long about realy soldering the IHS back to the CPU like Intel did it in the past. This thread will be my worklog for this task. I will keep updating this thread during my testing.

 

Let's start!

 

1. Theory:

 

Why soldering?

During my huge

TIM roundup

I also did tests at -150°C monitoring the delta temperatures between IHS and pot. Additionally I kept CoreTemp open. After each test at -150°C I had to heat up the pot and I used wPrime1024 for that. At only 1,5 vCore CoreTemp already showed positive temperatures at a pot temperature of -110°C.

Usually we bench at 1,8-1,95 Volt on LN2 and on Load I'm pretty sure the CPU core will also hit positive temperatures very fast. Replacing the TIM we could lower the temperature by about 10 Kelvin but often the maximum clock was lower in the end. Unfortunately I have no explanation for this.

At the moment I can not foresee whether soldering will change this behavior. That's why I want to try it

 

Limitations

Temperature Part 1:

I know 99% of you are pretty experienced in soldering so you know that most solder will melt at 250-350°C depending on the alloy and purpose. Furthermore you know that these temperatures will kill your CPU for sure. So how can you even solder these CPUs?

Studying the Intel datasheets I found some interesting information. Intel does not recommend to store the CPUs over 125°C. Checking for the materials used for the CPU itself I found that they are using normal fiberglass PCBs with gold plated pads on the bottom which are soldered at about 200°C.

 

Temperature Part 2:

At which temperature does the CPU realy die and why? The CPU is made of Silicon, Silicon oxide, copper, gold, silicon nitride and other chemicals you need for the layering process. Silicon is a semiconductor at room temperature but it gets 100% conductive at high temperatures. Depending on the doping of the semiconductor a CPU can already become fully conductive at around 120-130°C. Once this happens you have a short at some place and your CPU is dead.

However if you don't apply a voltage to your CPU it can resist higher temperatures without any problems.

 

Limiting temperatures:

- PCB ~150-180°C

- Silicon 200°C +

- Soldered pads on the PCB ~200°C

 

Solder

The IHS is made of nickle plated copper and the DIE is made of silicon. The top layer of the DIE is not conductive so its either a layer of poly-silicon, silicon oxide (SiO2) or silicon nitride (Si3N4). Either way I had to find solder which is able to merge both components. The key is Indium.

Indium is a very very expensive noble metall which melts at about 155°C and it sticks to almost every material. It's condictive and often used for semiconductors such as photovoltaic cells to connect wires to the silicon.

After some hours if google-action I found a company which was able to create Indium solderpads for me.

With a price of ~500€/kg for raw indium this wasnt a cheap order but anyway did it.

I could select different Indium alloys which include tin, silver, bismuth and other stuff.

Unfortunately the company could not tell me which alloy would be the best as I don't know the top layer of the CPU DIE exactly.

The temperature range of these alloys went from 110°C up to 175°C. So pretty much in the temperature range we figured out above.

 

Flux?

Usually soldering tin has already flux included or you use a additional flux for easier soldering. I also got 4 different fluxes to test which one was best for me.

 

 

 

2. The real deal

 

Of course I don't want to ruin my good 3770K straight so I ordered some Pentium G1610 which costs about 35€ per piece. The DIE of it is a bit smaler but should be good enough to test whether it works.

 

So here's the Pentium G1610. Note: you can click on the images to view the full size

 

 

I did a quick test on air first to check whether CPU is alive and which core temperature it will reach after few minutes on Prime95. Average temperature with stock TIM was 45°C and 47° max.

 

Now I cut off the stock IHS and cleaned both - IHS and DIE with acetone.

 

 

The Indium is very thin and ductile. I noticed that the fat from your fingers already completely ruin the result. So I'm using a pincette to place the soldering pad

 

 

 

 

I'm using 0,125mm thick Indium pads which I can stack if I need a thicker layer between IHS and DIE. I used my dead 3770K first to check which alloy is the best to connect silicon and nickle plating of the IHS. I have 4 different alloys such as In97Ag3 or In52Sn48 or pure In which all have different melting points and heat conductivities.

 

TBD: Add table with different alloys and thermal conductivities

 

 

During soldering the dead 3770K I noticed that it's realy difficult to place the IHS on the DIE in the correct position. So I thought of creating something to fix the IHS on the PCB/DIE. The solution was quite simple

 

I took a old board which I don't need anymore and cut off the socket.

 

 

 

 

So now I have a pretty neat gadget to place and fix the CPU inside the socket

 

 

 

 

enough for now More tomorrow!

Edited May 10, 2013 by der8auer

[Leeghoofd]

Great Stuff Roman ! Looking forward to see if this will work out !

[TaPaKaH]

Not sure if a Pentium can be made hot enough for conclusive (in terms of serious benching) results ... but good luck nevertheless.

[der8auer]

Sam I just used it to test the maximum solder temperature and whether the soldering process itself works. After that I will use the 3770k for LN2 testing

[Splave]

good luck, have a sheet of indium from GTX-580 testing looking forward to your results.

 

PS make sure your foil is thick enough.

 

What do you think about using a C-Clamp with fabric on the one side tightened around the chip, then super heating the one arm of the c- clamp with a torch then quickly cooling with cool air.

[Alex@ro]

Did you actually tried naked die direct contact with pot base?How do you think it will compare to re-soldered ihs >?Surely interesting things you are up to

 

Also,what is the difference between getting that solderpad,compared to liquid pro between die and ihs?I mean liquid has indium too,at -190 should there be diff between your solderpad and liquid pro?

Edited May 9, 2013 by Alex@ro

[Alex@ro]

Also,i find this interesting:

 

"Usually we bench at 1,8-1,95 Volt on LN2 and on Load I'm pretty sure the CPU core will also hit positive temperatures very fast. Replacing the TIM we could lower the temperature by about 10 Kelvin but often the maximum clock was lower in the end. Unfortunately I have no explanation for this."

 

When pot is at negative temperatures,there is no way cpu core is at positive temperature,ihs should be one of the best things for isolation in this case

 

Best way to measure die temperature would be to sneak a probe under ihs near the die,core temp or real temp or whatever are not reliable under zero....

[hokiealumnus]

Subscribed, this sounds like fun. It's the perfect balance between science and insanity.

[der8auer]

Updated the first post!

 

 

good luck, have a sheet of indium from GTX-580 testing looking forward to your results.

 

PS make sure your foil is thick enough.

 

What do you think about using a C-Clamp with fabric on the one side tightened around the chip, then super heating the one arm of the c- clamp with a torch then quickly cooling with cool air.

 

Thanks! I did my first test in a normal oven as it's pretty easy to control the heat. Will take a look at these C-Clamps tho.

 

 

Did you actually tried naked die direct contact with pot base?How do you think it will compare to re-soldered ihs >?Surely interesting things you are up to

 

Also,what is the difference between getting that solderpad,compared to liquid pro between die and ihs?I mean liquid has indium too,at -190 should there be diff between your solderpad and liquid pro?

 

Direct IHS contact is pretty dangerous and the connection would be still TIM which as a thermal conductivity of about 8-9 W/m*K. LiquidPro is made of Gallium and not Indium. Gallium has a thermal conductivity of about 30 W/m*K. Soldered Indium has about 82 W/m*K.

 

I will do a test with LiquidPro between IHS and DIE tho so we can compare it Should be interesting

 

Also,i find this interesting:

 

"Usually we bench at 1,8-1,95 Volt on LN2 and on Load I'm pretty sure the CPU core will also hit positive temperatures very fast. Replacing the TIM we could lower the temperature by about 10 Kelvin but often the maximum clock was lower in the end. Unfortunately I have no explanation for this."

 

When pot is at negative temperatures,there is no way cpu core is at positive temperature,ihs should be one of the best things for isolation in this case

 

Best way to measure die temperature would be to sneak a probe under ihs near the die,core temp or real temp or whatever are not reliable under zero....

 

The temperature in the core will highly differ from the temperature on the corner of the DIE. And yes the core realy has positive temperatures even if you cool down the pot.

 

 

Subscribed, this sounds like fun. It's the perfect balance between science and insanity.

 

Thx

[xxbassplayerxx]

I remember seeing a video in the past on how to remove the soldered IHS from a CPU... They placed the CPU face down on a frying pan and heated it until the solder began to melt... then they applied some pressure and pulled it off. You may be able to do this in reverse (i.e. secure IHS to CPU and set it IHS-side down on a frying pan.

[Jimba]

ooh nice mod roman! Great stuff.

 

I look forward to seeing if this works..

[Coolfx]

Nice idea, do you plan to test a CPU without IHS to compare ? I know it's risky with heavy pot but it will be interesting to compare results.

[Alex@ro]

Updated the first post!

 

The temperature in the core will highly differ from the temperature on the corner of the DIE. And yes the core realy has positive temperatures even if you cool down the pot.

 

 

I can't imagine this man,think about it,when you have a pot at -100 + degrees your pcb area around the socket will usually be frost,and that happens sooner or later.Now why does this area freeze?Because it's the cold that passed through cpu ihs to socket and pcb.

 

You can't say cpu core can have positive temperatures under a pot full of ln2,i need solid proof to believe that

[TerraRaptor]

I do easily get positives with SS under full load although evaporator is still negative (delta between avaporator and cores is around 30 degrees @5,3 4/8 1,525v). I do beleive higher frequencies and voltages may lead to etreme deltas of 100-150C.

[MaJ0r]

Very interesting topic! Subscribed.

[teurorist]

I can't imagine this man,think about it,when you have a pot at -100 + degrees your pcb area around the socket will usually be frost,and that happens sooner or later.Now why does this area freeze?Because it's the cold that passed through cpu ihs to socket and pcb.

 

You can't say cpu core can have positive temperatures under a pot full of ln2,i need solid proof to believe that

 

now we start to think a few steps smaller there are those very small transistors which always have a temperature peak when they are switched (load) so it needs a a tiny tiny little bit until they are cold and if we increase the V to much those temperature changes are higher

 

so 2 problems with high temperature changes

 

1 high mechanical load to the core

2 if we are close to 130°C in the middle from the core in a transistor for a part of ns it could be enough time for our energy to kill the CPU

 

 

 

and for the frozen pcb there is also a unused gpu, less temperature producing cash, different bus interfaces, and a imc also there is a connection from the head spreader to the pcb from the cpu all together way more square mm then the cores

[knopflerbruce]

I do easily get positives with SS under full load although evaporator is still negative (delta between avaporator and cores is around 30 degrees @5,3 4/8 1,525v). I do beleive higher frequencies and voltages may lead to etreme deltas of 100-150C.

 

Hmmm... I think it sounds a bit unreasonable. The power draw goes down as the CPU gets colder. Would be nice to see some measurements - like... 5 GHz prime at fairly high voltage, perhaps starting at SS temps and then just lowering the temperature down to full pot with some measurements in between.

 

Also, if this delta was real we're left with two options: 1: temperature does not matter much for clocks (at low temps), as for example max 1m frequency is "only" 400MHz higher than... say wp1024m. The difference also includes whatever other factors that would increase instability when running 8 threads full load compared to just one.

[Viss]

Coretemp never been accurate for me below zero, did it actually show -150 in the prog before loading the chip with wprime again?

[der8auer]

Yes I agree that it does not work for core temperatures below zero.

 

My CPU was running wPrime at 1,5 Volt 5 GHz with Pot at -150°C. At this point CoreTemp always shows 0°C on each core. Also if you cool down to -180°C it still shows 0°C

Now heating up the pot CoreTemp will eventually show positive temperatures. This happened at about -110°C. CoreTemp then showed 3-5°C and raising.

Without load the CoreTemp shows again 0°C but with load it keeps raising again.

 

I lowered the clock and voltage to not damage the CPU and did the same test. At 1,2 Volt and 4 GHz CoreTemp showed +5°C at a pot temperature of -30°C. Heating up the pot to +20°C CoreTemp shows about 56-60°C which is pretty normal. (all on load)

 

Conclusion:

CoreTemp does read out correctly if the core temperature is positive.

Edited May 17, 2013 by der8auer

[Coolfx]

I agree with your results with CoreTemp, I tested the same way with HWMonitor and it's working like this.

 

Can you make a test with a cpu without IHS, to compare with indium fixed IHS.

[The Stilt]

What alloy are you using?

"Alloy 82" (Indium Corp)?

 

Did you remove the oxidation prior applying the sTIM?

Did you remove the nickel plating from the thermal interface area (core-IHS)?

 

Questions, questions

[Splave]

My ss -48c 3770k 2core hits 5c on xp loading screen soooo

[der8auer]

What alloy are you using?

"Alloy 82" (Indium Corp)?

 

Did you remove the oxidation prior applying the sTIM?

Did you remove the nickel plating from the thermal interface area (core-IHS)?

 

Questions, questions

 

 

I've got Alloy 1E, 4 and 290.

 

So I'm testing alloys with 100% In, 52In 48Sn and 97In 3Ag

 

52In 48Sn has lowest melting temperature with about 115°C but doesnt realy stick to the die. 100% In needs about 160°C soldering temperature and sticks properly to die and nickle of the IHS.

 

I did 2 tests with nickle plating and without nickle plating (pure copper). Seems to work better on the nickle tho.

 

Which oxidation do you mean?

 

My ss -48c 3770k 2core hits 5c on xp loading screen soooo

 

What voltage did you try?

[The Stilt]

Which oxidation do you mean?

 

Indium forms a very thin (nominal 100 Angstrom thick) oxide layer within a few hours under standard atmosphere. If you plan to use this foil as a laser mounting, cold welding or cryogenic gasketing media, it is important to remove the oxide layer prior to application.

 

- Degrease the indium with an organic solvent, such as acetone, to remove any organic contaminants.

- Mildly etch the indium in a solution of 5-10% hydrochloric acid (by volume) at room temperature for 1 to 5 minutes, depending on oxide thickness, until surface appears bright.

- Thoroughly rinse twice in DI water.

- Rinse off the water with acetone (preferred) or isopropyl alcohol.

- Blow-dry with dry nitrogen.

- Because this procedure etches the Indium surface, exposing it to potential oxidation, only the indium that is going to be used immediately should be cleaned by this procedure. If kept under standard conditions, the oxide will reform on the surface of the Indium in approximately 2-4 hours.

[Alex@ro]

So i tested on dry ice using inflection point the things you say.lowest pot temperature was -72 on 3770K and -70 on 4770K ,both at about 1.7V and 2c .At the end of the ssession i started prime 95 on both cpu's with 1.5V and 4c8t and monitored using realtemp.

 

From -70 to -35 pot temperature all 4 cores showed 0 0 0 0 .From -35 started going up.And this happened to 2 diff cpu from different generations.I find it absolutely hard to believe you had -110 pot temperature yet positive core temperature.Something is wrong.