CPU Diffusion Barrier?

[Daharen]

Is the CPU Diffusion Barrier a special material other then Silicon placed on top of the Silicon die? Or is the Silicon itself the diffusion barrier, and when people warn about not lapping your die, they're just saying 'removing' material could allow what limited diffusion already occurs to reach the underlying circuitry since there is 'less' material to diffuse through?

If it is a separate material, is it possible to lap the die and reapply your own diffusion barrier to get a flatter surface without losing the protection?

[Lord Nicoll]

3 minutes ago, Daharen said:

Is the CPU Diffusion Barrier a special material other then Silicon placed on top of the Silicon die? Or is the Silicon itself the diffusion barrier, and when people warn about not lapping your die, they're just saying 'removing' material could allow what limited diffusion already occurs to reach the underlying circuitry since there is 'less' material to diffuse through?

If it is a separate material, is it possible to lap the die and reapply your own diffusion barrier to get a flatter surface without losing the protection?

The silicon diffusion barrier on most CPU's is silicon nitride. You cannot redo this coating if you lap the die down. I don't really know if liquid metal can diffuse into silicon readily over time, it might be a case of it takes 10+ years, or it could be a case of "Well I got two years out of this thing at 1.4v 5.4GHz so I guess it was either this or it degrades in the same time span". 

Edited June 19, 2019 by Lord Nicoll
edit. Added some of my opinion to the comment

[Enderman]

The die already is flat.

The lapping is done to remove layers to reduce thickness and bring the heatsink closer to the silicon.

[Lord Nicoll]

2 minutes ago, Enderman said:

The die already is flat.

The lapping is done to remove layers to reduce thickness and bring the heatsink closer to the silicon.

Eh, it's not always flat flat, there are often cases of it being not flat flat. Now, when I say not flat, we're talking machinists not flat, not IHS not flat, might be a few thousands of inch higher on side than another, or have a bit of a shallow in the middle. The reduction in thickness is still the main benefit though. 

[Daharen]

Just now, Enderman said:

The die already is flat.

The lapping is done to remove layers to reduce thickness and bring the heatsink closer to the silicon.

Using a micrometer I've measured sizable differences in the height that of parts of the silicon. It's a 5.3 Silicon Lottery 8700k so it overclocks fine with paste, but with direct liquid metal application I could push it further, but liquid metal doesn't gap fill as well as thermal paste, so core #4 which happens to be located under the low part of the silicon die not recieving proper contact rapidly fluctuates in temperatures, hitting up to 90 c, while all the rest of my cores never exceed 40 c, and the average temperature of all cores remains around 26c. core #4 also averages 26 c when not fluctuating, but I'm guessing that's because of lateral heat transfer, and when under load it would benefit from better direct contact with the heat spreader. 

To that end, I'm planning on lapping the die to make it closer to perfectly flat. In Der8aur's case the difference was up to 0.04 mm in height variation, for me it's more severe at 0.06 mm over core #4. Ideally I would find a way to not compromise the diffusion barrier, but I might just risk it since I may contemplate upgrading to AMD if the specs pan out. 

[Enderman]

2 minutes ago, Lord Nicoll said:

Eh, it's not always flat flat, there are often cases of it being not flat flat. Now, when I say not flat, we're talking machinists not flat, not IHS not flat, might be a few thousands of inch higher on side than another, or have a bit of a shallow in the middle. The reduction in thickness is still the main benefit though. 

The pressure from the heatsink and the flexibility of the silicon and fiberglass means that it will flatten regardless of a few thou variation.

[Daharen]

 

[Enderman]

Just now, Daharen said:

Using a micrometer I've measured sizable differences in the height that of parts of the silicon. It's a 5.3 Silicon Lottery 8700k so it overclocks fine with paste, but with direct liquid metal application I could push it further, but liquid metal doesn't gap fill as well as thermal paste, so core #4 which happens to be located under the low part of the silicon die not recieving proper contact rapidly fluctuates in temperatures, hitting up to 90 c, while all the rest of my cores never exceed 40 c, and the average temperature of all cores remains around 26c. core #4 also averages 26 c when not fluctuating, but I'm guessing that's because of lateral heat transfer, and when under load it would benefit from better direct contact with the heat spreader. 

To that end, I'm planning on lapping the die to make it closer to perfectly flat. In Der8aur's case the difference was up to 0.04 mm in height variation, for me it's more severe at 0.06 mm over core #4. Ideally I would find a way to not compromise the diffusion barrier, but I might just risk it since I may contemplate upgrading to AMD if the specs pan out. 

That sounds like you have an air bubble and bad liquid metal application because that would not be happening even if there was a thick layer of thermal paste and large surface variation.

 

Could also be your heatsink mounting.

You're using the der8auer delid die bracket, correct?

[Daharen]

1 minute ago, Enderman said:

That sounds like you have an air bubble and bad liquid metal application because that would not be happening even if there was a thick layer of thermal paste and large surface variation.

 

Could also be your heatsink mounting.

You're using the der8auer delid die bracket, correct?

Yep, I'd also agree with you by the way, except I've remounted it and reapplied now a total of 3 times with the same result in the same spot. Each time maximizing the pressure just short of using tools to increase mounting pressure (Though it's so thumb tightened I need to use something to unscrew the mount often for reapplication). I doubt it's mounting pressure given that fact, and I'm fairly confident in saying that if a bubble was appearing consistently in the same spot that would not likely be application error, but something to do with the shape between the two surfaces, as I can't imagine being that consistently unlucky. 

With paste, it doesn't happen, I get higher average temps, around 30c, and max out around 50-60c under maximum load, and there is no significant core to core variation. The issue only occurs with liquid metal. I am applying to both the block and the die, and getting full coverage with no pooling. In one instance (This was a 4th application I'm not counting in the three mentioned above), I tried a more generous application of liquid metal, but as expected that didn't have good general results, as liquid metal really doesn't gap fill even when overapplied, tending to eventually settle or be pressed out from between the mounting surfaces. 

Anyway, it may not be height variation, but it's the only thing I haven't really tried. I was on the forums a week or so ago about this problem, and have been troubleshooting since then, this is the last step and if this doesn't work I really have no clue what is going on and might just have to switch back to paste since I know that works at least.