Thicker top metal same current density, but different resistance??

[ripoo82]

In I3T50 technology with 5 metals, the top metal is thicker as the internal metals. This results in a lower resistance, 55mOhm/sq for internal and 33mOhm/sq for the top.

The maximal current in those metals however is the same, 1,1mA/um. I should expect also a difference here, what can be the explanation for this?

 

[rangermad]

Are all the metals the same material? Some processes use aluminum as the top metal instead of copper.

 

[ripoo82]

As far as I can see, they are the same material.

 

[dick_freebird]

At 55mOhm/sq the thickness is not that different and you'd
expect only a ~50% higher current density all things being
equal. But "as far as you can see" doesn't really address that
aspect. It could be composition, or it could be the conduct
of the qualification testing (for example, top metal is the
furthest from the heat sink if this was done in a standard
package and considering the metal as the heat source, which
in accelerated wearout testing it is and if accelerated too
much, can be hundreds of degrees above case, and I have
seen more than once an experiment conducted and driven
all the way to groundrule documents that gave zero thought
or analysis to the in-conductor temperature, using case temp
alone in the regression despite it being the number one
acceleration factor).

You should not assume top metal is the same as the others.
Assembly concerns often make top metal a different composition.
For example AlCuSi in the lower layers and plain Al on the top,
maybe just because they have decades of experience in control
of intermetallics with one composition and dread the risk and
expense of qualifying an alternate, especially when it's really
someone else's process that needs controlling and nobody
wants a two-body problem.

Of course if you're outside the foundry you stand about zero
chance of getting them to revisit or even reveal the derivation
of these rules.

In modern CMP processes the electromigration performance
is often driven by the tungsten plug vias which "break the
chain" for metal migration, causing metal to pile up on one
side and erode away from the other in direction of current
flow. This is really a close-in via problem but often imputed
(wrongly in my view) to the metal line in clear field as well.
The via is also a bad heater, very localized and at exactly
the worst point (where metal moves away from the
"immovable object") and again if the experiment was done
without analysis (or better yet thermal imaging at specific
points, under experiment stress) bad outcome analysis and
derivation of rules will be the result. Top metal vias may be
different geometry, different height, different resistance
and thus different local heating (lumped blindly into the
result).

I've also seen tests where layers were each stressed at
different current densities resulting in discrepant (and again
un-comprehended) local heating. One test I saw at 1E7A/cm2
and simple thermal stack calculations indicated the aluminum
was running near liquidus. Oh, i got some love for bringing
that up. It took me a few years of bitching to get that
do-over.

 

[ripoo82]

Thank you for your detailed answer, that gave some insight.