which current produces Electromigration? AC or DC?

[ms_90]

My doubt is whether AC or DC current produces EM?

As per my knowledge we use only DC current to calculate the width of metal. Is this right? or we need to consider both currents?

and what will be the scenario if AC signal is passing through the metal?


Thanks in advance...

 

[dick_freebird]

At high enough frequency AC current is often ignored, or
assigned a higher limit current density value. When not
provided one I will use 10X the DC limit, for AC RMS. A
single-pulse (e.g. ESD) or sparse-pulse (<10% duty) I
will also apply the 10X-published-DC rule to.

Now, the picture has changed from the old days of all-
aluminum interconnect stacks. Systems using tungsten
plug vias or vias with other barrier layers can have the
via as the primary EM failure point, the barrier prevents
"make-up" material flow from below / above and gaps
form at the current-outflow edges of vias at lower
currents than would fail the attached line, in the field.
I've seen this misattributed to the minimum-metal line
when it is in fact a via / dogbone-extent issue. You as
an external foundry customer would probably not be
provided this insight. Point being that it is not always
just a metal linewidth calculation - weakest link, end
to end.

 

[cl10greg]

Well a quick Google search says the electromigration is more critical in DC applications. If you did use an AC signal, I would assume you would just use the RMS current to call it DC. But that is just a quick search and guess.

 

[dick_freebird]

AC current redeposits material back where it came from,
more or less. "Less" being the problem with the barrier
vias and potentially changing old rules-of-thumb that
once were conservative.

 

[ms_90]

At high enough frequency AC current is often ignored, or
assigned a higher limit current density value. When not
provided one I will use 10X the DC limit, for AC RMS. A
single-pulse (e.g. ESD) or sparse-pulse (<10% duty) I
will also apply the 10X-published-DC rule to.

Now, the picture has changed from the old days of all-
aluminum interconnect stacks. Systems using tungsten
plug vias or vias with other barrier layers can have the
via as the primary EM failure point, the barrier prevents
"make-up" material flow from below / above and gaps
form at the current-outflow edges of vias at lower
currents than would fail the attached line, in the field.
I've seen this misattributed to the minimum-metal line
when it is in fact a via / dogbone-extent issue. You as
an external foundry customer ould probably not be
provided this insight. Point being that it is not always
just a metal linewidth calculation - weakest link, end
to end.

It's getting to complicated. Can u please simplify it and give any example for both use of ac n dc currents for calculating the width in case of EM?

 

[dick_freebird]

You treat them as multiple cases for the same line, if the
current has both AC and DC character. Most common is
to use the time-averaged value if the current is unidirectional
against the DC rule. If time averaged value is low and peak is
high (like true AC) then an appropriate AC EM rule is wanted.
But this all comes down to cases, which need to be looked
at.

In any case J=I/A against a Jmax rule appropriate to the
current characteristics and application temperature, lifetime
goals. Your I will be figured variously, appropriately to the
current profile and your Jmax rule likewise.

The first, most useful thing to do is figure (or read) the
Jmax DC rule for a minimum width line, minimum contact,
etc. Then you can forget any net that falls below this.
That gets 90+% of the circuit off the table.

The rest of it, that's a simple boring slog through detail.

 

[ms_90]

You treat them as multiple cases for the same line, if the
current has both AC and DC character. Most common is
to use the time-averaged value if the current is unidirectional
against the DC rule. If time averaged value is low and peak is
high (like true AC) then an appropriate AC EM rule is wanted.
But this all comes down to cases, which need to be looked
at.

In any case J=I/A against a Jmax rule appropriate to the
current characteristics and application temperature, lifetime
goals. Your I will be figured variously, appropriately to the
current profile and your Jmax rule likewise.

The first, most useful thing to do is figure (or read) the
Jmax DC rule for a minimum width line, minimum contact,
etc. Then you can forget any net that falls below this.
That gets 90+% of the circuit off the table.

The rest of it, that's a simple boring slog through detail.

So now my question is, which current values we can get in the process manual? means Idc or Irms or both? and what is that appropriate AC EM rule you mentioned in your post?

 

[dick_freebird]

You should always see a DC rule. That's the outcome of the most
basic reliability qualification testing.

AC rules are often an afterthought, people looking for a way out
after the DC rules say their RF circuit is going to fail when they
use RMS current against a DC rule.

Properly, a second AC-stimulus electromigration lifetest would
be performed. Then you'd derive a righteous rule from the data.
But this costs money and time, and applying a generally agreeable
fudge factor to the DC rule is something I have done many a time.
I use 10X the DC number for balanced AC and also as a backstop
for very-low-duty-cycle (using peak value) waveforms where
a straight duty cycle average might say "no prob" but the
peak value exceeds DC limits. This wants somebody to buy off
on the predicate, from the customer or the internal reliability
organization, in absence of data and blessed rules.

 

[ms_90]

You should always see a DC rule. That's the outcome of the most
basic reliability qualification testing.

AC rules are often an afterthought, people looking for a way out
after the DC rules say their RF circuit is going to fail when they
use RMS current against a DC rule.

Properly, a second AC-stimulus electromigration lifetest would
be performed. Then you'd derive a righteous rule from the data.
But this costs money and time, and applying a generally agreeable
fudge factor to the DC rule is something I have done many a time.
I use 10X the DC number for balanced AC and also as a backstop
for very-low-duty-cycle (using peak value) waveforms where
a straight duty cycle average might say "no prob" but the
peak value exceeds DC limits. This wants somebody to buy off
on the predicate, from the customer or the internal reliability
organization, in absence of data and blessed rules.

Can you please explain me about EM rule,DC rule and AC rule?