( ESNUG 293 Item 10 ) --------------------------------------------- [6/10/98]
Subject: Electromigration Failures, Reliability, & Metal Coverage
> Can someone, please, explain me OR point me to articles on the subject
> WHY metal coverage is needed. I know that if such not done, metal etching
> will cause minimum-width metals to disconect due to over-etch, BUT would
> like to understand WHY..?
>
> - Yehuda D. Yizraeli
> Zoran Microelectronics Ltd. Haifa, Israel
From: David Shain <dshain@nr.infi.net>
If Aluminimum is usually half to 1 micron thick & if there is a severe
step, like into a contact, or over some underlying topography, thinning
can occur. If you only have 10% step coverage then the metal will only
be .05 to .1 microns. There is a problem with this well before
"catastrophic" disconnect. The problem is a reliability issue associated
with high current density. Alumimum interconnects are subject to
degrading when subjected to high current densities. when a line is
thinned to 10% of it's designed thickness, that means the current
density can be 10 times higher than allowed. The aluminimum line will
fail as a result.
In the days of wet etch, over etch was a problem with thin lines, but
now with dry etch you usually will not overetch a thing line.
There are also specification on minimum step coverage for yield reasons
that must be met.
- David Shain
---- ---- ---- ---- ---- ---- ----
From: jws@billy.mlb.semi.harris.com (James W. Swonger)
Electromigration failures are the probable basis of the question. High
reliability circuits generally are designed to meet a maximum current
density in interconnect. The metal thickness and width are of interest;
typically the highest current density is found where the interconnect metal
traverses an oxide step topographic feature (contact sidewall, diffusion
oxide cut, poly+ILD, etc.).
There are really two fundamental requirements for interconnect: mechanical
integrity and electrical integrity. The mechanical (ability to hang
together through thermal cycling/shock, stress voids from creep due to
internal tensile stress, etc.) requires some minimum thickness (approx 2kA
is my rule of thumb). Electrical integrity is design & application
dependent; for a decent aluminum system you are looking for a worst case
avg. current density of 2E5 A/cm2 taking into account duty cycle,
topography, & a peak current density of no more than 10X that, maybe less.
The ability of materials to carry high currents for long periods of time
varies with:
- composition: refractory metals are not seen to fail at all; gold
and aluminum are known to electromigrate. Metals of intermediate
melting point and hardness can be expected to have some electro-
migration behavior, I think; e.g. copper will probably fall
between aluminum and molybdenum somewhere. Aluminum interconnect
often contains trace amounts (1-2%) of copper to improve its
current density tolerance. Maximum actual current density is
determined by long term life test under stress, but the Mil
specs impose a fixed limit. Thus in many cases we require only
that the process we are designing in exceeds that value.
- temperature: high temperature is a strong accelerating factor in
electromigration wearout; this mechanism is the primary reason
for most IC processes' 175C junction temperature rating. Search
"Black's Law", "electromigration" in reliability physics periodicals.
- area: notching and thinning of metal lines, designed widths and
processing nonidealities.
The Mil specs call out step coverage percentages as a means of setting some
standards for interconnect step coverage processing acceptance. This is
somewhat wrongheaded, since you can have a thin metal at 70% that's got less
cross-sectional area than thicker metal at 50%. But rules are rules, until
you negotiate.
- James W. Swonger
Harris Semiconductor Melbourne, Florida
|
|
