
FBGA User’s Guide
76
Version 4.2, November 1, 2002
MSG
Figure 3 shows a cross-section of the die-up configuration of an FBGA package.
same size problem. Thus, with
die shrinks occurring regularly,
μ
BGAs are not an attractive CSP
solution for flash memories –
either for us or for our customers
– because the form/fit of the pack-
age must change with every die
shrink. This was becoming pain-
fully apparent to several top-tier
OEMs who were attempting to
use the
μ
BGA package for flash
memories.
FBGA CONSTRUCTION –
DIE SHRINK FRIENDLY
In the FBGA construction, the
die is wirebonded to a substrate
and then overmolded with epoxy,
a construction very much like that
of a standard BGA (see Figure 3).
The minimum package size winds
up being about 1.2 times the size of
the die, slightly larger than a com-
parable density
μ
BGA. But this size
advantage wins no points when it
comes to die shrinks. Because of
the FBGA construction, it can do
what the
μ
BGA cannot: accommo-
date a reduction in die size with
no change to the package dimen-
sions. This renders a die shrink
transparent mechanically to both
AMD’s and our customers’ manu-
facturing lines – clearly, a win-
win situation.
BUILDING INFRASTRUCTURE
For any new package technol-
ogy, it is critical that there be a
supporting industry infrastruc-
ture. We already knew that in the
flash market OEMs want:
the smallest, most powerful
flash memory they can get in a
package that has the largest
pitch;
multiple supply sources;
package designs that do not
change with every die shrink.
So we set our sights on qualify-
ing the FBGA and on fortifying
the infrastructure for it, the latter
of which was no small task in a
market where the
μ
BGA was al-
ready synonymous with CSP.
With no appreciable tooling or
reliability data available at the
time for the FBGA, we knew we
needed to tackle these four pre-
requisites to market acceptance:
board level reliability;
board level rework ability;
socket suppliers;
alternative supply sources.
Board Level Reliability
We began testing the reliability
of the FBGA after it was mounted
onto a PCB similar to those in use
by our major customers. We soon
learned that the substrate of the
FBGA – the polyimide tape – was
too thin to absorb the stress in-
curred from the different CTE (co-
efficient of thermal expansion)
rates of the silicon versus the PCB
(using FR4 material). When
heated, the PCB expands at a
much higher rate than the sili-
con, and the package substrate
has to manage this difference.
Although the polyimide substrate
proved reliable enough for many
applications, it did not meet the
long-lifetime reliability that is
needed in commercial and indus-
trial outdoor applications, such
as in the telecommunications
infrastructure or automotive
environments.
■
Tape vs Rigid Substrate
While we were testing the tape
FBGA, we also evaluated an FBGA
design that had an organic sub-
strate of BT (Bismaleimide Triaz-
ine) resin. Because this material
was thicker and had a CTE closer
to that of the PCB, it could better
manage the stress than could the
thinner polyimide tape. So we
switched. The results of the ex-
tensive tests on the board level
reliability performance can be
found in the white paper, “Reli-
ability Evaluation of Chip Scale
Packages,” published in 1999 by
the following AMDers: Ranjit
Gannamani, Vis Valluri, Sidharth
Sidharth, and MeiLu Zhang. A copy
of this paper can be obtained from
MSD Engineering (x26415).
As for the
μ
BGA, its board-level
reliability performance is compa-
rable to that of the FBGA due to
the ability of its elastomer layer
to absorb the stress from the
different CTE rates. These reliability