Nondestructive 3D X-ray imaging

for advanced packaging failure analysis

[Zeiss Semiconductor Manufacturing Technology], and Daniel Nuez [Xilinx, Inc.]

P ackaging of integrated circuits

is growing more and more
complex – and housing multiple
die in a single package is just one challenge
chipmakers face. Typically, these die are
connected in complex ways, and chipmakers
must contend with shrinking feature sizes and
interconnects, escalating device density and
package size, thinner layers, and a widening
variety of materials. Figure 1: Acceptance of 3D X-ray microscopy is growing for failure analysis.
As a result, failure analysis (FA) on
nondest r uct ive met hod to f i nd a nd yellow dot in Figure 2a). The XRM detector
advanced packages is becoming increasingly
image defects in 3D. It thereby provides is composed of scintillator-coupled optical
difficult. The goal of FA is to isolate where
critical knowledge to guide next steps. microscope objectives combined with a
Appl icat ion of X R M t y pical ly f it s charge-coupled device (CCD) camera. The
it is and why it happened – its root cause.
between fault isolation and root cause X-rays pass through the sample and hit the
Visualization of defects aids determination
determination (Figure 1). scintillator mounted on the objective lens. The
of the root cause. Packages are essentially
Once the fault location is isolated, scintillator converts the pattern resulting from
opaque boxes cont ai n i ng elect r ical
traditionally, a next step is a visit to the X-rays transmitted through the sample into
connections. Often, to visualize a defect in the
the optical image captured on the right (Figure
electrical path, physical failure analysis (PFA)
that dest roy the sample are used to 2b). The sample is then rotated slightly,
is applied.
investigate the root cause of the failure. the image captured again, and this process
Maintaining integrity of the defect site
The techniques cited on the far right side is repeated through up to 360 degrees of
is critical. If a sample is cut or reduced
in Figure 1 all involve physically cutting, rotation. The resulting group of projections –
in size, further electrical analysis may
drilling or otherwise altering the sample typically, between 1,000 and 2,000 – are then
not be possible, and the structure may
in some way. If the fault is not properly processed by algorithms to mathematically
be disrupted by introducing artifacts or
reconstruct the 3D volume (Figure 2c).
changing the stress profile from that of an
The time required for the entire process
intact sample. Conventional nondestructive
Providing 3D intelligence ahead of is variable – typically ranging between 30
methods have become less effective at
destr uctive analysis is a key benef it minutes and 8 hours – depending on the
visualizing defects in many of today’s
of X R M. It enables h ig her success number of projections and how much time
packages, creating a significant need for
rates in cross-sectioning and finding is spent per projection. From the resulting
new nondestr uctive approaches such
root causes. Visualization of defects 3D volume, one can view any number of
as 3D X-ray microscopy (XRM).
by 3D XR M can even eliminate the horizontal or vertical cross sections through
need to perform PFA, saving time and the sample (Figure 2d) – essentially,
Benefits of X-ray microscopy resources. The case study included in isolating any desired sliver of the 3D volume.
In the typical board- and package- this article illustrates the effectiveness Therefore, details of fault locations can be
level FA lab workf low, failu res are of 3D XRM in the FA workf low. visualized without destroying the sample. As
evaluated nondest r uctively pr ior to an example, Figure 2d shows a virtual cross
destructive analysis (Figure 1). The most section of a 2.5D interposer stack.
common nondestructive PFA techniques
Visualizing defects nondestructively
with virtual cross sections The virtual cross-section plane can be
for isolating and visualizing defects moved interactively through the 3D dataset in
are optical inspection, 2D X-ray, and The power of 3D tomography comes from
its ability to provide virtual cross sections, any of the three orthogonal directions (x, y, z).
scanning acoustic microscopy (SAM).
Due to increased package complexity, revealing the details inside structures. Figure
2 areas, such as the substrate-side or chip-side
these imaging techniques are becoming of a flip-chip bump, and aids understanding
less effective. tomography process. Figure 2a shows that
data is acquired by collecting 2D projection of the failure mechanism. Figure 3 provides
XRM, a relatively new FA technique, another look at how a virtual cross section is
uniquely provides a high-resolution, images from a rotating sample positioned
between an X-ray source and a detector (the obtained from the reconstructed 3D dataset.

Reprinted from Chip Scale Review September • October • 2017 [ChipScaleReview.com] 1

 XRM vs. micro-CT
M ic r o c omput e d t omog r a phy, or
m ic r o - C T, i s a n o t h e r a p p r o a c h t o
obt ai n i ng 3D i m age s. Howeve r, a s
F i g u r e 4 i l l u s t r a t e s , X R M of fe r s
s ig n i f i c a n t r e s ol u t io n a d v a n t a g e s
compa red to m icro - CT. To ach ieve
high resolution in micro-CT systems,
high-geomet r ic mag nif ication is
requi red. T his involves placing the
sample ver y close to the sou rce
(Figure 4b) and moving the detector
as far away as possible – this ratio
det e r m i ne s t he mag n if icat ion a nd ,
t hu s , t h e r e s olu t io n of t h e i m a ge.
W i t h m i c r o - C T, l a r g e s a m p l e s
a re challengi ng to i mage at h ig h
resolution. As samples become
larger, they must be moved f u r ther
away from the source so they can be
rotated without colliding with it. As
the sample is moved away, there is a
Figure 2: XRM 3D tomography yields highly informative visual information about failures, nondestructively.
linear reduction in the magnification,
which, in turn, lowers the resolution
(Figure 4b).
The advantage of XRM is that high-
re solut ion i m age s ca n b e obt ai ne d
from f ully intact large samples that
are positioned further away from the
source (Figure 4a). The scintillator-
coupled microscope objectives provide
the magnification necessary to retain
r e s olu t io n ve r s u s d e p e n d i n g u p o n
geometric magnification alone. With
X R M, resolution remains relatively
independent of the package size, and
h ig h re solut ion ca n b e m a i nt a i ne d
with large sample sizes. This
capability represents the core value of
Figure 3: Using XRM, any plane through the 3D data may be viewed as a virtual cross section.
XRM vs. micro-CT technology.

Increasing the success rates of PFA

The following case study
demonstrates the benefits that 3D XRM
offers to chipmakers. In this instance,
a 2.5D inter poser test chip with
micro-bumps was used for packaging
development and process optimization.
In the center of Figure 5 is the package
computer-aided design (CAD) layout,
showing micro-bumps and larger C4
bumps. A short has been isolated to the
spot depicted in the green box at left.
At right is the 2D X-ray image – the
three micro-bumps are visible inside
the C4 bump, in the same orientation,
but it’s impossible to see where the
short is actually located.
The red dot ted line in the middle
image indicates the direction of the
Figure 4: XRM is well-suited for package FA because it maintains resolution regardless of package size.

2 Reprinted from Chip Scale Review September • October • 2017 [ChipScaleReview.com]

physical cut perfor med with PFA in
an effort to visualize the short. As the
orange line labeled “solder extrusion”
shows – and as was later determined
using X R M – the shor t f rom one
micro-bu mp to another exists at an
a ngle. X R M also revealed t hat t he
size and mass of the short was below
the detection limits of the 2D X-ray
system.
The failure analyst repeatedly cut and
polished the sample to get as close as
possible to the failure site. An anomaly
Figure 5: In a sample 2.5D interposer chip, an electrical failure was found at pin BC14, but 2D X-ray inspection
in the for m of solder ext r usion was
failed to show any structural anomaly.
observed (see Figure 6) and suspected
to be the cause of the short. An optical expose the failure for
image is on the left, and a SEM image imaging. Moreover, the
is on the right. Visual evidence of a process may introduce
short across adjacent bumps is missing artifacts from cutting
in both optical and SEM images. The a nd polish i ng that
a n a ly s t c o nt i n u e d t o p ol i s h a b o u t can hinder root cause
10 microns further, and as Figure 7 determination. Defects
shows, polished through evidence of may be missed, leading
the short. Although the electrical data the failure analyst to
pointed to the short’s general location, conclude that no defect
more precise information was needed could be found. Figure 6: The first cut using PFA revealed the solder extrusion, but not the
to successfully conf irm the short by With its high- bump-to-bump connection, requiring further cutting and polishing.
destructive PFA. resolution and
3D X R M c a n r eve a l d e t a i l s of nondestructive
a solde r br idge (locat ion , si ze a nd properties, 3D
o r i e n t a t i o n) p r i o r t o d e s t r u c t i v e XR M imaging
analysis. This infor mation can then and analysis has
guide and enable successful execution become increasingly
of a precise cut into the solder bridge. com monplace i n FA
As Figure 7 shows (right image), there workflows, particularly
wa s ev id e nc e of s old e r ex t r u sion s for adva nced 2.5D
in adjacent bumps. Before attacking a nd 3D pa ck ag i ng
t he sa mple f u r t he r w it h cont i nue d architectures. By Figure 7: The second polish destroyed the physical evidence of a micro-
polishing, a defective area was imaged prov id i ng det ailed bump short on pin BC14. Nondestructive 3D XRM tomographic imaging was
by 3D XRM using sub-micron voxels. 3D images of failure performed on adjacent bump BB15 due to evidence of solder extrusion in the
The 3D rendering in Figure 8 shows locations, it is a cross section’s optical image.
the exact location and orientation of valuable precursor to
the short, which guided the subsequent – and in some cases,
destructive cross-sectioning angle and can completely replace
resulted in a successful FA report. – physical cross-
sectioning.
Summary
Package technology is growing in Biographies
complexity and the FA workflow needs Cher yl Har tf ield
t o a d a pt t o t he new r e q u i r e me nt s. r e c e i v e d h e r
Conventional FA involves cutting into MA and BS in
samples and polishing the edge to the M i c r o b i ol o g y f r o m
approximate location of the failure. Then UT Southwester n
SEM and/or optical micrographs are used Medical Center
to capture high-resolution 2D images a n d T e x a s A & M , Figure 8: 3D XRM data confirmed the defect with no destruction of the chip sample.
in order to help determine the failure’s respectively. She is
root cause. While valuable for some Solutions Manager for X-ray microscopy at Zeiss Semiconductor Manufacturing
applications, this approach is destructive Technology; email cheryl.hartfield@zeiss.com
– it provides a single chance to choose Daniel Nuez received his BS in Information Systems from U. of San Francisco
the right cutting orientation that will and is a Senior Device Analysis Engineer at Xilinx, Inc.

Reprinted from Chip Scale Review September • October • 2017 [ChipScaleReview.com] 3

PCS Group
Carl ZEISS SMT

www.zeiss.com/pcs
4385 Hopyard Road
Pleasanton, CA 94588 USA
Reprinted by Carl Zeiss SMT with permission from Chip Scale Review.
© 2017 Chip Scale Review