TechSolutionsWLCSPMarket

TECHNOLOGY SOLUTIONS FOR A DYNAMIC
AND DIVERSE WLCSP MARKET
Ravi Chilukuri
Amkor Technology
Research Triangle Park, NC, USA
Ravi.Chilukuri@amkor.com
ABSTRACT
There are multiple technology platforms in development to
cater to the cost and performance requirements of the
diverse application space for WLCSP. The cost and cycle
time considerations have necessitated technologies that
utilize fewer process steps, while maintaining quality and
reliability standards. This paper examines material options
i.e., polymers and solder alloys for these new structures and
will also examine the effects of die sizes and I/O counts on
product reliability. Board level reliability (BLR) data and
analyses of the failure modes is presented.
Key words: WLCSP, BLR, CSPn3
, CSPn2
, CSPnl
INTRODUCTION
The advent of WLCSP (Wafer Level Chip Scale Package) in
the semiconductor industry was driven by a strong push for
cost-reduction and miniaturization. The primary utilizers of
this technology have been handheld or portable products,
such as cell-phones, ipods, laptops and netbooks. However,
more recently, WLCSPs are gaining foothold in other
industries, such as servers and automotive electronics.
Initially, reliability performance limited the use of WLCSP
to smaller die sizes (<2.5mm), lower pin counts (< 25) and
mature silicon technology nodes, making WLCSP an
excellent match for the analog/mixed signal space. With the
maturity of WLCSP in this market, WLCSP has transitioned
from an advanced package to a commodity product. Lower
cost WLCSP solutions have become a requirement due to
growing price pressure and inflation of cost of materials and
equipment. The cycle time pressure has increased due to the
changing business models and supply chain strategies in the
new economic environment. To meet these growing market
demands, WLCSP providers are faced with the challenges
of providing faster cycle times and higher capacity without
significant increases in capital expenditure.
Simultaneously, several recent technology advances have
enabled use of WLCSP in products with pin counts up to
200 and die sizes up to 7mm, opening the application space
to RF, high speed, broadband and memory components as
well. Consequently, WLCSP is expanding to markets and
applications previously supported by QFN and flip chip
CSP. This expansion puts additional price and cycle time
pressure on WLCSP manufacturing. Further, many of the
above applications require advanced silicon technology
nodes using low-K dielectrics, which present unique
challenges of their own. Unlike the QFN and fcCSP
packages, where the low-K silicon die is fully encapsulated
and supported by underfill/overmold, the WLCSP has
exposed die in which the fragile low-K layer needs to
withstand higher effective mechanical stress and the die
chip-out needs to meet more stringent criteria. To address
these concerns, there has been development involving
requiring new materials, processing techniques and design
rules.
Hence, WLCSP technology development has taken the
approach of focus on cost reduction as well as on functional
integration. The focus on cost reduction applies more to the
mixed signal/analog space using 200mm wafers primarily
where WLCSP has become a commodity. The focus on
functional integration has involved developing technologies
compatible with advanced silicon nodes (90nm down to
45nm), technologies for die with higher pin count (~200)
and wafer level fan-out technologies. Much of this
application space uses 300mm wafers.
In addition, the board level reliability requirements have
become more stringent for both temperature cycling and
drop tests. With miniaturization and addition of
functionality for the smart phones and other advanced
electronic gadgets, power management and thermal
dissipation requirements have increased. Thereby,
significant improvement in temperature cycling
requirements have become a primary concerns for many
applications. Further, with the introduction of WLCSPs into
the smart power IC space involving memory components,
there has been a need for polymer materials with lower cure
temperatures that preserve the programming of memory
components during WLCSP processing. The RF and high
speed devices benefit from having polymer layers with low
dielectric constant and high breakdown strength. Many of
these needs can be catered by the use of PBO based
polymers.
This paper examines material options i.e., polymers and
solder alloys for new lower cost WLCSP structures as well
as the effects of die sizes and I/O counts on product
reliability. Typical failure modes are also presented for
some of the cases.
BACKGROUND
There are currently several flavors of WLCSP technologies
in the industry, most of which use a 4 mask approach
consisting of 2 layers of polymer, a redistribution layer and
1
Received "Best of Conference" Technical Paper award at the 7th Annual International-Wafer Level Packaging Conference
Oct. 11-14, 2010 Santa Clara, CA

a UBM (under bump metallurgy) under the solder bump,
similar to Amkor’s WLCSP offering, CSPnl
. As discussed
previously [10], the recent focus has been to develop cost
effective RDL based WLCSP options (CSPn3
and CSPn2
)
that are meet the reliability requirements for most WLCSP
applications. Both of these technologies are derivatives of
the current HVM offering of CSPnl
. A comparison of these
technologies is provided in Figure 1. Unlike CSPnl
, which
uses a 4 mask process, the CSPn3
is a three mask stack-up
that provides cost and cycle time savings through
elimination of process steps and lower material costs.
Further cost reduction can be realized through CSPn2
, where
the first polymer layer is left out of the device stack-up.
CSPnl
CSPn3
(CSPnl thick Cu UBM)
1) Polymer Coat 4) UBM
2) RDL 5) Ball Place
3) Polymer Coat
1) Polymer Coat
2) RDL/UBM 4) Ball Place
3) Polymer Coat
NO1st Polymer Coat
1) RDL/UBM 3) Ball Place
2) Polymer Coat
CSPn2
(CSPX3 minus Polymer 1)
Polyimide 2
Figure 1. Amkor WLCSP stack-up options: CSPnl
, CSPn3
and CSPn2
Early data demonstrated that the BLR (board level
reliability) performance of the above three technologies is
comparable [11] and meet the generally accepted guidelines
for reliability performance. CSPn3
has since been tested for
package level reliability using PBO, in addition to
Polyimide (PI). Extensive BLR studies have also been done
comparing PBO to PI on CSPn3
. Solder alloy comparisons
have been done using BLR on both CSPn3
and CSPn2
.
PACKAGE LEVEL RELIABILITY
The CSPn3
structure was built using PI for both the first and
second polymer levels. The die design used for the BLR
tests was a 5.5mm x 5.5mm die with a 10x10 full bump
array at 0.5mm pitch (Figure 2). The wafers were
background to 11mil silicon thickness prior to dicing the
parts. A number of package level reliability tests were
performed on these parts, as listed along with the passing
conditions in Table 1. All the parts were preconditioned at
MSL1 requirements. Parts were electrically and optically
tested prior to preconditioning, after preconditioning and
after reliability testing. Bump shears were also performed
following the reliability tests. The same process and tests
were repeated on parts built using PBO for both the first and
second polymer levels.
Table 1. Package level reliability tests performed on
Polyimide and PBO based CSPn3
structures.
TestItem Test/Pass Condition
Multiple Reflow 260C, 5X
Preconditioning at Level 1 85°C/85%, 168hrs, reflow @260°C peak
Preconditioning at Level 3 30°C/60%, 192hrs, reflow @260°C peak
Autoclave (PCT) 121C, 2 atm, 100% RH, 168 hrs
Unbiased HAST 130C, 85% RH, 168 hrs
Temp Cycle (TC) -55C/+125C, 1000 cycles
High Temp Storage (HTS) 150C, 1000 hrs
Figure 2. Die design used for package reliability tests:
10x10 I/O, 0.5mm pitch, 5.5mmx5.5mm, 0.3mm sphere.
PI VERSUS PBO BLR COMPARISON ON CSPn3
In addition to the package level reliability, board level
reliability (BLR) was also evaluated for PI based CSPn3
versus PBO based CSPn3
. The die design used for the BLR
tests was a 5.3mm x 5.3mm die with a 12x12 full bump
array at 0.4mm pitch (Figure 3). The solder alloy used was
SAC405.
Figure 3. Die design used for board level reliability tests:
12x12 I/O, 0.4mm pitch, 5.3mmx5.3mm, 0.25mm sphere.
For BLR, JEDEC drop test and thermal cycling were
evaluated. The drop test board was an 8 layer board, and
the thermal cycle board had 4 layers. The boards used were
NSMD with 250um diameter Cu pads.
The JEDEC thermal cycling condition used for this study
was JEDEC JESD22-A104C, Condition G [4]. Parts were
cycled between -40 to 125C at 1 cycle/hour. The
temperature ramp was 18min and the dwell time at
temperature was 12min. Three boards with 15 mounted
2

WLCSP devices were used for each test case. The
comparison for PI based CSPn3
versus PBO based CSPn3
shows > 2X improvement in the mean life with the use of
PBO (Figure 4). With the PBO, mean life was greater than
1000 cycles for this 144 I/O die with first fails of greater
than 600 cycles. The failure modes for both the PI and PBO
versions were noted to be in bulk solder.
ReliaSoft W eibull++ 7 - www.ReliaSoft.com
Cycles to Failure
Cumulative%Failed
10 10000100 1000
1
5
10
50
90
99
PBO
CSPn3 PI
CSPn3 PBO
PI
Figure 4. Temperature cycling performance comparing PI
and PBO versions of CSPn3
.
The setup of the board level drop tester follows the
guidelines of JEDEC drop test standards [1-3], with 15
WLCSP devices (3x5 matrix) assembled on a test board
(132x77mm). The test board is connected to a fixture with
the die side facing down and then mounted to a drop block
with screws in the four corners of the PCB. This procedure
insures that the board experiences maximum flexure. The
drop block is dropped from a certain height along two
guiding rods, onto a rigid base covered with a rubber layer.
A multi-channel in-situ high speed data acquisition system
is used to measure input/output acceleration, in-plane strains
on board, and resistance of daisy-chained components. The
JEDEC drop test standard recommends certain input
acceleration values and pulse shapes [1-3]. The testing
described in this paper applied the aggressive JEDEC
condition H (2900G/0.3ms). The widely used JEDEC
condition B (1500G/0.5ms) proved to be impractical for
these parts due to the extensive testing time that would have
been required to collect sufficient data for statistical
analysis. Four boards per test case were utilized for drop
testing.
Drops to Failure
Cumulative%Failed
10.000 10000.000100.000 1000.000
1.000
5.000
10.000
50.000
90.000
99.000
PBO
PI
Jedec JESD22-B111
Condition H (2900G, 0.3ms)
Grp A
CSPn3 PI
CSPn3 PBO
Figure 5. JEDEC drop test performance comparing PI and
PBO versions of CSPn3
.
Previously, the drop performance of CSPn3
for both JEDEC
drop condition H (2900G/0.3ms) and condition B
(1500G/0.5ms) was compared [10]. From failure analysis,
the primary failure mode was found to be separation through
the IM at the Pb-free / Cu interface. The acceleration factor
between the two different drop conditions for this type of
IMC failure was 3X. Consequently, if the first failure (FF)
of a design under the condition H test conditions is 100,
then the FF under condition B would be estimated to be 300.
A comparison of the condition H drop test for PI versus
PBO shows no difference in the drop test performance
(Figure 5). For both cases, the mean life was > 1000 drops
at condition H indicating an expected life of about >3000
drops at the industry wide typically used condition B using
the above acceleration factor.
SOLDER ALLOY BLR COMPARISON ON CSPn3
Board level reliability (BLR) was also evaluated for
multiple solder alloys for PI based CSPn3
. The solder alloys
compared were SAC405, SAC105 and LF35. SAC405 was
selected for its higher silver content, which has typically
provided better TC performance than the lower silver
content alloys such as SAC105 on standard WLCSP stack-
ups. SAC105 was chosen due to the better drop test
performance that it has typically exhibited on standard
WLCSP stack-ups. LF35 was chosen due to the additional
dopant it contains that is expected to improve drop test
performance when the failures are in the IMC. The die
design used for the BLR tests was a 5.3mm x 5.3mm die
with a 12x12 full bump array at 0.4mm pitch (Figure 3).
The boards used were NSMD with 250um diameter Cu
pads. The BLR test conditions used were the same as those
described in the previous section.
There was no significant difference in the thermal cycling
mean life with different solder alloys on CSPn3
(Figure 6).
The plots for SAC105 and LF35 overlapped, as may be
expected due to the similar silver content in these two
alloys. Interestingly, the SAC405 performance was
equivalent to SAC105 for the CSPn3
structure, within the
3

limits of standard test variation. Failure modes for thermal
cycle stressing were typically bulk solder failures.
Likewise, the drop test performance of the different solder
alloys on CSPn3
was equivalent as well (Figure 7). Despite
using the more aggressive drop test condition H, the first
fails were at > 600 drops. This performance far exceeds the
typical industry requirements for drop tests. The primary
failure mode for drop was noted to be at the copper solder
inter-metallic layer.
Cycles to Failure
Cumulative%Failed
10 10000100 1000
1
5
10
50
90
99
SAC405
SAC105
LF35
Figure 6. Temperature cycling performance comparing
solder alloys on PI based CSPn3
.
Drops to Failure
Cumulative%Failed
10.000 10000.000100.000 1000.000
1.000
5.000
10.000
50.000
90.000
99.000
SAC405
SAC105
LF35
Figure 7. JEDEC drop test performance comparing solder
alloys on PI based CSPn3
.
SOLDER ALLOY BLR COMPARISON ON CSPn2
Board level reliability (BLR) was also evaluated for
multiple solder alloys for PI based CSPn3
. The solder alloys
compared were SAC405 and SAC105. The die design used
for the BLR tests was a 5.5mm x 5.5mm die with a 10x10
full bump array at 0.5mm pitch (Figure 2). The boards used
were NSMD with 235um diameter Cu pads. The BLR test
conditions used were the same as those described in the
previous section.
mean life with different solder alloys on CSPn2
(Figure 8).
The SAC405 performance was equivalent to SAC105 for
the CSPn2
structure as well, within the limits of standard test
variation.
Likewise, the drop test performance of the different solder
alloys on CSPn2
was equivalent as well (Figure 9). Despite
using the more aggressive drop test condition H, the first
fails were at > 500 drops. This performance far exceeds the
typical industry requirements for drop tests.
Cycles to FailureCumulative%Failed
100 100001000
1
5
10
50
90
99
SAC405SAC105
solder alloys on PI based CSPn2
.
Drops to Failure
Cumulative%Failed
10.000 10000.000100.000 1000.000
1.000
5.000
10.000
50.000
90.000
99.000
SAC405
SAC105
Jedec JESD22-B111
Grp A
Figure 9. JEDEC drop test performance comparing solder
alloys on PI based CSPn2
.
BUMP PITCH/HEIGHT BLR COMPARISON
The most common bump pitches used for WLCSP are
0.4mm and 0.5mm. The 0.5mm pitch is fairly common on
the analog and power management ICs, and on devices
going into lower end phones. The 0.4mm pitch has become
main stream in the last couple of years for WLCSP with its
growth in RF/high speed devices. There has also been
development at the 0.3mm pitch as well, but there have been
concerns regarding the higher assembly and board costs. For
testing the impact of the different pitches on CSPn3
, test
vehicles were chosen with the same bump array at both
0.5mm pitch and 0.4mm pitch (Figure 10). The bump
height for 0.5mm pitch and 0.4mm pitch devices was
250um and 210um, respectively. The Board level reliability
(BLR) was evaluated using these test vehicles for PI based
4

CSPn3
. The solder alloy used was SAC405. The 0.5mm
pitch die design used was a 5.5mm x 5.5mm die with a
10x10 full bump array. The boards used were NSMD with
235um diameter Cu pads. The 0.4mm pitch die design used
was a 4.5mm x 4.5mm die with a 10x10 full bump array.
Figure 10. Die design used for package reliability tests:
0.5mm pitch die - 10x10 I/O, 5.5mmx5.5mm, 0.3mm
sphere; 0.4mm pitch die - 10x10 I/O, 4.5mmx4.5mm,
0.25mm sphere.
performance of the two test vehicle on CSPn3
(Figure 11).
Both samples exhibited first fails close to or greater than
1000 cycles, with mean life close to 2000 cycles.
ReliaSoft Weibull++ 7 - www.ReliaSoft.com
Cycles to Failure
Cumulative%Failed
100 100001000
1
5
10
50
90
99
0.5mm pitch,
250um height
0.4mm pitch,
210um height
0.4mm bump pitch to 0.5mm pitch die on PI based CSPn3
.
However, the drop test performance of the 0.4mm pitch
device was better than that the 0.5mm pitch device (Figure
12). Despite using the more aggressive drop test condition
H, the first fails were at > 400 drops for both samples. The
mean life for 0.4mm pitch devices was however 2X greater
than the 0.5mm pitch devices. This is currently under
investigation.
Drops to Failure
Cumulative%Failed
10 10000100 1000
1
5
10
50
90
99
Jedec JESD22-B111
Grp A
0.4mm pitch,
210um height
0.5mm pitch,
250um height
Figure 12. JEDEC drop test performance comparing 0.4mm
bump pitch to 0.5mm pitch on PI based CSPn3
.
TIME DELAY STUDY
For CSPn3
and CSPn2
structures, ball loading is done directly
on to the copper surface. Due to the potential for copper to
oxidize, time delay studies were performed between the
RDL/UBM layer processing and ball loading.
Board level reliability (BLR) was evaluated for samples that
had significant time delay between the above steps for PI
based CSPn3
. The solder alloy used was SAC405. The die
design used for the BLR tests was a 5.5mm x 5.5mm die
with a 10x10 full bump array at 0.5mm pitch (Figure 2).
Cycles to Failure
Cumulative%Failed
10 10000100 1000
1
5
10
50
90
99
With Delay
Standard Process
standard process on PI based CSPn3
with a process with
significant time delay between copper redistribution/UBM
and ball load.
For temperature cycling, the standard process for CSPn3
resulted in industry preferred performance with first fails >
1000 cycles and mean life > 1500 cycles for a 5.5mm die
(Figure 13). The typical failure mode was through the
solder IMC. As anticipated, the samples with significant
time delay exhibited much lower lifetimes than that
exhibited by the standard process. The failure mode in this
5

case was separation of the solder bump from the copper
trace at the copper – IMC interface.
For the drop test, the standard process for CSPn3
resulted in
excellent performance with first fails > 600 drops and mean
life > 1500 drops, using the more aggressive drop test
condition H (Figure 14). This performance far exceeds the
typical industry requirements for drop tests. The typical
failure mode was through the bulk solder, as typically
observed for temperature cycling. As anticipated, the
samples with significant time delay exhibited much lower
lifetimes than that exhibited by the standard process. The
failure mode in this case was separation of the solder bump
from the copper trace at the copper – IMC interface.
Drops to Failure
Cumulative%Failed
1 1000010 100 1000
1
5
10
50
90
99
With Delay
Standard Process
Figure 14. JEDEC drop test performance comparing
standard process on PI based CSPn3
with a process with
significant time delay between copper redistribution/UBM
and ball load.
CONCLUSIONS
The BLR drop and thermal cycling data for CSPn3
and
CSPn2
position these technologies as cost effective and
viable WLCSP options. A comparison of PBO and PI based
CSPn3
technologies exhibited >2X improvement in thermal
cycling life and comparable drop performance using PBO.
With different solder alloys, CSPn3
exhibited equivalent TC
and drop performance. The 0.4mm and 0.5mm pitch
devices showed equivalent TC performance. Processing
time delay between UBM and ball load demonstrated to be
significant for package reliability.
ACKNOWLEDGEMENTS
Appreciation is extended to our colleagues from the R&D
group of Amkor Korea: SS Park, ES Yang, DH Moon, SW
Cha, TK Hwang and WJ Kang, as well as former
colleagues, Rex Anderson and Boyd Rogers, for their
technical support and efforts.
REFERENCES
[1] JEDEC Standard JESD22-B111, Board Level Drop Test
Method of Components for Handheld Electronic
Products, 2003.
[2] JEDEC Standard JESD22-B104-B, Mechanical Shock,
2001.
[3] JEDEC Standard JESD22-B110, Subassembly
Mechanical Shock, 2001.
[4] JEDEC Standard JESD22-A104C, Temperature Cycling,
2005.
[5] R. Anderson, T.Y. Tee, R. Chilukuri, H.S. NG, C.P.
Koo, and B. Rogers, “Amkor’s CSPnlTM
: Comparison of
Bump on Pad and Cu Redistribution WLCSP Designs",
IMAPS Proc., Scottsdale, AZ, (2009).
[6] T.Y. Tee, H.S. Ng, A. Syed, R. Anderson, C.P. Khoo, B.
Rogers “Design for Board Trace Reliability of WLCSP
under Drop Test”, 10th Eurosime Conference,
Netherlands, (2009).
[7] T.Y. Tee, L.B. Tan, R. Anderson, H.S. Ng, J.H. Low,
C.P. Khoo, R. Moody, B. Rogers, “Advanced Analysis
of WLCSP Copper Interconnect Reliability under Board
Level Drop Test”, EPTC Proc, Singapore, 1086-1095
(2008).
[8] R. Anderson, T.Y. Tee, R. Moody, L.B. Tan, H.S. NG,
J.H. Low, and B. Rogers, “Integrated Testing &
Modeling Analysis of CSPnl™ for Enhanced Board
Level Reliability”, IWLPC Proc., San Jose, CA, 184-
190, (2008).
[9] R. Anderson, Robert Moody, Boyd Rogers, and Dan
Mis, “Board Level Reliability Results for Amkor’s
12x12 I/O CSPnl™”, IMAPS Proc., Scottsdale, AZ
(2008).
[10] R. Anderson, R. Chilukuri, T.Y. Tee, C.P. Koo, H.S.
NG, B. Rogers, and A. Syed, “ Advances in WLCSP
technologies for growing market needs”, IWLPC Proc.,
San Jose, CA (2009).
[11] R. Anderson, R. Chilukuri, B. Rogers, and A. Syed, “
Advances in WLCSP technologies to enable cost
reduction”, IMAPS Proc., Scottsdale, AZ (2010).
6

TechSolutionsWLCSPMarket

More Related Content

Viewers also liked

Similar to TechSolutionsWLCSPMarket

TechSolutionsWLCSPMarket