Guarantee Reliability with Vibration Simulation and Testing...o Random vibration –product...
Transcript of Guarantee Reliability with Vibration Simulation and Testing...o Random vibration –product...
Guaranteeing Reliability with
Vibration Simulation and
TestingDr. Nathan Blattau
© 2004 - 2007© 2004 - 2010
o Nathan Blattau, Ph.D. - Senior Vice President
Has been involved in the packaging and reliability of electronic equipment for more than ten years. His specialties include best practices in design for reliability, robustness of Pb-free, failure analysis, accelerated test plan development, finite element analysis, solder joint reliability, fracture, and fatigue mechanics of materials.
.
9000 Virginia Manor Rd. Ste. 290, Beltsville MD 20705 | 301-474-0607 | www.dfrsolutions.com2
© 2004 - 2007© 2004 - 2010
o High cycle fatigue (HCF) due to mechanical stress induced
by vibration
o Millions of cycles to failure
o Small changes in stress have large impacts on time to
failure
o According to U.S Air-Force statistics 20 percent of all
failures observed in electronic equipment are due to
vibration problems
Vibration Fatigue
Steinberg D.S. Vibration analysis for
electronic equipment.
John Wiley & Sons, 2000.
3
© 2004 - 2007© 2004 - 20104
Reduce Costs by Improving
Reliability Upfront
Designing in Reliability, Earlier is Cheaper
4
© 2004 - 2007© 2004 - 2010 5
Lifetime under mechanical cycling
is divided into two regimes Low cycle fatigue (LCF)
High cycle fatigue (HCF)
LCF is driven by inelastic strain
(Coffin-Manson)
HCF is driven by elastic strain
(Basquin)
Vibration Fatigue
b
f
f
e NE
2
c
ffp N2 -0.5 < c < -0.7; 1.4 < -1/c > 2
-0.05 < b < -0.12; 8 > -1/b > 20
5
© 2004 - 2007© 2004 - 2010
Vibration – Harmonic and Random
o Single frequency o Random vibration is a continuous
spectrum of frequencies
AN INTRODUCTION TO RANDOM VIBRATION – Tom Irvine
MIL-STD-810G
6
© 2004 - 2007© 2004 - 2010
o Exposure to vibration loads can result in highly variable
results
o Vibration loads can vary by orders of magnitude (e.g., 0.001 g2/Hz
to 1 g2/Hz)
o Time to failure is very sensitive to vibration loads (tf W4)
o Very broad range of vibration environments
o MIL-STD-810 lists 3 manufacturing categories, 8 transportation
categories, 12 operational categories, and 2 supplemental categories
Mechanical Loads (Vibration)
7
© 2004 - 2007© 2004 - 2010
Typical Vibration Levels
Harmonic
Steinberg D.S. Vibration analysis
for electronic equipment.
John Wiley & Sons, 2000.
Random
MIL-STD-810G Figure 514.6C-1
US Highway truck vibration
exposure
1 hour is equivalent to 1000 miles
8
© 2004 - 2007© 2004 - 2010
o Random Vibration
o 9.8 to 28 Grms
o 0.07 to 0.5 G2/Hz
o Natural Frequency
o 72 Hz
o Results
o With BGA’s, SnPb solder always outperformed lead-free
o Less conclusive for leadless/leaded parts
Excessive Vibration
All BGA-225 Woodrow, IPC/APEX 2006
9
© 2004 - 2007© 2004 - 2010
Fatigue Exponent –High Cycle Fatigue?
c
field
test
test
field
N
N
o Assume the solder
strain is directly
proportional to the
board level strain
o Coffin – Manson
10
Fit to data provides c = 1.5
Exponent value is too low; representative of low-cycle fatigue
High-cycle fatigue exponent is typically 4 to 6 or higher
MIL-STD-810, Steinberg
Low cycle fatigue behavior can not be extrapolated to HCF behavior
© 2004 - 2007© 2004 - 2010
o HCF failures typically occur in the lead or solder joint
Through Hole Solder Interconnect Vibration Fatigue
Component Motion Board bending
11
© 2004 - 2007© 2004 - 2010
Surface Mount Vibration Fatigue – Board Bending
SnPb failure @ 1200 µε SnAgCu failure @ 1200 µε
12
© 2004 - 2007© 2004 - 2010
SnPb Fatigue Crack
o Very fine cracks
o Secondary micro-cracks
o Evidence of phase coarsening
13
© 2004 - 2007© 2004 - 2010
SnAgCu Fatigue Crack
Well defined crack path
Shrinkage crack provided
crack initiation site
Shrinkage Crack
14
Using Simulation
© 2004 - 2007© 2004 - 2010
Sherlock Design Analysis Software
16
© 2004 - 2007© 2004 - 2010
Parts List
17
© 2004 - 2007© 2004 - 2010
Boundary Conditions (Mount Points)
18
© 2004 - 2007© 2004 - 2010
Board Properties (Stackup)
19
© 2004 - 2007© 2004 - 2010
Test Vehicles and Modeling
Finite Element Analysis in Sherlock
20
© 2004 - 2007© 2004 - 2010
o 1st Natural Frequency: 159.45 Hz
Sherlock - Modal Analysis
21
© 2004 - 2007© 2004 - 2010
Modal Analysis
One accelerometer located at the center misses 3 frequency modes
22
© 2004 - 2007© 2004 - 2010
o Board level strain (at components)o Green: predicted (peak values)
o Purple: experimental (averaged over strain gage region)
Board Strain Analysis
23
© 2004 - 2007© 2004 - 2010
Board or Component Motion - FEA
24
© 2004 - 2007© 2004 - 2010
o Simulation
o Identify critical components prior during plan development
o Identify critical frequencies or board responses
o Identify locations for accelerometer placement
o Mounting consideration
o Boundary conditions
o Mounting configurations
Simulation During Test Plan Development
25
© 2004 - 2007© 2004 - 2010
o HALT chambers utilize repetitive shock (RS) to generate
vibration input
o Type of random vibration
o Pneumatic hammers strike the chamber table a generate the
vibration
o HALT vibration is not designed to replicate field environments
o Designed to expose weak links
o Rapid assessment
o Input/Output is typically displayed as a Grms value
o What does this mean?
o Is it suitable for simulation, can it be used in Sherlock?
HALT Example - Vibration
26
© 2004 - 2007© 2004 - 2010
o HALT vibration is specified by a Grms level
o What does this represent
o The root mean square acceleration (Grms) is the square root of
the area under the ASD (acceleration spectral density) curve in
the frequency domain
o The Grms value is typically used to express the overall energy of
a particular random vibration event
o Can it be used for simulation?
o No
o A straight Grms value can represent an infinite number of
acceleration and frequency combinations
o The time history of the table excitation must be captured
and processed into a usable format for the simulation
HALT Chamber Settings
27
© 2004 - 2007© 2004 - 2010
HALT - Repetitive Shock Induced Vibration
Input HALT level set to 33 Grms
28
© 2004 - 2007© 2004 - 2010
Vibration Data – Post Processing
Even though both chambers are programmed to output 16 Grms the PSD profile is higher
across the full frequency range for the Qualmark, the content below 5KHz used as
Sherlock inputs
Know your chamber! HASS or HALT profiles can be very different
FFT performed on the time history data to generate a PSD (power spectral density)
0.0000001
0.000001
0.00001
0.0001
0.001
0.01
0.1
1
0 20000 40000 60000 80000 100000 120000
PSD
(G
2/H
z)
Frequency
Qualmark
Chart/Hanse
24 Grms
5.7 GrmsResponse below 5KHz
29
© 2004 - 2007© 2004 - 2010
FFT of the Time History
Different FFT
parameters
Frequency PSD
1.20E+01 7.50E-05
1.00E+02 2.00E-04
7.84E+02 4.81E-02
1.57E+03 5.33E-02
2.35E+03 1.54E-01
3.14E+03 2.52E-01
3.92E+03 2.78E-01
4.71E+03 1.39E-01
5.49E+03 9.59E-02
6.27E+03 6.18E-02
7.06E+03 2.48E-02
7.84E+03 3.70E-02
8.63E+03 4.55E-02
9.41E+03 6.39E-02
1.02E+04 9.85E-02
1.10E+04 1.06E-01
32.7 Grms
32.8 Grms
30
© 2004 - 2007© 2004 - 2010
o HALT vibration can cause high cycle fatigue of components
o Simulation can identify critical locations for accelerometer placement
o Simulation can also identify critical components before HALT testing
o Simulation prior to HALT requires the development of a suitable vibration profile
o Leverage prior HALT vibration data
o Product should have similar mass and mounting
o Need the time history data (time verses acceleration)
o Data should include table outputs for multiple Grms levels
HALT Vibration
31
© 2004 - 2007© 2004 - 2010
Boundary Conditions (Mounting)
32
© 2004 - 2007© 2004 - 2010
o X,Y,Z RS vibration
o HALT profile converted to PSD profile
Pre-HALT Example
33
© 2004 - 2007© 2004 - 2010
o Identify locations for accelerometers
o Identify critical components
HALT Vibration - Modal Analysis
263.15 Hz
216.14 Hz
34
© 2004 - 2007© 2004 - 2010
o Sherlock supports “Virtual Accelerometers”
o The modal analysis identifies the regions of greatest
board response which can be used to place the HALT
accelerometers
Accelerometer Locations
Nf = 263.2 Hz
Through hole crystal oscillator
may need to be staked prior to HALT
35
© 2004 - 2007© 2004 - 2010
Product Response to Multiply Axis Vibration
36
© 2004 - 2007© 2004 - 2010
Accelerometer Response
37
© 2004 - 2007© 2004 - 2010
Critical Components
Excessive lead stains (1.4%) will lead to a rapid HALT failure
X1 needs to be staked
38
© 2004 - 2007© 2004 - 2010
o Sherlock identifies critical components and they are
mitigated before HALT is performed
o Different mounting configurations are investigated in
Sherlock and modified to prevent unrealistic movement
during the HALT test
o The hard questions
o How much of a margin do I need
o What is a good vibration level, depends on the product and
use
During the HALT Planning Stage
39
© 2004 - 2007© 2004 - 2010
o Test time for HCF usually take days to weeks
o Type of vibration applied depends on the goals
o Random vibration – product qualification testing
o Harmonic vibration – fundamental understanding of fatigue
behavior (SN curves)
o Example - Harmonic vibration of 208 I/O BGAs
o Resonant frequency: ~160 Hz
o Testing at ~153Hz
o SAC 305, SnPb
o 80 mils, 90 mils, 95 mils, 105 mils
o 6.8 Gs, 7.8 Gs, 8.3 Gs, 9.8 Gs
High Cycle Vibration Fatigue Testing - Example
40
© 2004 - 2007© 2004 - 2010
High Cycle Fatigue Testing
High cycle fatigue testing can
take weeks on a electro-
dynamic shaker
41
© 2004 - 2007© 2004 - 2010
o One Accelerometer located at the center of the
PCB, Frequency sweep 20 – 500 Hz
Analysis
42
© 2004 - 2007© 2004 - 2010
o Stepwise failure behavior characteristic of differing stress levels along board length
Experimental - Cycles to Failure
43
© 2004 - 2007© 2004 - 2010
20
238.9
n
n
f
QfPSDZ
Prediction (Steinberg – Displacement Based)
o Step 1: Calculation of maximum deflection (Z0)
o PSD is the power spectral density (g2/Hz),
o fn is the natural frequency of the CCA,
o Gin is the acceleration in g
o Q is transmissibility (assumed to be square root of natural frequency)
20
8.9
n
in
f
QGZ
Harmonic
Random
44
© 2004 - 2007© 2004 - 2010
Lchr
BZc
00022.0
Prediction (Steinberg)
o Step 2: Calculate critical displacement
o B is length of PCB parallel to component
o c is a component packaging constant
o 1 to 2.25
o h is PCB thickness
o r is a relative position factor
o 1.0 when component at center of PCB
o L is component length
o At critical displacement,
component can survive a minimum
20 million cycles under random
vibration
o 10 million cycles for harmonic
45
© 2004 - 2007© 2004 - 2010
4.6
0
0
Z
ZNN c
c
Prediction (Steinberg)
o Step 3: Life calculation
o Nc is 10 or 20 million cycles
o Several assumptions
o CCA is simply supported on all four edges
o More realistic support conditions, such as standoffs or wedge locks, can result in a lower or higher displacements
o Chassis natural frequency differs from the CCA natural frequency by at least factor of two (octave)
o Prevents coupling
o Vibration occurs at room temperature
o Depending upon the configuration and loading, vibration at lower or higher temperatures can increase/decrease lifetime
o Does not consider the influence of in-plane displacement (i.e., tall components) or components located at areas of high curvature
46
© 2004 - 2007© 2004 - 2010
o 208 I/O BGA, 1.6 mm thick PCB, 159 Hz
o Distance between standoffo Y = 76.2 mm
o X = 152.4 mm = 6 inches
o Package dimensions 14.4 x 14.4 mm
o Center component (Critical deflection)o Zc = 0.00022 * 6 / (1.75 * 0.062 *1*SQRT(0.567)) = 0.0162 inches peak
o Nc = 10 million cycles to failure
o Board deflection Gin = 9.8 g’so Zo = 9.8*9.8*SQRT(159)/(159^2) = 0.0479 inches
o 95.8 mils peak to peak, close to measured value of 105 mils
o Cycles to Failure
Example Prediction (Steinberg)
Lchr
BZc
00022.0
20
8.9
n
in
f
QGZ
4.6
0
0
Z
ZNN c
c
97000479.0
0162.0000,000,10
4.6
0
N
47
© 2004 - 2007© 2004 - 2010
o During vibration, board-level strain is
proportional to solder or lead strains
and therefore can be used to make
time-to-failure predictions
o Requires converting
cycles-to-failure displacement
equations (Steinberg) to use strain
o The critical strain for the package
types is a function of package style,
size, lead geometry
High Cycle Fatigue (Prediction)
n
ccNN
0
0
Lcc
ζ is analogous to 0.00022B but
modified for strain
c is a component packaging function
L is component length
48
© 2004 - 2007© 2004 - 2010
o During vibration the board strain is proportional to the solder or lead strains and therefore can be used to make time to failure predictions
o This requires converting the cycles to failure displacement equations (Steinberg) to use strain
o The strain for the components is now pulled from the FEA results
o The critical strain for the package types is a function of package style, size, lead geometry
FEA Failure Prediction
n
ccNN
0
0
Lcc
ζ is analogous to 0.00022B but modified
for strain
c is a component packaging function
L is component length
49
© 2004 - 2007© 2004 - 2010
Sherlock Vibration Fatigue Prediction – 105 mils Deflection
Steinberg using computed deflection
Steinberg
using
measured
deflection
50
© 2004 - 2007© 2004 - 2010
Sherlock Vibration Fatigue Prediction – 90 mils Deflection
Steinberg
using
measured
deflection
51
© 2004 - 2007© 2004 - 2010
o Finite element based simulations for vibration can capture:o Complex boundary conditions and whether the PCB is adequately supported
o Complex mode shapes, modal analysis where do you put the accelerometers during test
o Facilitate strain or curvature based predictions
o Displacement based methods (Steinberg)o High displacement does not always indicate high stress
o Only based on the natural frequency of the board
o Utilizes Miles equation to equivalence random vibration to harmonic
o Can provide very conservative results
o Experimental measurements can be used to validate simulations and verify response, not typically done to failure but for qualification purposes
o Combining simulations and fatigue predictions with qualification test results provides the most valueo Is the vibration response as expected
o Confidence in the reliability predictions
Conclusions
52
© 2004 - 2007© 2004 - 2010
Any Questions
53