DfR Solutions

Reliability Prediction and Assurance of Power Electronics Components September 5, 2016

European Power Electronics

Karlsruhe, Germany

9000 Virginia Manor Rd Ste 290, Beltsville MD 20705 | 301-474-0607 | www.dfrsolutions.com

© 2004 – 2010

WHAT IS AN ELECTRONIC POWER SUPPLY? They don’t supply power

Power Supplies = Voltage Converters 9000 Virginia Manor Rd Ste 290, Beltsville MD 20705 | 301-474-0607 | www.dfrsolutions.com

FOUR BASIC CONVERSIONS AC to DC (rectifier)

o o

o

Conversion from the grid

DC to AC (inverter)

DC to DC (converter)

o o

o

On-board conversion

AC to AC (drives)


COMMONALITY AMONG POWER CONVERTERS?

o

Isolated vs Non-Isolated Linear vs Switch Mode

o

Various Topologies

o

o o o o o o o o

Buck Boost Buck-Boost Flyback Half-Bridge Full Bridge Zero-Voltage Switch (ZVS) Radio (VHF) Frequency Conversion


POWER ELECTRONICS VS CONTROL ELECTRONICS Definition of Power Electronics is soft, but

o o o

Typically needs active cooling (Liquid or Forced Air) Power level typically exceeds 240VA (UL & IEC Safety Threshold)

Historic Challenges

o o o o

Erroneous Control

Efficient Switches Reliable Switches Cooling

Current Challenges

o o o o o

o

Power Density Overall Efficiency Liability System Control / Integration Long term reliability in mobile environment Safety in mobile environment


Weekend Mechanic

WHAT IS RELIABILITY?

The Probability the Product… …Will Meet Customer Expectations… …Within the Use Environment… …Over the Desired Lifetime And Power Supplies Struggle with Reliability


WHY STRUGGLE WITH RELIABILITY? Cost-Constrained

o o

Space-Constrained

o o

o

Transients, Electrical Noise, Temperature, Variation in Supply

One-Size Fits All

o o

7

Precious real estate goes to the (silicon) brain

Connected to the Real-World

o

o

Not viewed as a product differentiator (bottom-dollar)

Application does not always fit the design

Miscommunication on Reliability Expectations 9000 Virginia Manor Rd Ste 290, Beltsville MD 20705 | 301-474-0607 | www.dfrsolutions.com

Industry Standards Focus on Safety (Not Reliability) ISO 26262 Functional Safety (Primarily European)

o o o o

Only industry vehicle level safety standard Mitigates liability of manufacturer if standard is followed Very, Very complex and expensive to implement

UL 458 Power Converters/Inverters for Land Vehicles UL 2202 EV Charging System Equipment UL 840 Insulation Coordination

o o o o

Simple but incomplete insulation requirement specification specified by UL2202

IEC 60664 Insulation-Low Voltage Systems

o o

“Mother” insulation requirement specification globally, but has no application guidelines

IEC 60950-1 Information Technology Equipment Safety

o o

Written for Telecom, but most widely specified requirement for High Voltage Safety


RELIABILITY TESTING OF POWER ELECTRONICS o

High Cost of Test

o

Few Test Sites → Very limited number of samples

o

High reliability goals → Long Test Time

o

Cannot reasonably measure reliability!

o

Can only find gross & early life failure modes


WHAT TO DO? o

Conservative Design Rules / Practices

o

Test to failure at high stress levels (i.e., HALT) o

Focus on Design for Reliability (DfR) methodologies

o

o

10

Try to identify pertinent failure mode

Reliability simulation that incorporates Physics of Failure (PoF)


Physics of Failure

11


What is Physics of Failure? How does it differ from Traditional Reliability Assessment and Prediction?


THE ANSWER IS ON LINKEDIN!


TRADITIONAL RELIABILITY APPROACH

o

Answer: Use information on what has already happened


PHYSICS OF FAILURE APPROACH

o

Answer: Use information from your design


WHAT IS PHYSICS OF FAILURE (PoF)? o

Also known as reliability physics

o

Common Definition: o

The process of using modeling and simulation based on the fundamentals of physical science (physics, chemistry, material science, mechanics, etc.) to predict reliability and prevent failures


The Basics of PoF


PHYSICS OF FAILURE: MODELING AND SIMULATION o

What are we modeling / simulating?

o

Reliability (t > 0) = Material Change or Material Movement

o

Fundamental Material Mechanisms o o o o

Diffusion Creep Fatigue Oxidation/Reduction (Corrosion)


MATERIAL MOVEMENT: DIFFUSION Motion of electrons, atoms, ions, or vacancies through a material

o

o

o

Typically driven by a concentration gradient (Fick’s Law)

Can be driven by other forces (electromotive force, stress)


Creep (Dislocation Creep or Creep Deformation) The tendency of a solid to permanently deform when subjected to a fixed load

o

o

Metals: Driven by movement of defects within the crystalline structure

o

o

o

o

Corollary: Tendency of a solid to relieve stress when loaded at a fixed displacement

Dislocations (edge or screw) Grain Boundaries

Creep = f(σ, , t, T) [stress, strain, time and temperature]


FATIGUE Fatigue is the accumulation of damage due to repeated (cyclic) applications of stress

o

o o

o

 Movement of dislocations (‘glide’) N f  0.5( Pileup of dislocations creates voids 2 f Agglomeration of voids creates stress concentration, formation of cracking

Areas of concern: Interconnections

o o o o o

Connectors Solder Joints Leads Wire Bonds


)c

ELECTRO-CHEMICAL (OXIDATION / REDUCTION) Electro-Chemical effects are degradation due to chemical attack of metals leading to Corrosion and Dendrites

o

o

o o

o

Requires moisture o Normally requires a liquid film o Silver can grow dendrites with only high humidity Requires unpassivated metal Strongly accelerated by ionic contaminants (e.g. Chlorides, Sulfites, Sulfates, Bromides, Acetates, etc.) Accelerated by Temperature and Voltage

Areas of Concern:

o o

Interconnections (e.g. connectors, PCBs, component leads)


MATERIAL MOVEMENT AND M&S o

All physics of failure models can be condensed into answers to three questions o How large is the stress? o At what rate is this stress driving material movement? o At what time will this material movement induce failure?


STRESS VS STRENGTH Stress

o o o o

Rarely normally distributed Always extreme cases Rarely able to be changed

Strength

o o

o o

Can be increased Variability decreased Can rarely afford 0 probability of failure


FAILURE RATE VS STRESS LEVEL

o

Failure rate depends on all the stress conditions that the component / system experience Example: Power module reliability for power pulse test is dependent upon Tjmax, Delta T

ReliaSoft ALTA 7 - www.ReliaSoft.com

Life vs Stress

ReliaSoft ALTA 7 - www.ReliaSoft.com

at Delta T =60C vs T J M ax

Life

1.E+7

I nfineon Power Module Temperature-N onThermal Lognormal 398| 60 F=12 | S=0 Median Line Top CB Median Bottom CB Median 338 Stress Level Points Median Point I mposed Pdf 348 Stress Level Points Median Point I mposed Pdf 358 Stress Level Points Median Point I mposed Pdf 368 Stress Level Points Median Point I mposed Pdf 378 Stress Level Points Median Point I mposed Pdf 398 Stress Level Points Median Point I mposed Pdf 423 Stress Level Points Median Point I mposed Pdf

1.E+6

1.E+5 320

Michael J . Varnau D elphi 11/ 6/ 2008 4:40:29 PM 348

376

404

432

460

Life vs Stress at T J M ax = 398K vs Delta T

1.E+7

Life CB@ 90% 2-Sided

I nfineon Power Module Temperature-N onThermal Lognormal 398| 60 F=12 | S=0 Median Line Top CB Median Bottom CB Median 40 Stress Level Points Median Point I mposed Pdf 50 Stress Level Points Median Point I mposed Pdf 60 Stress Level Points Median Point I mposed Pdf 70 Stress Level Points Median Point I mposed Pdf 80 Stress Level Points Median Point I mposed Pdf 100 Stress Level Points Median Point I mposed Pdf

1.E+6

1.E+5

1.E+4 30

TjM a x Std=0.1485; B=4380.8314; C=2.4124E+7; n=3.7512

Life CB@ 90% 2-Sided

Life

o

Michael J . Varnau Delphi 11/ 6/ 2008 5:40:41 PM 100

De lt a T Std=0.1485; B=4380.8314; C=2.4124E+7; n=3.7512


300

POF-BASED RELIABILITY PREDICTION Most physics-of-failure (PoF) based models are semiempirical

o

o o

The basic concept is still valid Requires calibration

Calibration testing should be performed over several orders of magnitudes

o

o

Allows for the derivation of semi-empirical constants, if necessary

The purpose of PoF is to limit, but not eliminate, the influence of material and geometric parameters

o

o

E.g., Solder: Testing must be re-performed for each package family (ball array devices, gullwing, leadless, etc.)

9000 Virginia Manor Rd Ste 290, Beltsville MD 20705 | 301-474-0607 | www.dfrsolutions.com 26

PHYSICS OF FAILURE (POF) ALGORITHMS

 ~ 0.51eV T f  exp   kT

   exp(~ 0.063% RH )  n

Ea  1 1  t1  V2          exp t2  V1  K B  T1 T2 

 L  2    hs hc L     2  1   T  L  F          E1 A1 E2 A2 As Gs Ac Gc  9  Gb a  

Can be mind-numbing! What to do? 9000 Virginia Manor Rd Ste 290, Beltsville MD 20705 | 301-474-0607 | www.dfrsolutions.com

SIMPLIFY – AGGLOMERATE – AUTOMATE

“Everything should be made as simple as possible, but not any simpler.” – Albert Einstein Find the simplest model that captures the effects that affect you. Intelligently combine effects into the whole. 9000 Virginia Manor Rd Ste 290, Beltsville MD 20705 | 301-474-0607 | www.dfrsolutions.com

Why PoF?


Before Physics of Failure: Traditional Reliability Growth

1) Design

2) Build

DESIGN - BUILD - TEST - FIX (D-B-T-F)

3) Test

5) Fix Whatever Breaks.

4) No Faults Detected ? Yes 6) REPEAT 3-5 Until Nothing Else Breaks Or You Run Out Of Time/Money.

Today, This Is Not Enough! 1) All design issues often not well defined. 2) Early build methods do not match final processes. 3) Testing doesn’t equal actual customer’s usage. 4) Improving fault detection catches more problems, but causes more rework. 5) Problems found too late for effective corrective action, fixes often used. 6) Testing more parts & more/longer tests “seen as only way” to increase reliability. 7) Can not afford the time or money to test to high reliability. 8) Incremental improvements from faster more, capable tests still not enough. It Is Time for a Change


PoF IS PART OF A ROBUST DFR PROCESS Failure Mode Analysis

o o

Failure Mode Effect Analysis (FMEA), Fault Tree/Tolerance Analysis (FTA), Design Review by Failure Mode (DRBFM), Sneak Circuit Analysis (SCA)

Reliability Prediction - Empirical Design Rules Design for Excellence

o o o o

Design for Manufacturability (DfM), Design for Testability (DfT)

Tolerancing (Mechanical, Electrical) Simulation and Modeling (Stress)

o o o

Thermal, Mechanical, Electrical/Circuit

Simulation and Modeling (Damage)

o o

EMI/EMC, EOS/ESD, Physics of Failure, Derating


Why Physics of Failure: Leverage in Product Design http://www.ami.ac.uk/courses/topics/0248_dfx/index.html

70% of a Product’s Total Cost is Committed by Design 32


WHY PHYSICS OF FAILURE: FASTER / CHEAPER o

Traditional OEMs spend almost 75% of product development costs on test-fail-fix

o

Electronic OEMs that use design analysis tools o

o

o

Hit development costs 82% more frequently Average 66% fewer re-spins Save up to $26,000 in re-spins

Gene Allen and Rick Jarman .Collaborative R&D; (New York John Wiley&Sons. Inc. 1999). 17. Aberdeen Group, Printed Circuit Board Design Integrity: The Key to Successful PCB Development, 2007 http://new.marketwire.com/2.0/rel.jsp?id=730231


WHY PHYSICS OF FAILURE: FASTER TIME TO MARKET

P. Smith and D. Reinertsen. Developing Products In Half The Time (New York Van Nostrand Reinhold. 1991). 4.


WHY PHYSICS OF FAILURE: LESS ROBUST TECHNOLOGY

35


Everyone Else is Doing It: Road to Lab to Math (RLM)

By 2006, GM was able to reduce vehicle road testing to the point that the Mesa Arizona Proving Grounds, formerly operated by 1200 people, was sold.

Test

CAE-M&S

As the use of CAE based modeling & simulation methods increase, dependence on physical testing can be reduced and refocused. 9000 Virginia Manor Rd Ste 290, Beltsville MD 20705 | 301-474-0607 | www.dfrsolutions.com

PHYSICS OF FAILURE JUST WORKS

37


How PoF?


IMPLEMENTATION OF POF Prediction and Failure Avoidance

Test Plan Development

Failure Analysis


How PoF? Prediction and Failure Avoidance


History of PoF and Prediction/Failure Avoidance World War II was historic for several reasons

o o

o

o

Electronics used extensively for the first time (torpedoes, radar, etc.) Reliability of electronics was HORRIBLE (failure rates of 10 to >50%)

Many of the technology (transistors) and reliability methodology (i.e., DFMEA) in use today came from US DoD-funded efforts after WWII to improve the reliability of electronics


History of PoF and Prediction/Failure Avoidance o

1960: Rome Air Development Center (RADC) identified a need for a forum on the Physical Processes which caused Electrical & Electronic “Components” to Fail o 1960 Component Reliability Metric: MTBF is 100-1000 hrs o 1960 Component Quality Metric: Defects per Thousand

o

The first symposium on the "Physics of Failure in Electronics" held in Sept. 1962 at Illinois Institute of Technology (IIT) o Evolved into International Reliability Physics Symposium (IRPS)

o

Improvements in component metrics through PoF o Today’s reliability: MTBF measured in >1M operating hours o Today’s quality: Defects per Billion 9000 Virginia Manor Rd Ste 290, Beltsville MD 20705 | 301-474-0607 | www.dfrsolutions.com

HISTORY OF PHYSICS OF FAILURE (CONT.) o

CONFLICT ALERT

o

The organization that founded PoF (RADC) also managed MIL-HDBK-217 o

o

First empirical handbook that specified constant failure rate

Whaaa??? 9000 Virginia Manor Rd Ste 290, Beltsville MD 20705 | 301-474-0607 | www.dfrsolutions.com

MIL-HDBK-217 The purpose of MIL-HDBK-217 was to provide a prediction of failure rates at the component-level

o

o

Philosophy of MIL-HDBK-217

o o o

o

This output allows for system-level safety and risk analysis

Failure modes and mechanisms are irrelevant Failure rate is constant

This philosophy is the direct driver of many of today’s component qualification tests


JEDEC QUALIFICATION TESTING N Samples x M Test Time

o o

No minimum test time, though 1000 hours has become the default

Extrapolation of test results (samples x time) is not failure mechanism specific

o

o

Agglomerate all failure mechanisms into 0.7 eV

No failures allowed!

o o o

Limited understanding of margin and weak points More of a marketing than engineering exercise


Accuracy of Steady State Reliability Predictions Loughborough found deviations greater than 500%

o

Board one

Board two

Board three

Bellcore (currently Telcordia)

Board four

CNET HRD

Board five

Mil-Hdbk-217 Siemens

Board six -100

0

100

200

300

400

% Deviation from Field Failure Rate

Senior fellow at NASA

o o

Predictions can be off by as much as 10,000X


500

600

Steady State Failure Rate: Pick a Number

Reliability Predictions (MTBF)

If you don’t like the numbers... ...give me five minutes, I will make up a better one


PoF – Three Failure Categories GENERIC FAILURE CATEGORY o

Errors - Incorrect Operations & Variation Defects/Weaknesses. o Missing parts, incorrect assembly or process. o Process control errors (Torque, Heat treat). o Design errors o Missing functions, o Inadequate performance. o Inadequate strength.

o

Overstress. o Overheating. o Voltage/Current o Electro static discharge. o Immediate yield, buckling, crack.

o

Wearout/Changes, via Damage Accumulation. o Friction wear. o Fatigue. o Corrosion. o Performance changes/parameter drift

TYP. FAILURE DETECTION Quality Assurance Immediate or Latent defects

Performance Capability Assessments

Stress-Life Durability Assessments


WEAROUT 3. Strain : Instantaneous changes (materials\structural) due to loading, different loads interact to contribute to a single type of strain.

2. Stress The distribution/ transmission of loading forces throughout the device.

1. Loads Elect. Chem. Thermal, Mech... Individual or combined, from environment & usage act on materials & structure.

6. Time to 1st Failure:

(Damage Accumulation verses Yield Strength A Function of: Stress Intensity, Material Properties, & Stress Exposure Cycles/Duration].

7. Rate of Failure (Fall out)

A function of variation in; Usage, Device Strength & Process Quality Control (i.e. latent defects).

5. Failure Site & Type:

Typically due to a designed in: stress concentrator , design weakness, material/process variation or defect.


Knowledge of how/ which “Key Loads” act & interact is essential for “efficiently” developing good products, processes & evaluations.

4. Damage Accumulation (or Stress Aging): Permanent change degradation retained after loads are removed. From small incremental damage, accumulated during periods/cycles of stress exposure.

WEAROUT (DAMAGE ACCUMULATION) STRESS INDUCED DAMAGE ACCUMULATION

STRESS/ STRENGTH

Design’s Strength Decay/Spreads Over Time / Usage

Material Decay Increases UNRELIABILITY OVER TIME How well do you Understand & Design For Strengths & Stresses?

4  | 9 9

3  | 9 3

% t i l e

% t i l e

STRESS EXPOSURE TIME or USAGE CYC’S

2

 | 6 9 % t i l e

INITIAL UNRELIABILITY

FREQUENCY OF OCCURRENCE


PHYSICS OF FAILURE AND WEAROUT What is susceptible to long-term degradation in electronic designs?

o

o o o o o o o o o o o o

51

Integrated Circuits (EM, TDDB, HCI, NBTI) Interconnects (Die Attach, Wire Bonds, Solder Joints, Vias) Ceramic Capacitors (oxygen vacancy migration) Electrolytic Capacitors (electrolyte evaporation, dielectric dissolution) Film Capacitors Memory Devices (limited write cycles, read times) Light Emitting Diodes (LEDs) and Laser Diodes Resistors (if improperly derated) Silver-Based Platings (if exposed to corrosive environments) Relays and other Electromechanical Components Connectors (if improperly specified and designed) Tin Whiskers


PoF and Integrated Circuits

52


WHY PHYSICS OF FAILURE: LESS ROBUST TECHNOLOGY

53


IC WEAROUT: A REAL ISSUE Guaranteeing the conventional 10-year life of ICs is going to become increasingly difficult….the progressive degradation of the electrical characteristics of transistors and wires will start dominating over abrupt functional failures. Furthermore, the mean-time-to-first-softbreak will significantly diminish." Antonis Papanikolaou, IMEC, Leuven, Belgium, 2007

…the progress of Moore's Law means that transistor wear-out and statistical performance issues are beginning to cross over from the realm of academic and hypothetical discussion to real-world R&D engineering. EE Times Europe, 2007 ”The notion that a transistor ages is a new concept for circuit designers,” says Chris Kim (U of Minnesota). Transistor aging has traditionally been the bailiwick of engineers who design the processes that make transistors; they also formulate recipes that guarantee the transistors will operate within a certain frequency and other parameters typically for 10 years or so…But as transistors are scaled down further and operated with thinner voltage margins, it’s becoming harder to make those guarantees… transistor aging is emerging as a circuit designer’s problem. IEEE Spectrum, June 2009


GENERIC CMOS WEAROUT MODELS

EM: HCI:

TDDB: NBTI:


IC WEAROUT AND POWER SUPPLIES o

The limited complexity of power supplies limits the relevance of these Physics of Failure models

o

Most power supplies do not require state-of-the-art controllers o

o

56

Process node is rarely below 180nm

Most wearout issues become prominent at feature sizes below 45nm


MULTIPLE MECHANISM THEORY o

Simultaneous influence of multiple failure mechanisms o Electromigration (EM) o Time Dependent Dielectric Breakdown (TDDB) o Negative Bias Temperature Instability (NBTI) o Hot Carrier Injection (HCI)

o

Each mechanism is driven by different conditions o Activation energy o Electrical and thermal conditions o Transistor use conditions o

switching vs. constant bias


MULTI-MECHANISM THEORY: VALIDATION STUDY Electrical Failures Distribution

o

Field return data was gathered from a family of telecommunication products o 56 different ICs comprised 41.5% of the failed part population The validation activity was utilized failure data from 5 integrated circuits

CAP 1%

TRANSISTOR 19%

RESONATOR 6%

CARD 44%

RESISTOR 8%

o

INDUCTOR 0.07%


IC 21% FUSE 0.54%

DIODE 0.13% FILTER 0.02%

MULTI-MECHANISM THEORY: VALIDATION STUDY (cont.) o

Results demonstrate the accuracy and repeatability of the multi-mechanism model to predict the field performance of complex integrated circuits


PoF and Interconnects (Die Attach, Wire Bonds, Solder Joints, Vias)

60


How to Perform PoF of Interconnects? Critical aspect of PoF of Interconnects is including primary parameters and excluding secondary effects

o

o

Can sometimes drive the semi-empirical nature of PoF


PoF and Interconnects (Die Attach)

62


PREDICTING THE RELIABILITY OF DIE ATTACH Die attach tends to have only one failure mode

o o

Thermal cycle fatigue due to power cycling

Frequency of the power cycle can play a very critical role (driven by thermal inertia)

o

o

o

If the power cycle frequency is high enough, the failure site will shift to the wire bond If the power cycle frequency is low enough, the failure site will shift to the solder joints

Key Challenge: Defining failure

o o o

Most die attach configurations do not conduct electricity Die attach is primarily a thermal path


POWER CYCLING Some people believe that thermal cycling and power cycling are equivalent

o

o

Wrong

While both conditions will induce thermo-mechanical fatigue, the ramp rates (or rate of change) can be very different

o

o

Especially for power cycling, this can change the failure site


FAILURE MECHANISM SHIFT Power cycle is 10-20 seconds

o o

o

Power cycle is 1-2 minutes

o o

o

Failure mechanisms is likely to be die attach fatigue 50K to 500K cycles

Power cycle is 10-20 minutes (or longer)

o o

o

o

Failure mechanism is likely to be wire bond fatigue 500K to >1M cycles

Failure mechanism is likely to be solder joint fatigue 500 to 50K cycles

This is mechanism shift is driven by the thermal transient

http://semiengineering.com/the-auto-industry-takingthings-into-their-own-hands/


DIE ATTACH FATIGUE (ENGLEMAIER, 1982)

( Ld  Wd )(CTEdie  CTEDBC )T 2

 

2

2h

Strain range at the die/die attach interface • h = die thickness • W = die width • L = die length

 = coefficient of thermal expansion (CTE) • T = change in temperature •


DIE ATTACH FATIGUE (TIME TO FAILURE) Coffin-Mason based low-cycle fatigue damage model

o

N f   f (  )

c

Tin-based solders (SnPb, SAC305, Sn3.5Ag, etc.) tend to have fatigue exponents around 2 to 2.5

o

o

SnPb: ~2; Sn3.5Ag: 2.2; SAC305: ~2.4

Fatigue exponents for newer, higher temperature die attach solders have not yet been widely validated

o

o

Nanosilver, BiAgX, Sn25Ag10Sb (“J” alloy)


NANOSILVER: IS IT BETTER THAN SOLDER? Sintering with nanosilver is challenging (and expensive)

o o o

o

Requires pressure For large die, post sintering stress relief step is often required Voiding must be controlled


NANOSILVER: IS IT BETTER THAN SOLDER? (CONT.) o

For the most part, the industry agrees nanosilver is more reliable, but there can be issues

L. Melchor, Doctoral Dissertation

M. Beierlein, Adv. Pack. Conf. 2013


NANOSILVER: IS IT BETTER THAN SOLDER? (CONT.) o

Different publications provide different values for fatigue constants and exponents

Author

Constant

Value

Exponent

M. Knoerr

1.6E-01

Plastic Strain

-3.0

Y. Tan

5.8E+11

Shear Stress

-9.4

X. Li

1.6E-09

Shear Strain

-7.6

Y. Tan, et. al., Conf. on Fracture, 2013 9000 Virginia Manor Rd Ste 290, Beltsville MD 20705 | 301-474-0607 | www.dfrsolutions.com

LOW CYCLE VS. HIGH CYCLE FATIGUE? Lifetime under mechanical cycling is divided into two regimes

o

o o

o

Low cycle fatigue (LCF) High cycle fatigue (HCF)

LCF is driven by inelastic strain (C-M) c

 p   f 2 N f 

-0.5 < c < -0.7; 1.4 < -1/c > 2 o

HCF is driven by elastic strain (Basquin) f 2 N f  b e  E -0.05 < b < -0.12; 8 > -1/b > 20

71


PoF and Interconnects (Wire Bonds)

72


Reliability: When do Wire Bonds Fail? Exposure to elevated temperature

o o

Intermetallic formation

Exposure to elevated temperature/humidity

o o

Corrosion

Exposure to temperature cycling

o o

Low cycle fatigue


RELIABILITY AT ELEVATED TEMPERATURES Not an issue in aluminum-aluminum wire bond system

o o

o

The lack of intermetallic formation and differential diffusion makes it relatively immune to purple plague Prior studies have found little change in resistance after 1000 hours at 300C

Bigger issue in mixed metal systems, like gold-aluminum

o

o

o

Formation of brittle AuAl2 (purple plague) at 350C Diffusion of gold into Au5Al2 causes Kirkendall voiding at lower temps


RELIABILITY PREDICTION (ELEVATED TEMPERATURE) o

Is an absolute reliability prediction of wire bond reliability at elevated temperature possible?

o

Short answer: NO o

o

Diffusion behavior is very sensitive to bonding temperature, quality of bond, aluminum alloy, aluminum bond pad thickness, and encapsulant chemistry o Low bonding temperature o Si in Al-Cu bond pad o Thin bond pad (~1 um) o Bromide-free flame retardants Can change absolute and relative (acceleration factor) time to failure


Reliability Prediction – Temperature (cont.) For gold-aluminum, prediction is primarily by extrapolation from test results using Arrhenius and a conservative activation energy (0.9 eV)

o

 H  t f  A exp    kT  However, there is some question as to the presence of a minimum temperature

o

o

o

Periodically reported as 125C for unencapsulated and 85C for encapsulated Observed in other systems (tin-copper and whiskers)


OTHER SYSTEMS (CU-AL) Copper-aluminum forms intermetallics at a much slower rate

o

o

o

Most common activation energy of 1.26 – 1.47 eV o Micron reported 0.63 eV L. England, ECTC, 2007

HJ Kim, IEEE CPT, 2003

Molding compound has little effect

L Levine, Update on High Volume Copper Ball Bonding

C. Breach, The Great Debate: Copper vs. Gold Ball Bonding


OTHER SYSTEMS (CU-AL)(CONT.) Au-Al

Cu-Al

Cu-Al can show improved performance over Au-Al

o o o

Not to the extent expected based on intermetallic growth Different failure mode (gradual vs. sudden)


SHEAR STRENGTH AT ELEV TEMP Gold Wire

Copper Wire Cu

a.

b. Cu

o o

Shear strength of Au and Cu ball bonds on Al pads At lower temperatures ( 50 V/ns

Microsemi believes dV/dt issues were related to surface defects at the junction

o

o

o

Measure the dV/dt on the diodes and verify that the values are significantly less than 10V/ns Silicon Schottky diodes are limited to 10V/ns (Onsemi)


Typical dv/dt failure characteristics of 600 V/6A 4H-SiC Schottky barrier diode (SBD) and 600 V/8A silicon merged pin – Schottky (MPS) barrier diode at a case temperature of Tc = 25°C; note that SiC SBD is the diode marked SW2 whereas the silicon MPS diode ...

Krishna Shenai et al. ECS J. Solid State Sci. Technol. 2013;2:N3055-N3063


Die-Level Intrinsic o

Flicker (2014) of Sandia Labs reported concerning degradation behavior on SiC MOSFET and JFET devices

o

At 100Hz, 50% duty cycle at 25C, threshold voltage (VT) dropped at a rate of 2.5 mV/hr

o

Under similar conditions, the sub-threshold leakage current (Id) increased over 1000X 9000 Virginia Manor Rd Ste 290, Beltsville MD 20705 | 301-474-0607 | www.dfrsolutions.com

Die-Level Intrinsic (cont.) Other studies of SiC MOSFET reliability have raised concerns on long term performance at elevated temperatures

o

o

o

o

Lelis measured unstable threshold voltage due to electron tunneling into and out of the oxide traps Gurfinkel showed that the conventional DC measurement technique underestimated the threshold-voltage instability as fast transient trapping / detrapping events could not be captured with slow sweep rate Yu reported hot carrier effects in SiC MOSFET operated at moderate drain bias


Die-Level Intrinsic and PoF One of the significant challenges in developing PoF for WBG is the rapid change in architecture, materials, process, and design

o

o

o

179

Can make PoF models obsolete in relatively short order

An excellent example is time dependent dielectric breakdown (TDDB)


TDDB (Gate Oxide Reliability) Early experiments of insulators on SiC suggested that SiO2 on SiC was not reliable

o

o

o

More recent SiC oxide reliability tests have showed significant improvements

o

o o

180

Lipkin and Palmour showed SiC gate oxides had lifetimes less than 15 minutes at 6 MV/cm and 350◦C. Maranowski and Cooper determined (based on intrinsic failures) that 6H-SiC oxides had sufficient lifetimes only at E < 5 MV/cm and temperature below 150 ◦C.

Yu reported lifetimes of 215 hours at 6.4 MV/cm and 375◦C. Matocha and Beaupre reported lifetimes of 2300 h at 6 MV/cm and 250◦C


Historical Improvement in Oxide Reliability


Current Oxide Reliability


Package-Level Intrinsic The degradation mechanisms of concern for WBG packaging are similar to standard power silicon devices

o

o

o o

Wire bond corrosion Wire bond thermo-mechanical fatigue Die attach thermo-mechanical fatigue

The risk of fatigue will likely increase

o o

Lower CTE mismatch (by ~2ppm) will be overcome by much higher moduli [up to 400 GPa for 4H-SiC compared to 100 GPa for silicon]


Package Level Intrinsic o

Big risk is also exposure to higher temperatures and larger changes in temperature due to wider operating range of WBG

o

May require evaluation of new packaging designs and interconnect materials o

o

E.g., movement to WBG is driving increased substitution of aluminum wire bonds for copper wire bonds Copper wire bonds have higher thermal conductivity (reduced temperature rise), lower CTE (better match to WBG) and experiences less plasticity under load (better fatigue resistance)


PoF and Case Studies

185


PoF Case Study – Fretting


Frequency Response – 927 Hz

Maximum Deflection 0.0105 mm


Fretting Motion o

Very little motion near edge connections o Deflection at board to board connectors approx. 1.2 µm

o

Fretting motion regimes o Stick - Movement between the contact surfaces is accommodated by elastic deformation of the members in the near-surface regions o Stick – Slip - There is central stick area surrounded by an annular slip region where there may be crack formation, fretting fatigue, and wear debris. Movements are of the order 5 µm o Slip - All asperity contacts are broken during each cycle. Asperities slide across several others of the opposing surface. Damage is extreme with delamination wear. Movements of 10 - 100 typically µm are involved ANTLER, IEICE TRANS. ELECTRON., VOL.E82C, NO.1 JANUARY 1999

o

Potential motion is in the stick regime and fretting is not expected to occur 9000 Virginia Manor Rd Ste 290, Beltsville MD 20705 | 301-474-0607 | www.dfrsolutions.com

PoF Case Study – Ceramic Capacitors Manufacturer designed a wireless module for the Internet of Things (IoT) market

o

o

Incorporated a number of state-of-the-art ceramic capacitors DESCRIPTION

MANUFACTURER

PART#

APPLIED VOLTAGE RAIL

CAPC,X5R,0201,2.2UF,6.3V,+/- 20%

MURATA

GRM033R60J225ME47D

2.5V, 1.8V, 4.17V, 3.0V

MURATA

GRM033C80G473ME12E

2.5V, 1.8V

MURATA

GRM033C80G473ME12F

TAIYO YUDEN

RM AMK063 BJ473MPEF

TAIYO YUDEN

RM AMK063 6473MPEM

TDK

C0603X6S0G473MTB0NN

TDK

C0603X6S0G473MTB0PN

MURATA

GRM155R60J226ME11D

2.3V

MURATA

GRM188R60J226MEA0D

4.17V, 1.2V, 1.1V, 1.0V

MURATA

GRM188R60J226MEA0L

SAMSUNG

CL10A226MQ8NRNE

TAIYO YUDEN

CE JMK107BJ226MAETD

TDK

C1608X5R0J226MT005E

MURATA

GRM188R60J226MEA0

CAPC,X6S,0201,47.00nF,4.0V,+/- 20%

CAPC,X5R,0402,22.00UF,6.3V,+/- 20%

CAPC,X5R,0603,22.00UF,6.3V,+/- 20%

CAPC,X5R,0603,22.00UF,6.3V,+/- 20%


4.17V, 1.1V, 1.8V

Initial Assessment o

As an initial assessment, DfR used manufacturers test conditions and recommendations on acceleration factors to assess potential time to failure in the field

o

Field conditions o o

Voltage: See previous table Temperature: Evaluated a range of ambient temperatures + 18C temperature rise due to proximity of CPU

Test conditions

o o o o

Voltage: Dependent on part number (1X to 2X of rated) Temperature: Maximum rated temperature (X5R: 85C; X6S: 105C) Time: 1000 hours


Test Voltage DESCRIPTION

MANUFACTURER

PART#

TEST VOLTAGE

CAPC,X5R,0201,2.2UF,6.3V,+/- 20%

MURATA

GRM033R60J225ME47D

150%

MURATA

GRM033C80G473ME12E

100% (based on 0.47uF)

MURATA

GRM033C80G473ME12F

100% (based on 0.47uF)

TAIYO YUDEN

RM AMK063 BJ473MPEF

150% (?)

TAIYO YUDEN

RM AMK063 6473MPEM

150% (?)

TDK

C0603X6S0G473MTB0NN

Contact sales

TDK

C0603X6S0G473MTB0PN

Contact sales

MURATA

GRM155R60J226ME11D

100% (based on 10uF)

MURATA

GRM188R60J226MEA0D

100%

MURATA

GRM188R60J226MEA0L

100%

SAMSUNG

CL10A226MQ8NRNE

100%

TAIYO YUDEN

CE JMK107BJ226MAETD

150% (?)

TDK

C1608X5R0J226MT005E

Contact sales

MURATA

GRM188R60J226MEA0

100%

CAPC,X6S,0201,47.00nF,4.0V,+/- 20%

CAPC,X5R,0402,22.00UF,6.3V,+/- 20%

CAPC,X5R,0603,22.00UF,6.3V,+/- 20%

CAPC,X5R,0603,22.00UF,6.3V,+/- 20%


Lifetime (yrs) at 80% Rated Voltage

Murata (0201/X5R/6.3V/2.2uF & 0402/X5R/6.3V/10uF) 10 9 8 7 6 5 4 3 2 1 0

60

62

64

66

68

Capacitor Temperature (oC)

Reference sheets for GRM033R60J225ME47D and GRM155R60J106ME11D show the same lifetime curve

o

o

At 5VDC, capacitor temperature should be limited to a maximum of 64C (46C ambient temp) to reach a five (5) year lifetime


70

Murata (cont.) Several concerns with lifetime curve provided by Murata

o

o

o

o

o

Does not provide a voltage acceleration factor. Cannot extrapolate from 5VDC to 4.2VDC Does not define the failure rate for ‘useful lifetime’. Useful lifetime could be a very low probability of failure (10 yrs

o

Under constant 40C ambient (realistic worstcase), capacitors with applied voltages of 3V and 4.17V are a potential risk

Useful Lifetime (Yrs)


Murata Algorithm with 1.39eV Activation Energy o

Capacitors with applied voltage less than or equal to 1.2V, should have a ‘Useful Lifetime’ of >10 yrs

o

Under constant 40C ambient (realistic worstcase), all capacitors should have sufficient lifetime

Useful Lifetime (Yrs)


PoF Case Study – Nickel Wire Bonds o

Component manufacturer evaluating nickel wire bonds as a potential low-cost alternative for high temperature performance

o

Concerned about potential changes in to fatigue lifetime

o

Requested information on damage models for pure nickel (>99%) o

197

Especially interested in using this information for test plan development


KOEBERL ET. AL. (2007) Evaluated pure nickel (Ni 200/201)

o o o

o

99% pure, solid solution strengthened, wrought nickel Ni 201 is the low-carbon version

Determined combined Coffin-Manson/Basquin equation at room temperature 𝜀𝐴 = 𝜀𝑒𝑙 + 𝜀𝑝𝑙

198

𝜎𝑓′ = 2𝑁 𝐸

𝑏

+ 𝜀𝑓′ 2𝑁


𝑐

KOEBERL ET. AL. (2007)(cont.)

199


KOEBERL ET. AL. (2007)(cont.) o

200

Did not observe any measurable change in cycles to failure at temperatures up to 225°C


NICKEL MATERIAL PROPERTIES o

Pure nickel melts at 1455°C (1828K)

o

As a rule of thumb, creep is not expected to initiate until use temperatures are 0.5Tm o

For pure nickel, this is 641°C

Shear deformation mechanism map (more relevant than tensile for the braze application) by Frost and Ashby demonstrates a more nuanced behavior

o

o

Shear stresses slightly higher than 100MPa can induce power-law creep at temperatures below 0°C

Frost, Harold J., and Michael F. Ashby. "Deformation mechanism maps: the plasticity and creep of metals and ceramics." (1982). http://engineering.dartmouth.edu/defmech/

201


NICKEL MATERIAL PROPERTIES (CONT.) For the purposes of this activity, we will assume that the shear stress being applied to the nickel is less than 20 MPa

o

o

202

The deformation mechanism maps suggests this is the minimum stress necessary to induce creep at 175°C


ACCELERATION FACTORS FOR NICKEL FATIGUE

203

Fatigue Constant

2.46

Low Temperature (oC)

-75

-55

0

25

35

High Temperature (oC)

165

125

100

75

55

Change in Temp (oC)

240

180

100

50

20

Accel. Factor

1

2

9

47

448


Using PoF Software Case Study 1 American Auto Manufacturer® Motors saves over $1,384,000 in test costs o

In 2013 American Auto Manufacturer Motors begin utilizing PoF Software (Sherlock) for electric/hybrid inverter in parallel with its current testing plan.

o

Over the course of a twelve-month period, Sherlock identified four (4) designs where the probability of failure during PTC exceeded the acceptable standard for American Auto Manufacturer. o

204

Had the design moved forward, four (4) additional PTC tests would have had to be performed at a cost exceeding $346,000 each or $1,384,000 over the year


Using PoF Software Case Study 2 Global Storage Manufacturer reduces time-to-market by six (6) months o

Global Storage Manufacturer is using PoF software (Sherlock) to eliminate one (1) to two (2) physical tests for EACH PCBA Design

o

Ten (10) PCBAs are designed annually o

o

Sherlock also eliminates two (2) design revs for EACH PCBA

o o

205

Engineering labor and hard test costs (Chambers, Samples) are approximately $750,000 per test Results in an Annual Savings of $7,500,000

Time-to-Market Reduced by Six (6) Months


Using PoF Software Case Study 3 o

Case Study: Replacing Testing with Rapid Assessment

o

Working closely with DfR, Tier 1 Automotive Manufacturer convinced a Major OEM to replace elements of Product Qualification (temperature cycling, vibration, mechanical shock) with combination of PoF Analysis and DfR-led Quality Assessment o

o

206

Initial focus on managing design changes and cost reductions

Time-to-market reduced by 10 weeks and estimated cost savings of $150,000 9000 Virginia Manor Rd Ste 290, Beltsville MD 20705 | 301-474-0607 | www.dfrsolutions.com

Using PoF Software Case Study 6 o

Case Study: Optimizing Test Conditions and Test Time

o

Major HVAC manufacturer was performing extensive thermal cycling and thermal shock on all new AC Drive designs o

Used PoF software (Sherlock) to demonstrate test time was equivalent to 2X desired lifetime

o

o

o

207

Actual test time was in excess of seven (7) weeks

Sherlock also demonstrated that current test was not stressing the components at greatest risk of failure

Time-to-market reduced by three weeks and test costs reduced by $50,000 9000 Virginia Manor Rd Ste 290, Beltsville MD 20705 | 301-474-0607 | www.dfrsolutions.com

CONCLUSION Physics of Failure is part of a larger trend towards modeling and simulation

o

o o

o

208

Testing is too late, takes too long Design rules are too conservative, not pliable to new technology

With new products on the market, excellent opportunity to incorporate PoF-based analysis into new product development of power supply design
