PEM Research

Alexander TeverovskyParts, Packaging, and Assembly Technologies

Office, Code 562, GSFC/ QSS Group, Inc.

Outline

Temperature limitations in PEMsHASTAcoustic microscopyWire bond qualityHow to improve PEM guidelines

July 2004 2

July 2004 3

What limits testing T for PEMs?

Maximum junctiontemperatureMax. operational temperature Max storage temperature Glass transitiontemperature of MC

• There is no standardized technique to determine Tj max and Tst(?)

• Absolute maximum ratings?• Mfrs. warn that exceeding these T

may cause permanent damage• Ex1: Tjmax=165 oC, Tst=150 oC• Ex2: Tjmax=100 oC, Tst=100 oC• Does exceeding Tg cause failures?

A methodology to determine BI temperature and conditions is needed.Does exceeding Tg cause failures?

July 2004 4

Tg Measurements: TMA

TMA2940PEM

Quartz probe

Thermocouple

July 2004 5

Does low Tg indicate Poor Thermal Stability?

TGA for MC with different Tg TGA measurements of MCs with different Tg showed that materials with Tg ~135 and 115 oC had higher stability compared to MC with Tg.The most thermally stable polymers, silicone rubbers and Teflon, have extremely low Tg of -20 to -120 oC for the rubbers and -90 oC for Teflon.

90

92

94

96

98

100

300 400 500 600temperature, oC

wei

ght,

%

170 C138 C173 C174 C133 CTg~173 oC

Tg~135 oC

Tg is not an indicator of thermal stability of MC

DTGA at 1 oC/min N2

-0.08

-0.07

-0.06

-0.05

-0.04

-0.03

-0.02

-0.01

0

250 300 350 400

temperature, oC

mas

s lo

ss ra

te, %

/min

Tg=169

green 115

Tg=146

Tg=165

July 2004 6

Effect of Tg on Failure Modes and Mechanisms

When characteristics of MC affect reliability of PEMs, the Tg effect might be expected.This is important for analysis of acceleration factors and for prediction of long-term reliability based on high-temperature stress testing.

Most probable case:Ts > Tc > Tg; λq >> λexper

Tg most likely will not affect results of standard S&Q testing

July 2004 7

Tg Effect SummaryChanges in failure rates at T>Tg are possible and should be considered for analysis of acceleration factors and prediction of long-term reliability of PEMsbased on high-temperature stress testing.

Experimental data did not reveal the effect of Tg on results of standard screening and qualification testing.

For a normal quality lot there is no immediate danger in exceeding Tg during standard reliability testing.

TMA of plastic packages has been proven to be an important tool for FA and reliability evaluation of PEMs.

July 2004 8

HASTBiased HAST is the most sever environmental test (~ 30% to 50% chance of failure)Is HAST adequate to normal conditions?(130 oC > Top, swelling, delaminations, acceleration: 100 hr >100 years)Biased HAST win-win testing for space applications?Still… moisture can penetrate to die and cause failures.Moisture concentration depends mostly on RH, not on T. Temperature significantly accelerates penetration.Unbiased HAST might not reveal moisture-related failures.We do need an adequate accelerated testing to assess the effect of moisture for space applications.

July 2004 9

Opamp SMT simulation and HAST results

Part Type Pack DC SMT HAST Comm.

OP1 Instr. SOIC8 0030, 0110 1/30, 0/30

30/30, 27/30 Param. failures

OP2 Quad SOIC16 0101 16/30 2/30 SMT: IOS and AOL HAST: init.+1 new

OP3 Instr. SOIC8 0022, 9628 0/16, 0/16 0/16, 0/16

OP4 Bandw SOIC8 0018 0/30 0/30

OP5 Bandw SOIC8 0041 0/30 0/30

OP6 Dual SOIC8 0033, 0109, 9946

0/30, 0/30, 0/30

0/30, 0/30, 1/30

Possibly reverse installation

OP7 Precis. SOIC8 0019, 0021, 0029

0/30, 0/30, 0/30

0/30, 0/30, 0/30

OP8 Quad SOIC16 9945 2/30 0/28

July 2004 10

Distribution of Biased HAST FailuresFailures during biased HAST

0

20

40

60

80

100

0 200 400 600 800 1000

time, hr

failu

res,

%

130 C

150 C

110 C

Arrhenius plot of median life

10

100

1000

2.3 2.35 2.4 2.45 2.5 2.55 2.6 2.65

1000/T, 1/K

med

ian

time,

hr

Ea = 0.59 eV

Ea is lower than in Peck-Hallberg model: 0.79 -1.1 eV.This increases the probability of failures at low T.

July 2004 11

Arrhenius-Weibull model

10.00

1.00E+5

100.00

1000.00

10000.00

450.00300.00 330.00 360.00 390.00 420.00

Life time prediction at 85% RH

temperature, K

time,

hr

2.2 years Ea = 0.59 eV

HAST at 85% RHT = 110, 130, and 150 deg.C

1% failures

July 2004 12

RT Testing after Unbiased HASTRT bias testing after

110 oC HAST for 800 hrs

0.3

0.35

0.4

0.45

0.5

0.55

0 500 1000 1500

time, hr

IPS,

mA

Failures during RT testing followed unbiased HAST

0

20

40

60

80

100

0 200 400 600

time, hr

failu

res,

%

after 150C/85%/130hr after 130C/85%/350hr after 110C/85%/800hr

RT testing followed unbiased HAST resulted in failures in 24 to 600 hours.PEMs retain moisture for some time after HAST.

July 2004 13

Moisture diffusion and retention in PEMs

Note: C - moisture concentration on the die surface;M - mass of moisture in the package;C/Co = 1 corresponds to moisture saturation at 100% RH.

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0.01 0.1 1 10 100

time, hr

C/C

o, M

/Mo C/Co soaking

M/Mo soakingC/Co testingM/Mo testing

Example:Package 2mm

Soak conditions: 110oC/85%/48hr, unbiased.

Test conditions: 85 oC, bias.

Soaking

Testing

July 2004 14

HAST Summary

Biased HAST is a too severe test for space applications.Unbiased HAST is far less severe and might not reveal failures. However, post-HAST testing at RT under bias causes failures.A combination of unbiased HAST with LT bias testing might be a good alternative to biased HAST for QA of parts intended for space applications.

July 2004 15

C-SAM: SWIFT StatisticsRejectable delaminations were observed in 4 out of 6 types of power devices and in 14 out of 23 types of linear devices.

The proportion of rejects varied from 2.3% to 28% for power devices and from 0.14% to 83% for low-power devices.

The cost of AM is relatively high, up to $8 per part, even for alarge quantity lot.

Out of 31,090 parts subjected to screening, 565 (1.8%) were rejected by electrical testing and 3,586 (11.5%) by CSAM.

Acoustic microscopy rejected far more parts than did electrical measurements. Are all these rejects potential failures, and if so, what is theconfidence in quality of a lot with ~10% rejects?

July 2004 16

CSAM: Which Part to Choose?Delaminations are due to stress relief in PEMs.

This part might operate reliably because

mechanical stresses have been relieved and

delaminations are stable.

This part might fail if the stresses have not been

relieved and delaminationsdevelop in critical wire bond

areas.

Typically rejected Typically accepted

CSAM is a nondestructive test, but can damage the parts.If delaminations are serious defects, is it possible to improve reliability of a lot with 5% to 50% rejects by screening?Is CSAM useful for screening?

Before stress. After stress?Before stress. After stress?

July 2004 17

Changes in Delaminations Due to Stress Testing (Power Devices)

Die surface Leads

Top paddle

0

20

40

60

init after SMT after 1000TC

after HT TC

dela

min

atio

ns, %

FET2FET1FET3

0

20

40

60

80

100

init after SMT after 1000TC

after HT TC

dela

min

atio

ns, %

0

20

40

60

80

100

init after SMT after 1000TC

after HTTC

dela

min

atio

ns, %

FET2FET1FET3

The proportion of delaminations can vary significantly after solder reflow and environmental stress testing

July 2004 18

Results of Temperature Cycling of Parts with Excessive Delaminations

Part SMT TC0 100 TC0 3000/30 0/30

0/300/300/300/300/300/300/300/30

0/300/300/300/300/300/300/300/30

TC0 1k TC1 300 TC2 300PEM1 0/30 0/30 0/15 0/15PEM2 0/30 0/30 1/15 1/15PEM3 0/30 0/30 0/15 0/15PEM4 0/30 0/30 0/15 0/15PEM5 0/30 0/30 0/15 0/15FET1 0/30 0/30 0/15 0/15FET2 0/30 0/30 0/15 0/15FET3 0/30 0/30 0/15 0/15

PEM6* 0/30 0/30 - -

*Parts had excessive delaminations at finger-tips after HAST.

July 2004 19

CSAM SummaryParts with excessive delaminations at the paddle and leads (at secondary bond locations) had no electrical failures after 1000 TC at –55 to +125 oC conditions.

The analyzed power devices in TO220-style packages are prone to formation of top-of-die delaminations; however, no wire bond fractures occurred even after multiple temperature cycling.

No correlation between the proportion of delaminations and flux-induced leakage currents were observed indicating a failure of CSAM examination to screen out potential failures.

For the part types used, CSAM examination likely is not a value-added technique for screening; however, it is very useful for qualification testing.

July 2004 20

Wire Bonding: problems recognized by manufacturers

Lifted bonds in various SOIC-type (14/16/20/24 pins) widebody package

parts assembled by NSEM between Feb and Nov 2002

R.Sampan, R Kuang(Actel), CMSE’03:Hermetic gold bonded devices with contamination on bond pads may pass bond pull test at as-bonded condition but fail after 80 hrs at 150°C.

Degradation and failures in poor quality wire bonds might happen with time even at relatively benign storage

conditions.

July 2004 21

Wire Bonding: examples of previously previously revealed problems

Failures in parts manufactured in 1999 were observed during board-level testing in 2002 (FA: contamination-induced WB degradation).

COTS FPGA failed DPA due to poor wire bonding (2003).

A detector failed in 2002, three years after manufacturing due to degraded WB (Al wire to Au post).

A lifted wire in an ASIC packaged in

PQFP was revealed during failure analysis

(2003).

July 2004 22

Wire Bonding: problems 2004SDO:MTK-WCG-DF-0412 ASIC ceramic packageLifting at 1.9 g.

VMI hybrid plastic XA-16 ASIC ceramic

STEREO

GLAST: photodiodes+all ASICsin plasticQFP

Substantial portion of reliability issues are now at the packaging level.Development of a non-destructive technique for WB qualification in PEMs is important.

July 2004 23

Examples of WB degradation after HTSLMost WB were lifted showing porous structure of intermetallics. The peripheral area of Al around the bonds had multiple voids. The appearance of voids suggested diffusion of Al atoms towards the WB along the crystalline grain boundaries. For LT the appearance of the voids suggested a corrosion attack (possibly H3C-Br).

July 2004 24

Technique for Evaluation of WB Quality

Rc measurements:VF of P-N junctions used in the input/output ESD protection circuits are measured before and after the stress

MeasureVF0

StressMeasure

VF1

δRc = (VF1 – VF0)/IF

This technique is simple, does not require full electricalcharacterization, and have an accuracy ~ 0.3 Ohm.

July 2004 25

Example of early WB failuresWeibull-Arrhenius plot

THS

1% failures after ~ 1 year at 100 oCFA results: failures were due

to WB degradation

Wire bond degradation might cause failures even at relatively low temperatures of ageing

July 2004 26

WB Infant Mortality Failures

100.00 10000.001000.000.10

0.50

1.00

5.00

10.00

50.00

90.00

99.00

0.10

time, hr

cum

ulat

ive

prob

abilit

y, %

infant mortality failure

HTSL at 190 deg.C99% confidence bounds

This part had also WB

failures after HAST

July 2004 27

Degradation of contact resistance

0

20

40

60

80

100

0.1 1 10 100Rc, Ohm

accu

mul

ated

per

cent

120 hr

210

346

530

690

970

Distributions of Rc in air

0.1

1

10

100

0 500 1000 1500 2000

time, hr

med

ian

Rc,

Ohm

OP vacOP airLT vacLT air

air

vacuum

Kinetics of Rc variations

WB degradation in vacuum is much lesser than in air.After ~1000 hrs in air ~70% of WBs had Rc >10 Ohm; in vacuum Rc < 3 Ohm even after 1500 hrs.

July 2004 28

WB problems: SummaryWB degradation limits HT reliability of PEMs

WB is a real problem; however, it is not addressed in the guidelines.

Forward voltage drop measurement technique might be useful for WB qualification, life prediction and revealing IM failures.

There is a critical temperature, after which the rate of WB failures changes.

Vacuum testing might provide high acceleration without introducing new failure modes.

July 2004 29

Evolution of Parts Engineering for PEMs

Rule-based(MIL parts)

Knowledge-based(COTS PEMs)Phase I, II, ..

No wear-out, no damaging testing.

Reliability is mostly limited by defects => the importance of infant mortality and screening.

Element and package size reduction easier to damage.

Reliability is limited by intrinsic wear-out mechanisms => qualification is more important than screening.

Government maintains the standards and is responsible for reliability provided all criteria are satisfied.

PEs are responsible for reliability evaluation and mitigation of risks associated with use of COTS PEMs.

July 2004 30

Catch-22 in ScreeningLiterature data and our experience:Screening might not add value and might negatively impact reliability of high-quality components.However, for low-quality components, screening should be maintained.

Do not screen lotswith low IM

Screen the lot to reveal

IM

testi

ng

testi

ng

To break this circle, make BI and CSAM optional, depending on results of extended qualification.Perform qualification on unscreened samples with interim measurements and let the parametric/functional failures to go through. This will identify possible failure modes, evaluate the risk of failures, show whether the screening is necessary, and reduceoverall S&Q costs.

July 2004 31

A Feedback System to Improve S&Q

Most project test data are wasted; obtaining statistics of failures is not easy. (Projects are often not concerned about failures <PDA)Test results might be deceptive. (Rejects might be due to poor electrical measurements and/or damage during testing.) FA results might be insufficient. (FA might be performed only tomitigate the risk for the mission.)Do not miss an interesting failure case (WB infant mortality)

Projects are generating a lot of data.Learn from our experience.

QASAccumulation and analysis

of project data LL

Analysis of literature data

Additional evaluation and testing

July 2004 32

Revision of PEM GuidelinesMake CSAM optional provided extensive qualification using acoustic microscopy is planned. Allow qualification of non-screened devices provided a larger quantity of parts with interim measurements, failures going through, and analysis of failed parts, is performed.The proportion of area where intermetallics at Au/Al wire bond are formed should be evaluated during DPA. If < 50 %, additional evaluations is necessary (WB package-level qualification).Remove Tg limitations for BI.Eliminate flux application during preconditioning.Replace HAST with a combination of HAST and biased testing at RT.

July 2004 33

Some Trends for Improvement of the Guidelines to Discuss in Future

Put more emphasis on evaluation rather than on DPA. Addressing concerns rather than rejecting suspicious parts.

More attention to qualification rather than to screening.

Replace existing 3-level S&Q test flows with specially designed application-dependent plans.Use alternate techniques for S&Q:

noise measurements;Non-linearityRc measurements for WB qualification.Thermal impedance for power devices.…