part-to-part and lot-to-lot variability study of tid ... · normal law parameters optimized with...

TRAD, Tests & Radiations 07/03/2017 – ESA/CNES Presentation Days

Part-to-part and lot-to-lot variability study of TID effects

in bipolar linear devices

J. Guillermin, N. Sukhaseum, A. Varotsou, P. Garcia and C. Poivey

Radecs 2016 Proceedings [PC-2]

TRAD, Tests & Radiations 2

Introduction

Definition of Radiation Design Margin (RDM) requirements for Radiation Verification Testing (RVT) on flight lots under Total Ionizing Dose (TID)

Critical for linear bipolar devices Likely to show part-to-part and lot-to-lot variation in TID sensitivity

Investigation on the within-one-lot and inter-lot variability on

three different references Experimental characterizations Data analysis

One-sided tolerance limit (KTL) method is usually considered

for TID testing Applied to our test data Compared to a Maximum Likelihood Ratio (MLR) method which is

investigated and provides an interesting accuracy

07/03/2017 – ESA/CNES Presentation Days


Devices Under Test

Three different references were irradiated Three lots were tested for each reference 30 devices were irradiated for each lot

Cobalt 60 TID test

Dose rate 210 rad/h GAMRAY facility (TRAD – Labège, France)


0539A 1136A 1306A

Texas Instruments 5962R9950401VZA LM124AWG

GE245074 GE334152 GE337030

ST Microelectronics - TL1431ACZ

1052A Analog Devices 5962-3812801VGA AD584SH

0125A 0226A

Analog Devices 5962R3812801VGA AD584SH

Lots Manufacturer Part number Device

Op. Amp.

Volt. Ref.


Study Framework

Radiation hardness assurance standards Intra-lot variability taken into account with the 3-sigma approach

Over a test sample, the electronic device parameter drift considered for the worst-case analysis is mean ± 3 sigma

Inter-lot variability taken into account with periodic testing RVT periodicity depends on space project prime-contractor

In most of the cases, the 3-sigma approach is representative

enough and well adpated (when the associated risk is accepted) to take into account the intra-lot variability for a space project This study focuses on the few other cases…

In this presentation, 3 worst-cases (1 per tested reference) are shown Investigate the application of another statistical methodology

(MLR) that could be applied when a higher confidence level is required



Statistical Analysis

First level analysis 3-sigma approach KTL = 2.742 for n = 5 (C=90, P=90) [1]

Second level analysis Maximum Likelihood Ratio Test data fitted to a normal law

2-level analysis based on the use of experimental parameters ‘m’ and

‘s’ as defined above, as well as statistical parameters ‘µ’ and ‘σ’


∑=

=n

iix

nm

1

1∑=

−−

=n

ii mx

ns

1

2)(1

1

( )

−

−=2

21exp

21,,

σµ

πσσµ xxfnm

Experimental mean ‘m’

Standard deviation ‘s’

Normal law (2 parameters µ and σ)


AD584SH (AD)

AD584SH Vrline1 (%/V) lot 0226A


-1.0E-02

-5.0E-03

0.0E+00

5.0E-03

1.0E-02

1.5E-02

2.0E-02

2.5E-02

3.0E-02

0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220

Dose (krad)

Vrlin

e1 (%

/V)

m m ± 3s Specification

-1.0E-02

-5.0E-03

0.0E+00

5.0E-03

1.0E-02

1.5E-02

2.0E-02

2.5E-02

3.0E-02

0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220

Dose (krad)

Vrlin

e1 (%

/V)


020406080

100120140160180200220

All the lots 0125A 0226A 1052A

Out

of s

peci

ficat

ion

TID

leve

l (kr

ad)

All components 5 best 5 worst 4 best 4 worst random selection 1 random selection 2

Mean ± 3 sigma calculated over the 30 tested devices

Mean ± 3 sigma calculated over 5 randomly selected devices

Estimated lot behaviour for different subgroups Out-of-specification TID level based on

m+3s limit All devices, 5 best or worst, 4 best or worst,

random selections 30 tested devices

5 random selection

(1 lot plotted in grey)





TL1431ACZ (STM)

TL1431ACZ Iref (µA) lot GE334152






random selections 0123456789

1011121314151617

0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220

Dose (krad)

Iref (

µA)


30 tested devices

0123456789

101112131415

0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220

Dose (krad)

Iref (

µA)


5 random selection 0

5

10

15

20

All the lots GE245074 GE334152 GE337030

Out

of s

peci

ficat

ion

TID

leve

l (kr

ad)




30 tested devices

5 random selection




LM124AWG (TI)

LM124AWG Vio3 module 3 (mV) lot 1136A






random selections

020406080

100120140160180200220

All the lots 0539A 1136A 1306A

Out

of s

peci

ficat

ion

TID

leve

l (kr

ad)


-4

-3.5

-3

-2.5

-2

-1.5

-1

-0.5

0

0.5

0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220

Dose (krad)

Vio

3 m

odul

e 3

(mV)


-4

-3.5

-3

-2.5

-2

-1.5

-1

-0.5

0

0.5

0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220

Dose (krad)

Vio

3 m

odul

e 3

(mV)


30 tested devices

5 random selection




Second Level Analysis

Second level analysis Normal law parameters optimized with max. likelihood

The maximum likelihood is also used to define a confidence interval for the normal law parameters


( )∏=

=n

iixfL

1

,, σµLikelihood formula ( )LL maxmax =

Maximum likelihood

( )2,121ln 2

max

αχ −−≥

L

L

Confidence interval formula (maximum likelihood log-ratio)

0

1

2

3

4

5

6

7

8

9

6 7 8 9 10 11 12 13 14 15

Measured value

Freq

uenc

y

Test data Normal law (10;2)

Confidence interval for the optimized data mean

Representation of the normal fit confidence interval calculation (frequency plotted as a function of the

measured value for the electrical parameter)


AD584SH (AD)

AD584SH Vrline1 (%/V) lot 0226A


Max. Likelihood Ratio method with the 30 devices in one lot

Comparison of MLR confidence interval and 3-sigma calculation with 5 devices Number of sigma to cover the confidence

interval delimited by « 90% + 1σ »

MLR method, 10 devices selected in each one of the 3 tested lots To take into account the inter-lot variability

-1.0E-02

-5.0E-03

0.0E+00

5.0E-03

1.0E-02

1.5E-02

2.0E-02

2.5E-02

3.0E-02

3.5E-02

4.0E-02

0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220

Dose (krad)

Vrlin

e1 (%

/V)

Specification µ µmax 90% µmin 90% µmax+3σmaxµmin-3σmin µmax+1σmax µmin-1σmin m±3s

-1.0E-02

-5.0E-03

0.0E+00

5.0E-03

1.0E-02

1.5E-02

2.0E-02

2.5E-02

3.0E-02

3.5E-02

0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220

Dose (krad)

Vrlin

e1 (%

/V)

Specification µ µmax+1σmax/µmin-1σmin m±3s m m±10s m±15s

MLR confidence interval

Comparison

-1.0E-02

-5.0E-03

0.0E+00

5.0E-03

1.0E-02

1.5E-02

2.0E-02

2.5E-02

3.0E-02

0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220

Dose (krad)

Vrlin

e1 (%

/V)

Selected Data Specification µ µmax 90% µmin 90%µmax+3σmax µmin-3σmin µmax+1σmax µmin-1σmin m±3s(1 lot plotted in grey)

(3 lots plotted in grey, 30 selected devices in blue)


Comparison

10 devices in each lot

(green curves)

TRAD, Tests & Radiations 11 07/03/2017 – ESA/CNES Presentation Days






Comparison




Comparison

TL1431ACZ (STM)

TL1431ACZ Iref (µA) lot GE334152

0

2

4

6

8

10

12

14

16

18

20

0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220

Dose (krad)

Iref (

µA)

Selected Data Specification µ µmax 90% µmin 90%µmax+3σmax µmin-3σmin µmax+1σmax µmin-1σmin m±3s

0

2

4

6

8

10

12

14

16

18

20

0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220

Dose (krad)

Iref (

µA)


0

2

4

6

8

10

12

14

16

18

20

0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220

Dose (krad)

Iref (

µA)


(1 lot plotted in grey) MLR confidence interval

Comparison








Comparison

LM124AWG (TI)

LM124AWG Vio3 module 3 (mV) lot 1136A

-5.5

-5

-4.5

-4

-3.5

-3

-2.5

-2

-1.5

-1

-0.5

0

0.5

1

0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220

Dose (krad)

Vio3

mod

ule

3 (m

V)


-4.5

-4.0

-3.5

-3.0

-2.5

-2.0

-1.5

-1.0

-0.5

0.0

0.5

1.0

0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220

Dose (krad)

Vio3

mod

ule

3 (m

V)


-5.5

-5

-4.5

-4

-3.5

-3

-2.5

-2

-1.5

-1

-0.5

0

0.5

1

0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220

Dose (krad)

Vio3

mod

ule

3 (m

V)

Selected Data Specification µ µmax 90% µmin 90%µmax+3σmax µmin-3σmin µmax+1σmax µmin-1σmin m±3s



Comparison

MLR confidence interval (1 lot plotted in grey)


MRL Method Accuracy Multiplication factor applied to sigma

‘s’ at each TID step to cover the confidence interval delimited by « 90% + 1σ » from MLR calculation 30 device lot (top graph) 5 device random selection (bottom

graph)

Indication of the accuracy with the MLR method corresponding to KTL = 5 (P > 0.999 for C = 90)

The impact of the lot/selection

homogeneity in terms of degradation can be observed AD584 Vrline1 exhibits atypical

devices Larger margin required to cover

the lot behaviour LM124 Vio3 more homogeneous

Smaller margin required to cover the lot behaviour

1

2

2

3

3

4

4

5

5

6

0 50 100 150 200Dose (krad)

Coe

ffici

ent V

alue

AD584 Vrline1 0226A AD584 Vrload1 0226A AD584 Ios 0226A TL1431 Ioff GE334152TL1431 Zka GE245074 LM124 CMRR 1306A LM124 Iio1 1306A LM124 Vio3 1136A

0

5

10

15

20

25

30

0 50 100 150 200Dose (krad)

Coe

ffici

ent V

alue

AD584 Vrline1 0226A AD584 Vrload1 0226A AD584 Ios 0226A TL1431 Ioff GE334152TL1431 Zka GE245074 LM124 CMRR 1306A LM124 Iio1 1306A LM124 Vio3 1136A


Conclusion

TID testing was performed on 3 lots for 3 bipolar device references Analysis of the lot-to-lot and within-one-lot variability

Data analysis objectives

Characterize the lot coverage based on 5-device sample size for TID testing Propose and evaluate a Maximum Likelihood Ratio (MLR) method based on

the use of confidence intervals MLR is an accurate way to estimate the lot behaviour

Large sample size needed (30 devices) Use of a confidence interval on the mean μ of the electrical parameter

measurement

Can be applied to data groups composed with different lots Increase the sample size, better calculation accuracy Takes into account the lot-to-lot variability

Perspective

Cost problem related to the large needed sample size Adaptation of the MLR method to smaller sample sizes but keeping similar

accuracy



References

[1] MILITARY HANDBOOK, Ionizing Dose and Neutron Hardness Assurance Guidelines for Microcircuits and Semiconductor Devices, MIL-HDBK-814, 08 February 1994.

[2] R. L. Pease, ‘Total ionizing dose effects in bipolar devices and circuits’, IEEE Transactions on Nuclear

Science, vol. 50, no. 3, pp. 539–551, Jun. 2003. [3] A. Namenson and I. Arimura, ‘A Logical Methodology for Determining Electrical End Points for Multi-Lot

and Multi-Parameter Data’, IEEE Transactions on Nuclear Science, vol. 34, no. 6, pp. 1726–1729, Dec. 1987. [4] I. Arimura and A. I. Namenson, ‘Hardness Assurance Statistical Methodology for Semiconductor Devices’,

IEEE Transactions on Nuclear Science, vol. 30, no. 6, pp. 4322–4325, Dec. 1983. [5] R. Ladbury, J. L. Gorelick, and S. S. McClure, ‘Statistical Model Selection for TID Hardness Assurance’, IEEE

Transactions on Nuclear Science, vol. 56, no. 6, pp. 3354–3360, Dec. 2009. [6] R. Ladbury and J. L. Gorelick, ‘Statistical methods for large flight lots and ultra-high reliability applications’,

IEEE Transactions on Nuclear Science, vol. 52, no. 6, pp. 2630–2637, Dec. 2005. [7] R. Ladbury, ‘Statistical Techniques for Analyzing Process or “Similarity” Data in TID Hardness Assurance’,

IEEE Transactions on Nuclear Science, vol. 57, no. 6, pp. 3432–3437, Dec. 2010. [8] R. Ladbury and B. Triggs, ‘A Bayesian Approach for Total Ionizing Dose Hardness Assurance’, IEEE

Transactions on Nuclear Science, vol. 58, no. 6, pp. 3004–3010, Dec. 2011. [9] R. L. Ladbury and M. J. Campola, ‘Bayesian Methods for Bounding Single-Event Related Risk in Low-Cost

Satellite Missions’, IEEE Transactions on Nuclear Science, vol. 60, no. 6, pp. 4464–4469, Dec. 2013. [10] ‘Part-to-part and lot-to-lot variability study of TID effects in bipolar linear devices’ ESA study report ref.

TRAD/ESA/IR/VAR/NS/241115 of 29/11/2015. [11] J. Guillermin, N. Sukhaseum ‘Part-to-part and lot-to-lot variability study of TID effects in bipolar linear

devices’, Radecs 2016 Proceedings PC-2


part-to-part and lot-to-lot variability study of tid ... · normal law parameters optimized with...

Documents