AEC - Q003 Rev-A

February 18, 2013

Automotive Electronics Council
Component Technical Committee

GUIDELINE
FOR CHARACTERIZATION
OF
INTEGRATED CIRCUITS
 AEC - Q003 Rev-A
February 18, 2013
Automotive Electronics Council
Component Technical Committee

Acknowledgment

Any document involving a complex technology brings together experience and skills from many sources. The
Automotive Electronics Counsel would especially like to recognize the following significant contributors to the
development and release of this document:

Characterization Sub-Committee Members:

Tim Haifley Altera

Sandra Healy Analog Devices
Michael O’Sullivan Analog Devices
Xin Miao Zhao [Q003 Team Leader] Cirrus Logic
Ramon Aziz Delphi Corporation
Paul Grosch Freescale
Nick Lycoudes Freescale
Laurie McTurk Freescale
James Stanley Freescale
Drew Hoffman Gentex
Richard Lauks Infineon
Harold Lorenz Infineon
Jochen Roeder Infineon
Tim Weyland Infineon
Banjie Bautista ISSI
Gary Fisher Johnson Controls
Bob Knoell NXP
Yizi Xing NXP
David Ownby SMSC
Joel Dobson Texas Instruments
Thomas VanDamme TRW
Daniel Chung Xilinx
 AEC - Q003 Rev-A
February 18, 2013
Automotive Electronics Council
Component Technical Committee

NOTICE

AEC documents contain material that has been prepared, reviewed, and approved through the AEC
Technical Committee.

AEC documents are designed to serve the automotive electronics industry through eliminating
misunderstandings between manufacturers and purchasers, facilitating interchangeability and improvement of
products, and assisting the purchaser in selecting and obtaining with minimum delay the proper product for
use by those other than AEC members, whether the standard is to be used either domestically or
internationally.

AEC documents are adopted without regard to whether or not their adoption may involve patents or articles,
materials, or processes. By such action AEC does not assume any liability to any patent owner, nor does it
assume any obligation whatever to parties adopting the AEC documents. The information included in AEC
documents represents a sound approach to product specification and application, principally from the
automotive electronics system manufacturer viewpoint. No claims to be in conformance with this document
shall be made unless all requirements stated in the document are met.

Inquiries, comments, and suggestions relative to the content of this AEC document should be addressed to
the AEC Technical Committee on the link http://www.aecouncil.com.

Published by the Automotive Electronics Council.

This document may be downloaded free of charge, however AEC retains the copyright on this material. By
downloading this file, the individual agrees not to charge for or resell the resulting material.

Printed in the U.S.A.

All rights reserved

Copyright © 2013 by the Automotive Electronics Council. This document may be freely reprinted with this
copyright notice. This document cannot be changed without approval from the AEC Technical Committee.
 AEC - Q003 Rev-A
February 18, 2013
Automotive Electronics Council
Component Technical Committee

GUIDELINE FOR THE CHARACTERIZATION OF INTEGRATED CIRCUITS

Text enhancements and differences made since the last revision of this
document are shown as underlined text.

1. PURPOSE

The characterization of ICs is an extremely important function during the development of a new IC or
the modification of an existing IC. The purpose of this document is to provide guidance by highlighting
important considerations that should be evaluated during development of a characterization
procedure. This document is not intended to be a specification on how to perform a characterization.
To ensure consistent characterization every company should have a structured and documented
characterization procedure. The characterization process should be used to ensure that the design
and wafer fab/assembly processes being utilized demonstrate sufficient capability of providing a part
that meets the requirements of the customer. The results of every characterization should be
documented in a Characterization Report.

2. SCOPE

This guideline document provides the basis for establishing a procedure for characterizing the
electrical performance of Integrated Circuit products. This characterization procedure should be used
for new technologies, new wafer fabrication processes, new product designs, and change in bill of
material, manufacturing location and significantly modified ICs. This full scale characterization is a
procedure to assess a new product line, including establishing supplier datasheet limits. For
qualification of a product with established supplier datasheet limits, please reference AEC-Q100-009
Electrical Distributions Assessment procedure.

3. DEFINITIONS

3.1 ANOVA

Analysis of variance between groups.

3.2 Characterization

The process of determining the fundamental electrical and physical characteristics of a device based
on statistical analysis of experimental data modeling.

The input is batches of newly manufactured integrated circuit parts subjected to characterization
sampling and the output is a data set. The result is to prove or disprove that the devices can and will
continue to perform their targeted functions according to the product definition. Data resulted from
characterization are used to specify datasheet limits and functions.

It is important to determine the ESD and Latch-up capability of a part early in the development cycle
as these parameters could have a significant impact on the performance of the part.

Page 1 of 18
 AEC - Q003 Rev-A
February 18, 2013
Automotive Electronics Council
Component Technical Committee

3.3 Cpk

A measure of the relationship between the specification limits and the capability. Reference: PPAP
Manual Fourth Edition, see SPC Manual.

3.4 Criteria for Characterization

A device that meets the following criteria should be considered a candidate for characterization.
These criteria should not be considered a limitation, but a starting point for factors to consider.
• New product design.
• New device layout or changes to an existing device that could impact electrical parameters.
• New cell structure(s) which have not been used in production components.
• New processing methods or materials.
• New operating bias condition requirements (re-characterize at new bias extremes).
• New operating environmental conditions (such as voltage and temperatures).

3.5 Device Electrical Parameters

Electrical parameters specified in the part specification for the device.

3.6 Guardband

The difference between limits used for screening (e.g., characterization, production testing, etc.) and
specification limits. Guardbands are used to account for repeatability and reproducibility issues.

3.7 LSL

Lower Specification Limit.

3.8 Matrix Lot

Matrix lot is a lot composed of wafers that are based on manufacturing site’s process of records and
manufactured to the process corners for identifying design/process weakness and improvement as
well as indicating yield sensitivity corners.

3.9 PPM

Part Per Million. In this document the PPM is the probability the parameter will be outside of product
specification limits.

3.10 Specification Limits (e.g., Data Sheet Limits)

Numerical values of maximum, minimum, and typical for electrical parameters specified in supplier’s
data sheet or customer’s part specification. These values are used to determine pass/fail criteria for
device characterization and production testing.

Page 2 of 18
 AEC - Q003 Rev-A
February 18, 2013
Automotive Electronics Council
Component Technical Committee

3.11 Standard Deviation

Standard deviation or sigma is a measure of how widely values are dispersed from the average value
(the mean).

3.12 USL

Upper Specification Limit.

4. CHARACTERIZATION PROCEDURE

An outline of factors that should be considered when establishing a characterization procedure is

provided here for every supplier to establish a characterization procedure. The Device
Characterization Process Flow is shown in Figure 1.

4.1 Device Characterization Plan

The characterization plan should include the following major activities for the device to be
characterized:
• Review the Characterization Checklist, see Appendix 1.
• Determination of if a matrix lot is necessary for the device characterization.
• Determination of the characterization method to be used.
• Establishment of the parameters and conditions to be characterized.
• Define format of the characterization report.

4.2.1 Matrix Lot Characterization

When characterizing a matrix lot, the number of split cells, samples per cell and the data analysis
methods should also be defined in the plan. Guidelines for designing, planning and analyzing the
split cells for a matrix lot are provided in Appendix 2 and 3.

4.2.2 Sample Sizes

When deciding on sample sizes for characterization, two important factors are to be considered:
confidence interval and confidence level. Please see details in Appendix 4 for help in determining
sample sizes.

4.3 Characterization Report

The characterization report should include the following:

a. A copy of the characterization plan.
b. A detailed discussion of the characterization methods used (see Appendix 1, 2, 4 and 5).
c. A listing of parameters and conditions used in characterization.
d. Characterization data analysis and conclusions.
e. Document simulation results including brief explanations on methods applied – for
parameters that are not measurable and/or tested in production and covered by design
simulation only.
f. Identify part weaknesses and reliability concerns and define corrective actions.

Page 3 of 18
 AEC - Q003 Rev-A
February 18, 2013
Automotive Electronics Council
Component Technical Committee

Product 1234

Is
Characterization
Required?

y New product
y New chip layout or changes to an existing
No y New cell structure
y New processing methods or materials
y New operating bias condition
requirements

Yes
y Review Characterization
Checklist
Internal Characterization y Determination of the
Procedure characterization method
y Establishment of the
parameters & conditions

Characterization Plan

Perform
Characterization

Approve by Characterization Modify Plan

Similarity Report

y Characterization methods
y Parameters & conditions
y Data analysis & conclusions
y Simulation results
y Part weaknesses & reliability
concerns

No
Approve Report?

Yes

Completion of Project

Figure 1: Characterization Flowchart

Page 4 of 18
 AEC - Q003 Rev-A
February 18, 2013
Automotive Electronics Council
Component Technical Committee

4.4 Characterization Data Analysis

There are many statistical methods that could be used to perform analysis of the data. One
commonly accepted is using Cpk when the output response could be assumed with a normal or
Gaussian distribution.

4.4.1 Cpk and PPM to Data Sheet Limits

Cpk computes the differentiation between the actual process average and the specification limit over
standard deviation. The two things which Cpk measures:
a. The interpretations of the mean as how close they are to the center of the upper and lower
spec limits?
b. How extensively the readings are spread over the measurement range?

The distance of the upper and the lower spec from the distribution mean on the basis of the Cpk
value could be expressed in multiples of the standard deviation.

In the absence of production test to screen out parts outside specifications, the Cpk of the
specification limits will give an indication of the expected ship PPM. Both short term and long term
Cpk can be considered.

4.4.1.1 Short Term Cpk in Relation to PPM

In quality control terms, PPM stands for the number of parts per million indicated by this fraction of
area. In a short term sense, a process could have a centered normal distribution with two sided
specification (Upper Spec Limit / Lower Spec Limit) and no sigma shift, the relation between Cpk and
PPM is as shown in Figure 2.

Figure 2: Cpk vs. Probability of parts outside of specification (centered symmetrical normal
distribution)

With this assumption, a Cpk value of 1.00 means that 2700 PPM (0.27%) of the manufactured parts
could be out of tolerance, while Cpk 1.67 means that 0.57 PPM (0.000057%) are potential rejects.
Examples of typical Cpk and PPM values are given in Table 1 for a parameter with min and max
specifications.

Page 5 of 18
 AEC - Q003 Rev-A
February 18, 2013
Automotive Electronics Council
Component Technical Committee

Table 1: Short Term CPK and PPM Estimation

Cpk Sigma PPM

0.67 2.00 45500
1.00 3.00 2700
1.33 4.00 63
1.67 5.00 0.57
2.00 6.00 0.002

4.4.1.2 Long Term Cpk in Relation to PPM

In most product manufacturing the process does shift. The relation between Cpk and PPM allowing
for a 1.5 sigma process shift over time is illustrated in Figure 3.

Figure 3: Probability of parts outside of specification for Cpk of 1.67 (left) and with 1.5 Sigma
shift toward USL and LSL (right)

With this 1.5 Sigma shift, the potential parts outside specification could become significant higher
compared to without shift. Examples of typical Cpk and PPM values with long term 1.5 Sigma shift
are given in Table 2. For comparison purposes, Cpk and Sigma values exclude shift are included.
Under this condition, a Cpk of 1.67 may deteriorate to a Cpk of 1.16 which will result in a 233 PPM
compared to 0.57 PPM from Table 1.

Table 2: CPK and PPM Estimation (with a 1.5 Sigma process drift)

Assuming a long term 1.5 sigma

Short term
shift
Sigma with Cpk with
Cpk Sigma PPM
shift shift
0.67 2.00 0.50 0.17 308,538
1.00 3.00 1.50 0.50 66,807
1.33 4.00 2.50 0.83 6,210
1.67 5.00 3.50 1.17 233
2.00 6.00 4.50 1.50 3

Page 6 of 18
 AEC - Q003 Rev-A
February 18, 2013
Automotive Electronics Council
Component Technical Committee

4.4.2 Cpk and PPM with Production Test

To maintain low PPM when the Cpk to datasheet limits is inside six sigma (Cpk<2.0), 100%
production test should be considered to screen out parts that are outside specification. Because of
test measurement error and parametric temperature and lifetime drift, appropriate statistical models
should be used to create test limits that guardband against datasheet limits to ensure the shipped
product meets the customer PPM requirement.

4.4.3 Acceptance Guidelines

Cpk is a typical measure of characterization data for acceptance. The acceptance value of Cpk
should be target PPM level dependent with the process deviation (sigma) from target process during
production should be taken into consideration as well. Please reference AEC Q100 and other
supplier and customer agreement for acceptable Cpk values.

4.5 Data Reporting

Characterization data reporting is an important part of the final report. A summary of the data should
be provided and raw data should be archived and available for customer review upon request.
Graphical presentation of the statistical analysis of the experimental data and modeling is strongly
encouraged. For significant changes made to existing devices, comparison of data from the new
revision to the previous revision is recommended. The data could be presented in different ways with
two examples Correlation plot (Figure 4) or box plot (Figure 5) given below. Other methods such as
histograms, cumulative plots, and raw data plots are often used as well.

4.5.1 Hyper–Ellipsoid Correlation Plot

This includes the distribution of an electrical parameter as a function of other parameter(s) variation.
The graphical presentation of the characterization for a typical electrical parameter is a multivariate
hyper-ellipsoid, see Figure 4 (frequently called a Correlation Plot).

Each dot is a
measurement
Test Limit from one part

Specification
Limit
Parameter 2

Guardband

Parameter 1

Figure 4: Correlation Plot

Page 7 of 18
 AEC - Q003 Rev-A
February 18, 2013
Automotive Electronics Council
Component Technical Committee

4.5.2 Box Plot

This includes the distribution of an electrical parameter as a function of operating condition variations
(voltage, temperature, frequency, pad load conditions, input timings) or across process window matrix
cells. The graphical presentation of the characterization results for an electrical parameter in relation
to multiple operating conditions could be presented in different ways such as Correlation plot (Figure
4) or box plot (Figure 5). Statistical data from device characterization (e.g., target sample size, min,
mean, median, max, sigma / cpk (nominal group)) should be provided in addition for each box plot.

Upper Spec Limit

25th Percentile
Parameter

Minimum

Mean

50th Percentile

Maximum

75th Percentile

Lower Spec Limit

Group X Group Y Group Z Group Q

Group / Operating Condition Variations

Figure 5 – Box Plot

5. APPROVAL AND ACCEPTANCE

The characterization report should be approved by a responsible supplier representative. The results
of the characterization should be shared with the customer per request (this sharing may include a
copy of the characterization report or selected portions of the characterization report).

Page 8 of 18
 AEC - Q003 Rev-A
February 18, 2013
Automotive Electronics Council
Component Technical Committee

APPENDIX 1: CHARACTERIZATION CHECKLIST

A1.1 The following points should be evaluated during the planning stage for product characterization:

• Have all the cell structures used in this product been characterized?

• Has the wafer fab process changed since the cell structure characterizations?

• If the wafer fab process has changed, do the simulation models that will be used for
characterization account for all wafer fab process changes?

• If matrix devices will be needed, which process windows should be involved in the matrix?
What is the worst case variation in these process windows? (Note: The worst case
processing variation should be based on the worst case conditions observed during the last
six months or expected during normal manufacturing.) Who should participate in making
these decisions?

• Are there matrix cell interactions that should be considered?

• How can simulation models be used to simplify and expedite the characterization process?
How good are the available simulation models? What is the confidence in the simulation
models? What are the risks if the available simulation models are used?

• Have the drift characteristics of the cell structures been characterized? Is a device parameter
drift analysis needed as part of this characterization?

• Are there stresses associated with the product package that could affect the initial and late
life electrical parameters of the product?

• If matrix units are required, how many (sample size) from each matrix cell? How many
process variations need to be characterized? Will the product be characterized beyond the
required part specification requirements (hotter or colder temperatures, higher or lower
frequencies, higher or lower bias voltages)? Will the software required for the
characterization be available when needed?

• Do we know the junction temperatures (hot and cold) that must be evaluated in this
characterization? Have the thermal characteristics of the measurement system(s) been
considered, see Appendix 5?

• What form will matrix devices be in for the characterization – packaged? wafer level? Have
packaging requirements been included in the characterization plan?

• Will the ESD and Latch-up capability be evaluated early in the development cycle?

Page 9 of 18
 AEC - Q003 Rev-A
February 18, 2013
Automotive Electronics Council
Component Technical Committee

APPENDIX 2: MATRIX LOT CHARACTERIZATIONS

A2. For a new device, in particular in a new technology/process, a cross-factored experiment called a
Matrix Lot should be prescribed to estimate the effect of long-term process drifting. The Matrix Lot
design should be chosen to maximize the estimation of all desired factors and their interactions. The
center value run of the Matrix Lot is called the nominal split. The other cells are called corner cells or
splits.

Characterization data measured on parts taken from different cells of the matrix lot will have different
values. Each part parameter, then, will have a performance range, the result of parts being tested
from different cells of the matrix in conjunction with matrix cells are displayed in Figure 2.1.

Figure A2.1: Typical Matrix Cells (top) and Parameter Performance Range (bottom)

A2.1 Various Methods for Matrix Lot/Process Characterization

The performance of the Matrix lot could be evaluated using Cpk but with expected values that are
different for corner cells from nominal. The Cpk assessment criteria limits for process corner cells
should be based on the range of applied process shift and therefore a de-rating factor shall be
applied to reflect the long term probability of the process running at this condition.

A2.2 Characterization Statistical Model

To understand the performance characteristics of the Matrix Lot, a statistical model is built. The
location parameter and variances associated with that model will be estimated. These estimates will
be used to assess the process capability over time. This assessment is often in the format of a Cpk
calculation and the model is built using Analysis of Variance techniques (ANOVA). Several examples
of statistical models for analyzing the data are provided here.

Page 10 of 18
 AEC - Q003 Rev-A
February 18, 2013
Automotive Electronics Council
Component Technical Committee

A2.2.1 Cell Means Model

Could be used when corner cell states are assumed to be stable states into which the process may
drift over time and will stay there unless adjusted back. Cell means model is a one-way ANOVA and
it assumes homogeneous variance across all cells. CPK and a 95% confidence interval on CPK
could be estimated from the cell means model. Here focus is on the worst performing cell, based on
its CPK. It uses the worst corner cell performance as an indicator. If it stays within specifications,
then the process overall will. In the case the variance is not homogeneous across cells; the data from
each cell should be analyzed separately. The CPK from each cell should be evaluated against
chosen CPK criteria.

A2.2.2 Mixed Model

Could be used when corner cell states are interpreted as unstable states among which the process
will drift into over the long term. Mixed model ANOVA uses more than one variance component.
There is the variance within the cells and the variance from cell to cell. Each of these two is modeled
as a random effect within the mixed model. If that is your company’s interpretation, it might be good
to get your professional statistician involved when building the models.

A2.2.3 Random Effects Linear Model

Uses ANOVA to estimate the variation within the lots, and between the lots, and the total variance is
the sum of these two. This has general application for analysis of data when multiple lots and/or
matrix lots are included.

A2.2.4 Other Possible Models

There are other possible statistical models that may be more appropriate than the ones suggested
here. The statisticians in your company may have ideas much more mathematically involved than
those suggested in this document.

Page 11 of 18
 AEC - Q003 Rev-A
February 18, 2013
Automotive Electronics Council
Component Technical Committee

APPENDIX 3: DESIGN INDEX

A3. The idea of representing the goodness or badness of a product's Device Parameter performance
(over the worst case expected variation in Key Fab Process limits) with a single value is the reason
for establishing a Design Index (DI). A Design Index of 1 or greater for a particular parameter
requirement means that the device meets that parameter requirement at all matrix points. (Note: A
Design Index of less than 1 indicates that a particular parameter will probably cause a higher yield
loss during normal production.) This section will discuss the theory behind the DI and how to use it.

A3.1 In a matrix characterization (or simulation), processing variables are forced to certain values (to form
matrix cells) and the product performance is evaluated. The goal of the characterization is to
determine if the device performance will stay within specification limits when processing variables are
forced to their worst case values, see Appendix 1.

The Device Parameters are the values measured (or modeled), or tests performed, to ensure that the
device meets all of the electrical requirements defined in the Part Specification. In general, the
Device Parameters measured on parts taken from different cells of the matrix will have different
values. Each part parameter, then, will have a performance range, the result of parts being tested
from different cells of the matrix.

A3.2 In defining the DI it is assumed that the Device Parameter population that is sampled in each matrix
cell is normally distributed and independently distributed. The index also assumes that the product
would yield equally well when processed at any point within the Design/Process window. Thus, there
is no weighting of the matrix data.

The actual definition of DI is very similar to that of the Cpk index. The important difference is that the
Cpk is based on a single normal distribution (grouping all matrix cells into one distribution) while the
DI is based on the worst case of several individual matrix distributions, see Figure A3.1.

See Figure A3.2 for definition of WL, WU

Figure A3.1: Comparison of Cpk and Design Index

Page 12 of 18
 AEC - Q003 Rev-A
February 18, 2013
Automotive Electronics Council
Component Technical Committee

A3.3 The DI (one number for each Device Parameter upper and lower specification limit) is determined as
follows:

a. Statistically analyze the electrical distribution characteristics for each matrix cell, device
parameter, and test temperature. (The statistical analysis should include an estimation of the
mean and the worst case edges for each distribution, and should be based on the confidence
interval (ci) on the estimated mean, see Figure A3.2.) For example, if there are 5 matrix cells
(as shown in Figure A2.1) and three test temperatures, there will be 15 distributions (5 at
each temperature) for each Device Parameter, see Figure A3.3.

b. Using the Extreme WL (A) and Extreme WU (B), as shown in Figure A3.3, calculate the mean
[M = (A + B) / 2].

c. Then, determine the DI values (DI lower and DI upper) for each parameter) as shown in
Figure A3.3.

Figure A3.2: Statistical Analysis, Including the Confidence Interval

Page 13 of 18
 AEC - Q003 Rev-A
February 18, 2013
Automotive Electronics Council
Component Technical Committee

Figure A3.3: Analysis of Five Matrix Cells at Three Temperatures

Page 14 of 18
 AEC - Q003 Rev-A
February 18, 2013
Automotive Electronics Council
Component Technical Committee

APPENDIX 4: TEST DEVICE AND SAMPLE SELECTION

A4. Test Device Selection

One goal of characterization is to discover potential problems during early development when
changes can be made quickly, easily, and less expensively. Therefore, caution should be exercised
so that pretesting before characterization is minimized, since it has the possibility of truncating the
normal matrix cell distribution and hiding potential problems.

A4.1 One approach is to pre-test to only eliminate non-functional devices from the characterization sample.
Example: to help define non-functional, the following would be considered test limits for functional
devices:
• Override specification required parametric failures and only perform relaxed spec
functional tests.
• Widen all specification required parametric limits by a significant digit.

A4.2 A second method for sample population selection is the following. The data should be pre-screened
to remove statistical outliers before processing. One suitable method is using box-plot Inter Quartile
Range (IQR) to generate outlier ‘fences’. The inter quartile range can be calculated by ordering the
data in ascending order, and taking the delta between the 25th percentile point (called the first
quartile Q1) and the 75th percentile point (called the third quartile Q3). The IQR is the delta between
Q1 and Q3. The outlier fences are then set to:

Lower fence = Q1 – 3* IQR

Upper fence = Q3 + 3* IQR

Any data-points that fall outside these fences can be removed for calculating the mean and sigma of
the population.

A4.3 The method of selecting device can have a large effect on the amount of inherent process variation
that is included in the data. The main concern is to understand and design for the largest sources of
inherent process variation. Due to the number of different process technologies it is the responsibility
of the supplier to develop a combination of lots, wafers, and device locations that will provide the
inherent process variation.

A thorough analysis of sample size should also consider the following:

• Which wafers are selected for a cell?
• Which device locations are selected on a wafer?

A4.4 Sample Size Selection and Margin Error

During characterization, random samples are chosen. The measured results deviate in the mean and
limit ranges from the population Normal Distribution depending on the parts sampled. Larger sample
sizes will help to reduce such deviation/error. Two important factors are to be considered when
deciding on sample sizes for characterization.

A4.4.1 The confidence interval which is also called margin of error. When a measurement is taken on a
parameter on the samples, one could use a confidence interval to state that the measurement from
the entire relevant population will be between the mean minus the interval and the mean plus the
interval.

A4.4.2 The confidence level tells how sure one could be. It represents how often the true measurement
lies within the confidence interval. A 95% confidence level, a value used as customary, means one
could be 95% certain.
Page 15 of 18
 AEC - Q003 Rev-A
February 18, 2013
Automotive Electronics Council
Component Technical Committee

A4.4.3 If the data conforms to the normal distribution, the two-tailed confidence interval may be calculated
using the following equation:

Lower confidence bound = x - 1.96 s / √n

Upper confidence bound = x + 1.96 s / √n
Where: x = mean value of sample
s = standard deviation
n = (random) sample size
1.96 corresponding to 95% confidence level of Normal Distribution

The magnitude of error in the mean of the sampled distribution is given by 2*(1.96 s / √n). For a Cpk
value of 1.33 for the limit ranges, the expected percent of errors in the limit ranges are plotted with
respect to sample sizes in Figure A4.1.

Figure A4.1: Error in limit range vs. sample sizes *

(* This graph has been generated using the students-t distribution, which is required for statistical
accuracy when the sample size is small. The t distribution approaches the normal distribution
approximation as the sample size increases.)

A4.4.4 Note that with a sample size of 30, as shown in Figure A4.1 above, the percent of error in limit range
is 5.66%, which drops to 2.24% for a sample size of 100 and levels off to less than 1% with a sample
size of 300-400 units.

Page 16 of 18
 AEC - Q003 Rev-A
February 18, 2013
Automotive Electronics Council
Component Technical Committee

APPENDIX 5: CHARACTERIZATION TESTING

A5. The device selected for characterization shall be tri-temperature tested. Test temperatures need to
be established for room, hot, and cold test. The test setup temperature extremes should be able to
duplicate worst case product application junction temperatures.

It should be stressed that the junction temperature of the device in the selected operating condition is
the actual characterization target. For example, a low test temperature limit of -55ºC might be
required for a -40ºC instantaneous customer need (i.e. operate on cold wake-up) to account for a
15ºC steady state heating during testing. Likewise, a high test temperature limit of 135ºC might be
required for a 150ºC instantaneous package test to account for a 15ºC steady state heating during
testing.

The bias conditions and subsequent power consumption will determine if there is sufficient energy
dissipation to necessitate the need for test temperature correction(s). In some situations an
instantaneous condition is assumed that would null the effects of thermal dissipation, while in other
conditions a high energy steady-state condition is assumed that would necessitate the use of
temperature correction(s).

Simulate a Customer Junction Temperature of 155ºC

If the starting temperature of a customer system is: ...................................................................... 125ºC

And a thermal gradient (w/bias condition) causes a
temperature increase of: ..................................................................................................... 30ºC
Then the customer junction temperature requirement
is: ...................................................................................................................................... 155ºC

Characterization Test Temperature

The junction temperature to simulate is: ........................................................................................ 155ºC

If the steady state heating during testing causes a
thermal gradient (w/bias condition) that
results in a temperature increase of: ................................................................................... 15ºC
Then the required final test setup
(Characterization test) temperature is: ........................................................................... 140ºC *

* Note: Additional temperature may be added to this value for guardbanding.

Page 17 of 18
 AEC - Q003 Rev-A
February 18, 2013
Automotive Electronics Council
Component Technical Committee

Revision History

Rev # Date of change Brief summary listing affected sections

- July 31, 1997 Initial Release.

A Feb. 18, 2013 Complete Revision. Revised Acknowledgement and Sections 2, 3.2, 3.4,
3.5, 3.10, 4, 4.1, 4.2, 4.3, and Appendix 1 through 5. Added Notice
Statement; new Sections 3.1, 3.3, 3.7 to 3.12, 4.2.1, 4.2.2, 4.4, 4.4.1,
4.4.1.1, 4.4.1.2, 4.4.2, 4.4.3, 4.5, 4.5.4, 4.5.2, and 5; new Figures 1, 2, 3,
4, 5, and A4.1; and new Tables 1 and 2. Deleted signature block.

Page 18 of 18