1、EIA/JEDECPUBLICATIONSignature AnalysisJEP136JULY 1999ELECTRONIC INDUSTRIES ALLIANCEJEDEC Solid State Technology AssociationNOTICEEIA/JEDEC standards and publications contain material that has been prepared, reviewed, andapproved through the JEDEC Board of Directors level and subsequently reviewed an
2、d approvedby the EIA General Counsel.EIA/JEDEC standards and publications are designed to serve the public interest througheliminating misunderstandings between manufacturers and purchasers, facilitatinginterchangeability and improvement of products, and assisting the purchaser in selecting andobtai
3、ning with minimum delay the proper product for use by those other than JEDEC members,whether the standard is to be used either domestically or internationally.EIA/JEDEC standards and publications are adopted without regard to whether or not theiradoption may involve patents or articles, materials, o
4、r processes. By such action JEDEC does notassume any liability to any patent owner, nor does it assume any obligation whatever to partiesadopting the EIA/JEDEC standards or publications.The information included in EIA/JEDEC standards and publications represents a sound approachto product specificati
5、on and application, principally from the solid state device manufacturerviewpoint. Within the JEDEC organization there are procedures whereby an EIA/JEDECstandard or publication may be further processed and ultimately become an ANSI/EIA standard.No claims to be in conformance with this standard may
6、be made unless all requirements stated inthe standard are met.Inquiries, comments, and suggestions relative to the content of this EIA/JEDEC standard orpublication should be addressed to JEDEC Solid State Technology Association, 2500 WilsonBoulevard, Arlington, VA 22201-3834, (703)907-7560/7559 or w
7、ww.jedec.orgPublished by ELECTRONIC INDUSTRIES ALLIANCE 1999Engineering Department2500 Wilson BoulevardArlington, VA 22201-3834This document may be downloaded free of charge, however EIA retains thecopyright on this material. By downloading this file the individual agrees not tocharge or resell the
8、resulting material.PRICE: Please refer to the currentCatalog of JEDEC Engineering Standards and Publications or call Global EngineeringDocuments, USA and Canada (1-800-854-7179), International (303-397-7956)Printed in the U.S.A.All rights reservedPLEASE!DON”T VIOLATETHELAW!This document is copyright
9、ed by the Electronic Industries Alliance and may not bereproduced without permission.Organizations may obtain permission to reproduce a limited number of copiesthrough entering into a license agreement. For information, contact:JEDEC Solid State Technology Association2500 Wilson BoulevardArlington,
10、Virginia 22201-3834or call (703) 907-7559JEDEC Publication No. 136- i-SIGNATURE ANALYSISCONTENTSPage1 Introduction 12 Purpose 23 Scope 24 Definitions 35 Signature analysis procedure for an ongoing process 35.1 Overview 35.2 SA process flow 45.3 Signature analysis definition phase 45.4 SA validation
11、55.5 Revalidation 56 Signature analysis procedure for a finite population 57 Examples 67.1 Finite population case 67.2 Ongoing process 6Annex A An approach of signature analysis risk assessment proceduresfor an ongoing process 7A.1 Statistical terms and variables 7A.2 Statistical derivation of an es
12、timate for 8A.3 Statistical derivation of confidence intervals for A 9Annex B An approach to signature analysis risk assessment proceduresfor a finite population 11Annex C An approach to signature analysis risk assessment proceduresfor a process with correlated failures 13C.1 Statistical terms and d
13、efinitions 13JEDEC Publication No. 136-ii-JEDEC Publication No. 136Page 1SIGNATURE ANALYSIS(From JEDEC Board Ballot JCB-98-124, formulated under the cognizance of the 14.6Committee on Failure Analysis.)1 IntroductionDevice Analysis is a limited resource that should be targeted at solving unknown pro
14、blems,rather than analyzing repetitive failures. Signature Analysis (SA) is a lab resource managementtool in that it provides a formal method of reducing the number of these repetitive failuresrequiring full analysis. Two statistical models are presented, both of which assign a confidencelevel to an
15、y number of units with the same failure characteristics (i.e., signature) and same failuremechanism, allowing the statement “I am A% sure that greater than B% of other parts with thissignature will also have this failure mechanism”. One model is for use when the failingpopulation size is finite and
16、known, and will be referred to as finite population analysis (FPA).It would be used, for example, to define the sampling plan for a low yielding wafer lot, i.e., todetermine how many failing dice from each failure category must be fully analyzed in order toinfer information about that whole wafer lo
17、t. The other model, referred to as ongoing processanalysis (OPA) assumes the population of future fails is unknown or infinite, and would be usedto infer information about failures collected over time from multiple events, lots or labs. The useof two models also allows us to build-in the engineering
18、 bias that we would expect higherconfidence levels when we do failure analysis by inference on finite population failures, than onones collected over time.The backlog of analysis labs often includes device failures for which both the product engineerand analyst know what will be found during analysi
19、s. However, corporate specifications and/orcustomer requirements often dictate that the analysis be performed anyway. There is a need to beable to assign a failure mechanism (with a statistically determined confidence level) to devicesthat are analyzed by inference to historical data.There is a desi
20、re to promote a common definition of this system of analysis by inference, usingthe same statistical techniques, and to recognize that it is a formal means of doing failureanalysis.JEDEC Publication No. 136Page 22 PurposeThe purposes of these guidelines are the following:(a) To define a process of a
21、nalyzing failures by inference to previous traditional analyses ofdevices with the same failure characteristics, using a statistical and quantitative approach.(b) To provide statistical models for assigning confidence levels to analysis by inference.(c) To help gain acceptance of analysis by inferen
22、ce techniques through consensus ofterminology and methodology.(d) To provide examples of its use.3 ScopeThis method of analysis by inference may be applied to customer returns, failures from qualityconformance testing, reliability failures, qualification failures, and devices from engineeringexperim
23、ents and yield issues.Analysis by inference may be used either when the failing population is known and finite, (todefine a yield analysis sampling plan of a wafer lot, for example), or when the failing populationis unknown and assumed to be infinite.Using SA does not necessarily imply that the root
24、 cause is known or understood, nor does itnecessarily imply that a corrective action will or should take place, but SA can be used inconjunction with other programs that address these needs.It is not the intention of this document to do the following:(a) Define acceptable confidence levels.(b) Defin
25、e an SA database, reporting or tracking system.(c) Provide SA implementation details.Individual companies may create the policies and procedures to formalize the SA process in linewith these guidelines.JEDEC Publication No. 136Page 34 DefinitionsSignature: The necessary and sufficient information ab
26、out a failure that establishes a strongrelationship between failure characteristics and failure mechanism. This necessary and sufficientinformation could include emission microscopy results, morphology data, test data, IV-curves,environmental history, etc. and therefore could be either electrical or
27、 physical in nature. Thescope of application may be time-based, lot-based, package-based, design-based, etc.Signature Analysis (SA): A method to reduce the number of comprehensive failure analyses byapplication of statistical inference techniques.Ongoing Process Analysis (OPA): The application of Si
28、gnature Analysis to an unknown(assumed to be infinite) population of failures collected over time from multiple lots, events, orlabs.Finite Population Analysis (FPA): A special case of Signature Analysis where the signatureoccurs in a particular finite population of devices.5 Signature analysis proc
29、edure for an ongoing process5.1 OverviewThe Signature Analysis process consists of the following phases:Signature Definition - Determining which characteristics of repetitive failures define thesignature.Validation - A sufficient number of devices should be analyzed using traditional physicalanalysi
30、s methods in order to say “we are A% confident that at least B% (theproportion) of all devices with the same signature will reveal the same failuremechanism”. The annexes describe statistical techniques that can be used todetermine confidence levels.Application - Devices may be assigned a failure me
31、chanism without physical analysis, butby inference to previously analyzed devices with the same signature.Revalidation - Periodic reverification of the relationship between the signature andmechanism using traditional physical analysis methods.JEDEC Publication No. 136Page 45 Signature analysis proc
32、edure for an ongoing process ( contd)5.2 SA process flowREPETITIVE FAILURE OBSERVATIONSA DEFINITIONSA VALIDATIONAPPLICATION OF SA TO ASSIGN FAILURE MECHANISMSYESREVALIDATIONOK?NOACTION REQUIRED BEFORE SA APPLICATION CANCONTINUE5.3 Signature analysis definition phase(a) Repetitive failures will be us
33、ed to initiate a Signature Analysis definition. The failuresshould show electrical or physical evidence supporting the claim that they have the samefailure signature and mechanism.(b) The repetitive failures can represent FPA or OPA signatures.(c) Ongoing process SA samples should be from randomly s
34、elected lots.(d) Finite population SA samples imply that the scope of the resulting SA report is only forthis population. In other words, only additional parts from this same population may beanalyzed by inference.(e) Historical data should be studied to identify trends that could help define the sc
35、ope of thefailure.JEDEC Publication No. 136Page 55 Signature analysis procedure for an ongoing process ( contd)5.4 SA validationAfter signature definition, a sufficient number of devices must be completely analyzed byphysical analysis methods with the same failure signature and mechanism to satisfy
36、desiredconfidence level requirements.5.5 RevalidationThis procedure is to guard against the possibility that some hidden variables might have changed,causing the reported relationship between the failure signature and the failure mechanism tobecome invalid.A revalidation process consists of performi
37、ng a traditional physical analysis of at least one devicewith the signature in question to confirm that the inferred failure mechanism remains valid.Revalidation may be performed based on frequency of units with a given signature or based ontime.If the failure mechanism of the revalidation device do
38、es not match the inferred mechanism, thenrevalidation fails and action should be taken. Suggested actions include(a) Reevaluation of the original scope of application of the signature.(b) Traditional physical analysis of other devices analyzed by inference to this signature.6 Signature analysis proc
39、edure for a finite populationThe SA Procedure for FPA would consist of the Signature Definition, Validation and Applicationsteps described in the previous section, with a different statistical model used to determineconfidence levels and number of required analyses. The Revalidation step would not a
40、pply.JEDEC Publication No. 136Page 67 Examples7.1 Finite population caseTen units in lot 123 have all failed the 168 hour life test readpoint. The failure mode wasreported as functional failures, and further work in the failure analysis lab on one unit showed aphoton emission site at the input prote
41、ction circuitry of pin 1. Curve tracing the ten devices atdifferent biasing conditions shows they all have the same leakage characteristics for pin 1. It ishypothesized that all ten failures have the same signature, defined as the failure mode plus theother observed characteristics. Traditional phys
42、ical analysis techniques (in this case, layeretching and SEM) are performed on three of the units and all have the same mechanism, polyfilamentation at the same location in the pin 1 input protection circuitry, indicating ESD damage.According to Annex B it can be conclude that “we are 90% confident
43、that 70% of the remaining7 units will also have this same mechanism”. Therefore, the decision as to whether or not toanalyze more units depends upon our prescribed level of risk.7.2 Ongoing processEngineering has noticed that over the last three months they have analyzed 13 units of deviceABC from v
44、arious lots. In each case the device failed after 500 temperature cycles, and thereported failure mode, observed characteristics and failure mechanism are the same. The lab thenreceives two more units of device ABC, also temperature cycle rejects, with the same signature.Before performing traditiona
45、l deprocessing analysis on these two latest units the lab can conclude“we are 90% confident that 85% of other units with this signature will also have thismechanism”. (See Annexes A and C).JEDEC Publication No. 136Page 7Annex A An approach of signature analysis risk assessment procedures for an ongo
46、ingprocessThe following section defines and documents the procedure for determining the risk of assigninga failure mechanism to a series of analyses of devices of one particular failure signature. Thefailing population size is unknown and assumed to be infinite.A.1 Statistical terms and variables“n”
47、 denotes the number of devices analyzed with a particular signature; or the number proposedhypothetically in an inference calculation.“x” denotes the number of analyses (among the n) resulting in the primary failure mechanism:also denoted x A .“ ” (a number, 0 1) is the true, long term proportion of
48、 devices with the signature in questionthat would result in the primary failure mechanism, if analyzed: also denoted A .For a given , x is statistically distributed as a Binomial distribution with parameters n and .This is denoted( ) ( ) ( )f x nx B nx n x = 1 ,The statement that is distributed as a
49、 Beta distribution with parameters and is denoted( ) ( )( ) ( ) ( ) ( )g Be = + 1 11 ,Note: Notice that the Binomial and Beta families of distributions have the same form:K 1 K 2 (1- ) K 3 , for different “K” values in the two distribution families. To keep from confusingthem, remember . . .(1) The Beta is a distribution for and ranges from 0 to 1 continuously. It integrates to 1:( ) 110 = qq dg .(2) The Binomial is a distribution for x (the power of ) and ranges over the integers from 0to n discretely.It sums to
