IEEE P2427:Proposing the Essential Framework for Measuring Defect Coverage in

Analog Circuits

… not officially speaking for the IEEE P2427 working group

Jeff RearickSenior Fellow, AMD

Full Disclosure / Disclaimer

� This presentation reflects work done by an IEEE working

group, but it is not an official statement from that group.

The opinions expressed are merely those of this author

(however reasonable they may seem ☺).

2

Outline

� Background, Mission, Guiding Principles

� Six important decisions

� Framework for analog defect coverage analysis

� Conclusions

3

Outline

� Background, Mission, Guiding Principles

� Six important decisions

� Framework for analog defect coverage analysis

� Conclusions

4

Background: analog defect coverage

� Digital tests are graded using fault coverage metrics

� Digital fault coverage measurement is

well-defined (EDA tools, IEEE 1804)

(…yet!)

The IEEE P2427 Working Group is now underway to tackle this challenge and define the standard

� Analog tests are, well, complicated

� There have been many approaches used

in the last 3 decades; nothing standard

5

Mission: IEEE P2427

� Define the following processes:

• Given a circuit, produce the universe of defects• Given a test for that circuit, determine which

defects are detected

� Standardize the following information:

• Definition and communication of defect models• Definition and communication of detection criteria• Reporting of defect coverage

6

Guiding principles

� Keep it simple

� Keep it real

7

Outline

� Background, Mission, Guiding Principles

� Six important decisions

� Framework for analog defect coverage analysis

� Conclusions

8

Decisions that P2427 needs to make

Either this: … or this?

Defect-oriented Spec-based

Behavioral Circuit-level

Faults Defects

Single Multiple

Manufacturing Reliability

Catastrophic Variation

9

123456

Decision tree

Defect-oriented or Spec-based?

�Defect-oriented: requires a defect model applied

to a circuit netlist, then simulation to determine

detection; coverage is ratio of detected / total

�Spec-based: requires documentation and

simulation of performance-based measurements

for complex specifications; what is “coverage”?

1

Keep it simple 10

Behavioral or Circuit-level?2

�Behavioral: uses a high-level model which

mathematically/logically mimics circuit-level

behavior

�Circuit-level: requires a schematic or layout-

extracted representation of the circuit which

matches the physical design

Keep it real 11

Faults or Defects?3

�Faults: abstractions of defects into unwanted

significant behavioral changes from the

intended function of the circuitry

�Defects: unintended significant changes in a

Design-intent primitive circuit element

{R,L,C,M,Q,D} or connections between them

Keep it real 12

Single or Multiple?4

�Single: the “single fault assumption” has been

fundamental to digital fault coverage

measurement for decades

�Multiple: this would add complexity and impair

scalability, but does not seem to add much value

to the quantification of coverage: detecting single

defect (often) covers multiple defects

Keep it simple 13

http://central.oak.go.kr/journallist/journaldetail.do?article_seq=11532&tabname=abst&resource_seq=-1&keywords=null

Manufacturing or Reliability?5

�Manufacturing: “time-zero” flaws determine

yield (follow the money), and a test that can

detect time-zero defects can be re-used later

in the life cycle

�Reliability: latent defects are far less

common, far harder to predict, and (almost by

definition) impossible to detect at time zeroKeep it simple 14

reliabilitysolutions.co.uk

Catastrophic or Variation?6

� This is a tough question! ??

15

� Keep it simple

� Keep it realProcess variations

� Process variations are not simple, but they are real

• Hundreds of process variables• Some are correlated, some are not• Many circuit elements can vary at once• Analog designers spend lots of time in Monte Carlo• We can’t not consider some aspect of variations…

Catastrophic or Variation? Both!6

�Catastrophic: hard defects are simple to

enumerate, and they are representative of a

number of realistic manufacturing flaws

�Variation: the trick here will be to specify a

scalable, enumerable, yet interesting set of

variation defects which will indicate the

quality of a test (our fundamental mission)

Keep it real, find a way to keep it simple 16

Decisions that P2427 has made

Either this: … or this?

Defect-oriented Spec-based

Behavioral Circuit-level

Faults Defects

Single Multiple

Manufacturing Reliability

Catastrophic Variation

17

123456

[Provisionally, of course]

Outline

� Background, Mission, Guiding Principles

� Six important decisions

� Framework for analog defect coverage analysis

� Conclusions

18

Framework for analog defect coverage

19

1. Defect Models and Universe

3. Defect Coverage Reporting

2. Detection Criteria1. Circuit-level enumeration

of catastrophic and variation-based time-zero defects

2. Circuit-level simulation comparing responses of defect-free and defective circuits using measurable parameters

3. Summary of total, detected, and untestable defects

Proposed analog defect models

� Catastrophic (hard defects: change netlist topology)

20Photos: https://www.semiwiki.com/forum/content/3677-catching-ic-manufacturing-defects-slack-based-transition-delay-testing.html

�Extra interconnect material (i.e. shorts)�Unintended connection between …

�… terminals of an element (schematic) �… adjacent nodes (layout)

� Variation (soft defects: change element/model values)

• This is still under discussion in the Working Group (next slides)

�Missing interconnect material (i.e. opens)�Unintended disconnection of …

�… an element terminal from node (schematic)�… fanout stem and/or branch (layout)

Visualizing variation defects: simplistic R

21

Acceptable

process

variationDEFECTDEFECT

Principal

parameter

value (R)

Nominal

design-

intent

value (R0)

like

lih

oo

d

R0

Catastrophic

open defect

value

Ropen

R0

Catastrophic

short defect

value

RshortR0

High

variation

defect

value

r

R0+r

Low

variation

defect

value

r

R0-r

Implicit failImplicit fail

Variation defect models: options

1. Out-of-bounds variation of a principal parameter (R, L, C, W/L,

Ion, Ron, Vt, B, …) of a single circuit element {R,L,C,Q,M,D}

• Is this a realistic defect model?

2. Out-of-bounds variations of a principal parameter of all

circuit elements (i.e. a global parameter shift, not multiple defects)

• How would we handle correlation between parameters?

3. In- or Out-of-bounds variations of multiple different principal

parameters of many different circuit elements

• Doesn’t scale: can’t define a finite defect universe

• Others? P2427 WG work in progress22

Illustrating the P2427 framework (741 op-amp)

23

Untestable shorts 3

Testable defects 141 (?) [1]

Notes:[1] Proof of untestability is necessary for determining when test generation

is “done”: in digital ATPG terms: “fault coverage” vs. “test coverage”

Element Count Defects

Bipolar transistors 20 120

Resistors 11 22

Capacitors 1 2

Total defects 144

Nodes 24

TBD [2]

Inputs 2

Outputs 1

InOuts 2

Supplies 2

[2] Node-related defects have some open questions:a) Inject too many shorts [(n)(n-1)/2] in schematic?b) Inject too few opens [one per terminal per node] in schematic?c) What should we do with Primary Inputs / Primary Outputs?

Analog defect coverage flow

24

Defect list

Derive list of n defects

Responses different?

circuitnetlist*

Simulate defect-free circuit (once)

Simulate defectivecircuit (n times)

Test stimulus

More in list?

* Netlist could be schematic or layout-extracted; in fact, both should be employed

Detectedlist

Y

Y

Done: report Detected/n

N

Detection criteria

Outline

� Background, Mission, Guiding Principles

� Six important decisions

� Framework for analog defect coverage analysis

� Conclusions

25

Conclusions

� IEEE P2427 will enable uniform assessment of analog

tests by quantifying the coverage of analog defects

� This is crucial for automotive quality

� This framework is a long-overdue step toward standardized

analog defect coverage measurement

� The P2427 Working Group has made substantial progress

toward our mission, but still has tough decisions to make:

� Defining and enumerating variation defects� Pre-identifying untestable defects� Documenting detection criteria, defect models & values

� Join us if you’ve got some good ideas!26

Framework for analog defect coverage

27Work in progress, but we’ve got a great crew!

1. Defect Models and Universe

3. Defect Coverage Reporting

2. Detection Criteria