Sunteți pe pagina 1din 20

Fault Coverage Analysis of RAM

Test Algorithms

Marc Riedel
McGill University, Montreal, Canada

Janusz Rajski
Mentor Graphics, Wilsonville, Oregon

Outline

Motivation
Fault Models
Methodology and Complexity
Fault Simulation Results
Conclusions

RAMFLT

Motivation
Functional Memory Testing

A multitude of fault models and test schemes proposed.


Quality of fault coverage difficult to assess.

Coverage Measures Needed

To evaluate and rank existing test algorithms:

Deterministic/regular tests.
Pseudo-random/irregular tests.

To validate new test schemes for:

Embedded memories and BIST designs.


Specialized memory architectures (e.g., multiport, FIFO).
RAMFLT

Functional Cell Array Model


Bit-addressable 2-D array of binary storage elements:

m1,1 m1,2
m 2,1 m 2,2

..
.

..
.

mr,1 m r,2

...
...
..

...

m1,c
m 2,c

..
.

m r,c

ij mi , j {0,1, 0, 1, x}
0, 1:correct

0, 1: faulty
x:unknown

Operations: read, write-0, write-1.


RAMFLT

Functional Fault Behavior

Sensitized/desensitized by write operations.


Detected by read operations.
write

Unsensitized

Sensitized

0/1

1/0

read

Detected

write
RAMFLT

Cell Array Fault Models

Single Cell

stuck-at, transition, stuck-open, data-retention

Coupling

idempotent, inversion, state, dynamic


(2-cell and 3-cell versions)

Bridging

AND-type, OR-type
(2-cell and 3-cell versions)

Neighborhood
Pattern Sensitive

active, passive, static


(type I and type II neighborhoods)

RAMFLT

Fault Model Specification


Fault models are specified as inputs, not hard-coded.

Example

Format
sensitization

< write op. > < mem. pattern >


< write op. > < mem. pattern >

..
.

desensitization

..
.

< write op. > < mem. pattern >


< write op. > < mem. pattern >

..
.

..
.

0 1

0 1 sensitized fault
mem. pattern
write op.
RAMFLT

Ex.: 2-cell OR-type Bridging Fault

sensitization

desensitization

Operation

write-1, a
write-1, b
write-0, a
write-0, b

0
0
1
1

0
0
1
1

write-0, a
write-0, b
write-1, a
write-1, b

1
0
0
1

0
1
1
0

read to either cell


returns OR(a,b)

RAMFLT

Coverage Analysis
Simulation performed for arbitrary test sequences.
write-1, < add. >
read, < add. >
write-0, < add. >

case: write
Determine which faults are sensitized
or desensitized.

write-1, < add. >


read, < add. >

..
.

case: read
Classify all sensitized faults as
covered.

RAMFLT

Sensitization & Desensitization


A write operation can sensitize/desensitize several faults.
Example

faults in cells y1 , y2, y3, y4


sensitized by write operation

active NPSF
p1
p4

p2

p3
3

transition in nbh. pattern


p1, p2, p3, p4 sensitizes fault

y1
y4

y2
y3

RAMFLT

Delayed State Transitions


tD
0 1
tD
1 0

Sensitization/desensitization
occur after a time delay t D .

Used to model retention faults, e.g., "sleeping-sickness"


failures in DRAMS:
1

tD

RAMFLT

Multiple Faults

Error masking

CFid(x and
z
CFid(y and
z

Multiple
sensitizations

CFid(x and
y
CFid(y and
z

0 1

1 0

0 1

z
RAMFLT

Multiple Faults (cont.)

Sensitized faults change the memory pattern.


This affects the sequence of sensitization/desensitization of other faults.

Example
no faults
sensitized
1

fault A
sensitized

0A 0 C 1

0B 1

1
A

fault B
sensitized

0A 0 C 1

0B 1

faults A and B
sensitized

1
A

0
1

1
C
B

1
1

the pattern surrounding cell C is all 1s

a sleeping-sickness fault is sensitized


RAMFLT

Complexity
O(t)

with respect to the test sequence length t

NPSFs

O(2 )
k

with respect to the neighborhood size k

O(2 n )

k-cell coupling faults


with respect to the memory size n & number of coupled cells k

NPSFs: cells in physical proximity.


Coupling faults: cells located anywhere in memory array.
RAMFLT

Examples of Test Algorithms


March X

{(w ); (r, w ); (r,w ); (r)}

March C-

{(w ); (r, w ); (r,w ); (r,w ); (r, w ); (r )}

GALPAT

1
RAMFLT

Simulation Results for March X


Fault Class

FC (%)

global 2-cell
CFid
SCF
ABF

FC (%)

Type I NPSF
50.0
75.0
50.0

local 3-cell
CFid
SCF
ABF

Fault Class

Active
Passive
Static

6.25
6.25
15.6

Type II NPSF
25.0
100
50.0

256-bit memory (16 rows x 16 columns)

Active
Passive
Static

0.39
0.39
1.76
RAMFLT

Simulation Results for March CFault Class

FC (%)

global 2-cell
CFid
SCF
ABF

FC (%)

Type I NPSF
100
100
100

local 3-cell
CFid
SCF
ABF

Fault Class

Active
Passive
Static

12.5
12.5
31.2

Type II NPSF
50.0
100
100

256-bit memory (16 rows x 16 columns)

Active
Passive
Static

0.78
0.78
3.52
RAMFLT

Simulation Results for GALPAT


Fault Class

FC (%)

global 2-cell
CFid
SCF
ABF

FC (%)

Type I NPSF
99.7
100
100

local 3-cell
CFid
SCF
ABF

Fault Class

Active
Passive
Static

11.7
15.6
40.6

Type II NPSF
48.2
79.9
100

256-bit memory (16 rows x 16 columns)

Active
Passive
Static

0.81
0.98
4.10
RAMFLT

Trace of Simulation for ANPSF Test


SAF

TF

2-cell
CFid

(local)

3-cell
CFid
(local)

Type I
ANPSF

Type II
ANPSF
100
80
60
40

%FC

20

1000

262796

524028

785350

1046590

0
1307010

Test Sequence Entry


RAMFLT

Conclusions

General RAM fault simulation methodology.

Library of over 25 functional fault models.

Coverage statistics for over 40 test algorithms.

Application:
Evaluation of arithmetic BIST schemes for
memories.
RAMFLT

S-ar putea să vă placă și