This preview shows page 1. Sign up to view the full content.
Unformatted text preview: Outline – Automatic Test Pattern
Generation
• Test generation systems
• Test generation for combinational ckts
–
–
–
– DAlgorithm
PODEM
Boolean Satisfiability approach
Test compaction • Test generation for sequential ckts
–
–
–
– © K.T. Tim Cheng, 05_comb_tg, v1.0 1 © K.T. Tim Cheng, Timeframe expansion & Extended DAlgorithm
Ninevalued test generation
Potential detection
Issues of sequential ATPG 2 05_comb_tg, v1.0 The ATPG Problem Test Generation Systems • Problem definition: Given a logical fault model, and
a circuit, determine a small set of test vectors
that detect all faults in the circuit. Compacter • Problem complexity: Under the stuckat fault
model, the problem is NPcomplete even for
combinational circuits Circuit
Description
ATPG
With fault
simulator Test
Patterns – However, commercial test generators that
efficiently generate tests for >10Mgate ckts
are in use today. Fault
List
Aborted
Faults Redundant
Faults Undetected
Faults Backtrack
Distribution Ack: Bushnell and Agrawal, Essential of Electronic Testing, 2000
© K.T. Tim Cheng, 05_comb_tg, v1.0 3 Basics: Path Sensitization Method © K.T. Tim Cheng, A Simple Example
D Atomic operations:
1. Activation, fault excitation: Specify inputs so as to A
B
C generate the appropriate value at fault site for fault
excitation (I.e. set S to 1 for Sstuckat0 fault) G5
f1 G1 a X
s.a.1 G6
f2 E propagate the fault effect from the fault site to the
outputs/observation points (1) Fault activation : A = B = C = 1
(2) Have a choice of error propagating: through G5 or G6 3. Line justification: Specify input values so as to produce (a) Propagating through G5 requires G2 = 1 ⇒ A=D=0 ⇒ contradiction the signal values specified in (1) or (2) (b) Propagating through G6 requires G4 = 1 ⇒ C=1, E=0 4. Value implication: unique determination of values at ⇒ Test ABCDE=(111x0) other signals due to value assignments made in (1), (2), or (3)
05_comb_tg, v1.0 G2 G3 2. Error propagation: specify additional signal values to © K.T. Tim Cheng, 4 05_comb_tg, v1.0 5 © K.T. Tim Cheng, 05_comb_tg, v1.0 6 1 Line Justification B
H Completeness of ATPG Algorithms G3 A
F • E sa1 ⇒ E =0 C
D
E • A test generation algorithm is deemed complete iff
it will find a test for a fault if exists or prove that
there exists no test, given sufficient time. G4 G2 G1 – Complete algorithms can identify untestable faults • C = D = 1 to propagate through G1.
• To propagate G4, need G2 = G3 = 1
Attempt to line justify G2 = G3 = 1
– G3 = 1possible if A = F = 1 or B = H = 1 • Major complete algorithms for comb. ckts
– Dalgorithm (Roth, 1966)
– PODEM (Goel, 1981) • If A = F = 1 ⇒ inconsistency since C = 1 so G2 = 0. – FAN (Fujiwara, 1983) ⇒ Therefore, G3 = 1 ⇒ B = H = 1
– G2 = 1 need A = 0 or F = 0 – Socrates (Schulz, 1988)
– BooleanSATbased ATPG (Larrabee, 1992) ⇒ Tests are ABCDEH, BCDEFH
© K.T. Tim Cheng, 7 05_comb_tg, v1.0 Single Path Sensitization Is NOT
Complete
C
G1 G3 x d s.a.0 G6 G4 d f G2 G5 E C • d sa0 : ⇒ A = B = 1
• Propagate along G3, G6 requires C =1, G2 = G4 = G5 = 1.
– In order for G4 = 1 either E = 0 or G1 = 0 ⇒
inconsistency
⇒ E = 0, B = 1 ⇒ G5 = 0 ⇒ inconsistency
• Propagate along G4, G6 ⇒ E = 1 & G2 = G3 = G5 = 1
– G2 = 1 ⇒ C = 1, ABC = 111 ⇒ G3 = 0 inconsistency
⇒ No test © K.T. Tim Cheng, 05_comb_tg, v1.0 A
B G2 A
B 1
1 G1 1 9 E 1 0/1 G4
0 G5 1 © K.T. Tim Cheng, G6 G3
d
sa0 1 G6 G4
0 G5 0/1
1 10 05_comb_tg, v1.0 0/1 D 1 D D
D 05_comb_tg, v1.0 0/1 1/0 1/0
1/0 • D represents a signal which has value 1 in normal
(faultfree, good) ckt, and value 0 in faulty ckt.
(I.e. D ≡ 1/0); Similarly, D ≡ 0/1 • Two paths G3, G2 and G4, G6 are sensitized, i.e. error is
propagated along both paths. © K.T. Tim Cheng, 1 0 • Need to be able to deal with multiple “errors” at
the inputs to a gate.
1/0 1/0
1/0 G1 1 The DAlgebra 1 0
G3 d
sa0 1 E Multiple Path Sensitization
C 8 05_comb_tg, v1.0 But Boolean Difference Method
Finds Test ABCD=(1111)!! G2 A
B © K.T. Tim Cheng, D D D D
D D
D D 0 D
D D 1 D • Behaves like a Boolean variable.
11 © K.T. Tim Cheng, 05_comb_tg, v1.0 12 2 8
H
J9 An Example Primitive Dcubes
• Specifies the minimal input conditions which must
be applies to a logic element E in order to produce
an error signal at the output of E G1 sa0 :
1 3 4 0 X X 2 0
X X
0 5 6 1 1 1 7 G6 G2 8 X D 0
1 X 1 X 1 1 0 0 0 primitive cubes of G4 1 X 1 1 0 0 0 X 0 0
1 primitive cubes of G5 X
1 0
0
1 0 primitive cubes of G6 X 0 X 13 05_comb_tg, v1.0 © K.T. Tim Cheng, The DAlgorithm 0 0 1
© K.T. Tim Cheng, 13 F 11 6 9 10 11 12 13 1 12 1
1 1 primitive cubes of G3 • The propagation Dcubes of a logic element E
specify minimal input conditions which are
required to propagate an error signal on an input
(or inputs) to the output of that element. s.a.0 7 1 X
X primitive Dcube 1 G5 10
G3 5 C3
E4 primitive cubes of G2 Propagation Dcubes G4
G1 A1
B
2 1 1
14 05_comb_tg, v1.0 DAlgorithm Example 1) Select a primitive Dcube of the fault
2) Implication and checking for inconsistency.
If inconsistency occurs, go to (1).
3) Ddrive: selects an element in Dfrontier & attempts to propagate
D or D in its inputs to its output.
Dfrontier consists of set of all elements whose output values
are unspecified but inputs have some signals with D or D.
Ddrive is done by intersecting the test cube with a
propagation Dcube of the selected element.
Backtrack, i.e. select another propagation Dcube, if
intersection is null.
4) Implication of Ddrive: perform implication for the new test cube.
5) Repeat 3) & 4) until faulty signal propagated to an output.
6) Line justification: Consistency check on input conditions required.
© K.T. Tim Cheng, A1
B
2 10
5 G2 5
s.a.0
G2 C3
E4 © K.T. Tim Cheng, G3 G5 G3 G5 G6 6 12 7
13 F 11 16 05_comb_tg, v1.0 G6 6 • Now Dfrontier is G5 & G6 12 • Select G5 and a propagation Dcube of G5 7
13 F 11 1 2 3 4 5 6 7 8 9 10 11 12 13
perform implication 1 1 1 1 D 0 D
propagation Dcube of G5
1
D
D
1
D
Test cube after Ddrive 1 1 1 1 D 0 D 1 2 3 4 5 6 7 8 9 10 11 12 13
initial test cube ti 1 1
D
propagation Dcube of G3
D0D
test cube after Ddrive 1 1
through G3 = tc' D0D 1
Line justification 1 1 1 1 D 0 D 1
or
1111D0D
11 perform implication 1 1 1 1 D 0 D • Check implication: Dcube of G1 does not imply any
other signal.
• Dfrontier : G3
• Get propagation Dcube for G3
© K.T. Tim Cheng, 10
G1 G4
G1
s.a.0 C3
E4 G4 A1
B
2 15 05_comb_tg, v1.0 8
H
J9 8
H
J9 05_comb_tg, v1.0 8
H
J9
A1
B
2 17 D
D G4
10
G1 C3
E4
© K.T. Tim Cheng, 5 s.a.0
G2 G3 G5 G6 6 12 7
13 F 11 05_comb_tg, v1.0 18 3 Line Justification Flowchart for Dalgorithm
start Begin Initialize test cube (tc) Is there any
line in tc which
are not justified Select a primitive Dcube
of that as C Is there a D
or D on any PO ?
no yes Line Justification
impossible 19 primitive Dcube (1)
implication (2) 1
implication (4) 1
111
X
implication (5) G4 G1 2
3 6
sa1 4
G3 7 G5 G7 111 0D 9
12 10
11 inconsistent &
backtrack to (6) 1 1 1 D
D 0 1
1D
1D11D
1D11D H A
B 00D01D11D
00D01D11D
0
1
0D
1
D 20 05_comb_tg, v1.0 • Since the assignment of values is allowed to internal
lines, more than one choice is available at such internal
line/gate and backtracking could occur at each gate
• Could result in inefficiency for large ckts and some
special classes of ckts.
• Example: An ECAT (errorcorrection&translation) ckt D 0
1
0D01D11D
D
1 1
Implication
Line justification (7) 1 1 1
Implication
Line justification (8) 1 1 1
implication (8) 1 1 1
1 G8
G6 0D Select Dfrontier G8 (6) 5 G2 0
0D
0 111 8 Select Dfrontier G5 (3) 1
111 D
D
D © K.T. Tim Cheng, Potential Problems with DAlgorithm 1 2 3 4 5 6 7 8 9 10 11 12
11
D
11 inconsistent consistent Backtrack to the last
Line
point a choice exists
justification
none exists
impossible 05_comb_tg, v1.0 A DAlgorithm
Example 1 Intersect C with previous
test cube tc done Test has
been
generated Select a gate from Dfrontier
and a propagation Dcube
of the selected gate as C none exists © K.T. Tim Cheng, Select an unjustified
line and a primitive cube
C to justify the line consistent Backtrack to
the last
point a choice
exists Test has
been
generated yes Dintersect C with previous
test cube tc and perform implication inconsistent no s.a.0 C
E J D
Select Dfrontier
G6 (9)
11110D
1DD
1
1
0
implication (10)
0 N K F
G 1D P R Q
L
M 11110D01DD1D
Test is found : 1 1 1 1
© K.T. Tim Cheng, 21 05_comb_tg, v1.0 The PODEM Algorithm (Goel 1981) © K.T. Tim Cheng, PODEM Decision Tree for the Example • Only allows assignment of values to primary inputs
• The values assigned to primary inputs are then
propagated toward internal lines by the implication. Start
A=0 © K.T. Tim Cheng, 05_comb_tg, v1.0 A=1 A
B
C
E B=1
C=1 A
B P E=0 J
K F
G H s.a.0 C
E H s.a.0
P B=0 • Example:
First, a binary value is
assigned to an unassigned PI
to provide a fault effect at
fault site: A = 1
Determine the implications
of assigned PIs (only
forward implication): A= 1
cause no implication
Next, assign B = 1
A = B = 1 ⇒ imply H = D 22 05_comb_tg, v1.0 E=1 N
R
Q L J
K F
G N F=0 R
Q F=1 M G=0 L G=1 Test has been
Successfully
generated M 23 © K.T. Tim Cheng, 05_comb_tg, v1.0 24 4 Flowchart of PODEM PODEM Start • Essentially a process of finding a PI & a binary Assign a binary value
to an unassigned PI value for initial assignment. Determine implication of all PIs • Continue assigning PI values, checking to see if the
error is being propagated to outputs (after each
PI assignment, perform forward implication) yes Test
is found • If at any stage, either the fault cannot be excited
or the error cannot be propagated further, backtrack to the most recent PI assignment and No test
exists change it. © K.T. Tim Cheng, 25 05_comb_tg, v1.0 Is
objective line
A
fed by a
yes
no (fed by PI ?
gate G)
Current
OR/NAND & V=1
objective value V
AND/NOR & V=0
and type of gate driving
Exit
Next obj line is
objective
the input of G
line ?
which is at x
OR/NAND & V=0
AND/NOR & V=1
and is the
easiest to control
Next obj line is
the input of G
Is G
which is at x
a NAND/NOR
and is the
gate ?
hardest to control
yes objective is directed toward providing the fault
effect on the faulty line.
(2) Given the initial objective, a PI & a logic value are
chosen that have a good likelihood of meeting the
object.
– Done using the backtrace procedure. no Next obj value
is the same as
the current
objective value
27 05_comb_tg, v1.0 • Initial objective: (0, G2) • Implication: G2 = D © K.T. Tim Cheng, G4
G1
X1
X2
X3 G5
G2 X4 • Dfrontier is { G5, G6 }
• Attempt to propagate through G5
– Require X1 = 1 sa1 G8
Z1 G6 •
•
•
•
• 28 G4
X2 = 1 G7 G1
X1
X2 X3 = 1 Decision tree • Attempt to propagate D through G8 X4 = 1 29 © K.T. Tim Cheng, X3 X1 = 1 test is found
05_comb_tg, v1.0 A 05_comb_tg, v1.0 Initial objective: ( 1, G6 )
Backtrace to set X4 = 0
Implication: G3 = 1, G7 = 0 & G8 = 1 ⇒ failed in propagating error
Backtrack to most recent PI assignment ⇒ reassign X4 = 1
Implication: G3 = 0, G6 = D, G8 = D ⇒ Test is generated G3 • Implication: G1 = 0, G4 = 1, G5 = D © K.T. Tim Cheng, Next obj value
is the complement
of the current
objective value A PODEM Example – Cont’d A PODEM Example • Backtrace: X3 = 1 26 05_comb_tg, v1.0 begin effect has not appeared at fault site, the initial • Initial objective: (0, G2) Is there an
untried combination
of values on assigned
PIs?
no
yes
Set untried combination
of values on assigned PIs Found PI
initial assignment
is the current
objective value (1) Determine an initial objective. If the fault • Backtrace to PI : X2 = 1 no
Test
possible with additional
maybe
no assigned PIs ? Flowchart of Backtrace Steps in PODEM © K.T. Tim Cheng, © K.T. Tim Cheng, Is there a
D or D on any
P0 ? 05_comb_tg, v1.0 X4 = 0 G5
G2 X4 conflict &
backtrack G8
sa1 Z1
G6 G3
G7
30 5 Cost of ATPG Accelerating Comb. ATPG (A) How long? (Time complexity) Basic goals: (B) How much RAM? (Space complexity) • Reduce number of backtracks (C) How many vectors generated? (Test application time) • Reduce processing between backtracking • Theoretical result (Ibarra & Sahni, 1975) :
– Generating a test for comb. ckt is NPcomplete
• Worstcase time ∝ constantG Basic tools:
• Topological analysis (G = # of gates) • Emperical result (averagetime behavior): • Multiple backtrace – Total ATPG time ∝ constant • G2 • Learning – Test length ∝ G
© K.T. Tim Cheng, 31 05_comb_tg, v1.0 Socrates: Static Learning bX
a1
c X bX X
d X
f X
e c X During ATPG: aX
c © K.T. Tim Cheng, X bX X
d 0
f X
e ⇒ • Step 1: Construct a formula expressing the
Boolean Difference of a circuit with respect to
the fault a0
c X • Step 2: Apply a Boolean Satisfiability (SAT) solver
to the resulting formula 1
f 1
e Learned
information (a=1⇒f=1)⇔(f=0⇒a=0)
bX Given a fault, it consists of two steps: 1
d ⇒a 1 32 05_comb_tg, v1.0 Boolean Satisfiability Approach Preprocessing the ckt:
(1) Assign a logic value to a certain signal of the ckt
(2) Perform all implications from that assignment
(3) Learn from the results of implications
– Using law of contraposition: (A ⇒ B ) ⇔ (!B ⇒ !A)
Preprocessing: © K.T. Tim Cheng, X
d 0
f X
e 33 05_comb_tg, v1.0 © K.T. Tim Cheng, 34 05_comb_tg, v1.0 Step 1: Extracting the formula – Cont’d Step 1 : Extracting the formula (a) Construct the formula of the good circuit output X
A • Each node of the ckt is tagged with the logic formula in
3element conjunctive normal form, or 3CNF B (D+A) *
(D+B) *
(D+A+B) D
(X+D) *
(X+E) *
(X+D+E) • The formula is true iff the values assigned are
consistent with the truth for the logic element X E C (C+E) *
(C+E) (b) Construct formula of faulty circuit output X’ for fault D A
B C (C + A) * (C + B) * (C +A + B) A
B stuckat0 (no need to repeat the part identical to (a)) C (C + A) * (C + B) * (C + A + B) A C A B (D+A) *
(D+B) *
(D+A+B) 1 D (C + A) * (C + A ) D'
(X'+D') *
(X'+E) *
(X'+D'+E) X' E C © K.T. Tim Cheng, 05_comb_tg, v1.0 35 © K.T. Tim Cheng, (C+E) *
(C+E) 05_comb_tg, v1.0 36 6 Step 1: Extracting the formula – Cont’d Step 2: Satisfying the Formula
 Boolean Satisfiability (c) Construct the formula of the Boolean Difference:
XOR of (a) & (b) and the output of XOR should be 1
X
X’ From (a): • Given a suitable representation for a Boolean function f(X): BD=(X+X’) (X+X’)=V1 V2=1 – Find an assignment X* such that f(X*) = 1
– Or prove that such an assignment does not exist (i.e. f(X) = 0
for all possible assignments) (X+D) (X+E) (X+D+E) (D+A) (D+B) (D+A+B) (C+E) (C+E) From (b): • In the “classical” SAT problem, f(X) is represented in productofsums (POS) or conjunctive normal form (CNF) (X’+D’) (X’+E) (X’+D’+E) D’ From (c): • Many decision (yes/no) problems can be formulated either directly
or indirectly in terms of Boolean Satisfiability (V1+X) (V1+X’) (V1+X+X’) (V2+X) (V2+X’) (V2+X+X’) (BD+V1) (BD+V2) (note: BD=X⊕X’=(X+X’) (X+X’)=V1 V2) (BD+V1+V2) BD © K.T. Tim Cheng, 37 05_comb_tg, v1.0 Conjunctive Normal Form (CNF) © K.T. Tim Cheng, 38 05_comb_tg, v1.0 Literal & Clause Classification ϕ = ( a + c ) ( b + c ) (¬a + ¬b + ¬c ) violated satisfied unresolved satisfied ϕ = (a + ¬b)(¬a + b + ¬c )(a + c + d )(¬a + ¬b + ¬c ) Positive
Literal Clause Negative
Literal a assigned 0c and d unassigned
b assigned 1 * Source: João MarquesSilva and Karem A. Sakallah, “Boolean Satisfiability in EDA”, CAV2000 tutorial.
© K.T. Tim Cheng, * Source: João MarquesSilva and Karem A. Sakallah, “Boolean Satisfiability in EDA”, CAV2000 tutorial.
39 05_comb_tg, v1.0 Basic Backtracking Search 1
2
4
5 (a + c + d)
(¬a + c + d) 6 (¬a + c + ¬d) 7 (¬b + ¬c + ¬d) 8 • An unresolved clause is unit if it has exactly one
unassigned literal a (¬a + b + ¬c) 40 05_comb_tg, v1.0 Unit Clause Rule  Implications (a + b + c ) 3 © K.T. Tim Cheng, (¬b + ¬c + d) (a + b + ¬ c ) b
c
d b
c d c
d d ϕ = (a + c)(b + c)(¬a + ¬b + ¬c)
• A unit clause has exactly one option for being
satisfied a b → ¬c
i.e. c must be set to 0. d * Source: João MarquesSilva and Karem A. Sakallah, “Boolean Satisfiability in EDA”, CAV2000 tutorial.
* Source: João MarquesSilva and Karem A. Sakallah, “Boolean Satisfiability in EDA”, CAV2000 tutorial.
© K.T. Tim Cheng, 05_comb_tg, v1.0 41 © K.T. Tim Cheng, 05_comb_tg, v1.0 42 7 Basic Search with Implications 2
3 (¬a + b + ¬c) 4
5 (a + c + d)
(¬a + c + d) 6 (¬a + c + ¬d) 7 (¬b + ¬c + ¬d) 8 (¬b + ¬c + d) • CSAT: Combinational Circuitbased SAT Solver a (a + b + c )
(a + b + ¬ c ) 1 Public Released SAT Solvers by UCSB b – F. Lu, LiC. Wang, K.T Cheng, and R. Huang, “A Circuit SAT solver with
Signal Correlation guided learning,” DATE, March 2003. b c – F. Lu, L.C. Wang, K.T. Cheng, J. Moondanos and Z. Hanna, "A Signal
Correlation Guided ATPG Solver and Its Applications for Solving
Difficult Industrial Cases," DAC, Jun. 2003. c • Satori & Satori2: Sequential Circuitbased SAT Solver
a
b 4
3
5
7 b
c 4
3
5
7 d
c 6
5
8
5
6
8
8
6 6
d – M. K. Iyer, G. Parthasarathy, and K.T Cheng, “SATORI—A Fast
Sequential SAT solver for Circuits,” ICCAD, Nov. 2003. 6
6 – F. Lu, G Parthasarathy, M. K. Iyer, L.C. Wang, K.T. Cheng, "A Efficient
Sequential SAT Solver With Improved Search Stategies", UCSB
Technical Report, Dec. 2003 * Source: João MarquesSilva and Karem A. Sakallah, “Boolean Satisfiability in EDA”, CAV2000 tutorial.
© K.T. Tim Cheng, 05_comb_tg, v1.0 43 © K.T. Tim Cheng, 05_comb_tg, v1.0 44 Test Compaction for Comb. Tests
• Fault simulate test patterns in reverse
order of generation
– ATPG patterns go first
– Randomlygenerated patterns go last (because
they may have less coverage)
– When coverage reaches 100%, drop remaining
patterns (which are the useless random ones)
– Could significantly shortens test sequence –
reducing test application time © K.T. Tim Cheng, 05_comb_tg, v1.0 45 8 ...
View
Full
Document
This note was uploaded on 01/16/2012 for the course ECE 255a taught by Professor Timcheng during the Fall '11 term at UCSB.
 Fall '11
 TIMCHENG

Click to edit the document details