Testability Measure

Download Report

Transcript Testability Measure

,,
,,
VLSI Testing and DFT Course
Testability Measure
What do we mean when we say a circuit
is testable?
Definition: A fault is testable if there exists a
well-specified procedure to expose it
within a reasonable cost. A circuit is
testable if each and every fault in its
specified fault set is testable.
Testability Measure
1
,,
,,
VLSI Testing and DFT Course
Keys To Testability
1. Controllability
2. Observability
3. Predictability
Testability = Controllability + Observability
+ Predictability
Testability Measure
2
,,
,,
VLSI Testing and DFT Course
Design for Testability
To constrain the design to make test
generation and diagnosis easier.
Test Cost
Unconstrained
DFT
Circuit Size
Testability Measure
3
,,
,,
VLSI Testing and DFT Course
Testability (Controllability/Observability)
Measures
1. TMEAS [Stephenson & Grason, FTCS, 1976; DAC,
1979]
2. SCOAP [Goldstein, IEEE TCAS-26(9), 1979]
3. TESTSCREEN [Kovijanic 1979]
4. CAMELOT [Bennetts et al., 1980]
5. VICTOR [Ratiu et al., ITC, 1982]
Testability Measure
4
,,
,,
VLSI Testing and DFT Course
,
Stephenson & Grason s Approach
Developed for register-transfer-level
(RTL) circuits, but can also be applied
at the gate level.
The measures are normalized between
0 and 1 to reflect the ease of controlling
and observing the internal nodes.
Testability Measure
5
,,
,,
VLSI Testing and DFT Course
,
Stephenson & Grason s Approach
1. For each signal line s, we denote the controllability of s
as CY(s) and the observability of s as OY(s).
2. The values for the CYs and the OYs of all the signal lines
are derived by solving a
system of simultaneous
equations with the CYs and the OYs as unknowns.
x1
x2
xn
.
.
.
DUT
.
.
.
Z1
Z2
Zm
The expression used to calculate CY for each output
zj is
1 n
CY
z   CTF  n  CY  x ,
j
i 1
i
where CTF is the controllability transfer factor of the
component.
Testability Measure
6
,,
,,
VLSI Testing and DFT Course
,
Stephenson & Grason s Approach
Let Nj(0) and Nj(1) be the numbers of input
combinations for which zj has value 0 and 1,
respectively. Then
1 m 
CTF   1 
m j 1 
N
j
 0  N j 1 
2
n


.
0  CTF  1.
Each output controllability is assigned the same
value.
Testability Measure
7
,,
,,
VLSI Testing and DFT Course
,
Stephenson & Grason s Approach
The expression used to calculate OY for each input
xi is
1 m
OY  x i   OTF   OY
m j 1
z ,
j
where OTF is the observability transfer factor of the
component.
Testability Measure
8
,,
,,
VLSI Testing and DFT Course
,
Stephenson & Grason s Approach
Let NSi be the numbers of input combinations for which the change of
xi results in a change of output. Then
1 N NSi
OTF   n .
n i 1 2
NSi also means the number of input combinations that can
sensitize a path from xi to the output.
The OTF measures the probability that a faulty value at any
input will propagate to the outputs.
0  OTF  1
Each input observability is assigned the same value.
Testability Measure
9
,,
,,
VLSI Testing and DFT Course
,
Stephenson & Grason s Approach
3. Fanouts: Let s be a fanout stem and k be the number of its branches. Then the
CYs of each fanout branch is
CY  s
CY 
.
1  log k 
The observability of the fanout stem s is
k


OY  s  1   1  OY  bi  ,
i 1
where bi are fanout branches of s.
4.
Sequential components: Sequential components are modeled by adding
feedback links around the components that represent internal states .
Testability Measure
10
,,
,,
VLSI Testing and DFT Course
,
Goldstein s Approach---SCOAP
Sandia Controllability Observability Analysis
Program.
The measures reflect the difficulty of controlling
and observing the internal nodes; higher
numbers indicate more difficult to control or
observe.
The measures are, in a sense, minimum cost
values for controlling and observing.
Testability Measure
11
,,
,,
VLSI Testing and DFT Course
,
Goldstein s Approach---SCOAP
Combinational 1- and 0-controllabilities of Y =
AND(A,B,C):
CC1(Y) =CC1(A) + CC1(B) + CC1(C) + 1;
CC0(Y) = min{CC0(A),CC0(B),CC0(C)} + 1.
The result is incremented by 1 so that the number
reflects (in part) the distance to the PIs.
Working breadth-first from PIs toward POs, we calculate
the CC of the output line of each logic cell as a function
of the CCs of its input lines.
Testability Measure
12
,,
,,
VLSI Testing and DFT Course
,
Goldstein s Approach---SCOAP
The sequential controllability provides an estimate of the
number of time frames needed to provide a 0 or 1 at a particular
node.
For Y=XOR(A, B):
CC0(Y) = min{CC0(A) +CC0(B), CC1(A) + CC1(B)} + 1
CC1(Y) = min{CC0(A) + CC1(B), CC1(A) + CC0(B)} + 1
SC0(Y) = min{SC0(A) + SC0(B), SC1(A) + SC1(B)}
SC1(Y) = min{SC0(A) + SC1(B), SC1(A) + SC0(B)}
When computing the sequential controllabilities through
combinational circuits, the values are not incremented---no
additional time frames needed.
Testability Measure
13
,,
,,
VLSI Testing and DFT Course
,
Goldstein s Approach---SCOAP
When deriving equations for sequential circuits, SCs are
incremented by 1, but CCs are not incremented.
Positive edge-triggered DFF with active low reset:
CC0(Q) = min{CC0(R), CC1(R) + CC0(D) + CC0(C) + CC1(C)}
CC1(Q) = CC1(R) + CC1(D) + CC0(C) + CC1(C)
SC0(Q) = min{SC0(R), SC1(R) + SC0(D) + SC0(C) + SC1(C)} + 1
SC1(Q) = SC1(R) + SC1(D) + SC0(C) + SC1(C) + 1
Testability Measure
14
,,
,,
VLSI Testing and DFT Course
,
Goldstein s Approach---SCOAP
Observabilities:
P
Q
R
N
CO(P) = CO(N) + CC1(Q) + CC1(R) + 1
SO(P) = SO(N) + SC1(Q) + SC1(R)
Positive edge-triggered DFF with active low reset:
CO(R) = CO(Q) + CC1(Q) + CC0(R)
SO(R) = SO(Q) + SC1(Q) + SC0(R) +1
To watch R, then have to watch Q and drive FF to ,1, and then reset it to ,0,.
Testability Measure
15
,,
,,
VLSI Testing and DFT Course
,
Goldstein s Approach---SCOAP
May define testability as follows:
T(l/o) = CC1(l) + CO(l)
T(l/1) = CC0(l) + CO(l)
Initial Condition
SO
1
-
1
-
1
0
1
0
8
1
-
CO
8
1
-
SC1
8
PI
PO
IN
SC0
8
CC1
8
CC0
8
Node
Testability Measure
16
,,
,,
VLSI Testing and DFT Course
CAMELOT
Controllability:
CY(output) = CTF(output) x f(CYs(inputs));
controllability transfer factor
N (0)  N (1)
CTF  1 
,
N (0)  N (1)
where N(0) and N(1) are the numbers of input
combinations for which the output has value 0 and 1,
respectively.
Testability Measure
17
,,
,,
VLSI Testing and DFT Course
CAMELOT
CTF(NOT) = 1; CTF(NAND2) = 1/2;
CTF(NAND3) = 1/4; CTF(ORn) =1/2n-1
CTF(XORn) = 1.
Testability Measure
18
,,
,,
VLSI Testing and DFT Course
CAMELOT
P
PRESET
Q
Q
C
SN7476
K
K RESET Q
Q
J
J
+
+
R
Q+ = [(JQ + KQ)C + QC]PR + P
= JQCPR + KQCPR + QCPR + P
Minterms in the final expression is 40; 24 invalid
states; 16 common terms in this set of invalid states
and the set of minterms derived above. N(1) = 40 16 = 24, and N(0) = (26 - 24) - N(1) = 40 - 24 = 16.
CTF = 1- (24-16)/40 = 0.8. By symmetry, the CTF for
Q+ is also 0.8.
Testability Measure
19
,,
,,
VLSI Testing and DFT Course
CAMELOT
CY(output) = CTF(output) x f(CYs(input)),
where
1
f (CYs(inputs)) 
( (CY : i async inputs) +
i j
 (CY : clock) (CY : j sync inputs)).
Testability Measure
20
,,
,,
VLSI Testing and DFT Course
CAMELOT
Observability
OY(at output) = OTF x OY(at input) x
g(CYs(supporting inputs)),
Where OTF is the observability transfer
factor of the component for the input
concerned.
Testability Measure
21
,,
,,
VLSI Testing and DFT Course
CAMELOT
N(SP:I  O)
OTF( I  O) 
,
N(SP:I  O)  N(IP:I  O)
where N(SP:I-O) is the total number of
distinct sensitive paths from I to O, and
N(IP:I-O) is the total number of insensitive paths.
Testability Measure
22
,,
,,
VLSI Testing and DFT Course
CAMELOT
OTF(NOT) = 1;
OTF(NAND2) = 1/2 for each
input;
OTF(NAND3) = 1/4 for each
input (1 PDC & 3 NPDCs);
OTF (XORn) = 1 for each input
(0 NPDC).
Testability Measure
23
,,
,,
VLSI Testing and DFT Course
CAMELOT
Let OY(A-B) denote the observability of node A
at node B, then
OY(I-O) = OTF(I-O) x OY(I-I) x CY(av),
where
1
CY (av) 
( (CY: i async supporting inputs)
i j
 (CY: clock) (CY: j sync supporting inputs)) .
Testability Measure
24
,,
,,
VLSI Testing and DFT Course
CAMELOT
Fanouts:
k
OY (stem) = 1-  1  OY (branchi ) .
i 1
For a reconvergent fanout, select the shortest
I-O path which is to be sensitized (while
others are blocked), which is more likely to
result in a higher OY.
For feedback paths, the strategy is similar to
that for reconvergent fanouts with unequal
path lengths (i.e., select the shortest path).
Testability Measure
25
,,
,,
VLSI Testing and DFT Course
CAMELOT
Testability
TY(node) = CY(node) x OY(node).
TY(circuit) =
 (TY(nodes))
No. of nodes
Testability Measure
.
26
,,
,,
VLSI Testing and DFT Course
CAMELOT
CAMELOT()
1. input, check, and initialize circuit;
2. calculate nodal CY values from PIs
to POs;
3. calculate nodal OY values from POs
to PIs;
4. calculate nodal TY values;
5. calculate TY and interpret the results;
Testability Measure
27
,,
,,
VLSI Testing and DFT Course
Importance of Testability Measures
• They can guide the designers to
improve the testability of their circuits.
• Test generation algorithms using
heuristics usually apply some kind of
testability measures to their heuristic
operations (e.g., in making search
decisions), which greatly speed up the
test generation process.
Testability Measure
28
,,
,,
VLSI Testing and DFT Course
Bibliography
(1) J. E. Stephenson and J. Grason, ,,A testability measure for register transfer level
digital circuits,,, in Proc. Int. Symp. Fault tolerant Computing (FTCS), (Pittsburgh,
PA), pp. 101-107, June 1976.
(2) J. Grason, ,,TMEAS-a testability measurement program,,, in Proc. IEEE/ACM
Design Automation Conf. (DAC) , vol. 26, no. 9, pp. 156-161, 1979.
(3) L. H. Goldstein, ,,Controllability/observability analysis for digital circuits,,, IEEE
Trans. Circuits and Systems, vol. 26, no. 9, pp. 685-693, Sept. 1979.
(4) P. G. Kovijanic, ,,Testability analysis,,, in Proc. IEEE Semiconductor Test Conf.,
pp. 310-316, 1979.
(5) P. G. Kovijanic, ,,Computer-aided testability analysis,,, in Proc. IEEE Autotestcon,
pp. 292-294, 1979.
(6) R. G. Bennetts, C. M. Maunder, and G. D. Robinson, ,,CAMELOT:a computeraided measure for logic testability,,, IEE Proc. Pt. E, vol. 128, no. 5, pp. 177-189,
1981.
(7) R. G. Bennetts, Design of Testable logic Circuits. Reading, MA: Addison-Wesley,
1984.
Testability Measure
29
,,
,,
VLSI Testing and DFT Course
Bibliography
(8) I. M. Ratiu, A. Sangiovanni-Vincentelli, and D. O. Pederson, ,,VICTOR: a
fast VLSI testability analysis program,,, in Proc. Int. Test Conf. (ITC),
(Philadelphia, PA), pp. 397-401, Nov. 1982.
(9) W. C. Berg and R. C. Hess, ,,COMET : a testability analysis and design
modification package,,, in Proc. Int. Test Conf. (ITC), (Philadelphia, PA), pp.
364-378, Nov. 1982.
Testability Measure
30