Introduction

Download Report

Transcript Introduction

Introduction
VLSI Testing
1
Overview
•
First digital products (mid 1940's)
Complexity: low
MTTF:
hours
Cost:
high
•
Present day products (mid 1980's)
Complexity: high
MTTF:
Perhaps centuries?
Cost:
low
2
Observations

Testing is a cost burden, people buy digital devices to provide
computation, control and/or communications.

The percentage of product development dollar allocated to testing
continues to increase.

Test problems have changed, but the need for testing continues.

Test emphasis changes over time. As product cost declines,
maintainability is less important than it once was.

There is no one single solution to the testing problem.
3
Focus

What makes circuits difficult to test (why do the algorithms fail)?

How can the complexity of the problem be reduced?

How can algorithms be made more effective?

What are the trade-offs between the various existing strategies?

What are the likely future directions?
4
DEFINITIONS
•
Fault:
•
Design Fault: a design characteristic of either hardware or software
which causes or materially contributes to device
malfunction independent of the presence of physical
faults.
•
Failure:
the termination of the ability of a chip to perform its
required function.
•
Error:
functional manifestation of a fault.
•
Test Or
Test Pattern:
•
a physical condition that causes a device, component,
or element to fail to perform in required manner.
a specified primary input stimulus plus the expected
fault-free primary output response.
Fault Detection: application of test patterns which discover or are
designed to discover the existence of faults.
5
DEFINITIONS (Continue)
•
Fault Isolation: where a fault is known to exist, a test sequence
which identifies or it is designed to identify the
location of that fault within a specific circuit.
•
Fault Coverage: An attribute of a test or test expressed as the
percent of faults of the total fault population which
that test procedure will detect.
•
Fault Masking: The ability to avoid a fault by concurrently detecting
and correcting all faults.
6
Failures are caused by defects such as:
A.
B.
C.
D.
E.
F.
G.
Contamination.
Metallization Defects.
Implant Defects
Wafer Defects
Oxide Defects
Interconnect Defects
Design Defects Such As:
•
Too narrow conductors; high voltage drops.
•
Too high voltage across oxide; hot electron injection.
•
Too critical dimensions
7
FAILURES OBSERVED BY DIRECT INSPECTION OF 4-BIT
MICROPROCESSOR CHIPS*
SHORT BETWEEN METALLIZATIONS
39%
OPEN METALLIZATION
14%
SHORT BETWEEN DIFFUSIONS
14%
OPEN DIFFUSION
6%
SHORT BETWEEN METALLIZATION
AND SUBSTRATE
2%
INOBSERVABLE
10%
MISCELLANEOUS
15%
ALMOST ALL FAILURES ARE DUE TO
SHORTS AND OPENS
* GALIAY, CROUZET, AND VERGNIAULT, IEEE TOC JUNE 1980.
8
A WELL TESTED INTEGRATED CIRCUIT
Is As IMPORTANT AS
A WELL DESIGNED INTEGRATED CIRCUIT
MOS/CMOS has emerged as an important technology
9
COST
A STANDARD AMONG PEOPLE FAMILIAR WITH THE TESTING PROCESS IS:
If the cost for detecting a fault at the chip level is:
$X
Then to detect that same fault at the board level is:
$10X
At the system level:
$100x
At the system level but when it has to be found in
the field:
$1000X
10
Test Economics
Shipped Product Quality Level
SPQL  Y (1T )
Y-Process yield
T-quality of test (fault coverage)
•
Given the desired SPQL, and the process yield, the required test
effectiveness, T, is fully determined.
•
In logic circuits, T is computed by means of fault simulation.
•
Defect level (DL) is often used as the measure of goodness, where:
DL=1 –SPQL
11
MEAN FAULT CYCLE
Time
 System
Recovery
Manifests
MTTD
Detected
Isolated
MTTR
Corrected
System
Recovery
MTBF
System
Available
Fault Occurs
 Manifests
12
SIGNIFICANCE OF FAULT MODELS
•
A fault model is a hypothesis representing the fault mechanism in a
circuit.
•
The reliability of the product is determined by the accuracy and
effectiveness of the fault model.
13
COMPLEXITY
•
•
If a network contained N nets, any net may be good; s-a-1 or s-a-0. Thus all possible
network state combinations would be 3N. Assume a network with 100 nets, then there are
5x1047 different combinations of faults.
Test generation and fault simulation is approximately proportional to the number of gates
to the power of 3.
T=kN3
Computer
Run time
•
Constant
Number of
gate
For functional testing if a network has N inputs (combinational) then 2N patterns are
required for complete functional test. If the network has N inputs and M latches then 2N+M
patterns are required.
For VLSI assume
N = 25 and M = 50 then
#Patterns = 275  3.8×1022
Assume test rate of 1 µ sec, then test time over 109 years
14
THE TESTING PROBLEM
GIVEN A SET OF FAULLS, OBTAIN TEST VECTORS
Q1: WHICH FAULTS? (FAULT MODELS)
Q2: HOW IS TEST DERIVED?
• MANUALLY
• AUTOMATICALLY
o ALGORITHMS (ATG)-PODEM, SOFTG
o KNOWLEDGE-BASED - HITEST
Q3: HOW IS TEST QUALITY MEASURED?
• FAULT SIMULATION
o CONCURRENT METHOD
o FAULT SAMPLING
• FAULT COVERAGE AND PRODUCT_QUALITY
15
WHY MODEL FAULTS?
•
I/O FUNCTION TESTS INADEQUATE FOR MANUFACTURING
(FUNCTIONALITY vs. COMPONENT & INTERCONNECTION TESTING)
•
FAULT MODEL IDENTIFIES TARGET FAULTS
•
FAULT MODEL MAKES ANALYSIS POSSIBLE
•
EFFECTIVENESS MEASURABLE BY EXPERIMENTS
16
SOME FAULT MODELS
 SINGLE STUCK FAULTS
•
TRANSISTOR OPEN / SHORT FAULTS
•
MEMORY FAULTS
•
PLA FAULTS (STUCK, CROSS-POINT, BRIDG1NG)
•
FUNCTIONAL (PROCESSOR) FAULTS
•
DELAY FAULTS
•
ANALOG FAULTS
17
SINGLE STUCK FAULTS
1
Faulty Response
1
1
Ture Response
Test Vector
0
(1)
0
0
0

STUCK-AT-1
ASSUMPTIONS:
1. ONLY ONE LINE IS FAULTY.
2. FAULTY LINE PERMANENTLY SET TO 0 OR 1.
3. FAULT CAN BE AT AN INPUT OR OUTPUT OF A GATE.
18
FAULT EQUIVALENCE
TWO EQUIVALENT FAULTS ARE DETECTED BY EXACTLY THE SAME TESTS
s-a-0

s-a-1

s-a-1

THREE FAULTS SHOWN ARE EQUIVALENT
19
EQUIVALENCE FAULT COLLAPSING
s-a-1
s-a-1
s-a-0
s-a-1
s-a-0
s-a-0
s-a-1
s-a-0
N+2 FAULTS IN N-INPUT GATE
20
DOMINANCE FAULT COLLAPSING
F1: s-a-1

F2: s-a-1

IF ANY TEST FOR F1 DETECTS F2 BUT CONVERSE IS NOT TRUE,
THEN F2 DOMINATES F1.
ONLY N+1 FAULTS IN N-INPUT GATE
s-a-1
s-a-0
s-a-1
21
The Sensitized Path Method
Procedure:
1. Create a Sensitized Path from the fault to the primary output.
2. Justify the assignment of values to the outputs of internal gates.
Example:
G1
S/0
1
G5
0
1
G2
G7
Z
1 good
0 faulty
G7
Z
1 good
0 faulty
1
G4
G6
G3
Sensitized Path
1
1
G1
S/0
1
G5
0
X 0
0 X
G2
1
0
1
G4
X
X
1
G3
G6
X
Justify the assignment
22
The Sensitized Path Method (Continue)
Problems with the Sensitized Path Method
1. Making Choices
2. Reconvergent fan-out Paths
Making Choices
1
X1
X2
X3
X4
X5
2
6
G3
7
G4
Z
G1
3
4
G2
Y
5
23
The Sensitized Path Method (Continue)
Reconvergent Fan-out Paths
The sensitive path method is not guaranteed to find a test for a fault, even
where such a test does exist.
Example:
G4
G1
1
X1 0
X2
X3
0
G5
0
S/0
G2
G8
1
0
Z
G6
X4 1
0
G3
0
G7
0
1 Inconsistent
•
•
•
Try to propagate through G5  Inconsistent
Try to propagate through G6  Inconsistent
It appears that there is no test for the fault. However, such a test does exist
{0,0,0,0}
24
Redundancy and Undetectability
X1 1
X2
4
2
X3
•
•
3/1
6 Z
5
Fault 3/1 is undetectable because the gate is redundant.
Z = X1X2 + X1X2X3 = X1X2
25
The D-algorithm
Example:
S/0
26
The D-algorithm
Example:
0
D
1=D S/0
1 good
D
0 faulty
0
27
The D-algorithm
Example:
X1
X2
Z
X3
X4
S/0
28
The D-algorithm
Example:
X1
X2
×
1
1 good
D
0
Z
X3
X4
0 faulty
0
×
S/0
D
D
29
The D-algorithm
Example:
X1
X2
S/1
Z
X3
30
The D-algorithm
Example:
X1
1
1
1 good
D
X2
X3
1
0 faulty
D
Z
S/1
1
D
31