電機系

Chapter 1 Introduction
to VLSI Testing
超大型積體電路測試簡介
趙家佐
Goal of this Lecture

 Understand the process of Testing

 Familiar with terms used in Testing
 Testing as a problem of economics

2
Introduction to IC testing

 Introduction
 Types of IC testing

 Manufacturing tests

 Test quality and economy

 Test industry

3
 IC (SOC) Design/manufacture
Process
Specification

Architecture Design

Chip Design Chip design phase

Chip production phase

Fabrication

Test

 In chip production, every chip will be manufactured and tested.

 A chip is shipped to customers, if it works according to specification. 4
Tasks of IC Design Phases

•Function and performance requirements

•Die size estimation
•Power analysis
Specification •Early IO assignment

•High-Level Description Block diagrams

Architecture Design •IP/Cores selection (mapped to a platform)
•SW/HW partition/designs

•Logic synthesis
Chip Design
•Timing verification
•Placement, route and layout
•Physical synthesis
Fabrication •Test development and plan

•First silicon debug

Test •Characterization
•Production tests
5
Objectives of VLSI Testing

 Exercise the system and analyze the

response to ascertain whether it behaves
correctly after manufacturing
 Test objectives
 Ensure product quality
 Diagnosis & repair

 All considered under the constraints of

economics

6
Test Challenges
 Test time exploded for exhaustive testing
 For a combinational circuit with 50 inputs, we need
250 = 1.126x1015 patterns = 1.125x108s = 3.57yrs.
(10-7s/pattern)
 Combinational circuit = circuit without memory

Too many input pins  too many input patterns

7
More Challenges
 High automatic test equipment (ATE) cost
for functional tests
 Testing circuits with high clock rates
 Deep sub-micron/nano effects
 Crosstalk, power, leakage, lithography, high
vth variation…
 Test power > design power
 Integration of analog/digital/memories
 SOC complexities

8
Testing Cost
 Test equipment cost
 Analog/digital signal and measuring instrumentation
 Test head
 Test controller (computer & storage)
 Test development cost
 Test planning, test program development and
debug
 Testing-time cost
 Time using the equipment to support testing
 Test personnel cost
 Training/working
Testing Cost in Y2k

 Testing of complex IC is responsible for the

second highest contribution to the total
manufacturing cost (after wafer fabrication)
 0.5-1.0GHz, analog instruments, 1024 digital
pins: ATE purchase price
 $1.2M + 1024*$3000 = $4.272M
 Running cost (5-yr linear depreciation)
 = Depreciation + Maintenance + Operation
 = $0.854M + $0.085M + $0.5M
 = $1.439M/yr
Types of IC Testing (I): Audition of
System Specification

Specification  Translation of customer requirements to

system specifications is audited.
Architecture Design  The specification has to be reviewed
carefully throughout the
Chip Design design/production process.

Fabrication

Test

11
Types of IC Testing (II):
Verification
Specification

Architecture Design  The design is verified against the

system specifications to ensure its
Chip Design correctness.
 Verification is an essential and integral
Fabrication part of the design process.
 Especially for complex designs, the time
Test and resource for verification exceed
those allocated for design.
12
Types of IC Testing (III):
Characterization Testing
Specification
 Before production, characterization
testing are used.
Architecture Design
 Design debug and verification.
 Determine the characteristics of chips in
Chip Design
silicon.
 Setup final specifications and production
Fabrication
tests.

Test

13
Types of IC Testing (IV):
Production Testing
Specification
 In production, all fabricated parts are
subjected to production testing to detect
Architecture Design
process defects.
 To enforce quality requirements
Chip Design
 Applied to every fabricated part

 The test set is short but verifies all

Fabrication relevant specifications, i.e., high

coverage of modeled faults
 Test cost and time are the main drivers.
Test

14
Types of IC Testing (V):
Diagnosis
Specification
 Failure mode analysis (FMA) is applied
to failed parts.
Architecture Design
 To locate the cause of misbehavior after
the incorrect behavior is detected.
Chip Design
 Results can be used to improved the
design or the manufacturing process.
Fabrication
 An important step for improving chip
production yield.
Test

15
 Multiple Design Cycles
Specification

Design Verification Architecture Design

Chip Design

Fabrication

Test Failure analysis

Debug and Diagnosis

Long iterations  Late time-to-market/production

16
A Broad View of Chip Design
and Production Phases

Characterization

Design FAB Debug

Re-design FAB Production test

Diagnosis
Time to Market

Time to Yield
17
What Are We After in Testing?

 Design errors (first silicon debug)

 Design rule violation
 Incorrect mapping between levels of design

 Inconsistent specification

 Manufacturing defects
 Process faults/variation
 Time-dependent failures (reliability)
 Packaging failures

18
Various Design Errors

Breakdown of design errors in Pentium 4.

 Goof (12.7%) - typos, cut and paste errors, careless coding.
 Miscommunication (11.4%)
 Microarchitecture (9.3%)
 Logic/Microcode change propagation (9.3%)
 Corner cases (8%)
 Power down issues (5.7%) - clock gating.
 Documentation (4.4%)
 Complexity (3.9%)
 Random initialization (3.4%)
 Late definition of features (2.8%)
 Incorrect RTL assertions (2.8%)
 Design mistake (2.6%) - the designer misunderstood the spec

Source: Bentley, DAC2001 19

Methods to Find First-Silicon
Bugs
 Post-silicon debug requires a lot of efforts
 System Validation (71%).
 Compatibility Validation (7%)
 Debug Tools Team (6%)
 Chipset Validation (5%)
 Processor Architecture Team (4%)
 Platform Design Teams and Others (7%)

Source: Intel Technology Journal Q1, 2001

Validating The Intel Pentium 4 Processor 20
Defect Example: Particle

Source: ITC2004, D. Mark J. Fan, Xilinx 21

Defect Example: Metal breaks

Source: ITC2004, D. Mark J. Fan, Xilinx 22

Defect Example: Bridging

Source: ITC1992 Rodriguez-Montanes, R.; Bruis, E.M.J.G.; Figueras, J.

23
Systematic Process Variations

 Metal layer of NOR3XL standard Cell

24
Tests Before and After
Production
 (Before) Characterization Testing
 For design debug and verification
 Usually performed on designs prior to production
 Verify the correctness of the design & determine exact device
limits
 Comprehensive functional, DC and AC parametric tests
 Set final spec. and develop production tests
 (After) Production Testing
 To enforce quality requirements
 Applied to every fabricated parts
 Test vectors should be as short as possible under the constraints
of test costs and product quality
 Test costs are the main drivers

25
Test Items for Production Testing

 Circuit probe test (CP)

 Examine each part on the wafer before it is broken up into chips
 Final test (FT)
 Examine each part after packaging
 Usually FT includes
 Contact test
 DC parameter test
 AC parameter test
 Functional test
 Make sure circuits function as required by specification.
 Consume most test resources in production.
 Burn-in test (optional)
 Exercise chips in extreme conditions, e.g., high temperature or
voltage, to screen out infant mortalities
 Speed binning (optional)
 Determine the max speed of a chip and sell it accordingly
26
An Exemplary Test Flow

Objective: gross process defect

CP Metric: coverage of targeted faults
Patterns: functional / scan / BIST

Objective: process defect, package defect

FT Metric: coverage of targeted faults
Patterns: functional / scan / BIST

Objective: aging defects

Burn-in Metric: toggle coverage
Patterns: functional / scan (without comparison)

Objective: performance
Speed binning Metric: speed, delay fault coverage
Patterns: functional (mostly) / scan (rare)
Objective: Final quality screen
Quality Assurance
Metric: Adhoc
test
Patterns: Functional, System
27
Connectivity Test

 Verify whether the chip pins have opens or

shorts
 Also called open/short test
 Draw current out of the device and measure
voltage the input pin
 Utilize the forward bias current of the
protection diodes at the pin to determine
whether a short or open exists

28
DC Parametric Tests

 Tests performed by Parametric Measurement

Unit (PMU)
 Much slower than the normal operation speed
 Static (operating) current test
 check the power consumption at standby (operating) mode
 Output short current test
 Verify that the output current drive is sustained at high and
low output voltage
 Output drive current test
 For a specified output drive current, verify that the output
voltage is maintained

29
AC Parametric Tests

 To ensure that value/state changes occur at

the right time
 Some AC parametric tests are mainly for
characterization and not for production test.
 Test for rise and fall times of an output signal
 Tests for setup, hold and release times
 Tests for measuring delay times
 E.g. tests for memory access time

30
Burn-in Tests

 Early failure detection reduces cost

 Burn-in to isolate infant mortality failures
Infant
Normal Wear-out
mortality
lifetime period
period
Failure rate

~ 20 weeks 5 – 25 yrs
Time

Bathtub Curve of IC’s Failure Rate

31
An Example of IC Failure Rate vs.
System Operating Time With/Without
Burn-in
100,000

10,000
No burn-in
1,000 125C burn-in

100

10
150C burn-in
1
101 102 103 104 105 106
Time (hr)
32
 Functional Tests

 Selected test patterns are applied to circuits and

response are analyzed for functional correctness.
Test Manufactured Output
patterns response
Circuits

Acceptable/true Compare and

response Analyze

Test
result
33
Activities for Developing
Functional Tests
Specification

Architecture Design  Generate test pattern

 Evaluate the quality of test patterns
Chip Design  Design circuit with better test
efficiency
Fabrication

Test
 Apply test patterns

34
Key Issues of Functional Tests

 Where does patterns come from?

 Design simulation patterns (Functional patterns)
 Automatic test pattern generation (ATPG)
 How to evaluate the quality of test patterns?
 Fault coverage evaluation
 How to improve test efficiency?
 Design for Testability (DFT)
 How to apply test patterns?
 Automatic test equipments (ATE)
 Built-in self test (BIST)
35
Functional v.s. Structural Test
 Functional test
 Exercise the functions according to the spec
 Often require designers’ inputs
 Large number of patterns with low fault coverage
 Difficult to be optimized for production tests
 Structural test
 Use the information of interconnected components
(e.g., gates) to derived test regardless of the functions
 Fault modeling is the key
 Basis of current testing framework---ATPG, Fault
simulator, DFT tools, etc.

36
Fault Models
 Fault modeling is a way to represent the
cause of circuit failure.
 Model the effects of physical defects by the
logic function and timing
 Enumeration of real defects is impossible
 Makes effectiveness measurable by
experiments
 Fault coverage can be computed for specific
test patterns to reflect its effectiveness

37
Single Stuck-At Fault Model

 Assumptions:
 Only One line is faulty
 Faulty line permanently set to 0 or 1
 Fault can be at an input or output of a gate

• One of the gate input

a
f terminal was mistakenly
b connected to ground
• Fault: b stuck at 0
• signal b will always be “0”
38
 Logic Gate Basics

OR Gate AND Gate

A A
G G
B B

A B G A B G
0 0 0 0 0 0
0 1 1 0 1 0
1 0 1 1 0 0
1 1 1 1 1 1
Only binary values, 0 and 1, will be used.
A and B are inputs and G is the output. 39
Stuck-At Faults Example

Stuck-at 1 Stuck-at 0

A E
B
G
C F
D

Total Faults = Nf = 2* total number of signals =

2* 7=14
40
A Simple Simulation with Input
(ABCD)=(0111)

A=0 E=0
B=1
G=1
C=1 F=1
D=1

We use logic simulation to propagate (transfer)

input values to outputs.

41
What if F stuck-at-0 occurs
with (ABCD)=(0111)

A=0 E=0
B=1
G=10
C=1 F=10
D=1

We use logic simulation to propagate (transfer)

faulty values to outputs.
For this case, we say (0111) covers the fault F stuck-at-0.
42
Other Faults Covered By
(ABCD)=(0111)

A=0 E=0
B=1
G=1
C=1 F=1
D=1

By performing several logic simulation with faults

(fault simulation), we found (0111) covers four faults:
C, D, F, and G stuck-at-0.
43
Fault Coverage of
(ABCD)=(0111)

A=0 E=0
B=1
G=1
C=1 F=1
D=1

Since (0111) covers four faults: C, D, F, and G stuck-at-0.

And total number of faults is 14.
We say (0111) has a fault coverage of 4/14 ~ 28.6%
44
Fault Coverage of
(ABCD)=(0101)

A=0 E=0
B=1
G=0
C=0 F=0
D=1

Since (0101) covers four faults: A, C, E, F, and G stuck-at-1.

And total number of faults is 14.
We say (0101) has a fault coverage of 5/14 ~ 35.7%
45
Combined Fault Coverage of
(ABCD)=(0111) and (0101)

A=0 E=0
B=1
G=0
C=0 F=0
D=1

We know that both vectors cover different faults, so

the total number of covered faults are 4+5.
Therefore we have a total fault coverage 9/14 ~ 64.3%
46
Fault Coverage

 Fault Coverage T
 Is the measure of the ability of a set of tests to
detect a given class of faults that may occur on
the device under test (DUT)

No. of detected faults

T=
No. of all possible faults

 Fault simulation is used to evaluate fault

coverage for test patterns.

47
Meaning of Fault Coverage

 Our goal in testing is to find test patterns to

achieve 100% fault coverage.
 Under the assumption of the fault model (e.g.,
single stuck-at fault), we’ve done a good job!
 Remember the problem of testing a circuit with 50
inputs?
 Remember the problem of numerous defects that
can occur in a chip?
 Though single stuck-at fault model is very
simple, it is very effective.
 Other fault models is still needed to further
improve chip quality.
48
Automatic Test Pattern
Generation (ATPG)
 Generate test patterns to cover modeled
faults automatically.
 A complex process to determine the quality of
tests
 The most time-consuming process in test
development
 Very difficult for sequential circuits (circuits
has memory elements).

49
 An Example of ATPG for E
stuck-at-0
Step 2:
assign A=1 and B=1 Step 1: assign E=1
A E/0
Finally, we will see
B G=1 for fault-free circuits, and
G=0 for faulty circuits.

C F
D Step 3: assign F=0
Step 4:
assign (C, D)=(0, 0), (0, 1), or (1, 0)

We can have test vectors (A, B, C, D)=(1, 1, 0, 0), (1, 1, 0, 1),

(1, 1, 1, 0)
50
 The Infamous Design/Test Wall
30 years of experience proves that
test after design does not work!

Oh no!
What does
Simulation functionally correct! this chip do?!
We're done!

Design Engineer
Test Engineer
51
Design for Testability (DFT)

 DFT is a technique to design a circuit to be

easily testable
 Add the cost of area/performance, but
dramatically reduce cost for tests
 For example, use scan technique to make
test generation feasible on sequential circuit.
 A very important step in circuit design to
make sure a circuit is testable.

52
 Full Scanned Sequential Logic
---An Example of DfT

Scan_Ena
Scan
Test for SA0 fault here. Flip-Flop

Scan_In

53
Multiple Design Missions
 Chips have to optimally satisfy many constraints:
area, performance, testability, power, reliability, etc.

Performance

Area

Power Testability

54
Definition of BIST

 BIST is a DFT technique in which testing (test

generation , test application) is accomplished
through built-in hardware features.
 Advantages
 Better quality
 Reduce test application time
 Reduce test development time
 Costs
 Area increased
 Circuit performance degrade
 Yield loss
55
Tools for Developing
Functional Tests (Recap)
Specification

Architecture Design  ATPG

 Fault simulation
Chip Design  DFT
 BIST
Fabrication

Test
 ATE
 BIST
56
Testing and Quality

Shipped Parts
ASIC
Testing
Fabrication Yield:
Quality:
Fraction of Defective parts
Good parts Per Million (DPM)

Rejects

Quality of shipped part is a function of

yield Y and the test (fault) coverage T.

57
Defect Level

 Defect Level
 Is
the fraction of the shipped parts that
are defective
DL = 1 – Y(1-T)

Y: yield
T: fault coverage

58
Defect Level v.s. Fault
Coverage
Defect Level
1.0 Y = 0.01
Y = 0.1

0.8
Y = 0.25

0.6
Y = 0.5

0.4
Y = 0.75

0.2 Y = 0.9

0 20 40 60 80 100
(Williams IBM 1980) Fault Coverage ( % )

High fault coverage Low defect level

59
DPM v.s. Yield and Coverage

Yield Fault Coverage DPM

50% 90% 67,000

75% 90% 28,000
90% 90% 10,000
95% 90% 5,000
99% 90% 1,000

90% 90% 10,000

90% 95% 5,000
90% 99% 1,000
90% 99.9% 100

A chip with 100 DPM or below is considered of high quality.

60
Components of Test Costs (I)

 Determining the costs in each design phase is very

important for evaluating different test strategies
 Cost directly impacted by tests
 Test equipment
 Test development
 Test planning, test program development
 Test time
 Time using the equipment to support testing
 Test personnel

61
Components of Test Costs (II)

 Other costs associated with tests

 Design time
 Chip area (manufacturing costs)

 Time to Market

 Product quality
 Impact a company’s image and sales

62
 Cost Of Testing - The Rule of
Tens

1000
Cost
Per 100
Fault
(Dollars)
10 500
50
1 5.0
0.5

IC Board System Warranty

Test Test Test Repair

63
Implications of Rule of Tens

 Early detection can prevent costly diagnosis

and replacement later.
 For example, if a bad IC is not detected, the
cost to find a board including the bad IC is at
least 10 times higher.

64
 Test Economics

 Build an appropriate

Non-recurring costs
cost/benefits model based on $ Specification

empirical data of the

manufacturing process. $ Architecture Design
 Evaluate test strategies (DFT;
BIST) according to the model
$ Chip/test Design
 Customize the model for each
project
 Follow and review the model $ Fabrication
closely through careful
management Test
$

$Defect Level/Fail return 65

A Case Study for Test
Economics
 A BIST and Boundary-Scan Economics
Framework by JOSÉ M. MIRANDA
 Lucent Technologies Bell Laboratories
 IEEE Design and Test of Computers, JULY–
SEPTEMBER 1997

66
Conclusions

 Testing is used to ensure a chip’s quality.

 Testing is a complex and expensive task and
should be dealt with at early (design) stage.
 Test strategies should be evaluated with a
solid and overall economics model.

67