Design-for-Test Whitepaper

Design-for-Test

The Robustness of Various Test

Compression Techniques

Last Updated: December 27, 2007

Abstract
Larger designs and the growing population of non-stuck defects have led many companies to
adopt test compression techniques. In fact, the Embedded Deterministic Test (EDT) technology
within TestKompress has now been used in over one billion production chips. There has been a
surge of compression techniques promoted in the industry since TestKompress was released in
2001. So, why has TestKompress become the standard approach in industry? This paper will
explain the technology behind the various compression techniques and their robustness in the
presence of Xs, false and multicycle paths, low pin access, and other design factors.

www.mentor.com/dft
Copyright © Mentor Graphics Corporation 2007 All rights reserved
This document contains information that is proprietary to Mentor Graphics Corporation. The original recipient of this
document may duplicate this document in whole or in part for internal business purposes only, provided that this
entire notice appears in all copies. In duplicating any part of this document, the recipient agrees to make every
reasonable effort to prevent the unauthorized use and distribution of the proprietary information.
 Table of Contents

EXECUTIVE SUMMARY .................................................................................................... 1

INTRODUCTION................................................................................................................... 2
Why Compression .........................................................................................................................................2
Goals of Scan Compression ..........................................................................................................................2
COMPRESSION TECHNOLOGIES ................................................................................... 3
Virtual Scan ...................................................................................................................................................3
Adaptive Scan ................................................................................................................................................5
On-Product Multiple Input Signature Register (OP-MISR):....................................................................7
Embedded Deterministic Test (EDT)...........................................................................................................8
DESIGN CHARACTERISTICS AND THEIR IMPACT ON COMPRESSION........... 10
Compression Limits.....................................................................................................................................10
X Sources......................................................................................................................................................10
Design Flow Flexibility................................................................................................................................11
Area Overhead and Layout Constraints ...................................................................................................11
MANUFACTURING TEST................................................................................................. 12
Impact of Compression on Test Quality ....................................................................................................12
Diagnostics ...................................................................................................................................................12
SUMMARY ........................................................................................................................... 13

REFERENCES...................................................................................................................... 14

www.mentor.com/dft
Copyright © Mentor Graphics Corporation 2007 All rights reserved
This document contains information that is proprietary to Mentor Graphics Corporation. The original recipient of this
document may duplicate this document in whole or in part for internal business purposes only, provided that this
entire notice appears in all copies. In duplicating any part of this document, the recipient agrees to make every
reasonable effort to prevent the unauthorized use and distribution of the proprietary information.
Design-for-Test Whitepaper

Executive Summary
Scan compression has become a necessity for meeting test costs and quality requirements of
today’s nanometer designs. When considering a scan compression technology, several key
factors should be considered in order to ensure that the compression technology does not take
anything away from the existing high quality, low cost test. Some of the key areas are:
- Impact on test quality (test coverage)
- Data and time compression (tolerance to X sources)
- Low pin count testing (to enable multi-site testing)
- Design intrusion
- Area and layout overhead
- Diagnostics and impact on manufacturing flow
This paper examines several proposed and commercial compression solutions and determines the
advantages and limitations of each technology in these areas. An effective compression
technology must maintain the same high quality of test that can be obtained with uncompressed
patterns. Low test quality in terms of coverage and fault models considered, has a direct impact
on test escapes (DPM) and the ability to address yield issues. Additionally, achieving high
compression of test data and time should not be intrusive to the design nor result in large area
and routing overhead. Finally, in order to ensure successful and rapid defect analysis for parts
that fail test on the tester, fast and accurate diagnosis of compressed test patterns is essential.
Although all methodologies address some of these key issues, only TestKompress’ Embedded
Deterministic Test (EDT) technology is able to simultaneously address the requirements of high
data and time compression, ability to handle X states, and suitability of compressed patterns for
diagnostics while maintaining the same high quality test as uncompressed ATPG. TestKompress
can achieve more than 100x compression in test data volume and testing time, in addition to
maintaining high test coverages around 99%.

TestKompress (EDT)
Decompressor Encoding capacity Very High
Coverage loss Very Low
Compactor coverage X Masking None
loss due to Aliasing None
Compressed Pattern Diagnosis Yes
Minimum channels for 10x compression 1
Test Time Reduction 100x
Data Volume Compression 100x

TestKompress can deliver the same high quality test patterns through a single scan channel,
making it more suitable for modular and hierarchical DFT methodologies in large designs that
are pin limited for test. For more details and a summary of each technology’s performance, see
the table at the end of this paper.

The Robustness of Various Test Compression Techniques Page 1

Design-for-Test Whitepaper

Introduction
Why Compression
Originally, companies realized a need for compression because of rising tester costs. The test
pattern data volume exceeded the tester memory, requiring pattern reloads and excessive test
application time. Over time, that need has been supplemented with the necessity to improve test
quality. New fault models and additional test patterns are needed to detect new types of defects
and meet the quality levels of nanometer designs. The undesirable option of pattern truncation
results in lower test coverage and ultimately an increase in defective parts per million (DPM)
that are shipped to customers. Therefore, in order to avoid an increase in test escapes due to low
test quality, the industry has recognized an inevitable need for test pattern compression.

Goals of Scan Compression

Given that the goals of scan pattern compression are to lower tester costs and maintain high test
quality, we need to identify the specific requirements for an effective compression technique.

Test Cost
• Reduce the requirement of scan data memory
• Reduce test application time per part
• Reduce the number of required scan channels
• Reduce simulation time for serial load patterns
Test Quality
• Ability to support and compress all pattern types to fit within tester memory
• Ability to support and compress patterns for several different fault models
• Ability to maintain high at-speed test coverage in the presence of many X sources
• Diagnostics of compressed scan patterns
The key requirement of any compression technology is preservation of high test quality when
compared to standard (uncompressed) ATPG. In practical terms, this means that the same faults
should be detected regardless of the ATPG technique used. Fault detection is achieved by
specifying certain bits within the scan chains to deterministic values and fault simulating the
pattern in order to determine the detection level achieved. In standard ATPG, every scan cell can
be specified to the desired value simply by shifting in the scan cell value. In scan data
compression technology, fewer bits are loaded so the compression hardware must be able to
deliver the desired values to the appropriate scan cell in order to facilitate detection. Therefore,
higher compression can be achieved by compression technologies that have greater encoding
capacities, meaning that they are capable of specifying a larger percentage of the scan cells. We
will utilize the concept of “encoding capacity” in this discussion and examine the design
characteristics which may reduce a compression technology’s ability to specify bits and
consequently compress scan data and test application time.

The Robustness of Various Test Compression Techniques Page 2

Design-for-Test Whitepaper

Compression Technologies
Several technologies have been developed over the years in order to meet the compression goals
outlined in the previous section. We will briefly explore some of the better-known
methodologies and highlight their advantages and disadvantages in the following section.
Virtual Scan
The Virtual scan architecture consists of an Illinois scan type decompressor on the scan chain
inputs and an XOR tree based compactor on the scan chain outputs. Named after its birthplace at
the University of Illinois [3], the Illinois scan decompressor reduces test data volume by
broadcasting the same input data to several scan chains. Figure 1 shows 4 scan chains driven by
the same scan-in pin. If any of these 4 chains require a logic value different from the logic values
on the other 3 chains at the same scan-in cycle, then such a pattern cannot be applied on the
tester. This restriction may result in patterns requiring such faults to be left undetected, resulting
in a need for top-up ATPG using the uncompressed bypass mode. Whenever there is any such
constraint on pattern generation, it will also result in more patterns for the same fault list. The
biggest advantage of Illinois scan decompressor is its simplicity and low area overhead.
The scan chain outputs are compacted using an XOR tree as shown in Figure 1 below.

Scan

Channel 1 Channel 1

Channel 2

Channel 2

Figure 1: Virtual Scan Configuration

Fault effects captured on any of the four scan chains connected to an output channel can be
observed through the XOR tree compactor, as long as the there are no Xs captured on the same
scan_out cycle. Presence of Xs will result in loss of fault coverage, or result in generating more
test patterns. If a fault effect is captured in more than one scan cell, and an even number of those
fault effects appear on the inputs of an XOR tree in the same scan_out cycle, none of them will
be observed on the scan_out channel. This is called fault aliasing, and results in loss of coverage,
or increase in number of patterns.
Table 1 below shows results of regular ATPG with FastScan and the two compression techniques
using TestKompress and Illinois Scan decompressor. The scan chain outputs were observed
directly (without using any XOR tree compactor) in these experiments. In all cases, Illinois scan
shows higher data volume compared to TestKompress.

The Robustness of Various Test Compression Techniques Page 3

 Design-for-Test Whitepaper

Table 1: Experimental results using FastScan, TestKompress and Illinois scan

Normalized Normalized Normalized Normalized

Channel/Chain Test
Design Test Method Test Pattern Scan Compression
Ratio Coverage
Coverage Count Volume Ratio
FastScan 1600/1600 97.06% 96.75% 8215 578M 1.0
A TestKompress 16/1600 96.99% 96.75% 9965 7.7M 75.6
Illinois Scan 16/1600 96.75% 96.75% 14342 10M 57.3
FastScan 1600/1600 99.88% 94.92% 383 25M 1.0
B TestKompress 16/1600 99.86% 95.10% 1086 0.8M 32.1
Illinois Scan 16/1600 95.16% 95.16% 4168 2.7M 9.2
FastScan 1600/1600 99.04% 98.17% 672 59M 1.0
C TestKompress 8/1600 99.03% 98.16% 3232 1.6M 38.1
Illinois Scan 8/1600 98.17% 98.17% 6252 2.8M 21.5
FastScan 2400/2400 98.86% 97.00% 1056 193M 1.0
D TestKompress 12/2400 98.84% 96.96% 4480 4.3M 44.8
Illinois Scan 12/2400 96.95% 96.95% 12799 11.7M 16.5

Advantages of Virtual Scan

• Very simple architecture. No extra logic gates used on scan chain inputs.
• Good compression, very close to the ratio of scan chains to scan channels.

Limitations of Virtual Scan

• High fan-out nets between scan channels and internal scan chains may result in high
routing congestion.
• Linear dependencies between scan cells on the same scan cycle will result in loss of
controllability, leading to loss of coverage in compression mode.
• Top-up patterns in bypass mode may be required to improve test coverage.
• Loss of coverage in the XOR tree compactor due to X values captured in scan cells and
fault aliasing effects.

The Robustness of Various Test Compression Techniques Page 4

Design-for-Test Whitepaper

Adaptive Scan
The adaptive scan technology uses a modified Illinois scan structure at the inputs and a modified
XOR tree structure at the outputs. The decompressor is very similar to Illinois scan, but uses
multiplexers ahead of the scan chains [4,5]. The select data for the multiplexers are also loaded
from the tester allowing some flexibility in the data loaded onto the scan chains from a single
scan channel. Figure 2 below shows one implementation of adaptive Illinois scan.

Ch_in 1
Scan
Ch_out 1
Ch_in 2

Ch_out 2

Ch_out 3

Ch_in 3

Figure 2: Adaptive scan architecture

In the figure shown above, Channel 1 drives the select lines of the multiplexers, and is loaded
from the tester with different values in each cycle. The second and third scan chains can be
loaded with different values in each cycle by a proper selection of values loaded into Channel 1.
A general structure of adaptive scan will consist of one or more levels of muxes in front of every
scan chain, and one or more channels to drive the select lines of those muxes. In order to
improve coverage and compression, adaptive scan configurations require at least one scan input
to drive the mux-select lines and at least two other scan channel inputs.
A simpler approach is to change the mux-select line on a per pattern basis (rather than per cycle)
in order to allow for faster scan chain loading. In this technique, the mux-select line only changes
between pattern loads and remains static during scan shifting. This alternate scheme increases the
restrictions on the ATPG tool and will result in increased pattern count.
The XOR tree compactor on the scan chain outputs is modified [5] to observe certain scan chains
on more than one scan channel. This kind of fan-outs will enable fault effects to be observed on a
second scan channel if it gets masked on the first channel due to Xs, or due to fault aliasing
effects. The side effect of this fan-out structure is that, an X on a scan chain that fans out to more
than one channel will end up masking a lot more fault effects than a simple XOR tree.
Figure 3 below shows the rapid drop in test coverage when using XOR tree based compactors
like the one used in Adaptive Illinois scan architectures. This example used 1600 scan chains
compacted onto 13 scan channels. The effects of the coverage drop can only be minimized by

The Robustness of Various Test Compression Techniques Page 5

Design-for-Test Whitepaper

increasing the number of scan channels, resulting in a reduction of compression that can be
achieved.

100%

80%
Test Coverage

60%

FastScan

40% TestKompress

Adaptive Scan

20%

0%
0.0% 0.5% 1.0% 1.5% 2.0% 2.5% 3.0%
Percentage of Xs in Scan Cells

Figure 3: Effect of X values on test coverage

Advantages of Adaptive Illinois Scan

• Better encoding capacity and reduced linear dependencies on scan cells than regular
Illinois scan.
• Better observability of fault effects through the modified XOR tree compactor.

Limitations of Adaptive Illinois Scan

• Multiplexers and high fan-out nets on scan chain inputs resulting in some area overhead
and routing congestion.
• Some linear dependencies on scan cells may require top-up ATPG patterns to be generated
in the bypass mode.
• Commercial approaches require a minimum of 5 scan channels to reach as little as 10X
compression.
• Changing mux-select with each shift cycle can result in timing issues during shift.
• The XOR tree compactor still does not guarantee detection of all fault effects in the
presence of Xs and fault aliasing effects.

The Robustness of Various Test Compression Techniques Page 6

Design-for-Test Whitepaper

On-Product Multiple Input Signature Register (OP-MISR):

The OP-MISR architecture [1,2] is adopted from the logic BIST principles, but used for testing
devices on a tester (ATE). The architecture of OP-MISR is shown in the Figure 4 below:

Si/So Si/So Si/So

sdir

Scan Chain
Scan Chain

Scan Chain
mske Mask

mrun
mrst MISR

Figure 4: OP-MISR architecture

The key idea of this architecture is to use on-product signature generation to eliminate the
bandwidth and any data volume overhead for external response collection in the ATE while
continuing to supply ATPG test stimuli from the ATE. In the figure 4, Si/So is a bidirectional
scan pin whose directionality is controlled by “sdir”. The control signal “mrun” enables or
disables the MISR feedback connections. For mrun=1 the MISR is configured as a signature
register and compresses the scan chain outputs. When mrun=0 the MISR bits are connected in
series with the scan chains, such that the scan chains and the MISR contents can be unloaded to
the tester through the bidirectional scan pins. The Mask logic along with “mske” from tester will
provide the X logic masking capabilities, required to avoid corrupting the MISR signature.
The output response compaction with MISR can result in about 10-20x compression in data
volume. The use of bidirectional pins for scan chains can double the number of scan chains used
in a traditional arrangement. The resulting scan chains will be half in length, resulting in a test
time reduction of 2x [1,2].
OP-MISR+ [2] uses an illinois scan type of configuration on the input side resulting in better
compression in terms of testing time. OP-MISR+ improves the test time reduction from 2 to 11.

Advantages of OP-MISR
• Deterministic ATPG patterns delivered directly from the tester.

The Robustness of Various Test Compression Techniques Page 7

Design-for-Test Whitepaper

• No test points required, and no loss in coverage due to encoding capacities or linear
dependencies.
• Good compression of data volumes.

Limitations of OP-MISR
• Test time reduction limited to 2x in OP-MISR, and about 11x in OP-MISR+.
• Failure diagnosis requires retesting with MISR bypassed.

Embedded Deterministic Test (EDT)

EDT [6] was developed primarily to reduce the testing time and test data volume on large
multimillion gate designs that need testing for several fault models. TestKompress is the tool that
can generate the decompressor and compactor logic at the RTL level. The architecture consists
of a decompressor and a compactor embedded on the chip. The decompressor drives the scan
chain inputs and the compactor connects from the scan chain outputs. The TestKompress logic is
inserted only in the scan path as shown in Figure 5. In the presence of TestKompress logic we
can have a large number of very short scan chains.

D C
E O
C M
O P
M A
P C
R T
E O
S R
S
O
R

C o m p re s s e d C o m p a c te d
s tim u li A
A TTE
E re s p o n s e s

Figure 5: EDT architecture

Test patterns are generated by targeting individual faults in the core, and are converted to
compressed stimuli and compacted responses to be applied through the TestKompress logic. The
pattern generation algorithm and the fault models used are totally independent of the
TestKompress logic. Since the patterns are generated deterministically, we can obtain very high
coverage of the modeled faults with a very compact set of test patterns. There is also no need to
add any test logic in the core to improve testability, as the patterns are generated
deterministically.
The decompressor consists of a ring generator (unlike a Type I or Type II LFSR) as shown in
Figure 6.

The Robustness of Various Test Compression Techniques Page 8

Design-for-Test Whitepaper

15 14 + 13 12 11 10 + 9 8 7 6 + 5 4 3 2 + 1 0

16 17 + 18 19 + 20 21 + 22 + 23 24 + 25 + 26 27 28 29 + 30 31

Figure 6: Ring generator with polynomial x32+x18+x14+x9+1

The outputs of the ring generator flops will connect to scan chain inputs through a phase shifter
consisting of XOR gates. Creation of the compressed stimuli from a test pattern consists of
solving a set of linear equations based on the ring generator polynomial and the phase shifter
connections. Inputs to the ring generator are driven from the compressed stimuli on the ATE.
The output response compactor consists of an XOR tree and the masking logic, as shown in
Figure 7.

Chain 1
Chain 2
... ...
Chain N

decoder
decoder
pattern mask
data

Figure 7: Automated masking logic with compactor

The masking logic consists of a pattern mask register, decoder, and AND gates before the XOR
tree. The logic values for the pattern mask register are loaded from the compressed pattern data
on the ATE. The masking logic and the XOR tree compactor have the ability to handle any
number of unknowns (Xs) from the scan chains without any modifications to the functional
logic. The masking logic will also eliminate the effects of fault aliasing through the XOR tree.
The decompressor/compactor logic implemented in TestKompress can also perform fault
diagnosis using the same compressed patterns that is used on the ATE.
The TestKompress logic architecture is highly scalable. The ratio of internal scan chains to
external scan channels gives an indication of the possible reduction in testing time and test data
volume, compared to conventional scan testing with deterministic test patterns. There are other
factors that also determine the actual amount of compression that any design can obtain.

Advantages of EDT
• Best encoding capacity.
• Linear dependencies are completely eliminated by the phase shifters.

The Robustness of Various Test Compression Techniques Page 9

Design-for-Test Whitepaper

• A single scan channel can be used to obtain time and data compressions of more than
100x.
• No routing congestion, as there are no high fan-out nets. Modular implementation can be
used for easy block-level implementation.
• All faults that propagate to scan cells are guaranteed to be detected by the automated
masking capability of the compactor, even in the presence of any number of Xs and fault
aliasing.
• No need to generate any top-up ATPG patterns as the test coverage in the compressed and
bypass mode are the same.
• Compressed patterns can be directly diagnosed for failures found on the tester, and there is
no loss of diagnostic resolution compared with bypass mode patterns.

Limitations of EDT
• The ring generator, phase shifter, XOR tree compactor and the x-masking logic contributes
to an area overhead of up to 1% generally.

Design Characteristics and their Impact on Compression

Compression Limits
Is compression ever “good enough”? The limits on test data and test time reduction can be
estimated from the ratio of scan chains to scan channels. The test data reduction also depends on
whether one is using compactor logic, or MISR based compression logic. The actual value of test
data and time reduction depends on various factors, some of which are:
• Encoding capacity of the Decompressor logic
• Handling of unknown logic (X) values and fault aliasing by the compactor logic
• Number of X sources that propagate to scan cells in the design

X Sources
Unlike standard ATPG, any logic that produces unknown (X) states can result in difficulties for
compression technologies. All MISR-based methodologies, such as OP-MISR, require additional
masking logic in order to prevent any unknown states to be captured into the scan chains and
ultimately the MISR. Sources of Xs include un-initialized and uncontrollable flip-flops, bus
contention, floating buses, multiple clock domains, false and multicycle paths, and inaccurate or
incomplete simulation models.
Compactors without X masking are greatly impacted by X sources and will have lower coverage
because they must use bypass (uncompressed) patterns to achieve high test coverage.
Compression may be “good enough” for stuck-at faults but for transition faults, the introduction
of false and multicycle paths introduces many more X sources to the design. Compression will
greatly suffer as a result of the additional X sources. The following table 2 shows the actual
compression obtained by TestKompress logic, in row labeled Comp. for several industry designs
of various types and sizes. The table also shows the maximum number of specified bits and the
number of unknown value sources (#Xs) within the design. The the row labeled Channels:Chains
indicates the number of channels and chains used in that design.

The Robustness of Various Test Compression Techniques Page 10

Design-for-Test Whitepaper

Table 2: Experimental results with TestKompress logic

Design Flow Flexibility

TestKompress, incorporating the EDT technology, provides very flexible flow options. Given a
scan inserted netlist, the RTL TestKompress logic can be created by specifying the number of
external channels. The TestKompress logic can also be created early in the design flow by
simply defining the number of scan chains and the clocking for the first and last scan cells in
every chain. The TestKompress logic is pattern independent and need not be recreated following
ECOs as long as the number of chains and channels are not changed. During the TestKompress
RTL logic creation phase it is also possible to perform compressed pattern generation if the
synthesized design core is available at the gate level.
TestKompress logic can also be inserted at the block level, independent of other blocks in the
design, and not requiring any top level logic. The flow can be simplified as TestKompress can
have as few as one scan channel for each block. Modular TestKompress allows us to run ATPG
in parallel for all the blocks that have TestKompress logic as well as any uncompressed scan
chains, without requiring wrapper chains around blocks. In addition, TestKompress can be
implemented in a hierarchical manner where each block is tested individually using top level
muxing logic. Modular TestKompress has been successfully used in large industrial designs with
a volume production of 10 million chips [12].

Area Overhead and Layout Constraints

Another concern for designs with compression technology is the additional area overhead
required. If area overhead exceeds established limits, it may be required to move to the next
larger die size but that can be quite costly. It is important to note that area is effected by two
aspects, gate count, and routing area.
With regards to compression, difficulties can be encountered because many scan chains must be
connected to the decompression and compression logic resulting in huge fan-out nets.
The decompressor/compactor logic generated by TestKompress requires a small number of
simple gates for every scan chain in the design. The routing overhead is very minimal as there is
no huge fan-out within TestKompress logic or with its connections to scan chains. The routing
overheads (if any) can be minimized by either segmenting the decompressor/compactor, or using
modular TestKompress.

The Robustness of Various Test Compression Techniques Page 11

Design-for-Test Whitepaper

The Virtual scan architecture does not have any logic gates on scan chain inputs, but has a huge
fan-out net from every scan channel to the scan chains, resulting in a large area overhead due to
routing. The adaptive Illinois scan adds a multiplexer in front of every scan chain, in addition to
the huge fan-out nets and the routing overhead.
The area overheads due to the XOR tree-based compactor architectures depend on the number of
XOR gates added on the scan chain outputs. The adaptive scan compactor logic also has huge
fan-out nets from each scan chain output to multiple points in the XOR tree.

Manufacturing Test
Impact of Compression on Test Quality
All the test compression techniques discussed above rely on the fact that not all scan cells need a
defined logic value to detect any fault. The defined logic values on scan cells is also referred to
as a test cube for the targeted fault. As long as the number of defined scan cell values is a small
percentage of the total number of scan cells, most of the decompressor logic options discussed in
this paper will be able to deliver those values without much impact on test coverage and the
number of test patterns. The undefined scan cells will be defined by the values that can be
delivered by the decompressor logic used. The ability of the decompressor to deliver any set of
values on the defined scan cells is also known as the encoding capacity of the decompressor.
Decompressors with low encoding capacity will not be able to deliver certain test cubes required
to detect a fault, resulting in lower test coverage, and requiring the application of test patterns in
the compression bypass mode.
In addition to controllability, an effective compression methodology must be able to observe the
circuit responses through the compactor logic. The ability to handle a large number of unknown
states while maintaining high observability will greatly impact the compression solution’s ability
to achieve high test coverage while maintaining a high compression ratio.
Of all the compression technologies discussed in this paper the TestKompress decompressor
(ring generator) has the highest encoding capacity while the patented masking logic in
TestKompress provides for guaranteed observability of fault effects. This has been demonstrated
on hundreds of designs which show no loss in test coverage when generating highly compressed
patterns, compared to generating patterns in bypass mode.

Diagnostics
Diagnosis is the process of identifying defects or fault locations that resulted in failures caught
on the tester. The test patterns generated for different fault models are applied to the
manufactured device on an Automatic Test Equipment (ATE). If the measured logic value on the
device outputs does not match with the expected logic values in the test patterns, it is due to a
manufacturing defect. Diagnosis tools like YieldAssist can identify the fault locations that
resulted in those failures. Inputs to YieldAssist could be regular ATPG patterns, or compressed
ATPG patterns from TestKompress.
YieldAssist analyzes (by simulation) the actual responses from the tester and determines which
faults might have caused the observed failure effects. This method does not require building a
huge database fault dictionary, and is very effective in diagnosing both compressed and
uncompressed test patterns [7].

The Robustness of Various Test Compression Techniques Page 12

Design-for-Test Whitepaper

In an experiment on 70 manufactured devices and 10K failing patterns, it was found that 68 out
of 70 devices had perfect or good matching behavior between TestKompress and regular
uncompressed pattern diagnosis [8]. Several other experiments have shown that the diagnostics
resolution for direct compressed pattern diagnosis is at least as good as the diagnostic resolution
achieved in bypass mode with uncompressed patterns [8]. YieldAssist enabled engineers at
TSMC to improve the yield by 24% for one device. The diagnosis was completed in 2 days,
significantly improving the failure analysis productivity [11].

Summary
In this paper we discussed several test data and test time compression techniques. All the
techniques require on-chip decompression and compaction hardware adding to area overhead in
terms of logic gates or high fan-out routing nets or both. Any MISR based compression
technique, like the OP-MISR, also requires additional logic to block Xs from propagating to the
MISR, and bypass mode re-testing for failure diagnosis. Some of the compressors like adaptive
Illinois scan require more than one scan channel to achieve any compression of test data and
time. As shown in the section on adaptive scan (Figure 3), Xs captured on scan cells will result in
significant loss of coverage, requiring a top-up ATPG in compression bypass mode. Unknown
(Xs) will always be present in designs with false and multicycle paths, and cross clock domain
logic. Any pattern generated in bypass mode will result in an overall reduction of total
compression obtained.
The following Table 3 gives a quick overview of the topics discussed in the paper.
Table 3: Comparison summary of compression techniques discussed
Virtual Scan Adaptive OP MISR+ TestKompress
Scan (EDT)
Decompressor Encoding Low Moderate Moderate Very High
capacity
Coverage Moderate Low Low Very Low
loss
Compactor X Masking High* High* Very High* None
coverage loss
due to Aliasing High High Very High None

Top-up ATPG Yes Yes Yes No

Area Overhead Low Moderate Moderate Moderate
Compressed Pattern Possible Possible No Yes
Diagnosis
Minimum channels for 10x > 5 >5 >5 1
compression
Test Time Reduction 20x 50x* 20x 100x
Data Volume Compression 20x 50x* 100x 100x

The Robustness of Various Test Compression Techniques Page 13

Design-for-Test Whitepaper

* results may suffer significantly when using at-speed test due to Xs introduced with false and multicycle
paths, and cross clock domain logic.

As shown, the EDT based TestKompress logic can achieve very high compression of test time
and data even with a single scan channel, making it more suitable for modular compression
methodologies and very low pin count testing. Direct compressed pattern failure diagnosis is
possible with XOR tree based compactors. TestKompress effectively works with YieldAssist to
provide direct diagnostics of compressed patterns and is proven in production lines. For some
more information on different compression techniques please refer to [9] and [10].

References
1. Carl Barnhart et al., “OPMISR: The foundation for compressed ATPG vectors”, Proc.
Int’l Test Conf., pp. 748-757, 2001.
2. Carl Barnhart et al., “Extending OPMISR beyond 10x scan test efficiency”, IEEE Design
and test of computers, pp. 65-73, Sept-Oct 2002.
3. I. Hamzaoglu and J. H. Patel, “reducing test application time for full scan embedded
cores”, Proc. IEEE Int. Symp. on fault tolerant computing, pp. 260-267, 1999.
4. Samitha Samaranayake et al., “A reconfigurable shared scan-in architecture”, Proc. IEEE
VLSI test symposium”, 2003.
5. ESNUG article No: 442 item 11 on www.deepchip.com,
http://www.deepchip.com/items/0442-11.html

6. Janusz Rajski et al., “Embedded deterministic test”, IEEE Trans. on CAD of ICs and
systems, vol. 23, No. 5, pp. 776-792, May 2004.
7. Wu-Tung Cheng et al., “Compactor independent direct diagnosis”, Proc. Of the 13th
Asian test symposium, pp. 204-209, Nov 2004.
8. Andreas Leininger et al., “Compression mode diagnosis enables high volume monitoring
diagnosis flow”, Proc. Of Intl. test conference, session 7, paper 3, 2005.
9. Tom Lecklider, “Test pattern compression saves time and bits”, Evaluation engineering,
July 2005.
10. Michelle Lange, “Adopting the right embedded compression solution”, Evaluation
engineering, May 2005.
11. Will Hsu, Y. S. Chen, Wu Yang, Wu-Tung Cheng and Nagesh Tamarapalli, “Scan
Diagnostics in the Nanometer Design Era”, Semiconductor Manufacturing Magazine,
March 2006.
12. http://www.mentor.com/products/dft/upload/Renesas%20_TK_cust_success.pdf

The Robustness of Various Test Compression Techniques Page 14