You are on page 1of 6

Design Techniques and Tradeoffs in Implementing Non-destructive Field Test

Using Logic BIST Self-test


Amit Dutta, Malav Shah, Swathi G. and Rubin A. Parekhji
Texas Instruments, Bangalore, India.
Email:[amitd,m-shah1,swathig,parekhji]@ti.com
Abstract BIST self-test. Different architecture options are
Periodic testing of electronic devices on the field during reviewed and issues in interfacing the DUT with the test
application execution is becoming increasingly logic for supporting non-destructive field test are
important. In addition, some of these applications are discussed. Various tradeoffs in obtaining the desired test
embedded and real-time, requiring the system to be coverage in a given time interval are examined for
operational for extended periods. In such applications, different test data volumes. Some experimental results
field test must be cleverly interleaved with normal are presented for implementations on an IP core which is
operation, such that the latter is not impacted, while at currently being designed. It is shown thereby that a
the same time guaranteeing the correct operation of the proper choice of the test architecture, test schedule and
device, and identification of any malfunction or defects DUT interface can significantly improve the quality of
within a reasonable time. This paper discusses the design the field test solution. While the tradeoffs are illustrated
techniques and tradeoffs in implementing non-destructive and computed for a specific IP core design with a certain
field test using logic BIST self-test. Three specific time span (window) available for functional and test
designs aspects are discussed, namely (i) choice of logic mode operation, the methods described in this paper are
BIST self-test architecture, (ii) optimizations in test time generic and can be adopted on other design modules as
and additional memory requirements for attaining a well which require such form of field test support. They
given coverage, and (iii) DUT interface to self-test DFT also highlight the advantages and flexibility of adopting
logic to enable such form of test and application traditional logic BIST based self-test for the purpose of
interleaving. Data is presented for an IP core, which is field test, as also some limitations of such techniques in
presently being designed in Texas Instruments (India), meeting very high coverage requirements.
wherein such tests are being supported for use in
This paper is organized into six sections. Section 2
automotive applications.
provides the motivation for the non-destructive field test,
Keywords: Non-destructive test, field test, built-in self-test, and the associated application requirements. Methods to
logic BIST, system test. interleave functional mode and test mode of operation
1 Introduction are discussed in Section 3, together with specific design
and implementation considerations. Various architecture
Safety critical applications require that the hardware on options and their tradeoffs are presented in Section 4.
which they execute be fault free. While traditionally Different seed generation techniques are presented in
manufacturing testing with adequate life-time stress tests Section 5 together with experimental results. Section 6
has been adequate to guarantee the reliability of the concludes the paper.
electronic devices, such form of one-time testing is no
longer adequate for newer process technologies due to 2 Motivation
process marginalities and degradation. As a result, For the applications being considered, there are a few
periodic testing on the field during application execution important field test considerations to be met for SOCs
is becoming increasingly important [1]. Examples of being used in automotive systems.
such applications include medical electronic and
automotive electronic systems. (The latter are governed (a) The application is real-time and embedded. As a
by the SIL3 standard [2] to ensure functional correctness result, the functional mode of operation is always
by safeguarding against life-time reliability issues). The on. Hence offline testing is not an option. The test
embedded and real-time nature of these applications mode must hence be interleaved with the functional
requires them to be operational for extended periods. As operation. This implies that the two modes of
a result, it is necessary to interleave the normal operation be mutually non-destructive, and requires
functional operation with the test operation. In such context switching. Also, the window of operation for
situations, the field test may have to be performed across these two modes is not fixed. Hence the test
several intervals of normal application execution, scheduler must also support variable test execution
resulting in the need for a non-destructive field test time.
mechanism. (b) While it is desirable to concurrently perform the
This paper describes the design techniques and tradeoffs dual operations demanded by the application and
in implementing non-destructive field test using logic required for test, it is so possible only if the DUT

978-1-4244-4595-0/09/$25.00 
c 2009 IEEE 237
has natural partitions wherein selective portions of combines the advantages of logic BIST together
the logic can be manipulated in test mode, without with re-seeding [5,6]. (More details are provided in
disturbing the state of the remaining portions. This Section 4. The same solution is also used for
also requires a complex control mechanism to production test on previously designed cores and
switch between these portions of the DUT. For the SOCs, and hence there is no additional hardware
sizes of the modules being considered, such overhead). The DBIST engine can be driven to
partitioning was considered infeasible. Hence while generate patterns based on desired tradeoffs in test
test is performed online, the two modes of operation coverage (for a given test time and number of
are mutually exclusive. seeds), test time (for a given test coverage and
(c) The application takes priority over the test mode of number of seeds) and seed storage memory (for a
operation. Hence the test schedule must be created given test coverage and test time).
to provide a specified coverage across various
windows of test mode operation. For the same It is important to note that logic BIST (and DBIST)
reason, while there is no requirement on the latency techniques use the STUMPS architecture which naturally
incurred to report the detection of a fault, it is also provides test data volume compaction. While
expected that the application can roll-back to a safe compaction is indeed desirable since it directly
state on the first indication of such detection. As a influences the on-chip memory requirements, it also
result, the fault detection status is reported at the end forces the test patterns to be dense, i.e. with fewer dont
of each window of test operation. care-bits, so as to meet the overall entropy requirements
[7]. As a result, the seeds used in such re-seeding
Given the above requirements, the decision to support
techniques do not have many non-care-bits, and
field test based on logic BIST self-test techniques was
generation of new seeds from existing seeds is not easy.
driven by the following important observations:
Several re-seeding schemes are discussed in the
(i) High structural coverage is possible per window of literature [8,9,10,11,12,13]. However, in this paper, we
test operation, as compared to functional coverage. have considered a commercial BIST engine which
Such coverage is measurable, and can also be supports re-seeding, where a given seed is used to
selectively directed or augmented as required in generate a certain (variable) number of patterns, before
different windows of operation. Also, it is possible being replaced with a new one. (Refer to Section 5).
to create a test schedule where some portions of the
DUT are tested more often or on higher priority than 3 Architecture Considerations for Field Test
the others. (While this is possible using functional As explained previously, periodic field test is necessary
patterns [3], such forms of directed, selective and in safety critical applications. A power-on field test tests
improved coverage are more easily driven through the device only when it is powered up. (Schemes for
structural ATPG techniques. Besides scan overhead such form of logic and memory self-test are discussed in
has to be incurred to support context save and [14]). The solution proposed here is based on a hardware
restore given the homogeneity of the IP module BIST technique which can be periodically triggered
being considered). during an application execution. The technique must be
(ii) Logic BIST offers the advantages of a built-in non-destructive since the functional and test modes are
decompressor (PRPG) for generating test patterns interleaved. (In this paper, field test for logic alone is
and a built-in compactor (MISR) for capturing the considered, since there are different architectural
DUTs (with STUMPS architecture) response to the tradeoffs. Field test for memories can be implemented
test patterns [4]. As a result, both the test hardware using existing memory BIST techniques).
requirements and test data storage requirements are
minimized. However, a given sequence of logic 3.1 Sequence of Operation
BIST patterns generally provides high coverage Fig. 1 shows the field test controller block diagram and
improvement for the first few patterns, and then the architecture. The master controller controls the field test
coverage improvement rapidly tapers off. Higher activity. The context save/reload controller manages
coverage is possible only through re-seeding DUT content during field test. The self-test engine is
techniques. The re-seeding mechanism also provides based on deterministic built-in-self-test (DBIST) which
a fit into a window based test scheduling provides scan data to DUT. Fig. 2 depicts the steps
mechanism, wherein a new seed can be loaded at the involved in executing non-destructive field test.
start of a window and the DUT response can be 3.2 Interleaving Functional and Test Operations
compared at the end. There are several design considerations which must be
(iii) Deterministic BIST (DBIST) from Synopsys has met to optimally interleave test operation with functional
been considered for primary structural test pattern operation. These include:
generation and application mechanism, since it
(a) Time slots in the application firmware which are

238 2009 15th IEEE International On-Line Testing Symposium (IOLTS 2009)
available for test must be identified without frozen during the test mode. (iii) Reset and NMI
compromising on the response time of the embedded (non-maskable interrupt) inputs to the DUT should
application. A representative scheme is shown in not be gated off, thereby enabling the application to
Fig. 3. regain control of the DUT. (iv) The state of non-scan
(b) Different slots may not have identical duration and flip-flops must be preserved adequately during the
may not be equally spaced, given the real time test operation, since scan based context save and
nature of the application. Hence both the application restore is not possible. (v) There should be no
firmware and test schedule must interface with each pending instruction in the CPU pipeline or any
other at the window boundaries to save and restore pending request or acknowledgement for the DUT
the application as well as the test context of the from any external module before starting field test
DUT. For example, the application must resume at operation.
Time A2, from where it was halted at Time A1. (d) The DUT state (context) can be saved using scan
Also, the test schedule must resume at Time T2, shifts onto on-chip memory at the end of application
from where it was halted at Time T1. The best window. At the end of the test window, it can be
implementation requires the minimum number of similarly restored. At the end of test window, the
test intervals and minimum test time, (? T over all pass / fail status is generated by comparing the
intervals) without impacting the throughput as well signature in the MISR with the expected one. A new
as the application time ( ? A over all intervals). technique is proposed to ensure that the context
Field test master
restore is correctly done without any cycle mis-
controller matches or state corruption. Two MISRs are
incorporated at the DUT scan inputs and outputs
Context
(Refer to Fig. 4). The MISR at the scan outputs is
Self-test engine (BIST Context
+ control logic) save/reload save activated during context save and that at the scan
controller memory inputs is activated during context restore. A mis-
Scan out
match in the signature in the two MISRs indicates
that the restore operation was corrupted. (This points
Rest of DUT to incorrect data being read from the on-chip
memory used for context data storage). An
CPU additional cycling of the data in the STUMPS
Self-test or
context reload DUT channels can also be done to ensure that the value
shifted into the STUMPS channels of the DUT
Fig. 1: Architecture for non-destructive field test. during the restore operation is also correct.
Application running Application on Context reload/Self-test
L
F Scan in Scan out
No Idle DUT Context reload DUT
S Memory
Yes R (scan chains)
Configure field- Store result
MISR MISR
test controller
Yes
Self-test complete ?
Isolate DUT No Fig. 4: Two pass sanity check.
4 Implementation and Evaluation of Logic
Run
Context save self-test BIST Techniques
Fig. 2: Flow diagram for non-destructive field test. This section describes the implementation choices
available using DBIST to implement logic BIST.
Normal application Test application Various schemes for field test using logic BIST are
analyzed.
4.1 Implementation Considerations
A1 T1 A2 T2 A3 T3 A4
The non-destructive self-test mechanism is being
implemented for an IP core at 90nm process technology
Fig. 3: Interleaving application and test execution cycles. node, operating at 130MHz, as part of an SOC being
(c) There are other IP core specific requirements: (i) targeted for automotive applications. This core has a total
During test, the DUT I/Os are bound to prevent of 3500 flip-flops, which are re-grouped into a STUMPS
external inputs from impacting DUT test, and DUT architecture with 512 scan chains with an average scan
outputs from impacting other modules. (ii) Similarly chain length of 7 flip-flops.
the contents of the memory inside the DUT must be

2009 15th IEEE International On-Line Testing Symposium (IOLTS 2009) 239
As discussed in Section 2, re-seeding based DBIST from For a given coverage requirement, what is the
Synopsys is used for the implementation of field test. maximum test time affordable and minimum test
Fig. 5 shows the DBIST architecture. It has been time required.
augmented in multiple forms to support self-test [15,16]. For given bounds on test time and coverage, (i) how
The DBIST tool supported configuration has a PRPG many seeds are required, (since the methodology is
(and shadow register) of 257 bits, an MISR of 128 bits, based on expanding seeds to obtain as high field
and 256 / 512 STUMPS channels. The reduced STUMPS coverage in as minimal a time as is possible), and
length does not influence the result of the test since the (ii) how are the seeds generated and loaded in the
limits are set by the entropy bounds [7]. On the other core logic which is being tested.
hand, it is efficient in terms of test time compression
since with 512 scan chains, the chain length is reduced. Importantly, these options bring out the optimality trade-
The support for shadow register and MISR make DBIST offs in seed storage memory, test time and test coverage
a better candidate for field test. The shadow register is based on the various design and application constraints,
used to load a new seed at the same time that the existing resulting in different implementation scenarios.
seed is being unrolled into a pattern that is being shifted Based on the application, it has been determined that
into the STUMPS channels. Higher the width of the about 10% of the functional MIPS can be utilized for
shadow register inputs, the lesser is the seed loading test, and based on the firmware architected, an
time. Hence 12 shadow register inputs (maximum application window is about 40s. Accordingly, for each
permissible) have been chosen. (These are internal such window, 36s are used for functional operation, and
signals. For production testing with multi-site 4s for test operation. At 130MHz operation, this
capabilities, fewer pins are desirable, and hence fewer corresponds to about 520 cycles. An analysis was done
shadow register inputs may be desirable). A 257 bit seed on various feasible options with bounds on seed storage
is therefore composed of 12 segments each of 22 bits. memory, coverage and test time. The available 520
The MISR is a 128 bit signature register which is read by cycles per test window need to be shared amongst the
a 16 bit CPU bus. Thus, it takes 8 cycles to copy the following steps of self-test. (Refer to Table 1):
signature into internal memory for comparison. Several (i) Context save and restore operations: For IP core
enhancements have been made to the DBIST controller under consideration with STUMPS length of 7 flip-
to support field test. These include re-seeding through flops, context save and restore will require 7 shift
the device internal interface to the DBIST controller, cycles each to shift out all 3500 flops. Since an extra
programmable pattern counter for each seed, internal sanity check phase (as explained in section 3.2(d) is
clock selection, signature storage and comparison, etc. involved, additional 7 cycles are required, making a
Shadow scan in total of (3*7) = 21 cycles.
BIST_cl
(ii) New seed loading: This requires ~22 cycles (257 /
Shadow register BIST
12) for loading a single seed and 22*S cycles where
controller S is the number of seeds required per window.
PRPG (iii) Pattern shifting: Shifting the test pattern into the
STUMPS channels requires 7*P for shifting P
Phase shifter patterns.
(iv) Response capture: IP has 16 bit width data bus.
DUT
Core_Clk Unloading of 128 bit MISR (signature) register at
the end of each window (slot) hence takes 8 (128 /
16) cycles. The MISR shift out is done only once per
window to reduce the MISR storage memory and
Compactor
also to optimize on test time. This results in a
MISR detection latency of 4s in the worst case.
MISR_scan_out Based on these cycle count numbers, the general
Fig. 5: DBIST architecture.
equation to determine the number of patterns that can be
4.2 Evaluation of Different Schemes for Field Test applied for each of the five schemes in Tables 1 and 2
The analysis in this section gives an insight into how the can be given as:
logic BIST engine can be employed for performing the 21 (context save / re-load cycles) + 22*S (seed loading
non-destructive self-test at regular intervals during the cycles) + 7*P (scan shift cycles) + 8 (MISR unload
application. Such an analysis helps to make smarter cycles) =520
selections such as:
For a given bound on test time, (i.e. number of where S is the number of seeds per window and P is the
intervals * time per interval), what is the minimal number of patterns that can be applied in a given
coverage desirable and maximal coverage attainable. window. For parallel loading of new seed and pattern

240 2009 15th IEEE International On-Line Testing Symposium (IOLTS 2009)
scan shift, only one from the terms (ii) and (iii) of the for 85% coverage requires 150 seeds, and then using
above equation, whichever is greater, is applicable. Scheme B for 95% coverage requires another 2267
(Refer to methods D and E discussed below). seeds, resulting in a total of 2417 seeds. On the other
hand,3584 seeds are required with Scheme B standalone.
Five different schemes are possible with tradeoffs in seed
Table 1: Pattern and seed computation for different
storage memory, test time and coverage: techniques.
A. Load one seed per window and generate as many Number of cycles
Schemes P S
patterns as possible per window. Seed loading into (i) (ii) (iii) (iv)
shadow register and pattern shifting into the DUT A 3*7 22*S 7*P 8 67 1
are done serially. This is a simple implementation B 3*7 22*S 7*P(=S) 8 16 16
C 3*7 22*S 7*P(=2S) 8 26 13
for steadily improving coverage. D 3*7 22*S NA (P=S) 8 22 22
B. Load a new seed for every pattern. Seed loading and E 3*7 22*S NA(P=2S) 8 44 22
pattern shifting are done serially. This Table 2: Tradeoffs with different techniques.
implementation gives fast coverage. Memory Time
Schemes Coverage Seeds Windows
C. Load a new seed for every two patterns. Seed (Kbits) (ms)
loading and pattern shifting are done serially. The 75% 16 16 6.16 0.84
results of this implementation are between (A) and 80% 46 46 17.71 2.36
A
90% 555 555 213.67 20.84
(B) above. 95% 1782 1782 686.07 71
D. Parallel (Concurrent) loading of new seed and scan 75% 192 12 50.88 0.48
shift. Load a new seed for every pattern. 80% 352 22 93.28 0.88
B
E. Parallel (Concurrent) loading of new seed and scan 90% 1440 90 381.6 3.6
shift. Load a new seed for every two patterns. 95% 3584 224 949.76 8.96
75% 131 11 35.08 0.44
For these five schemes, Table 1 lists the number of C
80% 242 19 64.63 0.76
patterns (P) which can be applied and number of seeds 90% 1044 81 278.676 3.24
(S) required per window. The number of cycles required 95% 3046 235 812.9 9.4
75% 192 9 50.5 0.36
for each of the four steps described above is also shown. 80% 352 16 92.51 0.64
Table 2 lists the overall number of patterns and seeds D
90% 1440 66 378.53 2.64
required for obtaining a given coverage, and the 95% 3584 163 941.95 6.52
associated test time and seed storage memory 75% 131 6 34.44 0.24
requirements. The number of windows denotes the 80% 242 11 63.6 0.44
E
90% 1044 48 274.45 1.92
number of test slots required for self-test during the 95% 3046 139 800.61 5.56
application execution. The memory required includes the
seed and result (MISR) storage and excludes the context Using some of the novel techniques discussed in the next
save memory of 3500 bits (this is a constant, irrespective section, the overhead of seed storage using the additional
of the scheme selected). on-chip memory can be further alleviated.
As an example, for Scheme B, a new seed is loaded for 5 Handling Seed Generation and Loading
every pattern. As shown in Table 2, 192 seeds are
The seeds for logic BIST self-test can come from the
required to reach 75% coverage. Since the seeds per
following sources:
window is 16 it will take 12 (192 / 16) windows of
On-chip memory (RAM): The ATPG tool generated
application to reach this coverage. The total test time is
0.48ms (12 windows * 40s). The total memory required seeds can be stored in the on-chip memory. These
is 49.344 Kbits (257 bits per seed * 192 seeds + 128 bit seeds are loaded into the BIST engine during self-
MISR * 12 windows). test execution. As seen in Table 2, this requires
additional memory for test purposes.
The results in the Table 2 highlight the various tradeoffs Through application firmware: Instead of storing
that can be exploited to meet the requirements of seeds on chip, they can also be provided externally
coverage, test time and storage memory. For example, a through application firmware. Only the seeds to be
given coverage can be obtained with lower (higher) test used in a given window of execution must be
time with more (fewer) seeds. The overhead in terms of downloaded in a cache (RAM).
test time and seed memory can be suitably controlled to On-chip generator: This is described in Sec. 5.1.
obtain higher coverage. Combinations of the schemes
can also be used based on coverage needs and memory 5.1 On-chip Seed Generation Techniques
available on chip. A better scheme could be to use On-chip seed generation techniques can be used to attain
combinations of Scheme A for achieving a certain the desired coverage [12,13]. A small sub-set of seeds
coverage, followed by Scheme B, to save seed memory are stored on-chip and the rest of the seeds are derived by
requirements. Experiments show that using Scheme A various seed manipulation techniques.

2009 15th IEEE International On-Line Testing Symposium (IOLTS 2009) 241
Combinational operations: These include arithmetic combinational operations, nine simple logic operations
and logic operations like bit inversion, pair-wise and are chosen to be performed on the initial stored seeds.
group-wise bit rotation, or Boolean operations on This requires only ~700 2-input equivalent NAND gates.
pairs and groups of bits, (e.g. XORing groups of The equivalent memory overhead to store the number of
adjacent bits). seeds will be much larger.
Counter approach: A counter is initially loaded with The above techniques can help to reduce the number of
a stored seed. Once the counter runs for few cycles, seeds required to be stored on-chip to obtain coverage
the new counter state can be used as another seed. A upto a certain number. However, for coverage in excess
large seed (e.g. 257 bits) can also be split into of 87%-90%, deterministic seeds generated from the
smaller sets (e.g. 8 bits each), such that each set of ATPG tool may be required to be stored.
registers can be operated as counters independently.
LFSR approach: An additional LFSR can be used to 6 Conclusion
generate new seeds based on another seed. This is
This paper reviews the design implementation techniques
the classical logic BIST technique.
and tradeoffs incurred in implementing field test using
In this work, various combinations of the above
traditional logic BIST self-test techniques. Different
techniques have been considered, and their effectiveness
schemes for self-test for coverage improvement, test time
in terms of coverage improvements, test time
reduction and test data storage reduction are considered.
requirements, and reduction in the number of external
Experimental results are presented for an IP core in
seeds required is ascertained. With every one seed taken
which such support is being incorporated. The
from ATPG tool, nine new seeds are generated using
advantages and limitations of different schemes are also
these approaches. The results are shown in Table 3. The
highlighted. The paper suggests a framework together
second column gives the coverage results with the tool
with recommendations on how logic BIST self-test can
generated seeds, whereas the other columns give the
be extended to support field test. Future work in this area
coverage taking these seeds as well as internal
includes optimizations in test data volume and additional
derivatives of these seeds using above techniques. These
on-chip seed generation techniques so as to reduce the
results are obtained using stuck-at fault model.
overall memory requirements as also the test time.
As can be seen from Table 3, using a counter approach,
References
82%-83% coverage can be obtained with just 40 seeds, [1] E.J. McCluskey and S. Mitra, Fault-Tolerance, in Encyclopedia
as against 100 seeds required in the ATPG process. The on Computer Science and Engineering, CRC Press, 2004.
[2] International Electromechanical Commission, IEC-61508
graph in Fig. 6 shows how much improvement in Standard on Safety Integrity Level (SIL) for Automotive Devices,
coverage can be obtained for a given memory size, using http://www.iec.ch
[3] D.Gizopoulos, A.Paschalis and Y.Zorian, Embedded Processor -
on-chip seed generation techniques. Based Self-Test, Springer, 2004.
Table 3: Coverage results with on-chip seed generation [4] C.Stroud, A Designer's Guide to Built -In Self-Test, Kluwer
Academic Publishers, 2002
techniques. [5] P.Wohl, J.Waicukauski, S.Patel and M.Amin, "Efficient
ATPG Coverage Compression and Application of Deterministic Patterns in a Logic
tool ATPG BIST Architecture", Design Automation Conf., 2003.
Combinational Counter LFSR [6] Synopsys, Inc., SoCBIST User Guide, Version Z-2007.03, 2007.
generated tool [7] S.Alampally, J.Abraham, R.Parekhji, R.Kapur, T.W.Williams,
operations Approach Approach
seeds results Evaluation of Entropy Driven Compression Bounds on Industrial
10 71.33% 79.25% 79.47% 79.02% Designs, Asian Test Symp., 2008.
20 74.79% 80.95% 81.25% 81.01% [8] B.Koenemann, LFSR-coded test patterns for scan
design,European Test Conf., 1991.
30 76.62% 81.93% 82.28% 82.03% [9] C.Krishna and N.Touba, Reducing test data volume using LFSR
100 82.80% 85.65% 85.91% 85.78% reseeding with seed compression, Intl. Test Conf., 2002.
87% [10] A.A.Al-Yamini, E.J.McCluskey, Built-in reseeding for serial
BIST, VLSI Test Symp., 2003.
85% [11] Z.Wang, K.Chakrabarty, and M.Bienek, A seed-selection
83% method to increase defect coverage for LFSR-reseeding-based
test compression, European Test Symp., 2007.
Coverage

81%
[12] E. Kalligeros, X. Kavousianos, D. Bakalis and D. Nikolos, On-
79% the-Fly Reseeding: A New Reseeding Technique for the Test-per-
77% Clock BIST, Journal of Electronic Testing: Theory and
Direct ATPG tool Applications, June 2002.
75% [13] E. Kalligeros, X. Kavousianos and D. Nikolos, Multi -phase
73% On-chip seed genration
BIST: A New Reseeding Technique for High Test Data
71% Compression, IEEE Trans. on CAD, Oct. 2004.
[14] A.Dutta, S.Alampally, A.Kumar, R.Parekhji, A BIST
0 20 40 60 80 100 Implementation Framework for Supporting Field Testability and
Number of seeds Configurability in an Automotive SOC, Workshop on
Dependable and Secure Nanocomputing, 2007.
Fig. 6: Tradeoffs in coverage improvement with on-chip [15] S.Jain, J.Abraham, S.Vooka, S.Kale, A.Dutta and R.Parekhji,
Enhancements in Deterministic BIST Implementations for
seeds and ATPG tool generated seeds. Improving Test of Complex SOCs, Intl. Conf. VLSI Design,
The area required for the implementation of on-chip seed 2007.
[16] A.Dutta, S.Alampally, Prasanth V. and R.Parekhji, DFT
generation techniques is far less compared to the seed Implementations for Striking the Right Balance between Test
storage memory. For a sample implementation of Cost and Test Quality for Automotive SOCs, Intl. Test Conf.,
2008.

242 2009 15th IEEE International On-Line Testing Symposium (IOLTS 2009)

You might also like