Launch-on-Shift-Capture Transition Tests

Transcription

1 Launch-on-Shift-Capture Transition Tests Intaik Park and Edward J. McCluskey Center for Reliable Computing, Stanford University, Stanford, USA Abstract The two most popular transition tests are launch-on-shift (LOS) test and launch-on-capture (LOC) test. The LOS and LOC tests differ in their launch mechanisms, creating their own pros and cons. In this paper, new hybrids of LOS and LOC tests that launch transitions using both launch mechanisms are introduced. The new transition tests improved fault coverage without significant test length penalty. This paper presents the concepts and pattern generation methods of these new transition tests as well as experimental results that demonstrate the benefits of these tests. places timing requirements on the routing of scan-enable signal. Figure 1. The concept and waveforms of LOS 1. Introduction The transition test [Waicukauski 87] is one of the most widely used techniques to ensure the correct temporal behavior of the manufactured integrated circuits (ICs). It consists of a pair of vectors (V 1,V 2 ). The first vector, V 1, initializes a logic value at a fault site (a node in a network). Then, the second vector, V 2, launches a transition of logic values (0 1 or 1 0) at the fault site and propagates the transition to an observable point (a scan flip-flop or a primary output). Transition tests are categorized by how they launch transitions: launch-on-shift (LOS), launch-on-capture (LOC), and enhanced-scan transition tests. LOS and LOC tests do not require any additional hardware while the enhanced-scan transition test requires special types of scan flip-flops to apply the test vectors [Dasgupta 81]. The enhanced-scan transition is not considered in this paper; the focus is only on the first two types of transition tests. The launch-on-shift (LOS) test launches a transition of a logic value by the last clock pulse of the scan shift operation [Eichelberger 91, Savir 92], followed by a system clock pulse that captures the transition. Figure 1 illustrates the concept of LOS testing along with the waveforms of clock signal (clk) and scan-enable signal (SE). The time period between the launch clock pulse (cp1) and the capture clock pulses (cp2) determines the test application frequency. Note that the scan-enable signal must fully transition during this time period, and this Figure 2. The concept and waveforms of LOC Several methods exist to mitigate the timing requirements on the scan-enable signal of LOS testing. Level Sensitive Scan Design (LSSD) requires two fast clock signals that could be used to launch and capture transitions [Eichelberger 77]. Another method involves locally generating fast scan-enable signals for LOS testing [Ahmed 06, Wang 04]. For the rest of the paper, we assume that the designs considered in this work have capabilities to apply LOS test. The launch-on-capture (LOC) transition test launches a transition through a logic network which is controlled by system clock pulses [Eichelberger 91, Savir 94]. In LOC testing, both launch and capture vectors are initiated by system clock pulses. The concept and waveforms of the LOC test are described in Fig. 2. In Fig.s 1 and 2, it should be noticed that the second clock pulse (cp2) of LOS and both clock pulses (cp1 and cp2) of LOC are system clock pulses. In addition, the LOC test procedure also requires the scan shift operations and inherently contains the last scan shift clock pulse. However, unlike LOS test, the last shift of LOC need not be applied at fast speed. Paper 35.3 INTERNATIONAL TEST CONFERENCE /08/$ IEEE

2 Therefore, if an additional system clock pulse is appended to the LOS test and the last shift clock of the LOC test is applied at fast speed, the test applications of both tests are identical. This introduces the mix of these two test methods, or hybrids of LOS and LOC: launch-on-shiftcapture (LOSC) test and launch-on-capture-shift (LOCS) test. The launch-on-shift-capture (launch-on-capture-shift) test is a three-pattern test for transition faults that launches transitions by both launch-on-shift and launch-on-capture mechanisms. The concepts of LOSC and LOCS tests and their signal waveforms are illustrated in Fig. 3. LOSC (LOCS) tests consist of three vectors (V 1,V 2,V 3 ). The first vector (V 1 ) initializes a logic value at a fault site as in LOS. Then the second vector (V 2 ) launches a transition and propagates it. At the same time, V 2 also initializes a logic value at a different fault site in a way the first vector of LOC does. Finally, the third vector (V 3 ) performs two things; 1) it propagates the fault effect launched by vector pair (V 1,V 2 ), and 2) it launches a transition initialized by V 2 and propagates it. Figure 3. The concept and waveforms of LOCS and LOSC LOSC and LOCS are identical in terms of test application (last shift clock pulse followed by two system clock pulses), but differ in how they are generated. LOSC test is made by generating LOS test first with don t-care bits unspecified. These don t-care bits are filled with LOC test. LOCS test is generated in the same way, but the LOC test is generated first before LOS test is used to fill the don t-care bits. The don t-care bits are exploited in the context of test set compaction [Goel 79], test data compression [Hiraide 03], or power reduction during scan shift operations [Cho 07]. LOSC and LOCS share the same concept with the compaction scheme; they utilize don t-care bits to detect additional faults. However, LOSC and LOCS differ from existing compaction methods. LOSC (LOCS) targets additional faults either by LOC or LOS, whichever is easier, while compaction uses only one method to test additional faults. Therefore, in LOSC and LOCS, LOSuntestable faults can be tested by LOC and LOCuntestable faults by LOS. On the other hand, conventional compaction techniques leave untestable faults undetected. The rest of the paper is organized as the following. Section 2 discusses the limitations of conventional transition tests and previous works on the hybrids of LOS and LOC. Section 3 describes the test generation methods. Section 4 presents the experimental setup and results on benchmark circuits. Section 5 discusses the benefits and limitations. Finally, Sec. 6 concludes the paper. 2. Previous works In this section, possible problems related to the two transition test approaches as well as previous researches to alleviate these problems are introduced Untestable faults LOC test usually achieve lower fault coverages than LOS test. In many cases, it is due to functionally unsensitizable faults that are not tested by LOC [Savir 94, Rearick 01]. Many of these faults can be tested under LOS since it does not launch transitions through a functional logic network. The detection of these faults by LOS methods is suspected to cause over-testing (rejecting good devices by falsely testing under environments that would not happen in functional mode). Nevertheless, it is also shown that some of these faults are testable by LOC when more than two clock pulses are used for fault activation and propagation [Abraham 06, Zhang 06]. On the other hand, LOS test does not always achieve 100% fault coverage. There are also untestable faults in LOS testing as described in [Zhang 07]. Some of these faults are untestable due to shift-dependency, but may actually fail the system operation if present. To reduce the number of LOS untestable faults, scan chain reordering or test-point insertion techniques were developed [Li 05, Gupta 03, Wang 03]. It is also noticeable that some of these faults may be detected under LOC test approach [Zhang 06] Hybrids of LOS and LOC As discussed above, LOS-untestable faults may be tested by LOC while LOC-untestable faults may be tested by LOS. Naturally, there have been efforts to combine the advantages of both testing methods [Ahmed 06, Wang 04, Devtaprasanna 05]. In [Ahmed 06, Wang 04], circuits are partitioned into two regions. One region is controlled by slow scan-enable signals to be tested by LOC test. The other region is controlled by locally generated fast scan-enable signals and tested by LOS test. These methods partition circuits into many regions by a controllability measure or a developed cost function. The quality of the partitioning determines the effectiveness of the methods. Also, under these methods, portions of circuits are tested solely by LOS while other parts are tested by LOC only. Therefore, LOS-untestable faults in LOS-tested regions and LOC-untestable faults in LOC- Paper 35.3 INTERNATIONAL TEST CONFERENCE 2

3 tested regions are still uncovered. On the other hand, LOSC and LOCS do not depend on partitioning of circuits and they can detect faults by either LOC or LOS, whichever is easier. Another approach was to implement multiple scan-enable signals [Devtaprasanna 05]. These scan-enable signals do not require fast switching capability, but are controlled separately. Hence, while some scan-enable signals are de-asserted to perform LOC test, other scan-enable signals are kept high, keeping flip-flops in the shift mode. Therefore, some flip-flops perform launch and capture while other flip-flops only launch transitions by shift operations. In this method, a fault can be tested either by LOS or LOC. However, when some faults are tested by one test method, other faults in the same scan-enable region have to be tested by the same method in one test pattern. On the other hand, LOSC and LOCS can test faults in different methods simultaneously, which may increase the efficiency of a test pattern. 3. Test pattern Generation To generate LOSC and LOCS tests, Automatic Test Pattern Generator (ATPG) tools for LOS and LOC test are required. LOC ATPG can be used directly without any modifications, but LOS ATPG requires a slight modification. Unlike the conventional LOS test where fault effects are propagated only one time frame, the LOS test for LOSC and LOCS propagates fault effects over two time frames. However, in this paper, to generate LOS test with two system clocks without modifications to the existing ATPG, the netlists of the circuits were unrolled to emulate iterative networks of the circuits. The iterative network is a form that emulates multiple time frames of the circuit operation [McCluskey 58]. Using the unrolled netlists, conventional LOS ATPG could be used directly to generate LOSC and LOCS. The test generation flow is shown in Fig. 4 as a pseudocode. Note that the LOC test generation and fault simulation in the procedure use regular netlists while the LOS test generation and fault simulation use unrolled netlists. The generation of LOSC and LOCS consists of two ATPG processes: the primary ATPG and the secondary ATPG. The primary ATPG is an ATPG process that generates test patterns without any restrictions. The patterns generated by the primary ATPG contain don tcare bits that are not specified. For LOSC (LOCS), the primary ATPG produces test that launches transition by LOS (LOC) method. Test generation flow (pseudo-code) While ( undetected fault exists ) { Primary LOS (LOC) ATPG Extract care bits of LOS (LOC) pattern Secondary LOC (LOS) ATPG (with care bits from LOS (LOC) pattern) Fault simulate in LOS (LOC) method (with don t-care bits filled by LOC (LOS)) } Figure 4. LOSC and LOCS test generation flow The secondary ATPG is another ATPG process that exploits the unspecified bits of the test patterns from the primary ATPG to test additional faults. It can be regarded as a way to fill don t-care bits of existing test pattern. To keep the original care bits from the primary ATPG result during the secondary ATPG process, these care bits are extracted and used as constraints during the secondary ATPG. If there are don t-care bits left after the secondary ATPG process, they are filled pseudo-randomly. In LOSC (LOCS) test, the secondary ATPG uses LOC (LOS) method. The resulting test pattern is fault simulated again in LOS (LOC) method. This is necessary to identify additional LOS (LOC) detection by the care bits from the secondary ATPG. The iterative network for LOS ATPG would contain twice as many faults as in the original network. To prevent falsely testing faults in the second time frame, only faults that belong to the first time frame are considered in LOS ATPG. Table 1. Benchmark circuits circuit Gate count FF count No. of Faults s , ,940 s , ,738 s ,179 1,564 43,032 s ,253 1,275 47,074 b17 12,949 1,314 95,328 b18 35,884 3, ,746 b19 69,437 6, ,140 b20 7, ,542 b21 7, ,418 b22 11, ,664 Paper 35.3 INTERNATIONAL TEST CONFERENCE 3

4 Table 2. Test pattern counts and fault coverages of test sets circuit LOS LOSC LOC LOCS length cov Length cov length cov length cov s % % % % s % % % % s % % % % s % % % % b17 1, % 1, % 1, % 1, % b18 4, % 4, % 5, % 4, % b19 9, % 6, % 8, % 8, % b20 1, % 1, % 1, % 1, % b21 1, % 1, % 1, % 1, % b22 1, % 1, % 1, % 1, % 4. Experimental Results We evaluated the test length and fault coverage of LOSC and LOSC tests on some of ISCAS89 and ITC99 benchmark circuits shown in Table 1. To attain the most efficient LOSC or LOCS test sets, single test pattern should be generated by the primary ATPG and filled by the secondary ATPG before the next pattern is considered. In this way, the primary ATPG for the next pattern would not target faults that could be detected by the secondary ATPG for the previous pattern. However, the commercial ATPG tool is highly optimized for compaction. Due to the optimization, when the ATPG tool generates one test pattern at a time and repeats until all the faults are tested or tried, the resulting test length was abnormally longer than a test generated at once by the same ATPG tool. It is unfair to compare our approach with the LOS and LOC generated at once due to the optimization performed on the latter. Thus, we settled with something in between the two extreme cases. For the purpose of fair comparison, all the test sets in this paper were generated in a group of 32 patterns, so that both the proposed approach (LOSC and LOCS) and the base case (LOS and LOC) do not benefit from the optimization of the ATPG tool. In addition, all the test patterns were generated with the default backtracking limit (abort limit of 10) and the default compaction option (high compaction) LOSC and LOCS test sets LOSC and LOCS test sets were generated as described above. They are compared with LOS and LOC test sets in terms of test length and fault coverage in Table 2. In this table, columns under label length represent test length and columns labeled as cov represent the fault coverages of the test sets. In each row, the shortest test length and the highest fault coverage are bold-faced. Table 2 shows that LOSC tests achieved the highest fault coverage. However, the test length of LOSC was comparable to LOS test or shorter than LOS test in some cases. LOCS performs better than LOC, but worse than LOS and LOSC. That is because the LOCS test generation terminates when the primary LOC ATPG cannot produce any more tests for undetected faults, without invoking the secondary LOS ATPG even though some of these faults could be tested by LOS. The CPU time required to generate these test sets were not considered in this experiment. The commercial ATPG tool is optimized to generate pure LOS or LOC tests, but the generation of LOSC and LOCS tests requires invocations of both LOS and LOC ATPGs. In addition, the implementation of LOSC and LOCS is not integrated into the ATPG tool, but it is rather an augmented form. Therefore, in this work, the qualities of the test sets are compared without considering the test generation time Fault Coverage LOSC and LOCS tests achieve higher fault coverage compared to LOS and LOC respectively. These improvements in fault coverage are due to the undetectable faults under one method (LOC or LOS) being detected by the other method. Table 3 summarizes undetected faults in each method. Column 2 shows the total number of faults in each circuit. Column 3 and 4 represent undetected faults by LOS and LOC test respectively. Column 5 shows the number of undetected faults after applying both LOS and LOC. Finally, column 6 and 7 represents undetected faults of LOSC and LOCS test respectively. In each row, the entries with the fewest undetected faults are bold-faced. Paper 35.3 INTERNATIONAL TEST CONFERENCE 4

5 Table 3. Undetected faults of each test method circuit total faults LOS LOC LOS+LOC LOSC LOCS s , , s , , ,782 s , , ,824 s , , ,082 b17 95,328 17,437 28,726 11,641 11,871 29,592 b18 263,746 40,231 85,672 26,029 28,766 85,158 b19 511,140 81, ,142 53,981 54, ,789 b20 555,42 2,176 6,484 1,752 1,783 6,328 b21 56,418 2,167 7,067 1,802 1,871 7,249 b22 83,664 2,874 12,872 2,614 2,709 12,642 In most cases, the combination of LOS and LOC had the fewest undetected faults, or the most detected faults. Table 3 shows that some of LOS undetected faults were detected by LOC test and the some of LOC undetected faults by LOS test. Hence, when both LOS and LOC tests were applied, the number of undetected faults was the least. However, it should be noticed that LOSC detected the comparable number of faults to the combination of LOS and LOC. To show the portion of LOS detected faults and LOC detected faults in LOSC test, the numbers of faults in each category of b17 and b18 circuits are plotted in Fig. 5 and Fig. 6 respectively. Detected Faults 80,000 70,000 60,000 50,000 40,000 30,000 20, loc detect los detect Figure 5. LOSC test - detected faults by LOS and LOC (b17) Detected Faults LOC detect LOS detect Figure 6. LOSC test - detected faults by LOS and LOC (b18) In Fig.s 5 and 6, it is shown that the LOS detected faults are the majority of detected faults throughout the entire test set and LOC detection is insignificant in the beginning of the test set. However, at certain points (at pattern number 992 in Fig. 5 and at pattern number 1,312 in Fig. 6), the number of LOC detected faults increases while additional LOS detection is minimal. % don't-care bits % 95.00% 90.00% 85.00% 80.00% 75.00% Figure 7. Don t-care bit percentages of test patterns (b17) % don't-care bits % 95.00% 90.00% 85.00% 80.00% Figure 8. Don t-care bit percentages of test patterns (b18) It is well known that, in the last phase of test pattern generation, a significant number of patterns is required to gain the last small portion of fault coverage [McCluskey 89]. The patterns generated at the end are less efficient, or contain many don t-care bits, which give the secondary ATPG more freedom to produce tests for additional faults. The percentages of don t-care bits in scan loads of the primary ATPG (LOS) generated test patterns in LOSC are plotted in Fig.s 7 and 8. In these figures, each column shows the average percentage of don t-care bits of 32 test Paper 35.3 INTERNATIONAL TEST CONFERENCE 5

6 patterns, which would be fed to the secondary ATPG to generate additional LOC tests. The test patterns contain many don t-care bits (over 95% of a scan load) starting from pattern number 992 in Fig. 7 and pattern number 1,312 in Fig. 8. Notice the points where the percentages of don t-care bits increase and where the LOC detection increases (Fig. 5 and Fig. 6) are aligned Test Set Length LOCS test were shorter and achieved higher fault coverage than LOC test. However, the fault coverage of LOCS test was still lower than LOS or LOSC test. On the other hand, LOSC test improved fault coverage of LOS test, but it sometimes resulted in longer test length. The increased test length of LOSC over LOS is suspected to be caused by the two-time-frame propagation requirements of LOS test used in LOSC. To verify this assumption, LOS test sets generated for original circuits and the unrolled circuits (faults propagated over two time frames) are compared in Table 4. Table 4. Comparison on LOS of original and unrolled circuit LOS LOS Unroll LOSC circuit length fc (%) Length fc (%) Pat fc (%) s s s s B17 1, , , B18 4, , , B19 9, , , B20 1, , , B21 1, , , B22 1, , , In Table 4, columns under length represent the test lengths and columns under fc show the fault coverages. The test lengths for the unrolled circuits were always longer than the tests for the original circuits. However, LOSC tests were usually shorter than LOS for unrolled circuits and sometimes even shorter than LOS for original circuits. This is when the additional fault detection by LOC in LOSC test compensated the test length penalties from the two-time-frame propagations of unrolled LOS. The test length reduction of LOSC over unrolled LOS can also be coming from the mixed use of LOS and LOC. As discussed above, a significant number of patterns are used to test hard-to-detect faults at the end of test generations. However, hard-to-detect faults in LOS are not necessarily hard-to-detect in LOC. When this is the case, the pattern count can be decreased if they are targeted by LOC ATPG instead of LOS ATPG. Even though LOSC test achieved higher fault coverage compared to LOS test, the longer test length is not desirable. To fairly compare the test length and fault coverage of LOSC and LOS, the fault coverage of both tests were plotted in Fig. 9 and Fig. 10. Fault Coverage 90% 80% 70% 60% 50% 40% LOSC 30% LOS Figure 9. Fault coverage comparison of LOS and LOSC (b17) Fault Coverage 100% 90% 80% 70% 60% 50% 40% 30% 20% LOSC LOS Figure 10. Fault coverage comparison of LOS and LOSC (b18) In these figures, the fault coverage of LOS is higher than LOSC at the first portion of the test set. But, in the later part of the test set, LOSC achieves higher fault coverage than LOS. The cross-over points are at pattern number 992 in Fig. 9 and pattern number 1,056 in Fig. 10. These points are the same or very close to where the number of don t-care bits (Fig.s 7 and 8) and the LOC fault detections are increased (Fig.s 5 and 6). If the highest possible fault coverage is desired, LOSC test is better than LOS test. If the test length is the concern and the test has to be truncated, LOSC has higher fault coverage as long as the test set is truncated after the cross-over point (pattern number 992 of b17 LOSC test and pattern number 1,056 of b18 LOSC test) Launch-on-shift with Launch-on-capture top-off We think that LOSC and LOCS tests achieved higher fault coverages because they use both LOS and LOC methods. Hence, it would be natural to compare LOSC tests with combinations of LOS and LOC tests. Paper 35.3 INTERNATIONAL TEST CONFERENCE 6

7 LOS test achieved higher fault coverage and shorter test length than LOC test. However, like all the transition test generations, LOS test generation becomes inefficient when only hard-to-detect faults are left and many patterns are required to detect these faults to improve the last portion of the fault coverage. Therefore, it may be beneficial to switch to LOC test at certain point and generate top-off LOC patterns, which test hard-to-detect faults of LOS testing using LOC test. To compare this idea with LOSC test sets, LOS tests were topped-off with LOC tests at four different points. Each group of 32 patterns were fault simulated during the LOS test generation and when the increase of the fault coverage by the last 32 pattern was 1) less than 1%, 2) less than 0.5%, 3) less than 0.2%, and 4) less than 0.1%, the LOS test generation stopped and LOC test generation started on the remaining undetected faults. The resulting test sets are shown in Table 5. In Table 5, the test with the highest fault coverage is boldfaced. LOSC test always achieved the highest fault coverage. LOS tests topped-off with LOC tests achieved comparable fault coverages to LOSC test sets, but with more test patterns Experimental Result Summary LOSC and LOCS showed higher fault coverage than LOS and LOC tests respectively. LOSC test sets were sometimes longer and sometimes shorter than LOS tests while LOCS achieved more compact test than LOC test in most cases. Given the higher fault coverages of LOSC test set, aspects that affect the LOSC test length were investigated. First, when LOS fault coverage is relatively low (or more room to improve fault coverage), LOSC test set achieved better compaction with significant fault coverage improvement. Secondly, the more don t-care bits in a test pattern, the better the secondary LOC ATPG performed, detecting more faults by the LOC method. 5. Discussions The benefits of LOSC and LOCS tests are as the followings. First, they target untestable faults under one method (LOS or LOC) using the other method, improving fault coverage. Secondly, hard-to-detect faults in one method may be east-to-detect in the other method. Hence, testing each fault using the easier way can improve the efficiencies of test patterns, which may decrease test length. Lastly, they can be implemented on top of existing ATPG with compaction, leading to more efficient use of don t-care bits. LOSC and LOCS tests also have limitations. First, they require fast switching scan-enable signals. However, even without the fast SE signal, a LOCS test set has an advantage over a conventional LOC test. The LOS fault activation mechanism initiates transitions of logic values at fault sites and propagates the fault effects. This is analogous to a test for stuck-open faults when they are not applied at-speed. Hence, LOCS would have higher stuckopen fault coverage than LOC. Secondly, the LOS test in LOSC (LOCS) needs to propagate fault effects over two time frames rather than one time frame of the conventional LOS test, which penalizes the test length. However, when combined with LOC, LOSC and LOCS performed the same or the better in terms of fault coverage and/or test length compared to pure LOS, which justifies the use of two time frames. In this paper, possible improvements of LOS and LOC transition tests in terms of test length and fault coverage were investigated. However, the quality of the test set is not only measured by these two metrics. Other ways to Table 5. LOS test topped-off with LOC test LOS+LOC LOS (orig) (<1%) LOS+LOC (<0.5%) LOS+LOC (<0.2%) LOS+LOC (<0.1%) circuit length fc (%) length fc (%) length fc (%) length fc (%) length fc (%) length fc (%) s NA NA s s NA NA s b17 1, , , , , , b18 4, , , , , , b19 9, , , , , b20 1, , , , , , b21 1, , , , , , b22 1, , , , , , LOSC Paper 35.3 INTERNATIONAL TEST CONFERENCE 7

8 improve the transition test quality include applying the concept of TARO (Transition fault propagated to All Reachable Outputs) [Tseng 01, Park 05], N-detect [Ma 95] or employing multiple clock cycles [Zhang 06]. The idea of mixing two different tests in one test pattern should not be limited to the hybrids of LOC and LOS. Previous researchers have shown the importance of using multiple test metrics to test manufactured ICs [Maxwell 93, Nigh 97, Ferhani 06]. The hybrid concept of LOSC can be extended to the mix of different test metrics for this purpose. As long as don t-care bits are available in test patterns, they can be exploited to generate additional tests for different metrics. For example, LOC and single stuckat fault test or single stuck-at test and bridge fault test can be combined. 6. Conclusion In this paper, Launch-on-Shift-Capture and Launch-on- Capture-Shift tests are introduced. These tests exploits don t-care bits of existing LOS (LOC) test set to detect additional faults by LOC (LOS) launch mechanism, which may be more efficient for some faults. Experimental results showed that LOSC test achieved higher fault coverage than any other test or mix of test sets. 7. Acknowledgments We thank Rohit Kapur of Synopsys, Brion Keller of Cadence and Samy Makar of C-Switch for their helpful discussions and support. We thank all the members of CRC for their help. 8. References [Abraham 06] J. Abraham, U. Goel, A. Kumar, Multi-Cycle Sensitizable Transition Delay Faults, Proc. VLSI Test Symp., [Ahmed 06] N. Ahmed, M. Tehranipoor, Improving Transition Delay Test Using a Hybrid Method, IEEE Design & Test, vol. 23, issue 5, pp , [Avramovici 92] M. Abramovici, P. S. Parikh, Warning: 100% Fault Coverage May Be Misleasing!, Proc. Intl. Test. Conf., pp , [Butler 04] Butler, K. M., et al., Minimizing Power Consumption in Scan Testing: Pattern Generation and DFT Techniques, Proc. Intl. Test Conf., pp , [Cho 07] K.Y. Cho, S. Mitra, and E. J. McCluskey, California Scan Architecture for High Quality and Low Power Testing, Proc. Int l Test Conf., [Dasgupta 81] S. Dasgupta, R. G. Walther, T. W. Willams and E. B. Eichelberger, An Enhancement to LSSD and Some Applications of LSSD in Reliability, Availability and Serviceability, Proc. FTCS, pp , [Devtaprasanna 05] N. Devtaprasanna, A. Gunda, P. Krishnamurthy, S. M. Reddy and I. Pomeranz, A Novel Method of Improving Transition Delay Fault Coverage Using Multiple Scan Enable Signals, Proc. ICCD, pp , [Eichelverger 77] Eichelberger, E. B. and T. W. Williams, A Logic Design Structure for LSI Testability, Proc. Design Automation Conf., pp , [Eichelberger 91] Eichelberger, E. B., E. Lindbloom, J. A. Waicukauski and T. W. Williams, Delay-Fault Simulation, Structured Logic Testing, McCluskey (Ed.), Prantice-Hall, Inc., Englewood Cliffs, New Jersey, [Ferhani 06] F.-F. Ferhani and E. J. McCluskey, Classifying Bad Chips and Ordering Test Sets, Proc. Int l Test Conf., pp. 1-10, [Goel 79] P. Goel and B. C. Rosales, Test Generation & Dynamic Compaction of Tests, Dig. Papers Test Conf., pp , [Gupta 03] P. Gupta, et. al., Layout-Aware Scan Chain Synthesis for improved Path Delay Fault Coverage, Proc. ICCAD, pp , [Hiraide 03] Hiraide, T., et al., BIST-aided Scan Test A new Method for Test Cost Resutction, Proc. VLSI Test Symp., pp , [Li 05] W. Li, et. al., Distance Resticted Scan Chain Reordering to Enhance Delay Fault Coverage, Proc. Int l Conf. VLSI Design, pp , [Ma 95] S. C. Ma, P. Franco, and E. J. McCluskey, An Experimental Chip to Evaluate Test Techniques Experiment Results, Proc. Int l. Test Conf., pp , [Maxwell 93] P. C. Maxwell and R. C. Aitken, Test Sets and Reject Rates: All Fault Coverages are Not Created Equal, IEEE Design & Test, pp , [McCluskey 58] E. J. McCluskey, Iterative Combinational Switching Networks General Design Considerations, IRE Trans. Elctron. Comput., Vol. EC-7., pp , [McCluskey 89] E. J. McCluskey and F. Buelow, IC Quality and Test Transparency, Trans. IEEE Industrial Electronics, vol. 36, no. 2, May, [Nigh 97] P. Nigh, W. Needham, K. Butler, P. Maxwell, R. Aitken, and W. Maly, So What is an Optimal Test Mix? A Discussion of the SEMATECH Methods Experiment, Proc. Int l Test Conf., pp , [Park 05] I. Park, A. Al-Yamani, E. J. McCluskey, Effective TARO pattern generation, Proc. VLSI Test Symp., pp , [Rearick 01] J. Rearick, Too much delay fault coverage is a bad thing, Proc. Int l Test Conf., pp , [Savir 92] J. Savir, Skewed-Load Transition Test: Part I, Calculus, Proc. Intl. Test Conf., pp , [Savir 94] J. Savir and S. Patil, Broad-side delay test, IEEE trans. On CAD of IC and System, vol. 13, pp , [Tseng 01] C. W. Tseng and E. J. McCluskey, Multiple-output propagatioin transition fault test, Proc. Int l. Test Conf., pp , [Wang 03] S. Wang and S. T. Chakradhar, Scalable Scan-Path Test Point Insertion Technique to Enhance Delay Fault Coverage for Standard Scan Designs, Proc. Int l. Test Conf., pp , [Wang 04] S. Wang, X. Liu, S. T. Chakradhar, Hybrid delay scan: a low hardware overhead scan-based delay test technique for high fault coverage and compact test sets, Paper 35.3 INTERNATIONAL TEST CONFERENCE 8

9 Proc. Deaign Automation and Test in Europe, vol. 2. pp , 2004 [Waicukauski 87] J. Waicukauski, et. al., Transition Fault Simultaion, IEEE Design and Test, pp , April [Zhang 06] Z. Zhang, et. al., Scan Tests with Multiple Fault Activation Cycles for Delay Faults, Proc. VLSI Test Symp., [Zhang 07] Z. Zhang, S.M. Reddy, I. Pomeranz, Warning: Launch off Shift tests for Delay Faults May Contribute to Test Escapes, Proc. Asia and South Pacific Design Automation Conf., Paper 35.3 INTERNATIONAL TEST CONFERENCE 9