Power supply noise (PSN) has become a critical issue during high-quality at-speed testing. Discrepancies between the circuit's switching activity during functional and test mode can cause overtesting and lead to yield loss. Alternatively, reduced PSN effects around critical paths can result in undertesting the chip, causing test escapes. To achieve a high-quality at-speed test, it is necessary to solve these problems simultaneously. Our previous work introduced a noise index model (NIM), which can be used to predict the mismatch between expected and real path delays. This paper quantitatively investigates and compares NIM values for critical paths during functional and test mode. We then propose a test pattern modification method that harnesses the NIM. The method fills a subset of the don't care bits in partially specified test vectors such that the worst observed functional NIM for the targeted critical path is replicated during test mode.

As CMOS technology continues to scale, the accurate prediction of silicon timing through the use of pre-silicon modeling and analysis has become especially difficult. These timing mismatches are important because they make it hard to accurately design circuits that meet timing specifications at first-silicon. Among all the parameters leading to the timing discrepancy between simulation and silicon, this paper studies the effect of dynamic IR-drop on the delay of a path. We propose a noise index model, NIM, which can be used to predict the mismatch between expected and real path delays. The noise index considers both the proximity of switching activity to the path and physical characteristics of the design. To evaluate the method, we performed silicon measurements on randomly selected paths from an industrial 65 nm design and compared these with Spice simulations. We show that a very strong correlation exists between the noise index model and the deviations between simulations and silicon measurements.

With test quality being an imperative, this paper presents a methodology on how to apply volume scan diagnosis - known from the field of yield learning - to the domain of test quality learning. Volume diagnosis allows to drastically accelerate the learning cycle. We give guidelines on how to improve test pattern generation strategies and try to answer which defects can be addressed deterministically with adequate fault models versus where probabilistic methods such as N-detect need be applied. The paper is based on a detailed analysis of scan diagnosis data from a production volume of well over one million devices of a 90 nm product.

The demand for higher quality requires more effective testing to filter out the bad devices. It is already known that multi-detection of single stuck-at faults results in more fortuitous detections of defects not behaving as stuck-at faults, which increases the test quality. Existing multi-detect tests, i.e., the well-known n-detect tests, suffer from significant test size increases. This paper shows that embedding multi-detection of faults within regular ATPG patterns results in a higher quality without a significant increase in test set size. High-volume silicon measurement results demonstrate that embedded multi-detect tests detect 2.3% to 4.7% more defective devices than conventional single-detect stuck-at tests.

Improvement of diagnosis methodologies is a key factor for fast failure analysis and yield improvement. As bridging defects are a common defect type in CMOS circuits, diagnosing this class of defect becomes relevant for present and future technologies. Bridging defects cause two additional current components, the bridge and the downstream current. This work presents the effect of the downstream current on current signatures and its impact on the diagnosis of such defects. The authors demonstrate that the impact of downstream current is minimized at low power supply (Vdd) values. Therefore, current measurements at low power supply voltages are proposed to enhance bridge diagnosis. Experimental evidence of this behaviour is presented for real devices. Furthermore, current signatures measured at VVLVare used for the diagnosis of fifteen failing 0.18mum technology devices, which are demonstrated to contain a bridging defect.

Application specific ICs are typically designed to meet a given performance specification. For these ICs a higher performance does not add value and less performance makes the IC useless. This class of ICs is designed based on worst-case corner analysis. It is expected that this will become area costly in more advanced technologies. An alternative is to use statistical design techniques but this implies that the performance needs to be tested with, for example, path delay testing. Our experiments in 65 nm show that the actual delay depends on the global activity within an IC as well as effects in the local neighbourhood of the path. These global and local effects can independently cause about 15% of additional delay. Hence, their impact needs to be included during test and thr authors propose to create (close to) worst-case delay patterns. Individually, the patterns have an enhanced sensitivity for the most important local effects and combined they provide coverage for global effects. This makes them better suited as speed indicators than conventional path delay patterns.

A proposal for enhancing the diagnosis of full open defects in interconnecting lines of CMOS circuits is presented. The defective line is first classified as fully opened by means of a logic-based diagnosis tool (Faloc). The proposal is based on the division of the defective line into a number of segments. The selected group of segments is derived from the topology of the line and its surrounding circuitry. The logical information related to the neighbouring metal lines for each considered test pattern is taken into account. With the proposed diagnosis methodology, a set of likely locations for the open defect on the line is obtained. A ranking between the set of possible locations is presented based on the analysis of the quiescent current consumption of the circuit under test. Examples are presented in which the use of the diagnosis methodology is shown to discriminate between different locations of the full open defect.

Excessive power supply noise can affect path delay and cause overkill during delay test. This paper presents low-cost noise models for fast power supply noise analysis and timing analysis considering noise impact. Our prior work only considered array-bond chips. This work proposes a noise analysis methodology that can be applied to wire-bond chips as well as array-bond chips. Experiments were performed on an industrial design. Silicon results show as much as a 15% delay variation due to different don't care fill approaches. The power supply noise impact on delay must be taken into account when delay tests are applied

This paper presents the effectiveness of various stress conditions (mainly voltage and frequency) on detecting the resistive shorts and open defects in deep sub-micron embedded memories in an industrial environment. Simulation studies on very-low voltage, high voltage and at-speed testing show the need of the stress conditions for high quality products; i.e., low defect-per-million (DPM) level, which is driving the semiconductor market today. The above test conditions have been validated to screen out bad devices on real silicon (a test-chip) built on CMOS 0.18 /spl mu/m technology. The IFA (inductive fault analysis) based simulation technique leads to an efficient fault coverage and DPM estimator, which helps the customers upfront to make decisions on test algorithm implementations under different stress conditions in order to reduce the number of test escapes.

Resistive bridges not only cause a static faulty behavior in CMOS memories, but also lead to several dynamic faulty behaviors which are timing related failures. This paper introduces a new realistic dynamic fault model for random access-memories: the delay coupling fault, which models resistive bridges in the memory array. We show that well-known march tests do not cover delay coupling faults at the memory array. To cover the delay coupling faults, a new and efficient test algorithm (DITEC+) is presented. We have performed inductive fault analysis to validate this novel algorithm and shown a significant improvement on fault coverage. Also, experiment on silicon is carried out to show the existence of such dynamic faults and their detection by implementing DITEC+.

This work proposes an effective method for applying fine-delay fault testing in order to improve defect coverage of especially resistive opens. The method is based on grouping conventional delay-fault patterns into sets of almost equal-length paths. This narrows the overall path length distribution and allows running the pattern sets at a higher speed, thus enabling the detection of small delay faults. These small delay faults are otherwise undetectable because they are masked by longer paths. A requirement for this method is to have hazard-free paths. To obtain these (almost) hazard-free paths we use a fast and simple postprocessing step that filters out paths with hazards. The experimental data shows the effectiveness and the necessity of this filtering process.

Defects due to process-design interaction have a systematic nature. Therefore they can have a profound impact on yield. The capability to detect (and correct) them is a requirement to continue to follow Moore's law. Most of the systematic defects are detected during the process development. These defects are detectable with test structures or visual inspection tools. However some process marginalities are only show-up in the topology of 'real' designs. Moreover, these defects are often not detectable with stuck-at testing. We show two examples of process related defects which could only be detected with more advanced test methods such as transition fault testing and low voltage testing. To correct systematic problems, however, one should not only have the capability to detect defects but also to identify them. Our examples show that other tests could have been far more sensitive in detecting systematic issues. Therefore the detection of systematic defects gives new requirements to test suites and can only be achieved with a shift in the position of manufacturing test.

Intra-gate resistive open defects not only cause sequential behaviour in CMOS memory address decoders, but also lead to delay behaviour. This paper evaluates the fault coverage of the resistive open defects in the memory address decoders. It shows that both the strong and the weak open defects are not completely covered by applying the well-known March tests and special test pattern sequences published in the literature. We demonstrate that the fault coverage is increased by varying the duty cycle of the internal clock of the address decoder. For the self-timed memories, we introduce a simple DFT technique to control the duty cycle of the internal clock which activates/deactivates the word lines. Using defect-oriented test, we also created a fault dictionary based on the defect location, transistor types, the terminal name and also the faulty behaviour. The fault dictionary in combination with the bit-map fail data will facilitate the localization of the open defects.

The known methods of transition fault diagnosis usually suffer from the drawback of many candidates. The method presented in this paper aims at reducing the number of suspects. The transition fault patterns were generated by Philips in-house ATPG tool and applied on the tester. The fail information from tester was subjected to fault diagnosis resulting in a small list of faulty candidates. We then injected the delay faults into the golden netlist of the test chip and confirmed through simulation whether or not their behavior matched with the tester results. Upon successful matching, we proceeded with the selection of few testable paths through the suspect faulty node and created corresponding path delay patterns using the path delay ATPG (a prototype at the University of Bremen, developed in cooperation with Philips Semiconductors GmbH, Hamburg). Finally, we verified those path delay patterns on the tester to increase the confidence level of the diagnosis method The experimental results show the effectiveness of our novel approach for improving diagnostic resolution.

We propose a new coverage metric for delay fault tests. The coverage is measured for each line with a rising and a falling transition, but the test criterion differs from that of the slow-to-rise and slow-to-fall transition faults. A line is tested by a line delay test, which is a robust path delay test for the longest sensitizable path producing a given transition on the target line. Thus, the test criterion resembles path delay test and not the gate or transition delay test. Yet, the maximum number of tests (or faults) is limited to twice the number of lines. In a two-pass test-generation procedure, we first attempt delay tests for a minimal set of longest paths for all lines. Fault simulation is used to determine the coverage metric. For uncovered lines, in the second pass, several paths of decreasing lengths are targeted. We give results for several benchmark circuits.