A Correlative Microscopic Workflow For Nanoscale Failure Analysis and Characterization of Advanced Electronics Packages

3D X-ray Microscopy (XRM) is a key technique used in semiconductor product failure analysis due to its non-invasive nature and high-resolution capabilities. The process uses X-ray to scan the target region and generate 2-D images from different angles. It then combines these images to generate a 3-D cross-section picture of the target region, which is then visually inspected by failure analysis engineers on the possible cause of the failure. Despite the widespread adoption of 3D X-ray microscopy (XRM) as a valuable tool for failure analysis, it faces challenges such as extended scan times and substantial storage requirements. These issues stem from the necessity of acquiring a large number of measurements to achieve high-quality reconstructions. This thesis presents the development of novel techniques that are designed to make the 3D XRM measurement process to be more efficient in terms of time and storage requirements. The first technique leverages compressive sensing to capture 2-D images, hence reducing the time and storage spaces required when compared to the existing process. A novel measurement process, Sparse Matrix Deep Compressed Sensing (SM-DCS), is developed that is able to generate 2-D image from sparse (15%) version of the scanned image with quality that is comparable to the full-resolution image captured using the same equipment. The second technique, XRM-NeRF leverages Neural Radiance Field (NeRF) to reduce the number of 2-D projection images (16.67%) required to reconstruct an optimized 3-D model. It uses a deep learning process to obtain a set of fine-tuning parameters to mitigate measurement noise in the pose estimation of semiconductor samples. Results shows that it is able to achieve quality comparable to those using the existing 3D XRM process. iXRM-NeRF is then proposed to accelerate the learning process used in XRMNeRF. It leverages meta-learning technique to learn common knowledge shared across scenes, reducing the number of steps required for optimizing an iXRMNeRF model for a scene. Additionally, iXRM-NeRF model proposes a visualization technique that queries the model using orthogonal projection to generate the crosssectional view of a 3D XRM scene, assisting failure analysis engineers to identify defects. The work presented in this thesis hence introduced several innovative techniques that illustrate the promise of integrating compressive sensing and deep learning methodologies to expedite 3D XRM for semiconductor failure analysis. The integration of compressive sensing techniques, deep learning techniques, and specialized domain knowledge in this thesis highlights the potential to accelerate the product development lifecycle for semiconductor products.

As semiconductor manufacturing progresses to smaller process nodes, it is becoming increasingly difficult to climb the yield learning curve rapidly. The rate of yield learning dictates the growth and success of the semiconductor industry, and must be accelerated to fulfill competitive time-to-market, time-to-money and time-to-volume requirements. Software-based diagnosis plays a crucial role in yield learning. Diagnosis comprehends the test response of a failing circuit to determine the location, and sometimes, in addition, characterize the nature of a defect affecting the failing circuit. Besides identifying likely failure mechanisms and increasing the quality of chip testing, the feedback provided by diagnosis is used to select chips for physical failure analysis (PFA). PFA aims to visually examine a chip to characterize a defect, prevent similar defects in the future and, consequently, improve the design and manufacturing of a chip.However, PFA is often destructive, time- and cost-intensive, and not always successful. Diagnosis, on the other hand, is non-invasive and time- and cost-effective; moreover, it assists PFA and guides yield learning. The advantages of diagnosis, coupled with the diminishing performance of PFA with advancing technology, make it an encouraging facilitator for rapid yield learning.Therefore, the objective of diagnosis must be logic-level defect characterization to minimize (and ideally eliminate) the need for PFA, and accelerate yield learning. Logic characterization of a defect includes the derivation of its physical location and precise logic behavior. In this dissertation, a comprehensive diagnosis methodology is developed to actualize the aforementioned objective.The developed methodology comprises of three methods. LearnX/MD-LearnX is a physically-aware method that employs (a) the X-fault model to avert the elimination of a correct defect candidate and (b) machine learning to build a candidate-ranking model that learns the hidden correlations between the tester response and the defect candidates to pinpoint the correct candidate. PADLOC, which stands for Physically-Aware Defect LOcalization and Characterization, improves the physical location of a back-end defect (i.e., a defect that affects one or more interconnects and resides outside a standard cell) returned by LearnX/MD-LearnX by partitioning the defective net into physical subnets and identifying the subnets that influence defect excitation. In addition, PADLOC deduces the precise impact of a defect on the circuit functionality by examining its surrounding circuitry. NOIDA, which stands for NOise-resistant Intra-cell Diagnosis Approach, pinpoints the location of a defect within a failing standard cell implicated by LearnX/MD-LearnX. In contrast to prior work that typically constructs/employs a fault dictionary, NOIDA ascertains the location as well as the behavior of a front-end defect (i.e., a defect that resides inside a standard cell) by monitoring the logical activity of its intra-cell neighborhood. Additionally, NOIDA is resistant to circuit-level noise that may originate from potentially inaccurate transistor-level simulation. Results from numerous experiments reveal that our diagnosis methodology outperforms state-of-the-art commercial diagnosis. LearnX/MD-LearnX reports fewer defect candidates than commercial diagnosis for 69.4% silicon fail logs without losing accuracy. PADLOC implicates a smaller physical area for a defect for 47.2% silicon fail logs and attains at most 44X improvement. NOIDA reports an ideal diagnosis for 38.0% more front-end defects, when compared to leading-edge commercial diagnosis. In the presence of noise, NOIDA achieves an ideal diagnosis 7.6X more often.In summary, this dissertation endeavors to characterize a defect residing in a logic chip in terms of its precise physical location and logic behavior, which, consequently, most likely, enables rapid yield learning. The deployment of machine learning to pinpoint the correct candidate in LearnX/MD-LearnX, and the investigation of the neighborhood of a defect to determine its exact physical location and logic behavior in PADLOC and NOIDA are the novel components of this dissertation, and the reasons for its superiority over the state-of-the-art.

The line between packaging and silicon interconnect technology is blurring due to a reduction in package interconnect dimensions, which drives an increase in image resolution requirements. It is becoming more difficult to localize and image defects and structures throughout the packaging life cycle, from materials selection, through package and silicon design co-optimization, development, production, and field failure diagnostics. Characterization and failure analysis (FA) solutions must provide fast results for rapid development of packages meeting the required electrical, mechanical and reliability specifications with high yield and quality. Heterogeneous integration and complex packages containing multiple die drive new approaches to rapidly characterize structures, defects and processes. This paper presents new artificial intelligence (AI) developments in 3D X-ray microscopy (XRM) for non-destructive submicron-resolution imaging of packages. It also introduces the latest developments in focused ion beam (FIB) microscopes adapted with an integrated fs-laser for precise and fast analysis of deeply buried features in advanced packages.

As the complexity of integrated circuits increases, the electrostatic discharge (ESD) protection devices become critical to reliability issues. However, the physical failure analysis of ESD devices is destructive and time-consuming. In this work, by using the X-ray microscopy (XRM), we study the abnormal change in the leakage current of diode-triggered silicon-controlled rectifiers (DTSCRs) ESD structure under transmission line pulsing stressing. XRM is nondestructive physical failure analysis method by showing the in-depth morphology information of the devices. The results show that the proposed ESD devices have two parts, one is the trigger diodes part and the other is the SCR part. The SCR part was severely damaged and the contact and vias are melted under the electrical stressing, while the trigger diodes part remains intact. The failure analysis is nondestructive, multi-view, and time-saving. It is enlightening the design of novel ESD devices.

Xanthium sibiricum Patrin ex Widder and Xanthium italicum Moretti are morphologically similar fructus that are frequently misidentified. Xanthium italicum Moretti may possess inherent toxicity, and its adulteration of genuine medicinal materials poses a threat to clinical drug safety. Macroscopic observation and three microscopic techniques including stereo microscope, optical microscope, and 3D X-ray microscope were used for morphological identification of Xanthium sibiricum Patr ex Widder and Xanthium italicum Moretti in this study. 3D X-ray microscopy was applied as a novel tool for non-destructive, high-resolution discrimination of the two taxa. Intact fructus (n = 30 per species) were first screened macroscopically, then examined by stereo microscopy, optical microscopy, and 3D X-ray microscopy (0.3, 0.7, 1.5, 3.5, 18.06, 20.01 μm voxel size, Zeiss Xradia 520 Versa). The results showed that stereo microscopy, optical microscopy, and 3D X-ray microscopy collectively confirm the same conclusion from three distinct physical perspectives: surface topography, internal two-dimensional structure, and internal three-dimensional density distribution. The two Xanthium species differ significantly in burr spine morphology, fructus size and shape, the architecture and distribution of non-glandular and glandular trichomes, cotyledon conformation, and seed-coat cell patterning. In particular, 3D X-ray microscopy clearly resolves internal cotyledon spatial configurations and involucral cavity architectures, which furnishes critical endomorphic characters for taxonomic diagnosis. 3D X-ray microscopy provides unprecedented volumetric contrast of surface spines and internal seed architecture, permitting confident, non-destructive species identification. This study provides a basis for the safe clinical use of Xanthium sibiricum Patrin ex Widder. The frontier of 3D X-ray microscopy in plant systematics offers a novel, rapid, accurate and non-destructive protocol for the discrimination of morphologically elusive species.

This chapter discusses the application of scanning and transmission electron microscopy techniques, X-ray microscopy, as well as scanning probe microscopy for physical failure analysis. With the scaling-down of transistor features and interconnect dimensions and with the introduction of new processes and materials, physical failure analysis becomes more and more an essential task to exclude both yield-limiting process irregularities and reliabilityrelated failures in the semiconductor industry. Advanced imaging tools and techniques are needed to ensure high manufacturing yield and product reliability. Physical failure analysis is an essential task to exclude both yield-limiting process deviations and reliability-related failures in integrated circuit (IC) manufacturing industry. Its goal is to determine the root cause of failures or parameter deviations as a prerequisite for corrective actions. Polysilicon Gate Roughness, control of the transistor gate critical dimension with nanometer accuracy is a top priority task for leading-edge IC manufacturing.

Thermal compression bonding is being widely used for fine pitch copper pillar assembly as it provides high die placement accuracy and bonding suitable for mass production. In the current work, chip package interaction induced ILD integrity failures under thermo-compression assembly induced stresses were investigated. A step by step EFA (Electrical Failure Analysis) plus PFA (Physical Failure Analysis) methodology was developed to isolate the failing location and confirm the fail mode. The key steps include initially dataloging the fails and running the data through ATPG (Automatic Test Pattern Generation) tools to localize the failing nets. The PFA was then focused on the ILD sections in the vicinity of the Cu pillar interconnect (as the pillar is the primary load transference link between the die and substrate). Specific fail modes were then identified using a combination of Scanning Optical Microscopy (SOM), Scanning Acoustic Microscopy (SAM), layer by layer de-processing and Focused Ion Beam (FIB) cross-sections. The paper summarizes the key findings, failure modes from the study — it was found that the assembly process could result in damage initiation in the ILD which results in functional failures during subsequent reliability testing. The scale of induced damage makes it nearly impossible to detect using either traditional eFA, PFA or a combination of the two. Additionally, key assembly and design parameters, to resolve the fails are also discussed.

Physical failure analysis (PFA) is crucial to improving yield during the production life of an integrated circuit (IC). PFA begins with software-based diagnosis, which reports a set of likely failing candidates. A large set, which can result from test-set equivalent (TSE) faults, makes PFA success unlikely because each candidate may potentially have to be examined. Reducing the cardinality of a TSE fault class can reduce the time required for PFA since it likely decreases the number of candidates that must be examined. Additional (diagnostic) test vectors can be generated however to reduce class size. In this work, several layout-based physical metrics are defined and investigated for assessing the layout characteristics of a TSE fault class, which serves as an estimate of the time/difficulty required to perform PFA. In short, these metrics are used to define a physically-aware diagnostic resolution (PAR for short) for a fault class. A TSE fault class with a bad PAR likely requires more PFA effort than one with good PAR. To demonstrate the utility of PAR, a test-selection algorithm that chooses tests from an N-detect test pool based on the improvement in PAR is compared to techniques that use conventional definitions of diagnostic resolution. Comprehensive experiments reveal that fewer additional tests are needed when using PAR.

Time-dependent X-ray Microscopy (XRM) is an excellent technique to develop a fundamental mechanistic understanding of material behavior. Computational imaging plays a critical role in XRM, in a variety of ways. 2D projections are acquired and the resulting datasets are reconstructed using a filtered back projection algorithm. Several imaging artifacts are typically present, such as beam hardening, misalignment of the data, drift during time-evolved experiments (particularly at high temperatures and/or nanometer resolution scans), etc. Minimizing and removing these artifacts is, thus, very important. This is all the more important, because image segmentation is then done to quantify the statistics of the microstructure (often, as a function of time). The efficiency and accuracy of image segmentation is directly proportional to the quality of the initial reconstructed data. Thus, there is a need to develop efficient, robust algorithms that can handle large datasets obtained by 4D, time-dependent x-ray microscopy. In this paper, we will describe the challenges associated with computation imaging during x-ray microscopy. The use of Convolutional Neural Network (CNN) architectures based on a deep learning approach, as means of automating and handling x-ray microscopy data sets, in both lab-scale and synchrotron, will be discussed. The use of CNN techniques to robustly process ultra-large volumes of data in relatively small-time frames can exponentially accelerate tomographic data analysis, opening up novel avenues for performing 4D characterization experiments.

Cross sectional SEM specimens are routinely prepared for process monitoring/development, fabrication problem solving and failure analysis in semiconductor industries. A considerable variety of established methods are already in place for production of these specimens. They include: mechanical polishing, manual/automated cleaving, wet chemical etching and focused ion beam (FIB) milling. This report covers a new approach to the preparation of such specimens utilizing a broad ion beam technique. The technique consists of three steps, I) perpendicular “slope cutting” to expose the profile of the layers/features in the starting wafer, II) etching of the cut surface to delineate the microstructures and III) coating the specimen with a conductive thin film to prevent charging effects in the SEM. All three steps are carried out in a dedicated broad ion beam instrument, designed initially to etch and coat SEM specimens in one vacuum chamber [1]. The entire three-step process can be completed in one hour.

As the device technology is progressing in the nanometer regime towards atomic scale, the famous comment “There is plenty of room at the bottom” by Richard Feynman about 60 years ago needs to be reviewed carefully and understood in detail. A recent comment “There is plenty of difficulty near the bottom” by device manufacturers depicts another face of the advancement in technology. The difficulty observed is in the path for building reliable devices. Device reliability is the resultant of various analyses of the design, process and product and understanding innumerable phenomenon to control the extension of even atomic level defects, especially when the dimensions are at nanometer level. At the early stages of technology development, the reliability estimation was through specific tests prediction to weed out defects. Many such tests are a routine and has helped to develop technology. However, at that time itself it was known that understanding device physics and solving the related problems at smaller dimensions are the key to success. As such, the device development at any technology level has been through the critical understanding of the failures through physical analysis and hence solving the issues. This procedure is the Building-in reliability in devices. It could be noted that many fundamental physics problems such as hot carriers, issues of dielectric breakdown and strength, electro migration and stress migration, electrostatic discharge issues, etc which paused the threat at different stages were solved fully or partially through physical failure analysis and understanding the related failure mechanisms. However, as the technology progressed with new materials at new dimensions, the problems also became more challenging. The tools which helped and being used for analysis itself have limitations due to the dimensional miniaturization. Moreover, the diversification in the manufacture from design to product through different vendors and stages invites a separate approach for building the product reliability. This tutorial discusses the fundamentals of reliability and failure analysis in its basic concepts and from where it can be understood using a physics based approach. The progression in devices from the conventional MOSFET and its dimensional shrinkage to nano scale MOSFET and then to new FinFET is discussed with a view to understand how the physical failure analysis can reveal the failure mechanisms to improve reliability. While evolving these processes to build reliability in nano scale devices, it is also understood that the fundamental problem remains the same as in early 1960s. That is where the “the plenty of room” is emerged as “plenty of difficulty”.

It has been long recognized that SRAM memory is an ideal vehicle for defect monitoring and yield improvement during process development because of its highly structured architecture and simplified approach using memory bitmapping. However, the success rate of defect detection, especially for soft single-column failures, which is one of the most complex failure modes of an SRAM failure, is decreasing when using traditional physical failure analysis (PFA) with only the bitmap available for guidance due to a variety of invisible or undetectable defects that cause a leakage behavior in the device. In order to understand the leakage behavior in advanced high voltage (HV) processes, conductive atomic force microscope (C-AFM) [1-5] is introduced to perform junction-level fault isolation prior to attempting PFA. In this study, the quantified data extracted using the C-AFM can also be used to establish a connection between the failure mechanism discovered and the soft single column failure mode.

Fault Isolation / Failure Analysis (FI/FA) of increasingly complex embedded memory in microprocessors is becoming more difficult due to process scaling and presence of subtle defects. As physical failure analysis (PFA) is destructive and involves expensive and time-consuming processes, fault diagnosis needs to be as precise as possible to ensure successful physical defect sighting. This paper introduces a cache Fault Isolation methodology that focuses on exhaustive data collection to derive concrete hypothesis of physical fault location and to overcome the existing FA/FI challenges. The methodology involves a novel application of existing DFT techniques in combination with circuit analysis, pattern hacking, defect localization and PFA tools. Some of the techniques, for example pattern modification or circuit simulation, are applied repeatedly in order to obtain higher-level of isolation – from cell/logic level to transistor/gate level, and finally down to physical structure/layer level. This multi-level FI approach is the key to localize the failing area to greater precision, which had proven itself in Intel Itanium® II processor yield improvement process.

Laser Voltage Imaging (LVI) has become a well-established method for isolating scan-shift (i.e., scan chain integrity) tests failures [1, 2]. When LVI is coupled with time-domain information acquired using Continuous-Wave Laser Voltage Probing (CW-LVP) [3], the Physical Failure Analysis (PFA) success rate exceeds 90% for all types of failing conditions, from hard stuck-at fails to soft transition fails. This combination of Electrical Failure Analysis (EFA) techniques is effective because of its ability to pre-isolate the defect to a small enough area for successful PFA. While high PFA success rates are proven, there remains the issue of throughput: CW-LVP can be time consuming, and techniques that minimize the need for it are important. This paper introduces a novel LVI methodology that incorporates phase information [4] and reduces the need for CW-LVP for certain types of failures. Case studies will be presented.

Direct assessment of porosity and permeability in unconventional reservoirs such as shales and tight gas sands using imaging technologies is a challenge because of small pore throats, a broad pore size range, and sample heterogeneities with respect to mineralogy and grain size. This contribution presents the outcome of a range of Liquid Metal Injection (LMI) experiments in which we injected various fusible alloys into unconventional reservoir type of rocks, followed by Broad Ion Beam (BIB) polishing and SEM imaging to visualize and to better understand the pore connectivity and transport mechanisms in these tight rocks.In a series of experiments liquid metal was injected into a variety of organic-rich shales and other tight rocks at elevated temperature (approximately 90 °C) by increasing the pressure stepwise, up to a maximum of 500 MPa. Once the desired pressure was reached, the experiment was completed by turning off the heating and letting the samples cool down while maintaining the pressure. After solidification of the alloy, the sample was argon-ion polished to create a smooth surface for SEM imaging. Imaging sister samples on which we applied different intrusion pressures enables insight into how these tight rocks are progressively filled with the alloy.Firstly, injection of a non-wetting alloy followed by BIB surface polishing and SEM imaging allows comparison with Mercury Intrusion Porosimetry and visualizing which parts of the pore space are filled during intrusion of a liquid metal. SEM enables identifying alloy-filled pores below ten nanometers in size. High-resolution SEM images provide insight in preferred transport pathways and the variability of pore throat sizes. The experiments show that the non-wetting alloy does not intrude into samples in a uniform way. The relatively lowest porous sample investigated displayed the intrusion of alloy only into fractures, interpreted as induced micro-fractures, while others showed alloy within parts of the in-situ pore space. In a Posidonia Shale sample, intruded and non-intruded layers exist within the same sample, while the layers show no significant difference in mineral fabric at the large scale. At higher resolution, porous calcite fossils and framboidal pyrites were mostly filled with the alloys as was the clay-rich matrix. Injection of a sister sample with a more wetting alloy showed uniform intrusion of the alloy throughout the sample in the in-situ pore space and micro-fractures. The segmented SEM images allow quantification of the connected porosity. The question, which of the identified micro-fractures are induced or not, is subject of current research.Recent advances in LMI in combination with BIB milling technologies and SEM facilitate imaging connected porosity, both induced and in-situ, down to the nanometer scale in representative 2D cross-sections. Hence, LMI-BIB-SEM is a relatively easy method to verify whether unconventional reservoir samples contain good poro-perm and fracability properties.