Abstract Presentation slides from the ISTFA 2021 tutorial, “[Yield Basics for Failure Analysis].”

Abstract This paper explains how embedded assist and timing control techniques are being used to improve soft defect screening in nanoscale static random access memory (SRAM). The electrical stress test method is evaluated on advanced FinFET devices. As test results show, resistive and parametric defects that are difficult if not impossible to detect using conventional techniques become visible with the aid of assist and timing control circuits.

Abstract This paper presents a machine learning approach that uses genetic algorithms to optimize test program timing sets based on first silicon. The method accounts for test hardware differences, discrepancies in silicon processes, and IO pin interdependency. The general theory and implementation are covered in detail and the capabilities of the method, in terms of false fail discovery, elimination, and failure debug, are demonstrated using actual product test cases.

Abstract Fault localization using both dynamic laser stimulation and emission microscopy was used to localize the failing transistors within the failing scan chain latch on multiple samples. Nanoprobing was then performed and the source to drain leakage in N-type FinFETs was identified. After extensive detailed characterization, it was concluded that the N-type dopant signal was likely due to projections from the source/drain regions included in the TEM lamella. Datamining identified the scan chain fail to be occurring uniquely for a specific family of tools used during source/drain implant diffusion activation. This paper discusses the processes involved in yield delta datamining of FinFET and its advantages over failure characterization, fault localization, nanoprobing, and physical failure analysis.

Abstract This paper describes an electrical and physical failure analysis methodology leading to a unique defect called residual EG oxide (shortened to REGO); which manifested in 14nm SOI high performance FinFET technology. Theoretically a REGO defect can be present anywhere and on any multiple Fin transistor, or any type of device (low Vt, Regular Vt or High Vt). Because of the quantum nature of the FinFET and REGO occurrence being primarily limited to single Fins, this defect does not impact large transistors with multiple FINs; moreover, REGO was found to only impact 3 Fin or less transistors. Since REGO can be present on any multi-FIN transistor the potential does exist for the defect to escape test screening. Subsequently a reliability BTI (Bias Temperature Instability) stress experiment by nanoprobing at contact level was designed to assess REGO’s potential reliability impact. The BTI stress results indicate that the REGO defect would not result in any additional reliability or performance degradation beyond model expectations.

Abstract A BEOL compatible Metal-Insulator-Metal capacitor (MIMCAP) was successfully developed for GlobalFoundries 14nm technology node, and subsequently introduced on customer designs as decoupling capacitors. The lead production silicon wafers with MIMCAP showed good functionality at wafer SORT functional test. However, upon testing more wafers, it became evident that the wafer center was impacted by abnormal scan logic fallout. The observed yield loss did not correlate with the MIMCAP scribe line Health Of Line (HOL) structures and the failure root cause could not be directly pin pointed to the MIMCAP process integration. Product scan diagnostic was performed and several systematic failing logical nets were identified. Subsequent failure analysis showed open via contacts in the MIMCAP vicinity. A detailed layout analysis of the FA confirmed weak-points and repeating logic nets allowed identifying a chip design topography issue resulting in a narrower process window compared to the scribe line MIMCAP HOL structure. Thanks to this knowledge the MIMCAP process was further optimized and the wafer center fallout was fully recovered in volume production.

Abstract Electrical failure analysis (EFA) is usually time consuming and expensive. It is a critical step bridging data failure analysis (DFA) and physical failure analysis (PFA). This paper presents experiments on using volume scan diagnosis as a component in building up an efficient EFA solution for digital circuits and discusses practical considerations in the production yield flow. It provides a brief review of the existing methods and the requirement to build an efficient flow, followed by a discussion on the proposed solution. The proposed solution can be used systematically in different stages of the production yield flow. The paper then shows industrial case studies and their results and benefits.

Abstract This paper presents the success story of the learning process by reporting four cases using four different failure analysis techniques. The cases covered are IDDQ leakage, power short, scan chain hard failure, and register soft failure. Hardware involved in the cases discussed are Meridian WS-DP, a wafer-level electrical failure analysis (EFA) system from DCG Systems, V9300 tester from Advantest, and a custom cable interface integrating WSDP and V9300 with the adaption of direct-probe platform that is widely deployed for SoC CP test. Four debug cases are reported in which various EFA techniques are proven powerful and effective, including photon emission, OBIRCH, Thermal Frequency Imaging, LVI, LVP, and dynamic laser stimulation.

Abstract Process defects, either random or systematic, are often the top killers of any semiconductor device. Process defect learning and reduction are the main focuses in both technology development stage and product manufacturing yield ramp stage. In order to achieve fast defect learning, in-line defect inspection is implemented in critical layers during wafer manufacturing. In-line defect inspection is able to detect defects. However, in-line defect inspection alone cannot predict the impact of defects on device functional yield. Failure analysis is an effective method of finding the defects which really cause device functional failures. However, often, the defects found by failure analysis are very different from the original defects, making it difficult to understand the root cause. This paper will describe a methodology how to combine in-line defect inspection and failure analysis together to found the top killer defects and accelerate their root cause identification for fast defect learning and yield improvement.

Abstract In-line E-beam inspection may be used for rapid generation of failure analysis (FA) results for low yielding test structures. This approach provides a number of advantages: 1) It is much earlier than traditional FA, 2) de-processing isn’t required, and 3) a high volume of sites can be processed with the additional support of an in-line FIB. Both physical defect detection and voltage contrast inspection modes are useful for this application. Voltage contrast mode is necessary for isolation of buried defects and is the preferred approach for opens, because it is faster. Physical defect detection mode is generally necessary to locate shorts. The considerations in applying these inspection modes for rapid failure analysis are discussed in the context of two examples: one that lends itself to physical defect inspection and the other, more appropriately addressed with voltage contrast inspection.

Abstract This paper presents the successful use of the novel inline product-like logic vehicle (PATO) during the last technology development phases of IBM's 22nm SOI technology node. It provides information on the sequential PATO inline test flow, commonality analysis procedure, and commonality signature trending. The paper presents examples of systematic defects uniquely captured by the product-like back end of the line layout. Moreover, this complex logic vehicle also uncovered a rich Pareto of more than 20 types of systematic and random defect mechanisms across the front end of the line, the middle end of the line, and the back end of the line. And more importantly, the non-defect found rate was kept below 20%. This achievement was possible by: leveraging high volume inline test ATPG scan fail data through the novel commonality analysis approach; and selecting the highest ATPG confidence defects representing a known commonality signature to physical failure analysis.

Abstract The root cause deconvolution (RCD) provides an easy-to-understand defect Pareto, together with targeted physical failure analysis candidates. Unfortunately, even the RCD analysis also has some assumptions and limitations, and its result cannot always be interpreted literally. This calls for a variety of conventional yield analysis techniques to be adopted in parallel to improve the confidence in the RCD results. This paper briefly introduces the RCD analysis and explains how it distinguishes itself from other conventional volume diagnosis analysis techniques. Its typical inputs and outputs are discussed as well. Next, the paper focuses on two case studies where the authors leverage RCD for logic yield improvement together with other conventional analysis techniques. It then proposes a comprehensive analysis system that is backed up by accumulating RCD results over time and across different design IPs.

Abstract This paper presents a backside chip-level physical analysis methodology using backside de-processing techniques in combination with optimized Scanning Electron Microscopic (SEM) imaging technique and Focused Ion Beam (FIB) cross sectioning to locate and analyze defects and faults in failing IC devices. The case studies illustrate the applications of the method for 28nm flip chip bulk Si CMOS devices and demonstrate how it is used in providing insight into the fab process and design for process and yield improvements. The methods are expected to play an even more important role during 20-nm process development and yield-ramping.

Abstract With the microelectronic technology progresses in nanometer realm, like SRAM, logic circuits and structures are also becoming dense and more sensitive to process variation. Logic failures may have different root causes from SRAM failure. If these technology weak points for logic circuits are not detected and resolved during the technology development stage, they will greatly affect the product manufacturing yield ramp, leading to longer time of design to market. In this paper, we present a logic yield learning methodology based on an inline logic vehicle, which includes several scan chains of different latch types representative of product logic. Failure analysis for the low yield wafers had revealed several killer defects associated with logic circuits. A few examples of the systematic failures unique to logic circuits will be presented. In combination with SRAM yield learning, logic yield learning makes the technology development more robust thus improving manufacturability.

Abstract Scan chain integrity yield loss is a common concern, especially in early stage of product yield ramp. Typically, scan chain failure diagnosis can only proceed upon full silicon build and structural test. In this work, we propose a proactive methodology which enables failure debug step to be initiated as early as the onset of device fabrication, to bring forward yield learning. Scan chain cells and nets information are extracted from design data file and converted to inline optical wafer inspection care areas. In this way, the inspection recipe can be optimized for the detection of scan chain related defects. It is shown experimentally that such approach can potentially enhance general defect detection sensitivity by 50% and increase the defect hit probability on scan chain nets. Any findings serve as useful early data for process improvement feedback. Furthermore, marginal defects, which otherwise are not easily revealed using conventional approach, can also be detected to provide early warning for process drifts or variations.

Abstract This work presents the first application of a diagnosis driven approach for identifying systematic chain fail defects in order to reduce the time spent in failure analysis. The zonal analysis methodology that is applied separates devices into systematic and random populations of chain fails in order to prevent submitting random defects for failure analysis. Two silicon case studies are presented to validate the production worthiness of diagnosis driven yield analysis for chain fails. The defects uncovered in these case studies are very subtle and would be difficult to identify with any other methodology.

Abstract Yield on specific designs often falls far short of predicted yield, especially at new technology nodes. Product-specific yield ramp is particularly challenging because the defects are, by definition, specific to the design, and often require some degree of design knowledge to isolate the failure. Despite the wide variety of advanced electrical failure analysis (EFA) techniques available today, they are not routinely applied during yield ramp. EFA techniques typically require a significant amount of test pattern customization, fixturing modification, or design knowledge. Unless the problem is critical, there is usually not time to apply advanced EFA techniques during yield ramp, despite the potential of EFA to provide valuable defect insight. We present a volume-oriented workflow integrating a limited set of electrical failure analysis (EFA) techniques. We believe this workflow will provide significant benefit by improving defect localization and identification beyond what is available using test-based techniques.

Abstract The application of an individual failure analysis technique rarely provides the failure mechanism. More typically, the results of numerous techniques need to be combined and considered to locate and verify the correct failure mechanism. This paper describes a particular case in which different microscopy techniques (photon emission, laser signal injection, and current imaging) gave clues to the problem, which then needed to be combined with manual probing and a thorough understanding of the circuit to locate the defect. By combining probing of that circuit block with the mapping and emission results, the authors were able to understand the photon emission spots and the laser signal injection microscopy (LSIM) signatures to be effects of the defect. It also helped them narrow down the search for the defect so that LSIM on a small part of the circuit could lead to the actual defect.

Abstract Driving yield improvement activities to decrease baseline logic / scan yield fallout is often perceived as a challenging or complex activity. Much of this perception is based on yield or device teams historical interactions with the FA teams on these types of requests. This paper will present our methodology for generating a relatively high volume of scan FA results in a short time frame. This type of high volume scan FA enables real time FA paretos of what is causing scan / logic yield loss.

Abstract The authors present a detailed analysis of a full-thickness backside trenching and contact level circuit editing methodology used to achieve a success yield of greater than 90%. The methodology involves both full-thickness backside trenching and contact level circuit modifications on flip-chip parts produced with 90nm, 65nm, and 40nm foundry processes and mounted on laminated package technology. Having successfully edited >150 parts with this methodology, the authors prove that full-thickness trenching is a viable alternative to a traditional die-thinning process, and that circuit edits at the contact layer, to avoid milling into the copper metal layers, greatly reduces risk and uncertainty.