edfas.org 41 ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 26 NO. 1 to improve workflows. Libor Strakos, a Thermo Fisher representative, queried the audience if they would prefer to have the infrastructure to create tunable custom end user models, or be provided with a packaged solution that solves the specific problem directly. Haimson replied to this query stating his goal was to highlight the online tools that are becoming available to the general community and the opportunities derived if equipment vendors provided direct access to relevant signals to allow custom signal processing, such as endpoint signal traces. Valerie Brogden pointed out that with vendor API tools, it should be possible to access signals, such as the end point trace, and interface them with custom machine vision or machine learning modules. Steve Herschbein indicated that whole wafer defect review tools had basic automatic defect classification (ADC) capabilities 20 years ago and asked which aspects of those capabilities are transferable to today’s platforms. Edward Principe commented, having worked at Applied Materials and programmed their commercial whole wafer defect review tools for ADC, that the key to enabling the ADC workflow is the ability to seamlessly integrate the FA image database content as integral training data to create custom AI models at the end user level, where the nature of the problem is understood most intimately. Importantly, this approach of using data training sets which are open and transparent to the end users is also consistent with the ethical creation of AI models to ensure that the training data is not corrupted or manipulated by third parties. Another participant indicated he sees categories of AI and machine vision tools becoming future industry standards. An Infineon representative indicated they too are beginning to utilize their in-house FA imaging data as training collateral and pointed toward a specific use case of AI-based automated crack detection, which their team presented this year at ISTFA. A representative of Covalent Metrology highlighted the machine vision use cases that can be developed, as distinct from more complex AI-based models, especially to sort and classify large datasets of images. Scott Lockledge, co-founder of Tiptek LLC focused on the efficiency improvements in lamella transfer realized by employing a longer and higher aspect ratio than the traditional taper angle lift-out needle. They use a combination of technologies to produce the high aspect ratio tips including electrochemical etching and ion beam shaping. The core time-savings is related, in part, to reducing the time required reshaping or resharpening the lamella transfer needle. In a study conducted in cooperation with NIST, their model showed a 26% throughput increase (from 23 to 30 lamella transfers/24hrs). The fourth speaker, Valerie Brogen, presented a novel glove box which may be attached to the port of a SEM or FIB-SEM vacuum chamber. The new product is called Noble Dome, which she co-developed at the CAMCOR facility at the University of Oregon in collaboration with her colleagues. Applications include sample preparation for TEM/STEM, and other in-situ analyses, without exposure to air. Inert sample preparation and analysis can be especially important for studies on battery materials, as well as extra-terrestrial samples collected from space. The session was then opened to the attendees for general questions. The first question related to the use of alternate ion species for the purpose of TEM/STEM lamella preparation. The general consensus was that for standard lamella size (~10 µm x 5 µm), a gallium liquid metal ion source is still more efficient, when a sample does not suffer from gallium ion reactivity. In terms of the utility of a plasma FIB, there can be a benefit for preparation of large area lamella (e.g., > 20 µm across). The fundamental reasons gallium ion performance is more efficient for typical lamella dimensions is the source brightness and beam profile current density of a gallium liquid metal ion point source is superior to a collimated plasma ion source at moderate and lower beam currents. A plasma ion source has higher angular intensity, which delivers beam currents in the microamp range to more rapidly remove material over large areas with a better beam profile than a gallium ion source operating in the microamp current ranges. However, when a fine beam profile is required such as during the final milling steps of lamella preparation, the larger beam tails of a collimated plasma ion source operating in the picoamp range causes more rapid erosion of the protective layer and therefore less efficient milling, as compared to the tighter beam profile of a gallium ion point source operating in the picoamp current range. Backside (inverted) lamella preparation can mitigate beam profile erosion disadvantages of a plasma FIB in the low current regime. Rick Livengood from Intel highlighted the challenges of material contrast and end-pointing during “CIRCUIT EDIT CONTINUES TO BE A MONUMENTAL CHALLENGE AND INCREASINGLY REQUIRES POWER CONNECTIONS ON THE BACKSIDE, CONCURRENT WITH SIGNAL COLLECTION FROM THE FRONTSIDE OF AN ACTIVE DEVICE UNDER INVESTIGATION.”
RkJQdWJsaXNoZXIy MTYyMzk3NQ==