May_EDFA_Digital

edfas.org 13 ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 22 NO. 2 NOTEWORTHY NEWS 2020 ESREF The 31st European Symposium on Reliability of Electron Devices, Failure Physics and Analysis (ESREF 2020) will take place October 5-8, in Athens, Greece, at the Technopolis, an industrial museum and major cultural venue. This international symposium continues to focus on recent developments and future directions in failure analysis, quality and reliability of materials, and devices and circuits for micro, nano, and optoelectronics. It provides the leading European forum for developing all aspects of reliability management and innovative analysis techniques for present and emerging semiconductor applications. In light of COVID-19, video-recorded presentations will be accepted this year for authors unable to travel to the confer- ence site. For more information, visit esref2020.sciencesconf.org. GUEST EDITORIAL CONTINUED FROM PAGE 2 anomaly captured by an imaging screening method). By incorporating the raw data of multiple data streams into the predicted outcome confidence levels, ML can enable the critical ability to bin more complex failures and down select these cases for full FI/FA flows. Even after reducing the number of samples requiring FI/FA, labsmust still contendwith the increased complex- ity for those samples that do require root cause investiga- tion. To that end, AI methods provide broad opportunity in metrology and defect classification for imaging and material analysis. In general, AI-based image and spectral analysis can enhance detectability of defects, perform automated metrology, standardize measurements, and accelerate the pace of analysis. For destructive analysis, optical and electron imaging are typically used to visualize and measure features, and AI can be used to detect the interfaces betweenwhichmeasurements are taken. While traditional image analysismethods can do the same, they requirebounded customization for the applicationat hand andmust be rebuilt for products and structures that differ in any significant way. Machine learning and especially deep learning (DL) methods are more flexible in their applicationand canbe taught new functionsmorequickly. For nondestructive analysis, acoustic, optical (UV/Vis/ IR), and x-ray imaging are used to detect and visualize physical defects. Typically for FA applications, imaging of this type is analyzed manually because (a) the data is collected for a small number of samples and (b) the variable cases cannot be standardized for automation. However, with recent advances in open source ML and DL libraries pre-packaged for rapid image training and classification, it is now possible for FA labs to apply these methods to enable rapid, automated analysis of themore variable image data typically associatedwith nondestruc- tive FA. Today’s FA imaging tools are typically optimized for single sample analysis at high resolution—trading analy- sis time for resolution and detectability. For this reason, FA tools are not traditionally useful for high throughput metrologies. However, these AI methods now can enable FA metrologies. While factory metrologies typically take only a few seconds to a fewminutes to complete analysis for one sample, these FA metrologies may take several minutes up to a few hours per sample. These can become relevant or even critical to product and process develop- ment when new packaging architectures present issues that cannot be capturedusing factorymetrologies, or to FA workloads when complex failures are difficult to precisely fault isolate. Examples of tools that can be coupled with AI image automation to create FA metrologies include scanning acoustic microscopy, x-ray computed tomogra- phy, and scanning optical imagingmethods. In summary, methods of AI can and will pave new paths for FA labs to improve capability and productivity in support of rapidly changing packaging technology development.