We introduce machine learning-based solutions for differentiating wafers fabricated using trusted and untrusted mask-sets based on the typical metrology or wafer acceptance tests collected during semiconductor manufacturing and testing. Our methods leverage the systematic nature of process variation and capture the subtle causality between mask modifications and either physical dimensions or electrical characteristics of the produced silicon, which can then be used for the purpose of wafer-level mask-set attestation. Effectiveness of our solutions is demonstrated on a dataset of inline and e-test measurements from 8000 wafers fabricated with multiple variants of a mask-set in the GlobalFoundries 12LP FinFET technology node.

We demonstrate the value of integrating expert-level domain knowledge into Machine Learning (ML) model training, which becomes particularly important when modeling complex processes such as semiconductor manufacturing. Specifically, we discuss a machine learning-based methodology which correlates physical metrology measurements with process control monitoring electrical measurements by employing Multivariate Adaptive Regression Splines (MARS) and Non-Dominating Sorting Genetic Algorithm II (NSGA-II). Baseline effectiveness of this solution in predicting critical measurements for maintaining fabrication process integrity, such as yield shorts, ring oscillator active mode current (IDDA) and frequency differences, is assessed using actual High Volume Manufacturing (HVM) production data from an advanced FinFET technology node. Further improvements, however, can be obtained by leveraging domain-specific expertise. Indeed, as we demonstrate experimentally, model accuracy, training time, and explainability all improve when such expertise is integrated in the training process. Our results highlight the pitfalls of blindly applying machine learning and illustrate the value of including semiconductor experts in the development of machine learning models for process optimization-related tasks.

We discuss the problem of generating synthetic Process Control Monitoring (PCM) data and evaluating how accurately it reflects the distribution of actual measurements from manufactured wafers. PCMs are small test structures placed in the scribe lines of the wafer, on which electrical tests (E-tests) are conducted to monitor the impact of process variation on a manufactured wafer. Besides its immediate use in assessing and controlling wafer health, collective PCM data holds invaluable information for process engineers who seek to maximize yield and performance across process corners. Yet availability of such data is limited during the ramp-up phase of a process, when it is needed the most. To address this limitation, we introduce a methodology that leverages correlations across E-test measurements and across wafer locations to generate a large synthetic population from a small data sample. Furthermore, we discuss statistical metrics that can be used to evaluate the accuracy of the synthetically generated vis-à-vis the actual population. Effectiveness of our solution is experimentally validated using E-test data from ~8K wafers fabricated in an advanced GlobalFoundries FinFET node.

In this paper, we introduce an innovative super resolution approach to emerging modes of near-field synthetic aperture radar (SAR) imaging. Recent research extends convolutional neural network (CNN) architectures from the optical to the electromagnetic domain to achieve super resolution on images generated from radar signaling. Specifically, near-field synthetic aperture radar (SAR) imaging, a method for generating high-resolution images by scanning a radar across space to create a synthetic aperture, is of interest due to its high-fidelity spatial sensing capability, low cost devices, and large application space. Since SAR imaging requires large aperture sizes to achieve high resolution, super-resolution algorithms are valuable for many applications. Freehand smart-phone SAR, an emerging sensing modality, requires irregular SAR apertures in the near-field and computation on mobile devices. Achieving efficient high-resolution SAR images from irregularly sampled data collected by freehand motion of a smartphone is a challenging task. In this paper, we propose a novel CNN architecture to achieve SAR image super-resolution for mobile applications by employing state-of-the-art SAR processing and deep learning techniques. The proposed algorithm is verified via simulation and an empirical study. Our algorithm demonstrates high-efficiency and high-resolution radar imaging for near-field scenarios with irregular scanning geometries.