Abstract:Scanning Acoustic Microscopy is a dedicated powerful technology in NDT to analyze and characterize bonded interfaces for different technologies. As the complexity of the semiconductor package grows, even seemingly insignificant faults become quite important. Main applications include detection of voids, inclusions and delaminated areas in bonding interfaces and thickness variations in layer structures. The SAM operates with the pulse reflection method. The acoustic lens transforms high frequency electromagnetic vibrations that are transmitted inside the lens as a planar parallel wave field. Through the coupling medium, which is water, the cavity concentrates the sound field on the sample. The sound pulses reflected from the sample are gathered up by the acoustic lens and are converted into electromagnetic pulses by the transducer. These pulses are reconstructed into images with defined gray values. Pixels with certain gray levels are used to depict pulses. New developments in SAM combine high throughput inspection and high resolution down to the sub-micrometer range.The wafer bow is quickly eliminated by altering the array transducer focus point throughout the entire scanning procedure. New techniques for SAM's bonded interface protection keep the wafer almost completely dry throughout inspection. In order to discover and categorize failures with the least amount of data, research focuses on deep learning networks and scanning acoustic microscopes for automated failure assessment. High speed automated wafer-level inspections were made possible by the development of image and signal-based machine learning techniques. These operations can be carried out at both the wafer level and the die level without the need for sample preparation. The automatic handling, scanning, drying, and separation of good wafers from defected wafers based on acoustic data made possible by technological developments in SAM systems. Despite the high sensitivity, minute variations in the reflected signal that would point to a flaw can be lost during image reconstruction. More emphasis is being placed on creating AI-based algorithms for highly accurate automated signal interpretation for failure detection as a result of recent breakthroughs in AI (Artificial Intelligence). Since it requires less human work, automated failure analysis is becoming more and more popular for the detection and qualification of wafer bond integrity. This new technique trains the network for automated fault prediction using raw A-Scan data. In the first phase, die level prediction is reviewed and in the later phase, a wafer-level high volume inspection is performed. Combining SAM with an A-Scan-based Deep Learning approach primarily benefits non-destructive analysis with less labelling effort and training time. With minimal human effort, this process flow is perfect for automated high-volume inspection. SAM also uses image based machine learning approach for automated defect analysis. The U-net machine-learning algorithm serves as the foundation for the integrated image analysis program. Deep Learning technologies is integrated into a new modular image analysis (MIA) architecture with built-up integrated pipeline for Label Studio Software-based Deep Learning activities (labelling, training, analysis). The built-up pipeline is used to identify, train, and test various classes of defect in a bonded wafer. This method employs model confidence to manually mark marginal and defective predictions. High resolution and high throughput analysis for the defect characterization and defect analysis of various types of bonded wafers is improved in scanning acoustic microscopes using the array transducers, signal processing methods like SAFT, Wavelet filtering techniques, Artificial intelligence methods based on both image and signal. With all these advanced technologies, Scanning Acoustic Microscopy is capable of inspection of wafer bond integrity, TSV characterization, 3D package evaluation, Wafer Level Chip Scale Packaging (WL-CSP) and Crack inspection. For the metrology of mass production, from initial research and development, our equipment services and software algorithms assist manufacturers of nano electronics in better defect analysis and material characterization. Next Generation transducers were designed and developed in particular for WL-CSP applications.

Advances in machine learning and deep learning applications towards wafer map defect recognition and classification: a review

Observational and Experimental Insights into Machine Learning-Based Defect Classification in Wafers

Machine Learning Based Wafer Defect Detection

Review of Wafer Surface Defect Detection Methods

Inspection and Classification of Semiconductor Wafer Surface Defects Using CNN Deep Learning Networks

Wafer Defect Localization and Classification Using Deep Learning Techniques

Convolutional Neural Network for Semiconductor Wafer Defect Detection

A Deep Convolutional Neural Network for Wafer Defect Identification on an Imbalanced Dataset in Semiconductor Manufacturing Processes

Machine Learning-Based Detection Method for Wafer Test Induced Defects

Deep Learning-Based Detection of Defects in Wafer Buffer Zone During Semiconductor Packaging Process

Wafer map defect pattern detection method based on improved attention mechanism

AI classification of wafer map defect patterns by using dual-channel convolutional neural network

Application of Machine Learning Algorithm for Defect Analysis in Semiconductors Using High Resolved Scanning Acoustic Microscopy

A Deep Convolutional Neural Network-Based Multi-Class Image Classification for Automatic Wafer Map Failure Recognition in Semiconductor Manufacturing

Wafer Map Defect Patterns Semi-Supervised Classification Using Latent Vector Representation

Mixed-Type Wafer Failure Pattern Recognition

Auto-Labeling for Pattern Recognition of Wafer Defect Maps in Semiconductor Manufacturing

Recognition and Classification of Mixed Defect Pattern Wafer Map Based on Multi-Path DCNN

Semi-Supervised Framework for Wafer Defect Pattern Recognition with Enhanced Labeling.

Classification of Silicon (Si) Wafer Material Defects in Semiconductor Choosers using a Deep Learning ShuffleNet-v2-CNN Model

An Evaluation of Continual Learning for Advanced Node Semiconductor Defect Inspection