This survey reviews anomaly detection (AD) approaches in semiconductor manufacturing(SM). Analyzing six studies (from 2017 to 2025), methods are categorized by ML/DL approaches, data types, metrics, and semiconductor manufacturing processes. The findings highlight the evolution from classical ML to DL methods; osPCA, ABOD, and LOF are compared in [1]; in [2]VAD approaches are compared, with CFA and STFPM achieving best results; in [3] a LSTM autoencoder with SVDD objective function AD model is proposed; in [4]the effectiveness of DIFFI method for interpretability of IF is show; in [5] a pipeline composed of Encoder, Feature selection, and Ad approach is shown; in [6] MT-Former that learns new defects is proposed.
Questa survey esamina approcci di anomaly detection (AD) nel campo del semiconductor manufacturing(SM). Analizzando sei studi (dal 2017 al 2025), i metodi sono categorizzati in base ad approcci ML/DL, tipi di dati, metriche e processi SM. I risultati evidenziano l’evoluzione dai metodi ML classici ai metodi DL; osPCA, ABOD e LOF sono confrontati in [1]; in [2] sono confrontati approcci VAD, con CFA e STFPM che ottengono i migliori risultati; in [3] è proposto un modello AD basato su autoencoder LSTM con objective function SVDD; in [4] è mostrata l’efficacia del metodo DIFFI per l’interpretabilità di IF; in [5] è presentata una pipeline composta da Encoder, selezione delle feature e approccio AD; in [6] è proposto MTFormer che apprende nuovi difetti.
A survey on Anomaly Detection approaches in Semiconductor Manufacturing
PATRIZIO, RICCARDO
2025/2026
Abstract
This survey reviews anomaly detection (AD) approaches in semiconductor manufacturing(SM). Analyzing six studies (from 2017 to 2025), methods are categorized by ML/DL approaches, data types, metrics, and semiconductor manufacturing processes. The findings highlight the evolution from classical ML to DL methods; osPCA, ABOD, and LOF are compared in [1]; in [2]VAD approaches are compared, with CFA and STFPM achieving best results; in [3] a LSTM autoencoder with SVDD objective function AD model is proposed; in [4]the effectiveness of DIFFI method for interpretability of IF is show; in [5] a pipeline composed of Encoder, Feature selection, and Ad approach is shown; in [6] MT-Former that learns new defects is proposed.| File | Dimensione | Formato | |
|---|---|---|---|
|
Patrizio_Riccardo.pdf
Accesso riservato
Dimensione
764.78 kB
Formato
Adobe PDF
|
764.78 kB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/104191