Nikita Alexis Jacques Caudieux
Characterization, Simulation and AI implementation of pixel/photodiodes sensors.
Rel. Francesco Bertazzi, Andrej Suler, Jonathan Miquel. Politecnico di Torino, Corso di laurea magistrale in Nanotechnologies For Icts (Nanotecnologie Per Le Ict), 2026
Abstract
Image sensors are central to modern AI-driven applications such as facial recognition, autonomous driving, and Optical Character Recognition (OCR), yet physical CMOS sensors are inherently limited by electronic noise and manufacturing defects. Evaluating AI robustness directly on hardware prototypes is costly and difficult to control, motivating the need to bridge the “sim-to-real” gap between physical sensor imperfections and their downstream effect on AI performance. This internship, conducted within Apple’s Camera Sensing Technologies group in Grenoble, addresses this problem in three stages using the Maneki 2.25 µm dual-mode (Rolling Shutter / Global Shutter) image sensor as a case study. First, a full opto-electronic and dark-current characterization compares the RS and GS architectures across Charge-to-Voltage Factor, linearity, noise decomposition (temporal, fixed-pattern, row/column), dark current and DCNU thermal scaling (via an Arrhenius doubling-temperature model and activation-energy extraction), white-pixel and RTS defect populations, black-sun saturation behavior, and HDR operation.
Results show RS consistently outperforms GS in read noise, FPN, and dark shading, while GS offers a higher full-well capacity and eliminates motion artifacts at the cost of a higher noise floor and stronger RTS tail, driven by its additional storagenode/ transistor architecture
Relatori
Anno Accademico
Tipo di pubblicazione
Numero di pagine
Informazioni aggiuntive
Corso di laurea
Classe di laurea
Aziende collaboratrici
URI
![]() |
Modifica (riservato agli operatori) |
