PhD in Security of 3D/chiplet-based AI accelerators - TIMA

Descrizione

PhD in Security of 3D/chiplet-based AI accelerators - TIMA Sintesi in italiano (traduzione automatica): L'organizzazione TIMA offre un'opportunità di dottorato in ingegneria elettronica, focalizzata sulla sicurezza degli acceleratori AI basati su architetture 3D e chiplet. La sede del progetto è legata al PEPR-Agency Camelia, e il candidato dovrà affrontare sfide relative alla sicurezza, autenticazione e fiducia nei chiplet. Le principali mansioni includono la caratterizzazione sperimentale di architetture PUF su chiplet reali e la progettazione di algoritmi e protocolli per la sicurezza distribuita. È richiesta una laurea in ingegneria elettronica o un campo affine, con competenze in progettazione hardware e sicurezza informatica. Il candidato ideale avrà una solida comprensione delle proprietà dei PUF e delle tecniche di integrazione dei chiplet. 3D and chiplet-based architectures represent a promising direction for future hardware accelerators dedicated to artificial intelligence. They enable the heterogeneous integration of computing units, memories, communication interfaces, and specialized functions in close physical proximity, thereby improving performance, bandwidth, and energy efficiency. However, this distributed integration also introduces new challenges in terms of security, authentication, and trust, especially when the different chiplets may originate from distinct design, manufacturing, or assembly supply chains. This PhD thesis proposes to investigate a new hardware security approach based on the concept of chiplet-based PUFs. The principle consists in exploiting two identically designed chiplets, each integrating a set of ring oscillators. The PUF responses are generated by comparing oscillators located at corresponding physical positions in the two chiplets. This approach eliminates, by construction, biases related to placement, routing, and design asymmetries, while relying on local and random manufacturing process variations as a source of entropy. Main activities: - A first objective of the thesis will be to experimentally characterize this PUF architecture on real chiplets produced in the framework of the PEPR-Agency Camelia project on AI components. This will involve evaluating the classical properties of PUFs, including uniqueness, reliability, uniformity, entropy, and robustness against voltage, temperature, and operating-condition variations. - A second objective will be to design new algorithms and protocols for distributed security based on this inter-chiplet PUF. The central idea is that the secret should no longer be carried by a single component, but should instead emerge from the physical interaction between multiple chiplets. This approach could be used for mutual authentication of chiplets, detection of substitutions or counterfeits, generation of shared keys, and verification of the integrity of a 3D assembly after integration. Annuncio in inglese. Fonte: Euraxess (Commissione europea).