h1

TY  - THES
AU  - Staudigl, Felix
TI  - Towards trustworthy neuromorphic computing: an analysis of hardware security and reliability risks
PB  - Rheinisch-Westfälische Technische Hochschule Aachen
VL  - Dissertation
CY  - Aachen
M1  - RWTH-2025-01929
SP  - 1 Online-Ressource : Illustrationen
PY  - 2025
N1  - Veröffentlicht auf dem Publikationsserver der RWTH Aachen University
N1  - Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2025
AB  - The von Neumann bottleneck poses a significant barrier, limiting the computing performance and energy efficiency of conventional computing systems. Neuromorphic computing seeks to overcome this bottleneck by leveraging the intrinsic parallelism and efficiency of the brain-inspired Computing-in-Memory (CIM) paradigm. Nonetheless, the immature state of the foundational building block, memristive devices, introduces substantial challenges in terms of their reliability and vulnerability to hardware security threats, necessitating thorough analysis to ensure the development of secure and reliable computing systems for future applications. To provide an in-depth investigation of the reliability concerns, a fault injection platform is introduced evaluating the robustness of digital CIM operations. This platform marks a significant contribution to understanding and mitigating reliability issues in memristor-based systems by providing a holistic simulation framework operating on the crossbar and operational level. Moreover, the work presents NeuroHammer, a novel hardware security attack exploiting the unique properties of memristive crossbar arrays to compromise the integrity of neuromorphic computing systems. This attack underscores the susceptibility of Resistive Random-Access Memories (ReRAMs) to hardware security threats, highlighting the critical need for effective countermeasures. To perform a detailed reliability evaluation on actual devices, this dissertation unveils the NeuroBreakoutBoard (NBB)—a highly versatile instrumentation platform designed to examine the effects of memristive device nonidealities across various abstraction levels. The NBB’s ability in executing CIM operations on real memristive crossbar arrays sets it apart from previously proposed platforms, emphasizing its essential role in facilitating thorough analyses for the advancement of dependable neuromorphic computing platforms. In conclusion, this thesis delivers a comprehensive evaluation of memristive devices, spanning from reliability issues to hardware security threats, thereby paving the way for their broad adoption and integration into the next era of computing systems.
LB  - PUB:(DE-HGF)11
DO  - DOI:10.18154/RWTH-2025-01929
UR  - https://publications.rwth-aachen.de/record/1005679
ER  -