Datenbestand vom 13. Juni 2019
Tel: 089 / 66060798
Mo - Fr, 9 - 12 Uhr
Fax: 089 / 66060799
DER VERLAG IST IN DER ZEIT VOM 12.06.2019 BIS 23.06.2019 AUSCHLIESSLICH PER EMAIL ERREICHBAR.
aktualisiert am 13. Juni 2019
978-3-8439-2936-3, Reihe Informatik
Jan Micha Borrmann
Efficient Heterogeneous Multicore Architectures for Driver Assistance and Automated Driving
229 Seiten, Dissertation Eberhard-Karls-Universität Tübingen (2016), Softcover, A4
Highly and fully automated driving functions pose big challenges towards automotive processing platforms at various levels which have to be overcome to not delay their introduction to (mass) market.
In this thesis, the tiled heterogeneous processing platform HeMan together with prototypes and runtime management concepts is presented.
HeMan targets to narrow the different individually observable gaps in current processing systems with regards to processing performance, efficient hardware utilization and scalability of current automotive processing platforms.
Furthermore, the development of and for such platforms is addressed. HeMan specifically targets automotive assistance and automated driving applications, focusing on complex automotive camera-based perception and cognition functions considering their strong requirements with regards towards processing power as well as towards functional safety. This includes architectural support for heterogeneity where automotive domain-specific functionality can be to a large degree accelerated by architectural hardware support and functional safety features such as dual-mode redundancy. Architectural support is provided to enforce a separation of concerns paradigm, where individual processing tasks are required to not (accidentally) interfere with others. This includes hardware-implemented mechanisms for memory protection, fine-granular prioritization of on-chip communication as well as enforced communication characteristics.
A platform approach is used to manage the implementation of HeMan platform instances.
The design of HeMan was carried out after a thorough requirement analysis based on the results from three individual case-studies that were performed in advance. Individual case studies confirm the existence of the individual gaps towards efficient processing systems. Additionally, within these case studies, also the individual concepts that were later integrated into the single HeMan architecture were individually successfully evaluated.
As such, in this thesis, not only the proposed HeMan platform itself poses a contribution to the automotive research community, but also the individual case studies, which include multiple in-lab and vehicle demonstrators, posing relevant contributions on their own.