TIMA laboratory

PhD thesis defenses


« New performance evaluation methods for early and refined software development on SoC platforms ».

Candidate: P. Njoyah Ntafam

Advisor: F. Pétrot

President of jury: F. Rousseau

PhD: These de Doctorat, Université de Grenoble

Defense: Le 20/04/2018 à 10 h 00 GRENOBLE INP (Viallet) - Amphi Gosse

Abstract

The thesis aims to identify, develop and experiment methods, based on models that will allow the performance estimation of embedded software on-chip (System-On-Chip, SoC), on early stage of a SoC project. In fact, current design methods, that are not running software, are no longer sufficient for the complexity of new architectures, composed of both multi-processors with cache coherence and dedicated subsystems. The challenge is to determine whether a combination of models can provide both the accuracy of required estimation, and simulation speed expected at the beginning of a SoC project, by hardware, system, and software architects.

 

« Optimization of the operation of a generator of memory hierarchies for embedded vision systems ».

Candidate: K. Hadj Salem

Advisor: S. Mancini

PhD: These de Doctorat, Université de Grenoble

Speciality: Micro et Nano Electronique

Defense: Le 26/04/2018 à 14 h 00 ESISAR - Amphi A042 (50 rue Barthélémy de Laffemas - BP 54 - 26902 VALENCE)

Abstract

The research of this thesis focuses on the application of the Operations Research (OR) methodology to design new optimization algorithms to enable low cost and efficient embedded vision systems, or more generally devices for multimedia applications such as signal and image processing. For the case of non-linear image accesses, one solution has been proposed by Mancini et al. (Proc. DATE 2012) in the form of a software tool, called Memory Management Optimization(MMOpt), that creates an ad-hoc memory hierarchies for such a treatment. It creates a circuit called a Tile Processing Unit (TPU) that contains the circuit for the treatment. In this context, we address the optimization challenge set by the efficient operation of the circuits produced by MMOpt to enhance the 3 main electronic design characteristics. They correspond to the energy consumption, performance and size/production cost of the circuit. This electronic problem is formalized as a 3-objective scheduling problem, which is called 3-objective Process Scheduling and Data Prefetching Problem (3-PSDPP), reflecting the 3 main electronic design characteristics under consideration. To the best of our knowledge, this problem has not been studied before in the OR literature. A review of the state of the art, including the previous work proposed by Mancini et al.(Proc.DATE, 2012) as well as a brief overview on related problems found in the OR literature, is then made. In addition, the complexity of some of the mono-objective sub-problems of 3-PSDPP problem is established. Several resolution approaches, including exact methods (ILP) and polynomial constructive heuristics, are then proposed. Finally, the performance of these methods is compared, on benchmarks available in the literature, as well as those provided by Mancini et al. (Proc.DATE, 2012), against the one currently in use in the MMOpt tool. The results show that our algorithms perform well in terms of computational efficiency and solution quality. They present a promising track to optimize the performance of the TPUs produced by MMOpt. However, since the user's needs of the MMOpt tool are contradictory, such as low cost, low energy and high performance, it is difficult to find a unique and optimal solution to optimize simultaneously the three criteria under consideration. A set of good compromise solutions between these three criteria was provided. The MMOpt's user can then choose the best compromise solution he wants or needs.