Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz
AKPIK 2: Machine-learning methods and computing in particle physics
AKPIK 2.8: Vortrag
Dienstag, 26. März 2019, 17:10–17:20, H10
HfO2-based Memristive Navigation Processor — Karlheinz Ochs, •Enver Solan, Dennis Michaelis, and Leonard Hilgers — Ruhr-University Bochum, Bochum, Germany
Mathematically complex optimization problems are historically interesting subjects of research. That is because convergence time of such problems does not scale well with the complexity. One of these problems is the np-hard navigations process problem, where the shortest path between an entry and an exit in a maze is desired. Electrical circuits are proper tools to solve this problem efficiently and the reasons are twofold. First, a current naturally chooses the path of least resistance in any given circuit, which is a desirable feature in this context. Second, electrical circuits are inherently massively parallel which lead to fast convergence times. To this end, a self-organizing electrical circuit with HfO2-based resistive random access memory-cells (RRAM-cells) is proposed that obtains a solution within a convergence time that is linearly proportional to the length of the shortest path. RRAM-cells are known to have a fast switching behavior which is favorable for quick convergence. All necessary details on how to construct an arbitrarily sized maze and what occurs in the switching process of the RRAM-cells to obtain the optimal solution are presented. The authors also propose an emulator of the electrical circuit based on the wave digital method which could be implemented in embedded systems, e.g. digital signal processors or field programmable gate arrays. Different simulation scenarios confirm the previously derived theoretical results.