Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

DS: Fachverband Dünne Schichten

DS 40: Focussed Session: Memristive Devices for Neuronal Systems II

DS 40.3: Vortrag

Donnerstag, 23. März 2017, 16:00–16:15, CHE 89

Implementation of memristive devices in a crossbar-based pattern recognition scheme — •Mirko Hansen, Martin Ziegler, Finn Zahari, and Hermann Kohlstedt — AG Nanoelektronik, Christian-Albrechts-Universität zu Kiel, Germany

While several neuron-based learning concepts like Hebbian learning or spike-timing-dependent-plasticity have been successfully shown using single devices, their realization of whole systems using memristive devices remains challenging. Due to the progress in the field and the increase of devices per circuit, additional problems in terms of reliability and requirements in device quality arise.

We will present simulation results for a pattern recognition scheme using the MNIST benchmark dataset[1]. Parameters for these simulations were extracted from automated pulse measurements on niobium-oxide based double barrier memristive devices (Nb/Al-AlO_x/Nb_xO_y/Au)[2]. These devices show analog switching behavior, a high resistance and a strong I-V-nonlinearity, making them good candidates for the presented pattern recognition system. Aside from general pattern recognition performance, the impact of imperfect devices for the recognition rate will be shown. This includes the whole range of fabrication problems from shorted devices to devices with a high variability in switching, over to non-switching high resistance devices. Financial support by the German Research Foundation through FOR 2093 is gratefully acknowledged.

[1] F. Zahari et al., AIMS Materials Science 2: 203-216 (2015)

[2] M. Hansen et al., Scientific Reports, vol. 5, p. 13753 (2015)