Skip to main content
Video s3
    Details
    Presenter(s)
    Junren Chen Headshot
    Display Name
    Junren Chen
    Affiliation
    Affiliation
    University of Zurich
    Country
    Author(s)
    Display Name
    Junren Chen
    Affiliation
    Affiliation
    University of Zurich
    Display Name
    Chenxi Wu
    Affiliation
    Affiliation
    Institute of Neuroinformatics, University of Zurich and ETH Zurich
    Display Name
    Giacomo Indiveri
    Affiliation
    Affiliation
    University of Zürich and ETH Zürich
    Display Name
    Melika Payvand
    Affiliation
    Affiliation
    University of Zurich/ ETH Zurich
    Abstract

    Memristors are commonly used in crossbar arrays as “in-memory computing” elements to solve the von-Neumann bottleneck problem. However, they can also be used as “in-memory routing” elements to configure on-chip interconnection schemes and route signals among computing elements in configurable multi-core neuromorphic processors. While there has been a significant focus on the use of memristive devices as in-memory computing elements, to date, studies on the fundamental reliability properties of memristors as routing elements are still missing. In this paper, we analyze the reliability issues of using these devices in routing crossbar arrays, caused by sharing routing resources (collisions), and undesired pulses due to the leakage paths (on/off ratio requirement). We show that there is a trade-off between routing collision probability and the degree of connectivity (i.e., fan-in) of the receivers sharing routing channels. We provide specifications and guidelines based on a theoretical analysis for engineering the properties of memristive devices, and for designing routing systems based on memristor crossbars.

    Slides
    • Reliability Analysis of Memristor Crossbar Routers: Collisions and On/Off Ratio Requirement (application/pdf)