Performing Ultra-Low-Power Matrix-Vector Multiplications using Topological-Insulators
Abstract
Several key challenges present themselves to realize energy-efficient detection, localization and isotope identification. We will focus on drastically reducing the power consumption for the matrix-vector multiplications that lie at the heart of least-squares or modified least squares methods of the algorithmic detection of isotopes in spectra. Present-day computational approaches are increasingly limited by the power consumption associated with memory and transferring data between different memories. We propose to develop topological-insulator random-access memory (TIRAM) that is non-volatile, energy efficient, high-density, and does not require dedicated read-out circuitry. To realize such a device, we start from the exciting recent discovery of large-gap two-dimensional topological insulators (2D TIs). We will study 2D TIs coupled with 2D ferromagnets to exploit the ability of 2D TIs to inject electrons with a desired spin with 100% efficiency into ferromagnetic storage layers. We will study TIRAM intrinsic performance, the robustness against imperfections offered by topological protection, and the efficiency of performing matrix-vector multiplications using TIRAMs. In a second stage of the research, we will investigate the combination of the newly developed memory with TI-based field-effect transistors (TIFETs) which have recently been shown to exhibit lower power consumption and higher performance than any alternative emerging computing device. The co-integration of TIRAMs with TIFETs will yield immediate significant power savings and can eventually lead to non-von Neumann brain-inspired artificial neural networks that can perform pattern recognition tasks orders of magnitudes more efficiently compared to any conventional approach.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 16, 2019
- Source ID
- HDTRA11810018
Entities
People
- William G Vandenberghe
Organizations
- Defense Threat Reduction Agency
- University of Texas at Dallas