Memristor-Based Resistive Computing
This chapter reviews recent technology, circuits, and systems trends in memristive electronics, with particular attention to ultra-dense and energy-efficient resistive logic gates and signal processing. A reconfigurable nonvolatile computing platform that
- PDF / 1,409,393 Bytes
- 25 Pages / 439.36 x 666.15 pts Page_size
- 42 Downloads / 263 Views
Memristor-Based Resistive Computing Sung-Mo Steve Kang and Sangho Shin
10.1 Introduction In 1971, Chua published a seminal paper on memristor as a missing basic circuit element by explaining the constitutive relationship between electrical charge q and flux π linkage [1]. Chua demonstrated that the memristor can be physically realized by using other passive and active circuit elements and predicted that inherently passive memristors would be found. In 1976, Chua and Kang published a paper that defined a large class of devices and systems which they named memristive devices and systems to broaden the domain of useful nonlinear devices with memristive characteristics substantially and showed that many physical systems can be categorized as memristive devices and systems [2]. In 2008, almost 40 years later, Stan Williams and his research team at HP Labs unveiled a two-terminal titanium dioxide nanoscale device that exhibited memristor characteristics in a restricted operating range [3]. Continuing demands for more complex information processing require future systems integration to overcome various physical limitations of traditional CMOS technologies, including the lithographical limitation and the power density limit [4]. Such barriers to Moore’s Law call for revolutionary approaches to systems integration. To meet the increasingly difficult technological requirements, the emergent nanoscale resistive memory device technologies [3] have received much attention in order to help overcome the limitations of CMOS technologies. Memristive devices [2, 5] have been realized in a form of bipolar voltage-actuated nanoscale switches such that they can be used to build ultra-dense resistive memory arrays. Nanoscale memristor devices can be reconfigured into nonvolatile
S.-M.S. Kang () • S. Shin Department of Electrical Engineering, Jack Baskin School of Engineering, University of California, Santa Cruz, CA, USA e-mail: [email protected]; [email protected] R. Tetzlaff (ed.), Memristors and Memristive Systems, DOI 10.1007/978-1-4614-9068-5__10, © Springer Science+Business Media New York 2014
301
302
S.-M.S. Kang and S. Shin
memories, logic gates, programmable interconnects with a very high integration density and, more importantly, with CMOS compatibility [6, 7]. Thus, memristorsbased nanocomputing architectures offer promises for low-power and high-density computing applications, pushing Moore’s Law far beyond the present silicon roadmap horizons. A myriad of research opportunities have been opened by the memristive technologies. Besides the ultra-dense nonvolatile memory applications, they include self-adaptable analog/digital electronics [8–12], resistive signal processors [13], and synaptic neuromorphic networks [14]. One of the most plausible architectures is the CMOS/Molecular hybrid (CMOL) [6] which can be implemented by using Field Programmable Nanowire Interconnect (FPNI) [7, 15]. This is based on hybrid circuits composed of a conventional CMOS layer connected to multiple crossbar layers that contain memristive
Data Loading...