Modeling and simulating in-memory memristive deep learning systems: an overview of current efforts
Journal Publication ResearchOnline@JCUAbstract
Deep Learning (DL) systems have demonstrated unparalleled performance in many challenging engineering applications. As the complexity of these systems inevitably increase, they require increased processing capabilities and consume larger amounts of power, which are not readily available in resource-constrained processors, such as Internet of Things (IoT) edge devices. Memristive In-Memory Computing (IMC) systems for DL, entitled Memristive Deep Learning Systems (MDLSs), that perform the computation and storage of repetitive operations in the same physical location using emerging memory devices, can be used to augment the performance of traditional DL architectures; massively reducing their power consumption and latency. However, memristive devices, such as Resistive Random-Access Memory (RRAM) and Phase-Change Memory (PCM), are difficult and cost-prohibitive to fabricate in small quantities, and are prone to various device non-idealities that must be accounted for. Consequently, the popularity of simulation frameworks, used to simulate MDLS prior to circuit-level realization, is burgeoning. In this paper, we provide a survey of existing simulation frameworks and related tools used to model large-scale MDLS. Moreover, we perform direct performance comparisons of modernized open-source simulation frameworks, and provide insights into future modeling and simulation strategies and approaches. We hope that this treatise is beneficial to the large computers and electrical engineering community, and can help readers better understand available tools and techniques for MDLS development.
Journal
Array
Publication Name
N/A
Volume
13
ISBN/ISSN
2590-0056
Edition
N/A
Issue
N/A
Pages Count
9
Location
N/A
Publisher
Elsevier
Publisher Url
N/A
Publisher Location
N/A
Publish Date
N/A
Url
N/A
Date
N/A
EISSN
N/A
DOI
10.1016/j.array.2021.100116