View : 6 Download: 0

Segment-aware energy-efficient management of heterogeneous memory system for ultra-low-power IoT devices

Title
Segment-aware energy-efficient management of heterogeneous memory system for ultra-low-power IoT devices
Authors
Choi H.Koo Y.Park S.
Ewha Authors
박상수
SCOPUS Author ID
박상수scopus
Issue Date
2017
Journal Title
2017 2nd International Multidisciplinary Conference on Computer and Energy Science, SpliTech 2017
Keywords
Heterogeneous memoryInternet of ThingsPower consumptionSegmentUltra-Low-Power
Publisher
Institute of Electrical and Electronics Engineers Inc.
Indexed
SCOPUS scopus
Abstract
The emergence of IoT (Internet of Things) has brought various studies on low-power techniques back to embedded systems. In general, minimizing power consumed by program executions is the main consideration of system design. While running programs, executing data-dependent program code results in a large number of memory accesses which consume huge amount of power. Further, most embedded systems consist of multiple types of memory devices, i.e., heterogeneous memory system, to benefit the different characteristics of memory devices. In this paper, we conduct a research on low-power techniques to reduce the power consumption of heterogeneous memory to achieve ultra-low-power in the system level. This study proposes a segment-aware energy-efficient management to improve the power efficiency considering the characteristics and structures of the memory devices. In the proposed approach, the technique migrates program code from allocated memory device to another if the consumed power is considered to be less. We also analyze and evaluate the comprehensive effects on energy efficiency by applying the technique as well. Compared to the unmodified program code, our model reduces power consumption up to 12.98% by migrating functions. © 2017 Univeristy of Split, FESB.
ISBN
9789532900712
Appears in Collections:
엘텍공과대학 > 컴퓨터공학과 > Journal papers
Files in This Item:
There are no files associated with this item.


qrcode

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

BROWSE