Selective policies for efficient state retention in transiently-powered embedded systems: Exploiting properties of NVM technologies

Verykios, Theodoros D.; Balsamo, Domenico; Merrett, Geoff V.

doi:10.1016/j.suscom.2018.07.003

Cited by 11 publications

(6 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, the inference of a convolutional neural network requires a significant amount of memory to perform the computations, specifically for storing intermediate results and the network parameters. Non-volatile memory, such as flash memory, is typically used to store the parameters of the neural network but the number of read accesses to this type of memory should be minimized since the energy consumption is typically about 6x as high as reading from SRAM [42]. As a consequence the amount of memory accesses required for loading parameters should be reduced, for example by binarization of the network [23].…”

Section: Classification With Time Distributed Processingmentioning

confidence: 99%

“…The number of parameters is 38,403 which requires 153.6 kB of flash memory using 32-bit values. It is possible to store this amount in flash memory but read accesses to flash should be minimized due to the higher power consumption in comparison to reading from SRAM [42]. We therefore apply Incremental Network Quantization [48] which quantizes the parameters to power-of-two values in an iterative weight partition and quantization process.…”

Section: Implementation Challenges On Embedded Devicesmentioning

confidence: 99%

See 1 more Smart Citation

Event-triggered natural hazard monitoring with convolutional neural networks on the edge

Meyer

Farei-Campagna

Pasztor

et al. 2019

Proceedings of the 18th International Conference on Information Processing in Sensor Networks

View full text Add to dashboard Cite

In natural hazard warning systems fast decision making is vital to avoid catastrophes. Decision making at the edge of a wireless sensor network promises fast response times but is limited by the availability of energy, data transfer speed, processing and memory constraints. In this work we present a realization of a wireless sensor network for hazard monitoring based on an array of eventtriggered single-channel micro-seismic sensors with advanced signal processing and characterization capabilities based on a novel co-detection technique. On the one hand we leverage an ultra-low power, threshold-triggering circuit paired with on-demand digital signal acquisition capable of extracting relevant information exactly and efficiently at times when it matters most and consequentially not wasting precious resources when nothing can be observed. On the other hand we utilize machine-learning-based classification implemented on low-power, off-the-shelf microcontrollers to avoid false positive warnings and to actively identify humans in hazard zones. The sensors' response time and memory requirement is substantially improved by quantizing and pipelining the inference of a convolutional neural network. In this way, convolutional neural networks that would not run unmodified on a memory constrained device can be executed in real-time and at scale on low-power embedded devices. A field study with our system is running on the rockfall scarp of the Matterhorn Hörnligrat at 3500 m a.s.l. since 08/2018.

show abstract

Section: Classification With Time Distributed Processingmentioning

confidence: 99%

Section: Implementation Challenges On Embedded Devicesmentioning

confidence: 99%

Event-triggered natural hazard monitoring with convolutional neural networks on the edge

Meyer

Farei-Campagna

Pasztor

et al. 2019

Proceedings of the 18th International Conference on Information Processing in Sensor Networks

View full text Add to dashboard Cite

show abstract

“…This is especially wasteful if the power cycle is short, because only a small part of the program is expected to execute. This problem can be partially alleviated by techniques such as: comparing the current state to the latest snapshot, and only saving the difference [51,52]; or IC-specific memory management [50]. Another drawback is that reactive IC does not generally handle high-level atomicity constraints (C3) as gracefully as task-based or static methods.…”

Section: (Iii) Reactive Icmentioning

confidence: 99%

Energy-driven computing

Sliper

Cetinkaya

Weddell

et al. 2019

Phil. Trans. R. Soc. A.

Self Cite

View full text Add to dashboard Cite

For decades, the design of untethered devices has been focused on delivering a fixed quality of service with minimum power consumption, to enable battery-powered devices with reasonably long deployment lifetime. However, to realize the promised tens of billions of connected devices in the Internet of Things, computers must operate autonomously and harvest ambient energy to avoid the cost and maintenance requirements imposed by mains- or battery-powered operation. But harvested power typically fluctuates, often unpredictably, and with large temporal and spatial variability. Energy-driven computers are designed to treat energy-availability as a first-class citizen, in order to gracefully adapt to the dynamics of energy harvesting. They may sleep through periods of no energy, endure periods of scarce energy, and capitalize on periods of ample energy. In this paper, we describe the promise and limitations of energy-driven computing, with an emphasis on intermittent operation. This article is part of the theme issue ‘Harmonizing energy-autonomous computing and intelligence’.

show abstract

“…Other selective policies for efficient state retention with transient systems which exploit the properties of different NVM memories were recently proposed [21]. Due to the modular structure of the presented software, these polices can be added and evaluated as separate modules.…”

Section: B Snapshot Strategiesmentioning

confidence: 99%

Energy Harvesting Meets IoT: Fuelling Adoption of Transient Computing in Embedded Systems

Balsamo

Magno

Kubara

et al. 2019

2019 IEEE 5th World Forum on Internet of Things (WF-IoT)

Self Cite

View full text Add to dashboard Cite

The emerging class of transient computing systems enables computation to be sustained despite power outages due to the variable nature of energy harvesting. However, existing approaches are largely designed for specific architectures, and hence are not broadly applicable across different IoT devices. Emerging platforms based on portable, hardware-independent software should rely on lightweight operating systems (OSs) designed specifically for embedded IoT applications, such as Arm mbed OS and Contiki OS. To enable the widespread use of transient computing, transient approaches need to be integrated into these operating systems. In this paper, we discuss the challenges of providing software primitives for transient computing to facilitate hardware-independent implementation using standard OS APIs, and present the integration of a state-ofart transient approach, Hibernus into mbed OS. This OS is chosen due to the large community of developers and the open-source IoT code availability. Transient computing is offered through a modular and layered structure that uses the available mbed OS APIs, including different strategies for retaining the system state designed for different types of flash memory. To illustrate the applicability of the proposed design, we implemented Hibernus on two mbed platforms with different flash memories, which respectively requires 4.7mF and 4.9mF of additional storage.

show abstract

Selective policies for efficient state retention in transiently-powered embedded systems: Exploiting properties of NVM technologies

Cited by 11 publications

References 10 publications

Event-triggered natural hazard monitoring with convolutional neural networks on the edge

Event-triggered natural hazard monitoring with convolutional neural networks on the edge

Energy-driven computing

Energy Harvesting Meets IoT: Fuelling Adoption of Transient Computing in Embedded Systems

Contact Info

Product

Resources

About