Title: From Transient Computing to Transient Systems: Overcoming Challenges to Enable Real Applications
Abstract:
Sensor systems powered by energy harvesting usually include batteries or supercapacitors which impact the system cost and size, need time to be charged and are not environmentally friendly. In recent years, designers have proposed a new concept called transient computing that aims to remove these energy storage units and retain the system’s state between power outages, in order to cope with an unreliable energy source. However, retaining the system state is not the only problem a transient wearable application could have. In my research work, I detailed the different challenges that need to be addressed in order to make a wearable device transient, as well as the contributions and results obtained to overcome them.
Title: Autonomous Wearable Computing using Ambient RF Energy Harvesting
Abstract:
With the Internet of Things market exponential growth, great interest has arisen in power-autonomous computing at the network edge. Wearable electronics, a key emerging sector of the IoT market, impose additional design constraints such as ease of integration in wearable materials, and virtually infinite lifetime. Ambient Radio-Frequency power represents a reliable source for energy harvesting, utilising existing communication infrastructure. However, factors such as path losses, RF to DC conversion inefficiency and commercial Power Management Integrated Circuits (PMIC) imperfections have hindered the materialisation of integrated RF powered nodes. A system-oriented design approach, starting with textile antenna designs for RFEH and reconfigurable high efficiency rectifier is proposed towards realising a wirelessly-powered edge-computing system, integrated on-textile, for next generation pervasive wearable computing.
Title: Are You Sitting Down? Sensing postural changes with e-textiles
Abstract: Electronic textiles (e-textiles) are conductive fabrics and threads that can be used to form circuitry which can be directly integrated into wearable garments or other soft furnishings like seat covers. This talk will present recent work showing how e-textile pressure sensors can discriminate between social activities such as speaking and listening, but will also review the challenges in prototyping with this technology. In particular, highlighting the interdisciplinary expertise required from a broad range of disciplines — from signal processing to pattern cutting — that are needed in order to generate robust and reliable sensing systems.
Network-on-Chip (NoC) architectures emerged as a viable solution for the design of manycore embedded systems of the next generation. While bringing new opportunities and effective energy/performance tradeoffs, they also introduce new challenges: the design of NoC based systems involves several aspects, such as the partitioning and mapping of the application to the cores, the selection of an appropriate interconnection topology, together with an appropriate routing scheme for dispatching the packets among the nodes.
The assessment of NoC based systems by performing a low-level (e.g., RTL) simulation evaluation and/or a full system simulation of the whole NoC architecture, is an extremely time-consuming approach that makes unfeasible an exhaustive exploration of all the design alternatives. High level cycle-accurate NoC simulators are widely used to quickly get an estimation of the target requirements/objectives. However, they rely on the use of synthetic traffic patterns, characterized by specific statistical properties only (e.g., packet injection rate) and do not accurately model other important aspects of real traffic scenarios.
To overcome such limitations, different benchmark suites (e.g., PARSEC, SPLASH) were proposed with the aim of including a set of applications representative of new emerging workloads for massively parallel architectures. Nevertheless, they still assume a traditional shared memory mechanism, while a message passing mechanism based on the direct exchange of data packets between nodes would probably be a more appropriate and scalable choice for next generation NoCs.
This talk describes the experience and the challenges of developing an entire design flow that, starting from a single and slow traditional shared-memory full system simulation, allows a fast and multi-objective evaluation of real applications on several different NoCs using message passing.
Tailoring machine learning to textile embedded robotics
By their nature (i) soft robotics and (ii) wearable technology have many things in common. Both offer the opportunity for close contact with humans, in a naturalistic way -- wearable technology based on fabrics is closer to ordinary clothing, while soft robotics promise safety for ever closer human-robot interaction. However, from a robotic engineering standpoint, both also suffer from challenges in design, modelling and control. The materials and methods used in fabrication do not lend themselves well to traditional engineering approaches of signal processing or system identification due to high noise levels, unpredictable deformations and imprecise knowledge of the physical parameters. To start to address these challenges, the Robot Learning Lab at King's has been looking into the how to measure and understand human behaviour through the medium of textile-based sensors, in combination with statistical learning approaches. I will give an overview of our activities in this area, from our initial experiences working with textile craftspeople, to our latest activities bringing low-cost prosthetics to the developing world.
This talk, entitled "Developing flexible, washable and robust circuit demonstrators for yarn encapsulation and e-textile industrial applications", presents about the challenges of encapsulating flexible copper electrodes with PDMS material.