ECE 5/697LP: Design Principles for Low Power Embedded Systems

Introduction:

This course provides an overview of the core design principles used in the holistic design of low power embedded computing systems. Embedded systems may always benefit from lower power consumption because of the challenge of heat dissipation, energy is the ultimate bottleneck for resource constrained devices that rely on finite energy storage in the form of a chemical battery or voltage stored on a capacitor. Optimizing power consumption is fundamentally a circuit design problem, but optimizing energy consumption requires careful consideration of the end-to-end system that has many interactions that span the hardware / software boundary.

In this class, you will learn a host of state-of-the-art hardware and software design techniques used to optimize the overall power consumption / energy efficiency at the system level, which ultimately defines the battery life of an energy constrained system, or the impact of growing numbers of IoT devices on our power bill. Modern microcontrollers, radios, sensors, and power circuits have many dimensions of control and optimization and this course focuses on how we can dynamically tune these knobs for particular application requirements.

You will learn about these techniques in several ways:

• Research Literature Survey: You will get a chance to read and discuss recent research literature and see what directions modern embedded systems are going to operate more efficiently and help address global challenges including healthcare and climate change. Paper presentations will be student driven and are intended to be highly interactive. Students will be responsible for writing short reviews of each paper to be presented to encourage a lively discussion. What are the benefits and drawbacks of these new ideas?

• Introduction to new topics in embedded systems design: Part of the course will be dedicated to aspects of system design that you may not have had significant exposure to in other core classes. We will look at energy as the central optimization criteria for embedded systems and how this intersects with disciplines including network protocol design, signal processing, software scheduling, hardware design, algorithm design, and machine learning.

• Practical Tutorials: How do you know that your embedded system is low power? How can you simulate a system to ensure it meets performance objectives in the wild? We will have several tutorial assignments that will show you how to write code or optimize a hardware system for low power.

• Research Project: Each student (or group of students) will have the opportunity to design and evaluate an embedded system (or aspect of an embedded system) that can enable new applications or enhance existing applications because of the nature of its power efficiency. We will meet regularly during the semester and you will have a chance to present your system / findings at the end of the course.

Course Syllabus

A plan of topics to be covered in this course are listed below:

• Week 1: Applications

• Week 2: Low power hardware system design

• Week 3: Low Power Wireless Radios / Protocols

• Week 4: Backscatter Networking

• Week 5: Energy Harvesting Systems

• Week 6: Intermittent / Battery Free Computing

• Week 7: Energy Neutral System Design

• Week 8: Energy Aware Software and OS Design

• Week 9: Novel Sensing Systems

• Week 10: Device-Free Sensing

• Week 11: Edge Computing

• Week 12: Energy-Efficient AI

Grading

• Class Participation (10%)

• Small tutorial projects (20%)

• Reviews / Presentations (20%)

• Research Project (50%)

Reading List (Last Year's...to be updated)

Applications (01.30) :

• FarmBeats: An IoT Platform for Data-Driven Agriculture

• WISDOM: Watering Intelligently at Scale with Distributed Optimization and Modeling

• DualBlink: A Wearable Device to Continuously Detect, Track, and Actuate Blinking For Alleviating Dry Eyes and Computer Vision Syndrome

• EarBit: Using Wearable Sensors to Detect Eating Episodes in Unconstrained Environments

• Monoxalyze: Verifying Smoking Cessation with a Keychain-sized Carbon Monoxide Breathalyzer

Low Power Hardware Design (02.06)

• PowerBlade: A Low-Profile, True-Power, Plug-Through Energy Meter

• MBus: A Fully Synthesizable Low-power Portable Interconnect Bus for Millimeter-scale Sensor Systems

• System Architecture Directions for Post-SoC/32-bit Networked Sensors (*)

• Amulet: An Energy-Efficient, Multi-Application Wearable Platform

• FireFly: a cross-layer platform for real-time embedded wireless networks

Low Power Wireless Radios / Protocols (02.13)

• BLEach: Exploiting the Full Potential of IPv6 over BLE in Constrained Embedded IoT Devices

• Data Prediction + Synchronous Transmissions = Ultra-low Power Wireless Sensor Networks

• Understanding the Limits of LoRaWAN

• Challenge: Unlicensed LPWANs Are Not Yet the Path to Ubiquitous Connectivity (*)

• Orchestra: Robust Mesh Networks Through Autonomously Scheduled TSCH (*)

Backscatter / Battery Free Systems (02.27)

• Ambient Backscatter: Wireless Communication Out of Thin Air

• Self-Localizing Battery-Free Camera

• SkinnyPower: enabling batteryless wearable sensors via intra-body power transfer (*)

• PLoRa: a passive long-range data network from ambient LoRa transmissions

• Pushing the Range Limits of Commercial Passive RFIDs (*)

(* Denotes paper to read, review, and discuss)

Energy Harvesting Systems (03.05)

• Thermoelectric Energy Harvesting of Human Body Heat for Wearable Sensors (*)

• Capacity over capacitance for reliable energy harvesting sensors (*)

Intermittent Computing (03.10)

• An Energy-interference-free Hardware-Software Debugger for Intermittent Energy-harvesting Systems

• Timely Execution on Intermitently Powered Batteryless Sensors

• Intermittent Computation without Hardware Support or Programmer Intervention

• Energy-Harvesting Thermoelectric Sensing for Unobtrusive Water and Appliance Metering

Device Free Sensing (03.23)

• Synthetic Sensors: Towards General-Purpose Sensing

• Doorjamb: Unobtrusive Room-level Tracking of People in Homes using Doorway Sensors

Low Power Embedded Machine Learning (03.31)

• Wireless Computer Vision Using Commodity Radios

• A Low-Power Speech Recognizer and Voice Activity Detector Using Deep Neural Networks

Low Power Machine Learning and Edge Computing (04.07)

• MARVEL: Enabling Mobile Augmented Reality with Low Energy and Low Latency

• Neuro.ZERO: a zero-energy neural network accelerator for embedded sensing and inference systems

01.21 Introduction to Class Part I

01.23 Introduction to Class part II

01.28 How to Review / Read a Paper, Satisfying Application Requirements at Low Power

01.30 Paper Discussion: Applications - FarmBeats + DualBlink