ZERO: Towards Energy Autonomous Systems for IoT

ZERO summary

Today we witness an exponential growth of autonomously operating embedded devices such as Internet of Things (IoT) - incorporated into many physical systems, communicating wirelessly. IoT impact will be tremendous, with projected figures of 100 billion connected devices, and a global economic impact over $11 trillion by 2025. However, powering billions of these devices with standard solutions, like Li-ion batteries, causes huge maintenance and environmental problems; it severely limits their lifetime and functionality. Energy supply forms the Achilles heel of IoT.

ZERO makes IoT devices battery-oblivious, by creating energy-autonomous ultra-low power devices that scavenge their own energy. Hence, energy consumption must be drastically reduced, while energy scavenged should increase. This requires a holistic design approach, i.e., taking all design layers and system components into account.

ZERO provides necessary innovations in various areas, including energy scavenging, low-power circuit design, energy modulated computation, efficient energy conversion for storage in super capacitors, heterogeneous computing, low-power communication, and intelligent system-wide energy management.

ZERO solutions are widely applicable and lead to less waste, less maintenance, and better durability, but above all ZERO enables truly autonomous systems. Finally, ZERO is firmly embedded in the business needs of Dutch and international companies involved in this proposal.

ZERO structure

The program goals and focus lead to a number of scientific research challenges, grouped into six research lines that guarantee a strong coherence among the five projects:

R1. Energy harvesting.
R2. Ultra-low-power architectures and circuits.
R3. Energy-aware algorithms and self-learning applications.
R4. Efficient code generation for low-power and heterogeneous architectures.
R5. Ultra-low-power wireless communication.
R6. Multi-domain system-level modeling, energy management, and system integration
The applications within the projects are:

P1. Parking sensors, such that cars can find a free parking slot autonomously.
P2. Monitoring of traffic patterns using advanced audio beam processing.
P3. Autonomous roadside monitoring with video.
P4. Ultra-low power transponders for vulnerable traffic users.
P5. Dependable autonomous computing platforms supporting mobile traffic users.

The Research Lines and projects are structured according a matrix.

Research lines

R1: Energy Harvesting

There are a number of reasons for using energy harvesting within IoT devices, e.g., because it is difficult to reach devices for maintenance, wires are too costly because of the sheer number of devices, or because of environmental issues (Li-ion batteries are very toxic: some airlines refuse to transport IoT devices that contain Li-Ion batteries). Drawbacks of scavenging include: (i) each solution is dependent on the availability of harvestable energy sources, (ii) the upfront cost may be higher, and (iii) technology is less mature.

Energy can be harvested in many ways. The most optimal way depends on a number of factors, e.g., the environment an IoT device is located in, size and cost constraints, etc. Moreover, the amount of energy that can be harvested over time is usually not constant. For example, the kinetic energy from a moving train depends on the speed of the train, and is zero when the train has stopped. Therefore energy storage, e.g., with super capacitors, is needed to overcome periods of no or little harvesting activity. As can be observed from Figure 3, the amount of harvested power differs considerably between various sources. To substantially improve harvesting efficiency, R1 addresses the following scientific challenges:

  • Intelligent and more reliable energy harvesting from multiple sources in one device.
  • Harvest more energy than currently possible with new materials, with thermal harvesters as a promising solution.
  • Efficient DC/DC conversion suitable for non-resonant sources with maximum power tracking for adaptation to dynamic impedance characteristics of the harvester.

R2: Ultra-low-power architectures and circuits

Advanced processor platforms used for processing incoming signals, understanding them and taking appropriate actions, form the heart of autonomous devices. These processors, with the right software, make the sensor systems smart. State-of-the-art low-power signal processors typically employ an architecture that is either very specific or too generic; application specific architectures cannot be reused and generic architectures are not energy efficient (see right box). Therefore, autonomous devices need heterogeneous processing platforms, e.g., combining one or more general purpose CPUs, vector processing and more dedicated accelerators. These platforms are complex to design and very difficult to program efficiently (see R4). ZERO's research line R2 tackles several important low-power challenges:
  • Design of heterogeneous processing architectures, with adequate energy control options.
  • Research boundaries between architecture flexibility and application specificity.
  • Use approximate computing for ultra-low power.
  • Exploit new low power circuit techniques, like charge coupling.
  • Develop tighter integrated DC/DC conversion for improved efficiency and lower ripple, by feeding the converter with information about currents and voltage needed in the near future, such that sufficient flux/charge can be built-up in the converter.

R3: Energy-aware algorithms and self-learning applications

Energy should be a guiding principle for application coding as well as algorithm design of energy-autonomous systems. Algorithms should be sensitive to their environment, such that they can learn from it and react on changing circumstances [11]. Energy-aware hardware and system software can reduce the energy consumption of autonomous systems, however, energy-aware algorithmic engineering provides a complementary technique to reduce energy consumption beyond what hardware and system software can achieve. Challenges for R3 are:
  • Investigate new routes for designing efficient energy-aware algorithms. This includes understanding of the energy consumption of software/algorithms, establishing metrics for assessing the energy efficiency of algorithms/applications, and creating models for early estimating the energy consumption of software / algorithms.
  • Use of learning-based algorithms to make applications adaptive to changing environmental conditions and usage.
  • Learn and predict application usage scenarios.

R4: Efficient code generation for low-power and heterogeneous architectures

ZERO requires heterogeneous processing platforms. These platforms are extremely difficult to program efficiently. We have observed that the energy and performance efficiency of a straightforward mapping can be improved with at least a factor 10 (but often much more) by highly tuning the application and its mapping. However, this tuning makes the code incredibly complex, error prone, and difficult to maintain. In addition, every platform or application change requires redoing this process. Factors complicating efficient code generation include e.g.: different cache and scratchpad configurations, advanced DMA support, different instruction sets used, various types of parallelism to be exploited (SIMD, MIMD, threads, ILP), use of parametric accelerators, and Big-Little code migration. In addition, IoT devices often have real-time requirements, and, most importantly, they need special attention with respect to low power, exploiting all the energy knobs. Currently, most code generation methods ignore timing and power; these constraints are checked afterwards, leading to many long design cycles. It will be clear that R4 covers many challenges, among others:
  • Investigate code optimization opportunities for low power and measure their potential.
  • Implement advanced code transformations to perform above code optimizations automatically within an existing compiler flow. New is also that automatic optimization will be driven by power-estimation models.
  • Support automatic function recognition and offloading to more power-efficient function specific hardware accelerators.
  • Target various heterogeneous processing back-ends, including accelerators.
We will take LLVM tooling, with OpenMP4 (supporting accelerators), including state-of-the-art polyhedral optimizations (Polly), auto-vectorization, and existing optimizations for energy reduction like data locality and reuse improvement, as point of departure.

R5: Ultra-low-power wireless communication

To communicate with its surroundings, an energy autonomous node is facilitated with a wireless communications interface. Because of the limited amount of energy that is available, this interface has to operate extremely power efficient and, in case energy sources become exhausted, has to be able to cope with ultra-low voltages. Data-rates are expected to be (very) low for many applications ranging from a few bits per second to at most 10 kilobits per second (kbps). Current standards aiming at ultra-low power communications (such as IEEE 802.15.4 LR-WPAN, IEEE 802.15.6 WBAN, Bluetooth LE) provide data rates between 100-1000 kbps and the lowest published power consumption figures are around 500 µW. Besides this, interference rejection is poor and the reported circuits heavily rely on sophisticated analogue techniques where digital techniques have the advantage that they scale with technology. The scientific challenges within R5 are, therefore, to develop new digital solutions for transmitters and receivers:
  • Investigate new approaches to improve interference-robustness in low-power receivers.
  • Research new circuit techniques to cope with ultra-low voltages, even below 0.4V.
  • Define digital techniques for the mitigation of imperfections of ultra-low power analog RF circuits, that minimize the energy per transmitted and decoded bit.

R6: Multi-domain system-level modeling, energy management, and system integration

In order for systems to work energy-autonomously, it is important to have the right balance between potential energy use by the supported applications and the energy harvesting capabilities. On top of that, to overcome periods of heavy energy-use and/or low energy harvesting, locally stored energy will be needed. To be able to estimate the required energy storage capacity to assure continuous operation, a model-based approach will be taken to come up with good dimensioning rules. This is especially challenging as the required models will be so-called stochastic hybrid models that will have to encompass continuous state components (e.g., the amount of stored energy, or the harvesting rate given the environmental circumstances, the radio and the sensing and actuation devices), linear and non-linear environmental dynamics affecting both the continuous and the discrete state components, as well as stochastics, describing variations (disturbances) in the workload and harvesting potential. We build upon earlier work on the general theory of hybrid models; however, we have to extend it, and tailor it to be applicable in an energy-related context, as has been done in for wearable devices. R6 challenges are:
  • Development of model-driven, adaptive power management, using application scenario prediction, to guarantee continuous energy availability.
  • Develop stochastic models that can predict under what circumstances the node does not have enough energy to perform its tasks.


P1: APSN - Autonomous Parking Sensor Networks

Wirelesses embedded devices are the basic building block of the Internet of Things (IoT): systems that sense physical phenomena, process perceived information and actuate. And as the demand for mobility and sustainable operation of the IoT increases, powering IoT wireless sensors without batteries and cords becomes a necessity. The natural solution is to supply the energy to the IoT sensors through ambient energy harvesting. This enables the era of battery-less sensing and computation. However within the 3GPP-based cellular mobile ecosystem for the IoT, battery-less embedded systems are still unexplored. One of the reasons is that current energy provision through energy harvesting is not sufficient to enable communication using cellular standards for the IoT, such as NB-IoT or LTE-MTC, which are still not optimized for ultra-low-power (energy harvesting only) systems. For example, remote parking sensors (a classical IoT application) typically requires a bulky AA-battery packs: the size of the PCB itself. This brings three new avenues of research: (i) radically low-power communication methods (in addition to long-range existing IoT standards) such as ambient backscatter, (ii) improved energy harvesting techniques, and (iii) optimization of 3GPP IoT. Therefore, in the context of the whole ZERO program, the specific APSN project objective is to design an autonomous sensor network, anticipated to demonstrate the world’s-first completely battery-less parking system, that does not require any energy supply maintenance, and communicates with 3GPP base stations stationed kilometers away using harvested energy only.

P2: AAS - Autonomous Acoustic Systems

utonomous acoustic systems (AAS) can be found in various shapes and sizes, ranging from city wide acoustic monitoring systems to hearing aids worn by an individual. To deliver an improved quality and user experience, future generations of these systems should use adaptive signal processing algorithms, while staying within a stringent energy budget for autonomous operation. The AAS project uses two of these systems as driver cases to develop a novel programming paradigm and accompanying ultra-low power implementation platform for a wide range of autonomous acoustic systems.

P3: ARM - Autonomous Roadside Monitoring

oday we see an exponential growth in the use of monitoring systems. Many different sensing technologies are being used like road loops, passive infrared (PIR), laser, radar, and electro-optical (EO, vision) sensors. Especially, the laser, radar and EO sensors are becoming more-and-more intelligent; e.g. smart cameras with license plate recognition. Also information from multiple sensors need to be combined, e.g. speed measurement using radar combined with cameras. Typically they communicate wirelessly. We also observe that many monitoring systems have to be low cost and have no direct access to the mains power supply due to ownership and required permits and certification. Therefore they need to be rechargeable battery operated in combination with energy scavenging. The ARM project will develop the processing software and hardware for these systems, more precisely: The goal of the ARM project is the development of energy-autonomous programmable processing platforms for smart road-side monitoring systems that combine information obtained by cameras and radars. These systems will observe the road (see picture), interpret complex scenes, and communicate its findings with other road users and responsible authorities. For power-efficiency reasons we need to use heterogeneous computing hardware, including vector processing and accelerators, like embedded FPGAs and embedded GPUs. This heterogeneity can make programming of these platforms very time-consuming and error-prone because small changes in the software typically result in very significant changes in the real-time performance.

P4: ULPT - Ultra-Low Power Transponders for vulnerable road-side users

Energy autonomous communication devices that will not only enhance our comfort but will also improve our safety, will soon emerge on the market. Examples of such devices are autonomous WLANs for vehicular networks (IEEE 802.11p): transponders that can be mounted on bicycles to inform car-drivers about the presence of vulnerable road-users such as children going to school by bike. Advanced versions of these transponders should not only indicate presence but should be able to broadcast the trajectory of the bike. Given the available energy budgets and the energy inefficiency of currently available technology, initial versions of these transponders will not be equipped with a GPS and therefore will need to communicate using e.g. Bluetooth with the bike’s computer or the user’s smartphone to estimate speed and position in order to warn the drivers and prevent collisions.

These transponders will only become successful in the market if they are easy to install on every bike without wiring and are maintenance free. This implies that these transponder should be able to scavenge sufficient energy from the environment and store it in a supercapacitor. Initial estimates indicate that it is feasible to scavenge sufficient energy for a basic version of a transponder by making use of a small magnet mounted on a spoke like the ones used for wireless speed measurement devices. This basic version will only be able to transmit a very short data package with a fixed content over around 20 meters distance at a rate of a few Hz. Power management should enable that even if a biker stops, for example for a traffic light, there is still sufficient energy available in the supercapacitor to continue a few minutes at potentially a reduced transmission rate. The development of more advanced versions of the transponder that are capable of position estimation and trajectory transmission will require a significant improvement of energy scavenging capabilities and a reduction in energy usage of the transponder device. Similar technological improvements are needed to enable the development of energy autonomous transponders for pedestrians.

P5: DAMC - Dependable Autonomous Mobile Computing

Dependable Autonomous Mobile Computing equipment (DAMC) for mobile monitoring, sensor processing and communication requires maintenance free and very robust, compact systems. An illustrative example is a battalion of soldiers that are several days on a mission, without direct connection to a power source (mains). Such mobile equipment as illustrated on the right (helmet mounted camera and microphones, jacket with integrated equipment), with multiple embedded processors, has to run for a long period of time within a strict power budget. Currently, such equipment needs frequent maintenance, e.g., for replacement of batteries and replacement of vulnerable components, e.g., damaged by thermal stress. The main challenge for such dependable energy-efficient autonomous mobile systems is to achieve a maximum useful lifetime, while always having sufficient monitoring, communication and computation performance for critical tasks.