Energy sources and power management in IoT sensors and edge devices
How will remotely located IoT devices be powered in the future? In this except from “Internet of Things for Architects”, Perry Lea takes a look at the significant problems IoT designers will face when resolving issues about energy sources and power management for IoT devices.
This article is an excerpt from the book, Internet of Things for Architects, written by Perry Lea and published by Packt Publishing. With this book, learn to design, implement and secure your IoT infrastructure.
Powering sensors and edge devices reveal a significant problem. When we consider that sensors and edge devices number in the billions, and the fact they will be used in very remote areas, power becomes a challenge. Furthermore, there will be IoT deployments where sensors will be buried undersea, or embedded into concrete infrastructures, further complicating power sources. In this post, we will explore the concepts of energy harvesting and power management. Both are very important concepts in the overall IoT story.
Managing power is a very broad topic, and spans software and hardware. It is important to understand the role of power management in a successful IoT deployment, and how to manage power efficiently for remote devices and long-lived devices. The architect must build a power budget for the edge device, which includes:
- Active sensor power
- Frequency of data collection
- Wireless radio communication strength and power
- Frequency of communication
- Microprocessor or microcontroller power as a function of core frequency
- Passive component power
- Energy loss from leakage or power supply inefficiency
- Power reserve for actuators and motors
The budget simply reflects the sum of these power contributors subtracted from the source of power (battery). Batteries also do not have a linear power behavior over time. As the battery loses energy capacity while discharging, the amount of voltage will drop curvilinearly. This poses problems for wireless communication systems. If the battery drops below a minimum voltage, a radio or microprocessor will not reach the threshold voltage and brown out.
For example, the TI SensorTag C2650 has the following power characteristics:
- Standby mode: 0.24 mA
- Running with all sensors disabled: 0.33 mA
- All sensors on at 100 ms/sample data rate and broadcasting BLE: 5.5 mA:
- Temperature sensor: 0.84 mA
- Light sensor: 0.56 mA
- Accelerometer and gyros: 4.68 mA
- Barometric sensor: 0.5 mA
The TI SensorTag uses a standard CR2032 coin cell battery rated at 240 mAh. Therefore, the maximum life is expected to be about 44 hours. However, the rate of decline changes, and is not linear for battery-based devices, as we will see when we cover Peukert’s capacity.
Many power management practices are employed, such as clock gating components not being used in silicon, reducing the clock rates of processors or microcontrollers, adjusting the sensing frequency and broadcast frequency, back-off strategies to reduce communication strength, and various levels of sleep modes. These techniques are widely used in the computing business as a general practice.
The techniques described here reflect reactionary power management techniques. They try to minimize energy usage based on dynamic voltage, frequency scaling, and other schemes.
New techniques to consider on the horizon include approximate computing and probabilistic design. Both of these schemes rely on the fact that absolute precision is not necessary at all times in a sensor environment running at the edge, especially in use cases involving signal processing and wireless communication.
Approximate computing can be done in hardware or software, and simply reduces the level of precision in integers when used with functional units such as addresses and multipliers (for example, the value 17,962 is fairly close to 17,970). Probabilistic design realizes many IoT deployments can tolerate certain degrees of faultiness to relax design constraints. Both techniques can reduce the number of gates and power to an almost exponential drop over regular hardware designs.
Energy harvesting is not a new concept, but is an important concept for IoT. Essentially, any system that represents a change in state (for example, hot to cold, radio signals, light) can convert its form of energy into electrical energy. Some devices use this as their sole form of energy, while others are hybrid systems that use harvesting to augment or extend the life of a battery. In turn, energy harvested can be stored and used (sparingly) to power low-energy devices such as sensors in IoT. Systems must be efficient in capturing energy and storing power. Therefore, advanced power management is needed.
For example, if an energy harvesting system uses a piezoelectric mechanical harvesting technique embedded in a sidewalk, it will need to compensate when there is not enough foot traffic to keep the unit charged. Constant communication with energy harvesting systems can further drain the power.
Typically, these IoT deployments will use advanced power management techniques to prevent complete loss of functionality. Techniques such as low standby currents, low-leakage circuits, and clock throttling are frequently used. The following figure illustrates the area where energy harvesting is ideal, and the technology that it can power. Care must be taken by the architect to ensure the system is not underpowered, nor overpowered.
Harvesting systems, in general, have a low energy potential and low conversion efficiency. An architect should consider energy harvesting in situations where there is a large supply of untapped waste energy, such as in industrial settings:
Energy from light, whether it is natural or artificial, can be captured and used as an energy source. The same A photodiode can be used in greater quantities to build a traditional solar array. The capacity of energy generation is a function of the area of the solar array. In practice, indoor solar generation is not as efficient as direct sunlight. Panels are rated by their maximum power output in the form of watts.
Solar harvesting is only as effective as how much the sun shines, which varies seasonally and geographically. A region such as the Southwestern US can reclaim considerable energy from direct photovoltaic sources. The Photovoltaic Solar Resource of the United States map was created by the National Renewable Energy Laboratory for the U.S. Department of Energy shown as follows:
In the United States, Southwest regions fare particularly well with sun intensity, generally lack cloud light barriers, and have good atmospheric conditions. Whereas, Alaska has the weakest energy density. Solar photovoltaics are not typically efficient. One can expect an 8% to 20% efficiency, with 12% being typical. Regardless, a 25 cm2 solar array could produce 300 mW at peak power.
Another factor is the incidence of light. For a solar collector to achieve such efficiency, the light source must be perpendicular to the array. If the angle of incidence changes as the sun moves, the efficiency drops further. A 12% efficiency collector when the sun is perpendicular will be roughly 9.6% efficient when the sun is 30 degrees from being perpendicular.
The most basic solar collector is the solar cell that is a simple p-n semiconductor, and similar to the photoelectric sensors discussed earlier. As explained earlier, an electric potential is generated between the p and n material when a photon is captured.
Piezoelectric effects can not only be used as sensors, but can also be used to generate power. Mechanical strains can be converted to energy through motion, vibration, and even sound. These harvesters could be used in smart roadways and infrastructures to harvest and change systems based on traffic movement, even when embedded in concrete. These devices produce currents on the order of milliwatts, and thus are suitable for very small systems with some form of energy collection and storage. This process can be performed using MEMS piezo-mechanical devices, electrostatic, and electromagnetic systems.
Electrostatic harvesting incorporates Faraday’s law, which basically states that one can induce an electric current by changing the magnetic flux across a coil of wire. Here, the vibration is coupled either to the coil or a magnet. Unfortunately, this scheme in the IoT sensor area provides too little voltage for rectification.
Electrostatic systems use the change in distance between two capacitive plates held at a constant voltage or charge. As the vibration causes the distance to change between the plates, energy (E) can be harvested based on the following model:
Here, Q is the constant charge on the plates, V is the constant voltage, and C represents capacitance in the preceding equation. Capacitance can also be represented by the length of the plate Lw, the relative static permittivity as ε0, and the distance between plates d as shown:
An electrostatic conversion has the advantage of being scalable and cost-efficient to produce through micromachining and semiconductor fabrication.
The last method for mechanical-to-electrical conversion is piezo-mechanical. The same basic concept applies to energy generation. As the piezo-mechanical MEMS device attempts to dampen the mass attached to it, the oscillations will be converted into an electrical current.
Another consideration for the capture and conversion of vibrational or mechanical energy is the need for conditioning before the energy is used or stored. Normally, a passive rectifier is used for conditioning by incorporating a large filtering capacitor. Other forms of energy harvesting do not need such conditioning.
RF energy harvesting
Radiofrequency (RF) energy harvesting has been in production for years in the form of RFID tags. RFID enjoys the benefit of being a near-field communication that uses a transceiver that essentially powers the RFID tag due to its close proximity.
For far-field applications, we need to harvest energy from broadcast transmissions. Broadcast transmissions are nearly everywhere, with service from televisions, cell signals, and radio. Capturing energy from radio frequencies versus other forms is particularly difficult, as RF signals have the smallest energy density of all harvesting techniques. The capturing of RF signals is based on having the appropriate antenna to capture a frequency band. Typical frequency bands used are in the 531 to 1611 kHz range (all in the AM radio range).
Thermal energy can be converted into an electric current for any device exhibiting heat flow. Thermal energy can be converted into electrical energy by two basic processes:
- Thermoelectric: Direct conversion of thermal energy into electrical energy through the Seebeck effect.
- Thermionic: Also known as thermotunneling. Electrons are ejected from an electrode that is heated, and into an electrode that is cool.
The thermoelectric effect (Seebeck effect) is produced when a gradient of temperature exists in conducting materials. The flow of carriers from a hot to a cold region between two dissimilar electrical conductors creates a voltage differential.
A thermocouple, or thermoelectric generator (TEG), could effectively produce voltage simply based on the temperature difference of a human, based on their core body temperature and outside temperature. A temperature difference of 5 degrees Celsius could generate 40 uW at 3V. As heat flows through the conduction material, a hot-side electrode induces electron flow to a cold-side electrode-producing current. Modern thermoelectric devices use n or p-type bismuth telluride in series. One side is exposed to the source of heat (called the thermocouple), and the other is isolated. The energy harvested by the thermopile is proportional to the square of the voltage and equivalent to the temperature difference between the electrodes. One can model the energy harvested by a thermocouple by the following equation:
Here S1 and S2 represent the different Seebeck coefficients for each of the two materials (n and p-type) in the thermopile when there is a temperature differential, TH -TL. Since the Seebeck coefficients are functions of temperature and there exists a temperature difference, the result is a voltage difference. This voltage is generally very small, so many thermocouples are used in series to form a thermopile.
One substantial problem with current thermocouples is the poor efficiency of the energy conversion (less than 10%); however, their advantages are notable, including their small size and ease of manufacturing, resulting in fairly low costs. They also have a very long lifetime of over 100,000 hours. The main problem, of course, is finding a relatively constant source of thermal variance. Using such a device in an environment throughout multiple seasons and temperatures is challenging. For IoT devices, thermoelectric generation typically resides in the 50 mW range.
Thermionic generation is based on the ejection of electrons from a hot electrode to a cold electrode over a potential barrier. The barrier is the work function of the material, and is best used when there is a significant source of heat energy. While its efficiencies are better than thermoelectric systems, the energy required to jump the potential barrier makes it generally unsuitable for IoT sensor devices. Alternative schemes such as quantum tunneling could be considered, but it currently remains a research activity.
Typical storage for an IoT sensor will be a battery or supercapacitor. When considering the architecture for sensor power, one must consider several aspects:
- Volume allowance for a power subsystem. Will a battery even fit?
- The battery energy capacity.
- Accessibility. If the unit is embedded in concrete, limited forms of energy regeneration can be used, and the difficulty of replacing batteries can be costly.
- Weight. Is the unit intended to fly as a drone, or float on water?
- How often will the battery be recharged?
- Is the renewable form of energy constantly available, or intermittent, as in solar?
- The battery power characteristics. How the battery’s energy will vary over time as it’s discharged.
- Is the sensor in a thermally constrained environment that can affect battery life and reliability?
- Does the battery have a profile that guarantees minimum current availability?
Energy and power models
Battery capacity is measured in amp-hours. A simplified equation to estimate the life of a battery power source is given as:
In the equation, Cp is Peukert’s capacity, I represents the discharge current, and n is Peukert’s exponent. The Peukert effect, as it’s known, helps predict the life of a battery where the capacity of a battery decreases at a different rate as discharge increases. The equation shows how discharging at higher rates removes more power from the battery. Alternatively, discharging at lower rates will increase the effective runtime of the battery.
One way of thinking of this phenomenon is a battery manufacturer rated at 100 Ah, and discharging it fully in 20 hours (5A, for this example). If one were to discharge it faster (say, in 10 hours), the capacity would be lower. If one were to discharge it more slowly (say, over 40 hours), it would be greater. However, when represented on a graph, the relationship is nonlinear. Peukert’s exponent usually lies between 1.1 and 1.3. As n increases, we move further from a perfect battery to one that discharges faster as the current increases. Peukert’s curve applies to lead-acid battery performance, and an example is shown in this next graph:
One can see the difference in discharge rates for various types of batteries. One advantage of alkaline batteries is that the discharge rate is almost linear in a large portion of the graph. Lithium-ion has a stair-step function in performance, thus making battery charge predictions more difficult. That said, Li-ion provides a near steady and continuous voltage level over the life of the charge, and consistently powers electronics over the span of its charge:
The graph also illustrates that lead acid and Ni-Cd have less voltage potential and a curvilinear degradation in power that can be computed more reliably. The trailing slopes are also indicative of Peukert’s capacity.
Temperature greatly affects battery life, and specifically the electro-active carriers in a cell. As temperature increases, the internal resistance of a battery decreases when being discharged. Even when batteries are stored, they can self-discharge, which impacts the total lifetime of a battery.
A Ragone plot is a useful way to show the relationship between energy storage systems when trading off energy capacity and power handling. It is a log-based scale where the energy density (Wh/kg) of a power source is plotted against the power density (W/kg). This relationship shows devices that tend to have a greater lifetime (batteries) versus devices that tend to store more energy (supercapacitors):
Batteries like Li-ion have higher energy density and discharge rates than nickel-cadmium and nickel-hydride batteries. Capacitors produce very high-power densities, but are relatively weak energy density. Note the plot is log-based, and also shows the discharge time for various storage systems. Image courtesy of C Knight, J. Davidson, S. Behrens “Energy Options for Wireless Sensor Nodes”, Sensors, 2008, 8(12), 8037-8066.
Typically, a lithium-ion (Li-ion) battery is the standard form of power in mobile devices due to its energy density. In such a battery, lithium-ions physically move from the negative electrode to the positive. During recharge, the ions move back to the negative region. This is known as an ionic movement.
Batteries also develop memories with many charge-discharge cycles. This capacity loss is expressed as a measure of the initial capacity (for example, 30% loss after 1,000 cycles). This degradation is almost directly correlated to environmental temperature, and loss will increase in a high-temperature environment. Therefore, it is imperative that the architects manage thermals in a constrained environment if lithium-ion is to be used.
Another factor in battery life is self-discharge. When unwanted chemical reactions occur in a battery cell, energy will be lost. The rate of loss depends on chemistry and temperature. Typically, a Li-ion can last for 10 years (at ~2% loss per month), while an alkaline battery will only last 5 years (15% to 20% loss per month).
Supercapacitors (or supercaps) store energy at significantly higher volumes than typical capacitors. A typical capacitor will have an energy density of somewhere between 0.01 Watt-hours/kg. A supercap has an energy density of 1 to 10 Watt-hours/kg, thus placing them closer to the energy density of a battery that can be on the order of 200 Watt-hours/kg. Like a capacitor, energy is stored electrostatically on a plate, and doesn’t involve the chemical transfer of energy like a battery. Usually, supercaps are made of fairly exotic materials like graphene, which can impact on overall cost. Supercaps also have the advantage of charging to their full potential in seconds, whereas Li-ion batteries will charge within minutes to approximately 80%, but require a trickle current to safely get higher. Additionally, supercaps can’t be overcharged, whereas Li-ion can be overcharged and can result in serious safety concerns. Supercaps come in two forms:
- Electric double-layer capacitors (EDLC): Use an activated carbon electrode and store energy electrostatically
- Psuedocapacitors: Use a transition metal oxide and electrochemical charge transfer
Supercaps have an advantage over batteries in predicting the remaining time power will be available. The remaining energy can be predicted from the terminal voltage, which changes over time. Lithium-ion batteries have a flat energy profile from fully charged to discharged, thus rendering time estimation difficult. Since the voltage profile of a supercap changes over time, a DC-DC converter is needed to compensate for wide variations in voltage.
In general, the main issues with supercaps or capacitors are the leakage current and the cost. As can be seen in the following table, supercaps have their place. One will often see them in a hybrid solution with regular batteries to provide instantaneous power (for example, electric vehicle acceleration), while the battery supply sustains running power.
Radioactive power sources
A radioactive source with a high energy density (105kJ/cm3) can generate thermal energy due to the kinetic energy of emitted particles. Sources such as cesium-137 have a half-life of 30 years and a power capacity of 0.015 W/gm. This method can generate power in the Watt-to-kW range, but isn’t practical in low-power sensor levels for IoT deployments. Space vehicles have used this technology for decades. Promising developments using MEMS piezoelectronics that capture electrons and force a micro-armature to move can create mechanical energy that may be harvested.
A secondary effect of radioactive decay is the relatively weak power density profile. A radiation source with a long half-life will have weaker power density. Thus, they are suitable for bulk-charging supercaps to provide momentary energy when needed. The final issue with radioactive sources is the significant weight of lead shielding required. Cesium-137 requires 80 mm/W of shielding, which can add significant cost and weight to an IoT sensor.
Energy storage summary and other forms of power
Choosing the correct power source is critical, as mentioned earlier. The following table provides a summary comparison of different components in a system to be considered when picking the correct power source:
|Energy density||200 Wh/kg||8-10 Wh/kg|
|Charge-discharge cycles||Capacity drops after 100 to 1,000 cycles||Nearly infinite|
|Charge-discharge time||1 to 10 hours||Milliseconds to seconds|
|Operational temperature||-20oC to +65oC||-40 degrees Celcius to +85 degrees Celsius|
|Operational voltage||1.2 V to 4.2 V||1 V to 3 V|
|Power delivery||Constant voltage over time||Linear or exponential decay|
|Charge rate||(Very slow) 40 C/x||(Very fast) 1,500 C/x|
|Operational life||0.5 to 5 years||5 to 20 years|
|Form factor||Very small||Large|
|Cost ($/kWh)||Low ($250 to $1000)||High ($10,000)|
You just enjoyed an excerpt from the book, Internet of Things for Architects, written by Perry Lea and published by Packt Publishing. Use the code ORJEA09 at checkout to get recommended eBook retail price for $9 only until June 15, 2018.