As the power dissipated in passive components is negligible, the power consumption of an IoT essentially comes from its active components. The large majority of modern components relies on Complementary Metal-Oxyde-Silicium (CMOS) logic, which enables a lower power consumption. The power consumption of CMOS components is the sum of a static power consumption PS, a short-circuit power dissipation PSC [145]
and a dynamic power consumption Pdyn. The static power consumption is fixed and
independent of the component workload. This consumption is due to the fabrication process of the component, and notably the small leakages of each transistor which constitutes the IC. Thus, the reduction of PS is determined during the conception
of the component, when its architecture and fabrication process is determined. The short-circuit power dissipation PSC, illustrated in Fig. 3.1, is created during a gate
transition, when a direct path is created between the power supply and the ground by a pair or PMOS and NMOS transistor. This power dissipation can be reduced by ensuring that all switching signals have sharp rising and falling edges, reducing the duration of this short-circuit. The dynamic power consumption Pdyn varies with
the workload of the component. For each CMOS gate in the IC, the dynamic power consumption Pdyn can be calculated as (3.1), where CL is the capacitance of the gate
load, fν is the switching frequency of the gate and VG is the voltage of the gate. As
most gates are not always switching, the expression can be written as (3.2) where α is a coefficient between 0 and 1 which represents the duty cycle of the gate. In an
68 Reducing Power Consumption in IoT Nodes Vdd Vin I I Vin t t trise tfall
Figure 3.1: Short circuit power consumption in a CMOS inverter.
IC, CL is determined by the architecture and fabrication process of the component,
and thus cannot be reduced once the component is fabricated.
Pdyn = CL× fν× VG2 (3.1)
Pdyn = α × CL× fν × VG2 (3.2)
In usual server and workstation computing systems, the majority of the power consumption comes from the CPU. In IoT nodes, the microprocessor core is inte- grated in a micro-controller, which also integrates RAM and ROM memories and an array of analog and digital peripherals to interface the CPU with external compo- nents. The microprocessors used in WSN nodes are small scale device, which do not integrate a Memory Management Unit and can not run usual Operating Systems (OSs) such as Linux [146] or BSD derivatives. Instead, the memory is directly ac- cessed. Although the application can be run bare-metal, without any OS, a small OS such as Riot-OS [147] or Contiki [148], or a Real-Time Operating System (RTOS) such as FreeRTOS [149], is often used to ease the development process. As IoT use-cases become more complex, the use of more complex and featured OS such as NuttX [150] or uClinux [151] is also considered. A majority of micro-controllers embed a microprocessor from the ARM Cortex-M family, but other possibilities in- clude AVR cores, PIC cores, MIPS cores, the MSP430 CPU core or the older Intel 8051 processor. Micro-controllers implement several low power modes, which en- able the developer to disable different components of the micro-controller in order to lower the overall power consumption. Due to the variety of possible implemen- tations, the power consumption of micro-controllers can vary from several µW in low-power mode to a maximum or nearly 1 W. As the micro-controller does not run at full power all the time, its mean power consumption is often far lower than the maximum. In WSN nodes especially, the micro-controller mainly runs control tasks, while the heavier data processing calculations are usually done by a central remote server.
The wireless communications of WSN nodes require the use of a radio communi- cation IC, which integrates both transmission and reception chains. The reception chain integrates a low-noise amplifier to amplify the received signal and the trans- mission chain includes a power amplifier. Depending on the wireless communications protocol, the emission power can increase up to several hundreds mW. Some proto- cols also require the node to periodically emit control messages or to open reception windows to check for incoming messages. Due to this, the radio IC has to be period- ically turned on and consumes power, which raises the mean power consumption of the WSN node. Due to the relatively high power consumption of the amplifiers used in both reception and transmission chains, and the relatively low power consumed by micro-controllers, wireless communication accounts for a significant part of the total node power consumption. It can be noted that the use of wired communica- tions could significantly decrease the power consumption, but comes with significant constraints on the use-case of the node.
Due to the variety of use-cases for WSN, it is difficult to quantify a priori the power consumption of the sensing element. While some sensors, such as temperature or voltage sensors, are easy to integrate and consume little power, others, such as gas sensors, consume much more energy. Sensors usually integrate a transducer, which converts the sensed physical data to an electrical value (voltage or current), which goes through a signal conditioning circuit and is converted to a digital value with an ADC. Depending on the sensed data, the transducer can be a simple thermis- tor, or can be based on complex electro-chemical reactions. Moreover, the signal conditioning stage can be complex, including multiple filters and preamplification stages. Additionally, some sensitive sensors have to be calibrated before use, which consumes power. Although the impact of sensors can be minimal on the overall power consumption, it can also, in complex use-cases, be the most power consuming part of a sensor.
Finally, the quiescent power consumption of the power converters (SMPS and LDO) also has a significant impact on the system. This quiescent power consumption corresponds to the energy consumed by the active circuitry of the converter (feedback loop, comparators, voltage generator...) which is always powered regardless of the circuit activity. WSN nodes usually use small scale power converters, which have a low quiescent current consumption of several µA. Although a dozen µW power consumption is lower than the consumption of other active components by several orders of magnitude, this represents the majority of the power consumption of the device when all components are in a low power mode. Therefore, minimizing the quiescent current of the power converters reduces the mean power consumption of the node and has a significant impact on the power consumption of the platform.
Fig. 3.2, Fig. 3.3 and Fig. 3.4 show the result of a power consumption estimation obtained with Wisebatt [152] for a commercial WSN node, powered by a primary battery, over its whole lifetime. The node is based on Wi6labs LoRa platform, built around a STM32L0 micro-controller and a SX1272 RFIC. The device is used to periodically measure temperature and humidity values, and immediately transmit them over a LoRaWAN network. Since the device is in standby 99.99 % of the time, the power components account for a large majority of the spent energy, as shown in Fig. 3.2. If the SMPS converter power consumption is ignored (Fig. 3.3), the power consumption of communication and processing (micro-controller) tasks can
70 Reducing Power Consumption in IoT Nodes
���������������
���������������� ������������������� �����������������������
Figure 3.2: Energy consumption repartition per component type with main SMPS converter.
��������������� ����������������
��������������������
������������������������
Figure 3.3: Energy consumption repartition per component type without main SMPS converter.
be observed, alongside the consumption of other power components, such as load switches, which stay powered all the time. Although this can not be seen in this graph, the radio-communications IC has the highest instantaneous power consump- tion. Finally, Fig. 3.4 shows the power consumption repartition per task. Despite the very low power consumption of components in standby mode, the sleeping power consumption accounts for 44.5 % of the energy consumption. Radio-communications tasks amount to a total of 35.6 % of the consumed energy. In particular, using a more efficient MAC protocol could reduce the time spent listening to the radio channel, and could significantly reduce the energy consumed by this task.
������������� �������������������� ��������������������� ������������������ ��������������� ������������������������ ���������������
Figure 3.4: Power consumption repartition per task for a commercial WSN node.