Green IoT energy-efficient design and mechanisms

As IoT is adopted to addresses problems in the various sectors of the society or economy, the energy demand of IoT is increasing rapidly in an almost following an exponential trend. As the number of IoT devices increases, the amount traffic created by IoT devices increases, increasing the energy demand of the core networks that are used to transport the IoT traffic and also increasing the energy demand of data centers that are used to analyse the massive amounts of data collected by the IoT devices. The large-scale adoption and deployment of IoT infrastructure and services in the various sectors of the economy will significantly increase the energy demand from the IoT cyberphysical infrastructure (sensor and actuator devices) through the transport network infrastructure the cloud computing data center infrastructure. Therefore, one of the design goals of green IoT is to develop effective strategies to reduce energy consumption. These strategies should be deploy across the IoT architecture stacks. That is, energy-saving strategies should be implemented across all the IoT layers including:

• The perception or “things” layer: Consists of IoT sensors that collects data and send to computing platforms for analysis and actuators that manipulate physical systems based on the feedback from data analytic platforms.
• The network or transport layer: Consist of the network network (access and Internet core network) infrastructure that is used to transport the data collected by the sensors to fog or cloud computing platforms and the feedback or commands from the fog or cloud computing platforms to manipulate actuation that control cyberphysical systems at the percetion or things layer.
• The Application layer: For processing (analysing) and storing the data collected by the IoT sensor devices and transported through the transport layer to the data centers. The results of the computations can be made available to users through applications or send back to the things layer to manipulate actuators.
• The energy and sustainability management layer: It is an abstract layer that span across all the above three layers, as energy-efficiency and sustainability management is implemented across all the above layers.

At each layer, various energy-efficient strategies are implemented to reduce energy consumption. A large proportion of energy is used for performing computation and for communication at the various layers. A significant amount of energy is saved by deploying energy-efficient computing mechanism (both hardware and software mechanisms), low-power communication and networking protocols, and energy-efficient architectures. Energy-efficiency should be one of the main design, manufacturing, deployment and standardisation goal for green IoT systems. The energy-saving mechanisms may vary from one layer to another but they can be classified into the following categories:

• Green hardware
• Green communication and networking
• Green architectures
• Green software
• Green security
• Green policies

A realistic approach to significantly reduce the energy consumption in IoT systems or infrastructures is to significantly improve the energy efficiency of hardware systems, because a large proportion of energy is used to power the electrical and electronic hardware such as computing nodes, networking nodes, cooling (and air conditioning) systems, and power electronics systems, security, and lighting systems. Recently, a lot of attention is being made to improve the energy efficiency of hardware systems in ICT infrastructures, especially in the IoT industry. The energy-saving mechanisms in IoT infrastructures include:

• Reducing the size of hardware device
• Using energy-efficient materials
• Energy-efficient hardware design
• Turning off idle devices
• Energy-efficient manufacturing

To achieve the green IoT vision, it is essential to deploy the energy-efficient hardware in the entire IoT infrastructures (from the perception layer to the cloud) throughout the IoT industry. Green IoT hardware is not limited to energy-efficient hardware design and hardware-based energy-saving mechanisms in the IoT infrastructure but also includes sustainable hardware approaches such as

• Using disposal and recyclable materials to manufacture IoT hardware
• Incorporating energy harvesting systems into IoT systems or infrastructure

Reducing the size of hardware device

There have been a significant reduction in the size of electronic hardware from the times of the vacuum tube to modern day semiconductor chips. In the early days of electronics, computers occupied entire floors of build, radio communication systems were large systems integrated into cabinets, and the smallest electronic device at the time was a two way radio system that was often carried on the back ^[1]. As the sizes of electronics devices decreased, their energy demand also dropped drastically.

Over the past few decades, the sizes of computing and communication devices have decreased significantly, reducing the power required to operate them. Despite the significant progress made by the semiconductor industry to decrease the size of semiconductor chips while improving their performance, there is still a persistent drive to keep decreasing the sizes of semiconductor chips to decrease their cost, reduce energy consumption, and conserve the resources required to manufacture these chips.

One of the Co-founders of Intel, Gordon Moore observed that “the number of transistors and resistors on a chip doubles every 24 months” and it was adopted by the computer industry as the well-known Moore's law and become a performance metrics in the semiconductor or computer chip industry. As more transistors were being packed into a single small-sized chip, the sizes of computing and networks equipment decreased significantly which also translated to a significant decrease in power consumption. Although advanced chip manufacturing have decreased the transistor gate length significantly,current leakage have also increased, resulting to an increase in the power consumption and heat dissipation of chips. Thus, doubling the number of transistor on the chip could double the amount of power consumed by the chip^[2].

In some energy-hungry IoT devices, batteries with higher energy capacity are required. The energy capacity of a battery is correlated with its size. That is batteries with higher energy capacities may be larger and heavier, placing a limit to the extend to which the size of the device can be decreased. The energy capacity of the battery may be relatively small but an energy harvesting module is attached to the battery to continuously recharge the battery with energy harvested from the environment. The addition an energy harvesting module may increase the size of the IoT device but it increase the operational life or the lifetime of the device. It should noted that the energy harvested by energy harvesting modules is very small and that the power electronics components also consume energy.

Another approach to keep decreasing the sizes of IoT device and possibly decrease the energy consumption is to integrate the entire electronics of an IoT device, computer or network node into a single Integrated Circuit (IC) called System on a Chip (SoC) ^[3]. The components is the device or node that are often integrated into an IC or SoC include a Central Processing Unit (CPU), input and output ports, memory, analog input and output module, and the power supply unit. The SoC can efficiently perform specific functions such as signal processing, wireless communication, executing security algorithms, image processing, and artificial intelligence. The primary reason for integrating the entire electronics of a system into a chip is to reduce energy consumption, size, and cost of the system as whole. That is, a system that was originally made of multiple chips is integrated into a single chip that is smaller in size, may be cheaper, and consume less energy. External devices such as the power sources (batteries or energy harvesting, antennas and other analogue electronics components) can be integrated into a SoC to reduce size, energy consumption, and cost.

Using energy-efficient materials -energy-efficient sensors

Energy-efficient hardware design

At the IoT perception layer, some of the energy-efficient mechanisms include:

1. Energy-efficient sensors (Green sensors): designing IoT sensors to consume as minimum amounts of energy as possible. When selecting the sensors to be used during the design of IoT devices, energy consumption and sustainability should be among the considered design criteria.
2. Energy-efficient radio modules (Green radio modules): Radio modules are the major consumers energy in IoT devices and designing them to consume minimal amount of energy significantly decreases the energy consumed by IoT devices. When choosing the IoT device to be used for an IoT application, the energy consumption of the radio modules should be taken into consideration.
3. Low-power microcontroller microprocessors (Green MCUs and ICs): the energy consumption of the microcontroller or microprocessor is very important as these devices are often powered by batteries with limited energy capacity. In selecting IoT devices to be used for an IoT application, the performance and energy consumption of the devices should be prioritised rather than sacrifising one for the other. Some of the design strategies that have been develop to improve the energy efficiency of the microcontroller or microprocessor of IoT devices is
   • Duty cycling: Switching off the microcontroller or microprocessor when the device is idle and then switching it on only when there is needed for processing.
   • Using low power microcontrollers or microprocessors: Choosing very low-power microcontrollers or microprocessors with very limited processing power but consumes relatively small amount of energy.
   • Using energy-efficient CMOS ICs to manufactures MCUs or CPUs: Manufacturing the components of IoT devices using energy-efficient CMOS ICs can significantly reduce the energy consumption of IoT devices.
   • Hardware acceleration and SoC design: Using application specific integrated circuits (ASICSs) to implement hardwired functionalities in an energy-efficient way (e.g., DSP systems, System-in-package(SiP), System-on-Chip (SoC)), resulting in highly compact designs (combining sensors, MCU, batteries, and energy harvesters into a single chip). As tens of billions to trillions of IoT devices are being deployed in in various sectors (e.g., intelligent transport systems, smart health care, smart manufacturing, smart homes, smart cities, smart agriculture, and smart energy) of the society or economy, the amount of traffic generated by IoT devices and transported through the local network and the Internet to fog or cloud computing platforms is also growing rapidly. The amount of computing or processing required to analyse the massive amounts of data generated has also increased significantly. The increase in the amounting of traffic and computing or processing requirement also increases the energy consumption of hardware deployed in the networking and data center infrastructures handling the IoT traffic and data. Some of the hardware-based energy-saving strategies that can be leveraged to reduce the energy consumption of networking and computing nodes in IoT based-infrastructure ( some of which were discussed) in ^[4] include:

1. Custom systems-on-chip: A design approach that integrates some or all system components into a single chip which reduces the size of the system compared to the approach of designing the various components of the system separately. Although the size, weight and the energy consumption of the SoC devices may be relatively lower compared to devices designed using separate chips, their performance may be lower. For example, a Raspberry Pi that contains a Broadcom SoC may consume less than 5 W, it processing power may be less than that of computer processors. SoC are used in mobile phones to ensure acceptable computing or processing and networking performance while minimising the energy consumption to extend the battery life. Thus, the SoC design approach will enable a significant reduction in the size of the device and energy consumption without necessarily sacrificing the performance of the devices.
2. Dynamic frequency scaling: The processor, microprocessor, or microcontroller can be forced into a low-power mode by reducing it's clock frequency or voltage. Also, the power consumption of the peripheral components of the device can be dynamically reduced by dynamically powering down some of the peripherals that are idle (not used at all). The power consumption of the peripherals can be controlled in such a way that they consume power only when necessary. Dynamic frequency or voltage scaling scaling can be be implemented in a software which is then used to monitor and adjust the power and clock frequency or voltage of the processor. Frequency and voltage scaling can be implemented on computing and networking nodes from the IoT perception layer, through the networking or transporting layer to fog/cloud computing layers. Frequency or voltage scaling is a feature that has been implemented in some Intel process in the form of P-states and C-states. The P-states provide a mechanism to the scale the frequency and voltage at which the processor runs to reduce it's power consumption and the C-states are the states of the CPU when it has reduced or turned off some of its selected functions ^[5].
3. Low-energy displays: For applications that require the display of information, increasing the energy efficient of the display could decrease the energy consumption of the device.
4. Hardware data processing (e.g., (AI hardware): Rather than using the CPU for all types of computing or processing tasks, hardware acceleration is employed to shift unique data operation or some specific computing tasks into dedicated hardware. Hardware acceleration refers to the process by which an application offloads some specific computing tasks onto some specialised hardware components (e.g., GPUs, DSP, ASICs etc) within a system to achieve greater efficiency than it is possible using a software that is running solely on a general purpose CPU. ^[6]. Tasks such as visualization, packet processing, AI processing, cryptography, error correction, and signal processing can be offloaded onto specialised hardware, freeing up the CPU to perform other tasks. Such specific hardware often offer high performance and low energy consumption when compared to CPUs. For example running AI-based tasks on GPUs is more efficient compared to running them on a CPU, which justify why GPUs are more preferable then CPUs. AI specific hardware have been introduced especially for neural networking tasks. Thus, IoT hardware designers should always examine carefully if there are tasks that could be offloaded to specialised hardware to free up the microcontroller or processors, significantly improving performance and energy efficiency.
5. Cloud computing (remote processing): Cloud computing is a cost-effective and scalable computing paradigm that enables on-demand remote access of computing resources such as software, infrastructure, and platform over the internet. By adopting cloud-based services (software-as-a-service, infrastructure-as-a-service, platform-as-a-service) companies or organisations do not need to invest in hardware infrastructure to host their service, significantly reducing the energy demand of IT services. An interesting strategy that has significantly increased the performance and energy efficiency of IT infrastructure and services is virtualisation. Virtualisation refers to the hardware or software methods that enable the partitioning of a physical machine into multiple instances that run concurrently and share the underlying physical resources, and devices. It involve the use of Virtual Machine Monitor (VMM), also called a hypervisor, to manage the Virtual Machine (VMs) and enable them to share the underlying physical resources (hardware). The sharing of hardware resources by VMs that are hosting multiple services (data analytics, high performance computing, security, etc.) significantly reduces the energy demand from data centers. Several energy-efficient strategies (e.g., switching-off idle servers, energy-efficient task scheduling, and other optimization methods) have been developed and implemented in data centers. The exponential increase in the number of deployed IoT devices and the generation of massive amounts of data they generate and send to fog computing nodes or cloud computing data centers will likely increase the energy consumption of data centers significantly, requiring green cloud computing strategies.
6. -Photonic computing: In an attempt to increase processing performance and significantly decrease energy consumption researchers and experts in the electronics and computer industries are seeking for ways to use optical devices for data processing, data storage, and data communication. Optical or photonic computing offer high speed, high bandwidth, and low energy consumption benefits that can be exploited to meet the need for high performance computing, high speed communication, and low energy consumption, an can be considered as a promising technology for high performance or high speed computing and communication technologies for computing and networking nodes in the IoT networking/transport and fog/cloud computing layers. The main components of a photonic or optical computing systems are optical processing units (for data processing), optical connectors (for optical data transfer), and optical storage units (for optical data storage). In optical or phototonic computing, light waves (photons) produced by lasers or incoherent sources are exploited as a primary means for carrying out numerical calculations, reasoning, artificial intelligence, data processing, data storage and data communications for computing unlike in traditional computers where these functions are performed using electrons ^[7]. A major challenge in optical or photonic computing systems is the inefficiencies or performance bottlenecks introduced when converting electrical signals to optical and optical signals to electrical as there is still a need to interface them with existing digital computing and communication systems.
7. Improving the energy efficiency of mobile radio networks: The adoption of Low-Power Wide Area (LPWA)cellular technologies (e.g., NB-IoT, LTE-M) have enabled the deployment of IoT networking services over existing mobile network ^[8]. More than 50% of the energy consumption of cellular base station is consumed by power amplifiers. Improving the efficiency of the power amplifier of wireless access network nodes (e.g., improving the efficiency of the power amplifier of 4G/5G/6G base stations). Another strategy to reduce the energy demand of cellular mobile base station is to centralise or shift some of the base band processing to the cloud or a pool of base band units, the so-called Cloud Radio Access Network (C-RAN).
8. Turning off idle networking or computing nodes: The most popular energy-efficient management strategy is to switch off idle devices or components. This approach can be applied from the IoT perception layer to the fog/cloud computing layer.

The increasing proliferation of IoT devices in almost every sector or industry developing and developed economies have resulting in the increase in the amount of data collected from the environment, increasing the demand for processing or computing. IoT devices and traditional devices require high performance, QoS, and longer battery life which can be achieved primarily by developing strategies that can improve both the computing performance and energy consumption. Green or sustainable computing is the practice of developing strategies to maximise energy efficiency (minimise the energy consumption) and to minimise the environmental impact from the design and use of computer chips, systems, and software, spanning across the supply chain from from the extraction of raw materials needed to make computers to how systems are recycled ^[9].

Green computing strategies can be implemented in software or hardware. Some of the hardware-based green computing strategies have been discussed above on the section on Green IoT hardware. The software strategies will be discussed on the section on Green IoT software below. A major green computing strategy that is improving both computing performance and energy efficiency is hardware acceleration. Hardware accelerators such as GPUs and Data Processing Units (DPUs) are major green computing drivers because they provide high performance and energy efficient computing for AI, networking, cybersecurity, gaming, and High Performance Computing (HPC) services or tasks. It is estimated that about 19 terawatt-hours of electricity a year of electricity could be saved if all AI, HPC and networking computing tasks could be offloaded to GPUs and DPU accelerators. With increasing use of sophisticated data analytics and AI tools to process the massive amounts of data generated by IoT devices, green computing strategies such as hardware acceleration will be very essential ^[10].

Green computing is not only about devising strategies to reduce energy consumption. It also include leveraging high performance computing resources to tackle climate related challenges. For example the use of GPUs and DPUs to run run climate models (e.g., prediction of climate and weather patterns) and to develop other green technologies (e.g., energy-efficient fertilizer productions, development of battery technologies etc.). A combination of IoT and green computing technologies is providing powerful tools to scientists, policymakers, and companies to tackle complex climate related problems.

The data gathered or generated by IoT devices is often sent to processing node ( edge nodes, fog computing nodes or cloud computing data centers) that are often located at some distance away from the devices. As the data generated by the IoT devices increases, the traffic to be transported across the network infrastructure increases, requiring upgrades on the infrastructure to handle the growing traffic, resulting in a corresponding increase in the energy demand. Apart of computing, communication is the largest energy consumer in IoT infrastructures. In an IoT device, must of the energy is consumed by the wireless communication module. Some green IoT communication and networking mechanisms include:

1. Low-power networking and communication technologies
2. Energy-efficient data transmission
3. Network level offloading of computation
4. Energy-efficient communication

The development of advanced design and manufacturing processes to produce energy-efficient chips is one of the strategies that is currently being used to reduce the energy consumption to achieve the green computing and communication goals. Given the rapid adoptio of smart phones and IoT systems, producing energy-efficient chips is very important. An example to illustrate how advanced manufacturing may significantly reduce the energy consumption in Computing and communication devices is the A-series chips used in Apple's iPhones. The power consumption of the 7-nm A12 chip is $50\%$ less than that of its 10-nm A11 predecessor. Also the 5-nm A14 chip is $30%$ more power efficient than the 7-nm A13 chip, and the 4-nm A16 is $20%$ more power-efficeint than the 5-nm A15. ^[11].

A similar trend has been can be observed in the PC industry although there is no guarantee that more advanced chip manufacturing processes with keep improving the performance and energy efficiency of chips.

(discuss chips in 4G/5G base stations)