Unleashing the full potential of IoE with Edge Computing Technology

The IoT has made its way from a fancy, at times highly indistinct, concept to actual reality. Today, it is already empowering a real-world intelligent infrastructure, in which uniquely identifiable physical objects are capable of wirelessly communicating with each other to find the state of interests and take proper actions. The number of connected devices was estimated at 30 billion in 2020 growing to over 75 Bn by 2025, and there have been prognostics of one trillion IoT devices produced until 2035. Still, one can point to a couple of bottlenecks limiting the scalability of present and future IoT - clearly roadblocks for the sought-after IoE realm! The first is power, and particularly reliance on pervasively spread battery-powered devices. The second can be defined as data bandwidth, i.e. the massive amounts of data that need to be transmitted, stored, and processed at centralized computing facilities, so-called data centres enabling all kinds of “cloud” services. Let us dig a little deeper into each of these highly interrelated challenges.

Powering the IoE? Not with batteries!

A growing number of IoT-connected devices, including the ones with low-power consumption (1 microW to 1milliW), continue to rely on disposable or rechargeable batteries to satisfy their autonomy and data transmission/storage capacities. But, as we (and many others!) keep insisting, batteries and IoT are not a match! Small devices reliant on batteries are trouble - mostly because of the limited lifespan of batteries, the high costs, and burden associated with battery replacement and device maintenance, which can become a logistic nightmare in very large distributed sensor systems, and the social and environmental problems associated with the lifecycle of batteries and respective raw materials. In short, batteries are bound to represent a clear bottleneck in the massive deployment of IoT devices and a huge environmental problem.

Luckily, ONiO has cracked the power problem: ONiO.zero offers a core solution for powering all types of small wireless IoT sensors and devices using ambient energy - completely eliminating the need for batteries.

Challenges with big data and bandwidth

IoT devices can produce massive amounts of data that frequently need to be transmitted, stored, and processed. The most common approach is to transmit large data sets to backend infrastructure and process such data in the cloud/data centres. As the IoT scales, the sustainability of this centralized computing approach is unequivocally challenged. It has been estimated that the total data volume of connected IoT devices worldwide will reach 79.4 zettabytes by 2025. At this rate, with one trillion devices by 2035, the data volume would reach >1000 zettabytes. To put this into perspective, the current capacity of data centres worldwide is about 2.5 zettabytes. Hence, with centralized computing approaches, there is simply not enough data storage and processing capacity to keep up with the growth of IoT. A good way to tackle this problem is through edge computing.

Edge computing

As explained in our recent article dedicated to the topic, edge computing refers to computing processes that take place closer to the edge of a network - i.e information processing that is more distributed and takes place at multiple locations within the network, without having to go through the central data processing centre or cloud. For IoT networks, this means that IoT devices/nodes are endowed with capabilities to locally run computing operations, namely data processing, filtering-off the volume of data that is actually sent through the cloud to data centres and offloading processing from centralized computing resources. This has remarkable implications on network loads, both in terms of data volumes and communication bandwidths.

So, reaching this point it seems that all is good: ONiO.zero provides power from ambient energy, edge computing technology deals with data bandwidth - problem solved! Well, not quite there yet… there are a couple more chapters in this story!

Combining ultra-low-power and edge computing abilities

Computing processes take power. Hence, processing data and information at edge devices require substantial amounts of power. This is not a major problem for network nodes that are wired: power efficiency is certainly a consideration in any device, but there’s plenty of power available, and hopefully electricity grids will run mostly on renewable energy sources in the not-so-distant future. But if we talk about wireless nodes, then the game is completely different! Batteries aside, running smart wireless devices solely on ambient energy requires ultra-low-power approaches. And we mean real ultra-low-power, i.e. in the range of nano to micro Watts! This is why power management is a keystone for ONiO - with ONiO.zero shining as the world’s lowest power microcontroller. Yet, we want more (or actually less in this case!) - so we’re constantly looking for ways to push the envelope with respect to power economics and efficiency. So, what we will try to pull off is to do edge computing at ultra-low-power!

Let us first look into how far others have gone in addressing such a challenge. There have been numerous R&D initiatives focused on ultra-low-power edge computing devices - as reviewed e.g. by Brooks and Sartori (2017), Capra et al. (2019) and Gamatie et al. (2019). But most of these haven’t reached successful real-world implementations as of now. So, here we will focus on actual implementations in devices that exist in the market. Several major players in the semiconductor industry have developed and implemented different kinds of hardware accelerators, in particular autonomous peripherals, into their low-power microcontroller designs. Examples include the Event System (EVSYS) from Microchip Technology, the Peripheral Reflex System (PRS) from Silicon Labs, the Peripheral Interconnect Matrix from STMicroelectronics, the Programmable Peripheral Interconnect (PPI) from Nordic Semiconductor, and the Event Link Controller (EVC) from Renesas. Common to all these is the purpose of offloading tasks to the autonomous peripherals to lower CPU load and save on power consumption, with the potential to also minimize latencies and improve throughput. Yet, all these solutions are very primitive, basing their operations on routing signals/wires from one peripheral to another without any logic in the signal path. They should be considered a bus matrix where single bits of information are distributed - not doing any logic/decision functions. 

New approaches

ONiO is playing with new approaches integrating edge computing and neuromorphic concepts - namely to implement neural networks in ultra-low-power integrated circuits. The end goal is to implement extremely power-efficient hardware accelerators which can perform algorithmic operations based on peripheral/sensor data without the intervention of the CPU. This will allow offloading heavy software tasks on hardware, thus saving power by waking up the CPU only when needed. 

But what does this means exactly? In practical terms, this means that we can create wireless, batteryless smart edge nodes/devices capable of processing data and making simple decisions locally - without having to send data back and forth to the cloud, thus saving on data transmission (bandwidth) and computing loads in data centres. Obviously, we are bound by the constraints of ultra-low-power systems - so not really running full chess games there! So, what is it good for?

Applications

Typical applications of our new power-aware neural network solutions will include threshold-based decision making, keyword recognition in speech, filtering of radio packages, and metering applications.

Let’s take a typical case of “thresholding”: monitoring oil level on an engine. Instead of continuously sending level sensing data to the cloud/backend, the system can be tuned to report only “low” levels - i.e. when the level drops below a certain threshold, the system communicates to the cloud/backend and the user gets a warning sign. The system can be taught to take smarter decisions - e.g. report only when three consecutive measurements indicate low level, to avoid “false positives” (e.g. “low level” point measurements due to transitory engine position/inclination or the like). 

Another example is smart metering. Electricity and water meters are today largely relying on being always connected and heavily reporting to a back office system that implements the smartness in the system. This results in a lot of data traffic and long response times. Meters can be trained to only alert when certain events happen - such as a water leak, excessive power usage, etc. Basically avoiding that the heavy lifting always happens in the back office.  This is particularly important in occurrences where the network infrastructure is not available - where, in a traditional system, the smartness is then lost.

In a nutshell, power aware neural network solutions shall be capable of mixing inputs from several sources and detecting relevant events (when something happens) without CPU intervention, filtering-off non-relevant data points and report/transmit relevant data/events only, as opposed to continuously sending data to the backend/cloud.

Leading the way

ONiO has initiated an R&D project focused on new power-aware neural network solutions. The project received support from the Regional Funds in Oslo (RFF Oslo) - which allows us to take a first stab into developing and testing prototype implementations. The project ties nicely with the ambitions of Oslo as a leading “smart, green, and safe” city. We expect that the future solutions can be capable of impacting on areas such as smart buildings and smart cities - raising the bar on the ambition and reach of such initiatives. For instance, enabling infrastructure-embedded sensors - such as sensors embedded in concrete which measure moisture in walls or vibration in structures for condition monitoring and preventive maintenance. These are typical “deploy-and-forget” applications, where it is impossible to change battery or replace the sensors, and which only need to report relevant events - such as humidity levels or vibration/strain changes above a defined threshold. Other use-cases within smart cities involve smart metering, as mentioned above.

The market for this type of solution is huge. If we take smart buildings, estimates point to a global market value in the area of $70-80 billion, growing at a CAGR of approx. 20% over the coming 5-10 years. As for smart cities, estimates are highly variable, but the global market shall be worth at least $100 billion currently and growing in the ballpark of 20%. Durability, robustness, compactness and cost-effectiveness of distributed sensor systems (hundreds/thousands of sensors) are key drivers for these markets. All in all, these are extremely attractive market segments, where ONiO’s technology can make a real difference.

In conclusion

Edge computing is a vast concept, and we are far from pretending to be able to crack all challenges and unshackle its immense potential. Yet, we are pushing the boundaries of what can be done with extremely small amounts of power, using ambient energy only, and offering decisive contributions towards wireless, true batteryless IoT nodes. These, in turn, will make it possible for the future vision of the all-connected Internet of Everything world!

Are you ready to take your smart city and/or building initiatives to the next level? ONiO has initiated an R&D project focused on creating power-aware neural network solutions, supported by the Regional Funds in Oslo (RFF Oslo). These solutions can enable infrastructure-embedded sensors (such as moisture sensors in walls or vibration sensors for condition monitoring and preventive maintenance) to report relevant events without having to continuously send data to the backend/cloud.