Addressing The New Wave Of The IoT With Edge AI
2Shares
By processing data locally, edge AI can perform faster inference and support real-time use cases.
FEBRUARY 1ST, 2024 - BY:
OMAR CRUZ
Today, we are witnessing the exponential growth IoT is experiencing. Every second, 127 devices are getting connected with an expected forecast for 43 billion IoT devices by 2027. As this market grows and evolves, so does the demand for more sophisticated, powerful, energy efficient and accurate system solutions that can help enrich the way of life. Among the many crucial technologies enabling this exciting future of the IoT, Edge AI will help to enhance IoT’s capabilities by enabling data analysis, predictive insights and intelligent decision making at the edge of the IoT.
Let’s first begin with the basics – what is Edge AI?
While developers and users might be already familiar with Artificial Intelligence (AI) and/or Machine Learning (ML), there’s not yet that sense of familiarity to the edge AI terminology. Edge artificial intelligence (edge AI) is the implementation and deployment of AI applications in an edge computing environment or device close to where the data is located rather than in a central environment such as in a cloud computing facility.
In practice, edge AI involves collecting data from sensors or other sources such as trackers or health monitoring devices, processing that data locally on the edge device using an AI model, and then using the output of the model to trigger an action or send a notification. By processing data locally, edge AI can help perform much faster inference and support real-time use cases, reduce latency and network traffic, improve privacy and security as well as energy efficiency.
Today, developers are looking at a wide range of use cases and applications to be used in the edge AI such as facial/gesture recognition used in appliances and smart home systems; wearables and health monitoring devices; predictive maintenance in factory automation; as well as security cameras using AI to detect and handle suspicious activity in real-time, adding efficiency and less expensive services.
Common use cases to be deploying AI include smart speakers and voice assistant technology using speech recognition analysis relying on a set of complex AI technologies. These include the use of automatic speech recognition (ASR) to convert the sound waves into words and converting those words into actual meanings by using Natural Language Understanding (NLU) and then the smart speaker responding by using Natural Language Generation (NLG).
Additional trends include the smart home market where AI is set to improve the efficiency of edge devices and provide frictionless user experiences. Examples like adjusting water and detergent amount as well as rinsing and spinning times for more efficient washing in washing machines; or learning users’ preferred temperatures, detecting indoor and outdoor temperatures and recognizing the people in the room in thermostats; or ovens personalizing meals depending on users’ taste while being able to provide safety by making sure only adults are able to use the device; and even differentiating floor types and optimize cleaning and battery efficiency in vacuum cleaners. All of these use cases utilize complex AI algorithms at the edge.
Fig. 1: PSoC Edge target applications.
It is normal for a new trend or technology to bring new challenges and key aspects to be aware about, and edge AI is no exception to this. What we currently see is that edge AI has a particular impact on the following areas:
Higher performance (and low power): The growth of the IoT results in the need for more sensors, which in turn results in the sharing of more information, augmenting the complexity of the devices including the need for more computing capabilities. Now, with the added variant of being able to process ML operations on the device itself, the need for a high-performing core plus a neural net compute hardware accelerator to perform ML operations, (both) become essential for this new generation of edge devices. This requirement becomes even more challenging when we add power optimization to the equation so that battery-operated and power-conscious end devices are able to perform efficiently while consuming less power. The upside: Less energy is consumed by storing data and running algorithms on the edge device than would be needed for transferring everything to the cloud.
Security and privacy: While edge AI devices perform the majority of their operations and data processing locally, mitigating security and privacy concerns since less data is sent out to the cloud and external locations, this does not mean that all the data in the edge AI device is inherently protected against attacks. As security attacks continue to evolve, a robust, right-sized level of embedded security is needed across edge AI devices in order to maintain the integrity and privacy of the data.
Lack of expertise, time consumption, and comprehensive enablement: It is next to impossible to develop edge AI devices without the right expertise. This can come in the form of lack of hardware knowledge on how to use certain accelerators and processors optimized for AI/ML, or a lack of experience in how software is used in the development and deployment of AI models. A lack of knowledge and time consumption concerns could prevent developers from opting for the best approach or restrict management from taking the right decision.
Certainly, in these cases, having a well-rounded, robust hardware and software with an end-to-end toolchain solution provided by experts would help to reduce uncertainty, as well as speed up time to market for the next generation of edge AI devices.
Infineon’s PSoC Edge addresses the edge AI challenges by introducing a new family of microcontrollers with a broad spectrum of capabilities including high performance, low power, state-of-the-art security and comprehensive enablement for a faster time to market.
As stated, the increasing system complexity – with sensors aggregation as well as complex data being handled at the edge – is pushing the limits of performance in the microcontroller space. At the same time, low-power consumption and high energy efficiency continues to be the need of the hour across the IoT world. To support these requirements, Infineon has introduced a multi-domain architecture approach with PSoC Edge offering high performance capabilities, including a high-performance core and a 
hardware accelerated neural processor unit, while also supporting increased energy efficiency by featuring an ultra-low-power domain for Always-On applications. This allows an IoT device to remain in deep sleep mode while being able to detect acoustic events or face detection actions and trigger actions so the system can fully wake-up, perform the task required and go back to sleep, maximizing energy efficiency and resulting in longer battery life without sacrificing performance. Thus, edge AI does not only accelerate digitalization but also helps support decarbonization via power optimization.
As mentioned above, another key challenge is to safeguard data protection and minimize security threats. So, it becomes necessary for solution providers to equip their edge AI offerings with higher levels of security such that more secured devices are available for consumers. If not properly protected, Edge AI devices can be tampered with, subsequently becoming a point of entry within the network. Furthermore, in this case attackers can get hold of Edge AI enabled devices broadly available in the market like thermostats, smart speakers or smart locks, analyze them for vulnerabilities and create malicious software to compromise the technology and the network. For these reasons, having a robust, right-sized embedded security architecture becomes paramount for these new wave of Edge AI devices.
Lastly, Infineon understands the importance of time to market for an Edge AI device. With its recent acquisition of
Imagimob, Infineon also now adds the capability to offer an end-to-end ML platform with high flexibility and ease of use, with a strong focus on delivering production-grade ML models. With strong ecosystem partners, comprehensive documentation, evaluation kits with connectivity and HMI modules, as well as the industry-recognized ModusToolbox software, integrated with
Imagimob’s edge AI development platform and its
Ready Models for a faster and validated way to take ML-enabled microcontrollers to production, PSoC Edge provides the hardware, software and tools offerings for a friction-free design experience and an accelerated time to market.
By processing data locally, edge AI can perform faster inference and support real-time use cases.
semiengineering.com
www.linkedin.com
