Computing was very popular a few years ago, but as all applications were deployed and massive amounts of data flooded the cloud, corresponding problems arose, such as slow processing speeds and long processing times. Therefore, edge computing has become a research hotspot. In fact, edge computing originated from IoT applications, and its role lies in optimizing cloud computing within IoT applications. Its initial purpose was to handle some computations that don't need to be uploaded to the cloud remotely. So, what role does edge computing play in embedded system applications? And how can embedded engineers best utilize this technology? To this end, the editors of *Microcontroller & Embedded System Applications* invited industry experts to share their perspectives!
Industry voices
The Internet of Things (IoT) is a driving force behind edge computing.
Secretary-General of the Embedded Systems Association
He Xiaoqing
In his paper "Edge Computing: Vision and Challenges," Professor Weisong Shih of Wayne State University defines edge computing as follows: "Edge computing refers to a computing technology that can be performed at the network edge. Such technologies and platforms upload and download data between the cloud and IoT devices to balance the requirements of system computing, real-time performance, power consumption, and security." GE, a giant in the Industrial Internet of Things (IIoT) field, points out that "edge" refers to computing infrastructure located close to data sources, such as industrial controllers and data collections from various devices and sensors, which are typically located far from cloud computing centers.
International research reports indicate that edge computing is currently developing primarily in two areas: First, communication and industrial-related projects, such as 5G virtualization gateways, wireless access networks, and 5G-CORAL (combining research findings from telecommunications edge computing and fog computing). Second, smart IoT gateways, such as Bosch (Prosystgatewaysoftware), Siemens (IoT2000gateways), Microsoft (Azure IoT Edge), and Amazon (Greengrass and Snowball Edge). Open-source edge computing projects are also very active, with notable examples including EdgeXFoundry (IoT gateway) and ParaDrop (University of Wisconsin-Madison).
Embedded companies place great importance on edge computing. NXP launched the EdgeScale platform and Edge-Box development kit, providing edge computing with support ranging from high-performance embedded processor chips and open-source software operating systems to cloud and security support. ARM launched MbedEdge, which, along with ARMmbedCloud and MbedOS, forms a complete IoT solution supporting edge computing. Advantech of Taiwan launched the EIS intelligent edge server and the WISE-PaaS/EdgeSenser edge gateway, the latter integrating MbedOS and MbedEdge, as well as NXP edge embedded processors. EIS also includes Microsoft's AureIoTEdge service.
It's important to note that edge computing devices are typically installed at the system's sensing and execution ends, and most are embedded systems. However, from the perspective of an IoT system, edge computing devices are an integral part of the entire IoT system; they cannot exist in isolation and require the support and services of cloud computing and artificial intelligence technologies.
Research and product development in edge computing technology are currently in their early stages, attracting significant attention from cloud computing, communication equipment, and embedded systems companies. The rapid development of the Internet of Things (IoT) is the biggest driver of edge computing's rise. The development of IoT can be roughly divided into two periods: the first is the transition from interconnectivity to intelligence, and the second is the transition from intelligence to autonomy. Most IoT projects are currently in the first period; edge computing will play a crucial role in the second period, and it is expected to enter a new phase of rapid development in the coming years.
The value and future of embedded systems and edge computing
Sunsea AIoT Technology Co., Ltd.
Chief Technology Officer Zou Jun
Edge computing refers to providing computing and distributed processing capabilities at the network edge, bringing data processing and related applications closer to where the data is generated. It features low latency, fast response, and low bandwidth consumption, complementing cloud computing.
Edge computing plays several key roles: First, it enables rapid data processing and response. Edge computing and its applications are distributed at the network edge, closer to the data source than cloud computing, sometimes even just one hop away. This proximity allows for data processing, reducing latency. Second, it reduces the amount of data transmitted to the cloud, lowering network bandwidth requirements. With the explosive growth of the Internet of Things (IoT), devices generate massive amounts of data. According to predictions from consulting firms like IDC, 20% to 40% of future data will be processed by edge computing. Processing data at the edge, such as simple data filtering, significantly reduces bandwidth pressure and lowers data transmission and storage costs. Third, it avoids single bottlenecks and single points of failure. Because edge computing and its applications are distributed, it reduces or even eliminates business bottlenecks and single points of failure.
Because edge computing is distributed at the network edge, the computing and storage resources of edge nodes are limited compared to cloud computing/data centers. Therefore, edge computing and its applications naturally utilize embedded computing and embedded hardware units. Embedded systems can play the following roles in edge computing:
1. Data Filtering
Objects generate massive amounts of data, much of which is useless or noise. Simple computational processing can filter out a large portion of this data. For example, temperature sensors can periodically collect and report data. In the simplest scenarios, only abnormal situations need to be monitored. Therefore, in edge computing, a simple numerical comparison can filter out the vast majority of data. This simple data filtering function can effectively leverage the low resource requirements of embedded technology for edge processing.
2. Data Statistical Analysis
Edge computing can perform statistical analysis on data within a specific geographic area and transmit the results to the cloud. Similar functionality can also be achieved using embedded computing, thus ensuring that the analysis results closely reflect the actual data.
3. Complex Event Processing (CEP)
CEP is a relatively mature technology. When CEP is combined with embedded systems, it can be used as an edge application to process events quickly and locally, thereby greatly improving the response speed to events.
4. Applications of Artificial Intelligence
Cloud computing is a major driving force behind the rapid development of artificial intelligence. However, placing all AI in the cloud makes it difficult to achieve real-time processing of emergencies, such as in autonomous driving scenarios. Admittedly, AI training requires substantial computing resources, which edge computing struggles to handle. However, increasingly more AI inference can be performed at the edge, enabling AI applications as edge computing applications to have rapid response capabilities. Therefore, how to implement AI inference and applications in embedded systems is also a crucial topic and direction for development.
Based on the above technologies, Sunsea AIoT Technology Co., Ltd. (hereinafter referred to as Sunsea AIoT) launched the world's first AIoT mobile intelligent computing terminal MICD (Mobile Intelligent Computing Device) product at the Mobile World Congress Shanghai in June this year. This product is a perfect combination of artificial intelligence, edge computing, and mobile computing. For details about this product, please refer to the following related reports (Phoenix Network report http://tech.ifeng.com/a/20180703/45047508_0.shtml, etc.).
Founded in 2004, Sunsea AIoT is committed to becoming a leader in the field of AIoT (Artificial Intelligence of Things). Sunsea AIoT provides leading IoT "cloud + edge" solutions, cloud video, data centers, wireless communication, wired broadband, and energy-saving products and solutions to domestic and international operators, ICT equipment manufacturers, system integrators, and users across various industries. Since 2016, focusing on a global strategic layout, it has implemented proactive expansion, acquiring module manufacturers "Longsung Technology" and "SIMComm," becoming the AIoT company with the highest global module shipment share. Simultaneously, it introduced the Ayla cloud platform (the first in China to achieve Level 3 SOC security), providing innovative AI technology and becoming the world's first "cloud + edge" ecosystem IoT company with AI capabilities.
Edge computing makes the Internet of Things smarter
Director of Connectivity and Smart Home at Imagination Technologies
SimonForrest
In a world comprised of countless intelligent, connected devices, it's widely believed that available communication channels will soon be overwhelmed by the massive amounts of data generated by these devices. This is clearly unsustainable and requires a different approach. Edge computing has demonstrated that not all data needs to be transmitted to a central server infrastructure for processing in the cloud. Embedded edge technologies, by their very nature, leverage local computing power, enabling field devices to proactively transform "data" into "information" earlier in the delivery chain, with only critical information being transmitted to the cloud for storage and further processing. This significantly reduces the peak bandwidth requirements of the entire network and also shortens the total CPU cycles required in the cloud.
More importantly, the ability to process data quickly within the device itself provides a new level of autonomy. Today, we are seeing a new wave of products adopting artificial intelligence and relying on (edge) neural network acceleration embedded within the SoC. These products are often seen as "smarter," capable of performing actions locally based on input and processing data independently, especially when server-based (cloud) infrastructure is unsuitable or the device is offline.
For example, an IoT home security system that uses edge technology to process video frames locally can incur minimal processing overhead in the cloud. Instead of inefficiently sending all video frames to the cloud, as most are identical, the system uses embedded AI edge technology to identify only those frames that display suspicious activity. Once processed, the system may choose to send only the useful video excerpts to the cloud or simply send an alert to your smartphone. Using the previous analogy, we have now transformed gigabytes of redundant video data into a small number of bytes containing useful information, which can be achieved simply using edge technology.
Device scalability is also significantly improved because edge technology ensures that distributed computing resources scale proportionally across the system, with each additional device adding more capacity. Another key aspect is latency; data forwarding between the cloud and devices requires round-trip time associated with data transmission. If embedded edge technology is used to process only data within the device, latency can be minimized, or in some cases, eliminated entirely. This is particularly important for embedded systems that require absolutely guaranteed response times.
The demand for increased computing power and "intelligence" in IoT products remains constant. The introduction of new technologies such as natural language processing for voice control, gesture-based interfaces, advanced image classification, and visual recognition systems signifies that edge computing has become increasingly important on a large scale. For Imagination, this means higher performance and lower power consumption, thanks to the advanced neural network acceleration and GPU computing technologies we've developed for our SoC partners. These SoCs will have the ability to surpass existing performance metrics, moving beyond simply transmitting IoT data to the cloud to extracting valuable information from multiple inputs, all processed within the silicon chip.
eFPGA technology provides security for edge computing environments.
Senior Product Marketing Manager, Achronix Semiconductor
AlokSanghavi
Next-generation embedded applications are pushing processing tasks off the cloud and to the network edge. Simultaneously, building processing architectures around programmable logic provides new capabilities that make computing more data-centric. Programmable logic enables the construction of data processing pipelines; conversely, traditional processors require complex memory cache hierarchies to push data into their processing pipelines.
Supported by computing solutions built with programmable logic, data can flow seamlessly between nodes, with data units being manipulated as they pass through using a combination of custom logic circuits and DSP engines. Each unit, after being processed, is forwarded to the next node. As needs change, the programmable logic array can be rewired and configured to better support data-centric applications.
Standalone FPGA chips often incur additional power consumption and performance penalties due to the frequent data interaction with more specialized ASICs. Embedded FPGA (eFPGA) technology offers a way to meet energy efficiency, performance, and size constraints by integrating a programmable logic array within the ASIC. Using eFPGA technology, hardware acceleration capabilities can be brought into the chip. A prime example of these capabilities is reconfigurable processing unit arrays used for convolutional kernels or max-pooling computations required for machine learning applications. By implementing these functions within an embedded programmable logic array within the ASIC, higher overall chip performance can be achieved while reducing system cost and power consumption.
eFPGA technology offers another advantage in edge computing environments—containers and virtualization provide effective support for secure operations in the core cloud because these systems can leverage robust physical security. Devices at the network edge require a higher level of hardware protection because attackers can more easily break into chassis and tamper with system settings in roadside cabinets or service rooms. Physical security is crucial because edge computing systems receive less support from administrators.
Security features are integrated into the hard-wired logic surrounding the eFPGA core, enabling encrypted uploads of virtual circuits to the logic array and continuous monitoring for potential vulnerabilities. The hard-wired logic ensures the separation of programmable functions uploaded by different users and prevents them from eavesdropping on each other.
By integrating security circuitry and programmable logic onto the chip, physical access for attackers to eavesdrop on communications becomes extremely difficult, if not impossible. With the integrated CPU, the entire service's computational functions can be isolated from the outside world and routed directly to the eFPGA, thus limiting the amount of information transmitted from the chip. When communicating with other services, strong encryption units within the hardwired logic of the eFPGA can be used to support a robust security architecture suitable for edge computing needs.
