Smart homes, industrial IoT, wearable devices, and smart surveillance are currently hot topics of interest. Enhancing the AI processing capabilities of edge computing will significantly improve the intelligence level of terminal devices. In this context, edge AI has entered a fast track of development in recent years. The MCU is the most widely used processor in the field of edge computing, and its integration with edge AI has become an important development trend.
MCU Manufacturers Accelerate Their Layout in Edge AI
In recent years, the development of edge AI has been accelerating. Due to the explosive growth of data generated by terminal devices, in order to achieve rapid response and parallel processing with cloud big data, more and more manufacturers are turning their attention to edge computing, attempting to embed trained AI algorithms into terminal devices, thereby greatly enhancing the computational capabilities on the terminal side. However, to realize this vision, edge AI processors must not only possess a certain level of computational efficiency but also feature low cost and low power consumption. The MCU is undoubtedly one of the best choices to achieve optimal results, which is why an increasing number of MCU manufacturers are beginning to integrate MCUs with edge AI.
Renesas has launched the embedded AI technology “e-AI” for this purpose. For example, in the field of smart manufacturing, the “e-AI” unit solution can be added as an additional unit to devices, achieving the entire process from sensor data collection to data processing, analysis, and evaluation/judgment through pre-learned AI processing models. STMicroelectronics promoted the slogan “Let most STM32 products support AI deep learning” at the fourth STM32 summit held in Shenzhen last year. Silicon Labs has also partnered with AI innovation company Cartesiam to optimize algorithms for the IoT. In a demonstration observed by reporters, it showcased how edge AI can be applied in industrial IoT for motor control. For instance, a machine learning algorithm was trained on fan operating modes, and when it detected an anomaly, it would send an alert to a display via Bluetooth.
Domestic MCU manufacturers have also recognized this development trend. Reports indicate that Qianxin Technology is collaborating with GigaDevice to engage in AI ecological cooperation in the artificial intelligence and RISC-V ecosystem, jointly providing customers with AI algorithm chip-level acceleration solutions based on TensorChip and GigaDevice GD32 MCU technology. Through algorithm compression and chip collaboration technology, they aim to further enhance the advantages of the GD32V series MCUs in the AIoT field and provide more convenient integrated solution options for customers in relevant fields.
Application Implementation: A Surge of Intelligent Terminals
“MCU + Edge AI” has begun to be applied in an increasing number of fields. “Edge computing refers to moving computing and processing capabilities from cloud data centers closer to the user access network. For MCU manufacturers, in addition to continuously optimizing the integration, power consumption, cost, and security of the chips themselves, they also need to build a broad product platform and advance from multiple dimensions to meet the ever-evolving application demands,” pointed out Jin Guangyi, Product Marketing Director at GigaDevice. Integrating some simple AI algorithms into MCUs is a supplement and enhancement to the existing MCU product line. With the arrival of 5G, there will be higher and more extreme demands for latency and energy consumption from traditional products, making edge computing a very effective implementation method. The significance of implementing AI on MCUs lies in combining the low power consumption, low cost, real-time performance, stability, short development cycle, and broad market coverage of MCUs with the powerful processing capabilities of artificial intelligence, thus enabling a surge of intelligent terminals.
Image and voice processing is one of the target application areas for “MCU + Edge AI”. For example, graphic recognition, voice assistant wake word processing, and other sound classification applications used in various security systems. In the field of smart manufacturing, many factories have been affected during the pandemic, coupled with rising labor costs, the trend towards automation and intelligence in industrial manufacturing will deepen. Additionally, the proliferation of 5G will drive the development of IoT, vehicle networking, and Wi-Fi 6. At the same time, the wider coverage of 5G base stations will lead to a wave of “replacement frenzy,” generating more consumer electronics demand around the 5G ecosystem, smart homes, and IoT, thus the demand for MCUs will also grow rapidly in the coming years. Chen Guodong, Senior Marketing Manager of Cypress’s IoT Computing and Wireless Division in China, stated: “In recent years, the country has been vigorously supporting 5G, electric vehicles, high-end manufacturing, consumer electronics, and IoT. These directions will become the engines for sustained economic development in the future and are also markets that the MCU industry must focus on in the coming years. Under these broad directions, some subfields will emerge, such as smart locks and smart speakers.”
In the post-pandemic era, the market for wearable devices has also expanded. “Recently, NBA teams have started using smart rings to monitor players’ basic health parameters to provide early warnings for COVID-19. This smart ring is based on Cypress’s PSoC series MCUs. After the pandemic, people’s attention to personal health will increase, thus promoting the development of the wearable device market. Existing devices mainly conduct routine monitoring of heart rate, blood pressure, etc. In the future, they will encompass more functions, including backend big data analysis, mining, and disease correlation research, which will also require the application of edge AI technology,” said Chen Guodong.
Balancing Power Consumption and Performance is a Challenge
In the future, if AI applications are to be realized in various fields of life, MCUs are undoubtedly one of the best choices. However, MCUs currently exhibit relatively low performance, so there are still certain challenges in applying MCUs to AI computations.
“With the rapid development and popularization of cloud computing, edge computing is becoming increasingly important in the field of artificial intelligence. There are many reasons for this phenomenon, such as the heavy network burden caused by massive data flows, security issues, and system latency. To improve the overall performance of MCUs used for edge computing, the industry expects MCUs to possess high processing capabilities, ultra-low power consumption, ultra-small size, and enhanced security mechanisms,” pointed out Qiu Yi, Marketing Manager for IoT Products at Silicon Labs in the Asia-Pacific region.
How to achieve a balance between power consumption and performance is a question that manufacturers need to consider. “Currently, the development of the artificial intelligence industry needs to realize scene implementation, and the demand differences for different scenes and applications are very large. Simply achieving voice recognition and facial recognition is far from enough; the core competitiveness of the industry in the future will be high integration, ultra-low power consumption, and ultra-high cost performance. Taking smart locks as an example, while achieving facial recognition, can the processor also integrate fingerprint recognition, lock control, and other system functions? Artificial intelligence must be implemented in the scene, and then the functions must be refined for that scene. Therefore, it is necessary to improve the integration of MCUs based on this scene while reducing power consumption and improving cost performance,” said Chen Guodong.
In addition, manufacturing processes are a key consideration for MCU manufacturers during product design. “The next generation of MCU processes will develop to 40nm and 22nm. These nodes are not cutting-edge technology nodes, but they are sufficient for MCUs. At these two sizes, MCUs can achieve optimal cost. The more advanced the node, the better the dynamic power consumption of the MCU, but the static power consumption will worsen, so a balance point must be found. 40nm and 22nm are very suitable technology nodes for MCUs,” pointed out Chen Guodong.