Some thoughts on the development of embedded technology[Copy link]
In the 1970s, the emergence of microprocessors made it possible to embed microcomputers into an object system and realize intelligent control of the object system. People call computer systems that realize intelligent control of object systems embedded computer systems. Initially, people defined embedded as: "Application-centered, computer technology-based, software and hardware tailorable, suitable for special-purpose computer systems with strict requirements on functions, reliability, cost, volume, and power consumption". With the rapid development of microelectronics technology, computer technology has been rapidly integrated with the breadth and depth of applications in various industries, making embedded applications increasingly widespread, embedded products in various forms and ubiquitous, and also making great changes in the connotation and extension of the concept of embedded systems. Embedded technology involves microelectronics technology, application technology, computer technology, and software technology. The technologies interact, influence each other, and complement each other. "Ubiquitous computing" makes embedded technology a hot topic. This article will explain embedded technology from these four technical perspectives, point out the problems of domestic embedded talents, and give relevant suggestions.
1. Microelectronics technology is the basis for building embedded systems
The microelectronics industry is a basic, leading and strategic industry related to the overall national economic and social development. It is an important indicator of the country's development level and comprehensive national strength, and the core and key to the development of a new generation of information technology. Ultra-high capacity, ultra-small size, ultra-high speed, ultra-high frequency and ultra-low power consumption are the basis for solving "deep embedding", the endless pursuit of information technology, and the eternal driving force for the rapid development of microelectronics technology and industry. Looking back at the origin and development of embedded computers, we can clearly see that microelectronics technology is the foundation for the development of embedded technology. The earliest embedded system originated from the birth of Intel 4004 microprocessor. Various manufacturers have successively launched microprocessors. Embedded systems with these microprocessors as the core have been widely used in various industries such as instruments and meters, medical equipment, and household appliances, forming a broad embedded application market. Serialized, modularized, and standardized single-board computers with embedded processors as the core and easy for users to use have also appeared. While single-board computers were being embedded in applications, the microcontroller, which integrated the basic components of embedded applications such as microprocessors, IO (input and output) interfaces, A/D (analog-to-digital) conversion, D/A (digital-to-analog) conversion, serial interfaces, RAM (random access memory), ROM (read-only memory) into a VLSI (very large-scale integrated circuit), was born. This was the early single-chip microcomputer, which to a certain extent achieved the miniaturization, low power consumption, and high reliability requirements of embedded applications. This was also the early stage of SoC (system on a chip) technology. The continuous advancement of process technology has greatly improved the integration of chips, making it possible to integrate integrated circuits with more functions. Integrated circuits have therefore rapidly developed to the SoC stage. SoC brings low power consumption, low cost, miniaturization, intelligence and high reliability to embedded systems at the chip level, making many embedded applications that were originally limited by size, power consumption, weight, etc. possible. It can be seen that the development of SoC technology has further accelerated the speed of embedded system upgrades and the degree of miniaturization, and also determined the popularity of embedded systems, the depth of application and the degree of intelligence. With the improvement of chip integration, SoC chips have become the core of embedded systems. System-in-Package (SiP) technology is a technology that integrates the system functions of multiple IC chips, passive components (or passive integrated components), antennas, optical devices, biological devices, and micro-electromechanical systems (MEMS) made by different processes into a single package to form a microsystem device. SiP technology is an effective method to achieve miniaturization and high reliability of embedded systems based on packaging. In recent years, with the rapid development of AI technology, the acceleration of complex algorithms driven by application scenarios and more stringent requirements of "small, low, and light" have posed great challenges to the integration of Monolithic SoC (single-chip system-on-chip), and the scale, integration, and complexity have increased exponentially. For example, the flagship GPU Volta GV100 released by NVIDIA in 2019 is as high as 800mm. It is reported that the new nuclear-grade GPU (graphics processor) of the Ampere series to be launched this year will reach 826mm in the 7nm process; the AI chip of Silicon Valley startup Cerebras Systems has an area of 46 225mm, and the on-chip SRAM (static random access memory) is as high as 18GB. With such a large area of chips, the yield and cost are difficult to control. Monolithic SoC chips have reached the "end of the road" in the past few years. SoC may usher in an era with Chiplet technology as the core.
Chiplet technology is to split the original large Monolithic SoC single chip solution into a combination of multiple small chips, and then reassemble it through advanced packaging. Its essence is the 2.5D/3D packaging of SiP, but the early SiP only meets the link between chips of different processes, such as the heterogeneous integration of CPU/GPU and DRAM (dynamic random access memory). After Chiplet was proposed, different dies can be manufactured using different process nodes and can even be provided by different suppliers. Third-party chiplets can greatly reduce design time and manufacturing costs.
Although SoC will remain mainstream for a long time, Chiplet designs and implements different components on independent dies, providing a new approach to solving the manufacturability and cost issues of highly complex, ultra-large-scale heterogeneous systems. According to the statistical relationship between yield and area size, for chips smaller than 10mm, the yield of the monolithic solution and the chiplet solution is not much different. Once the chip area exceeds 200mm, the yield of the monolithic solution will be more than 20% lower than that of the chiplet solution. It can be expected that the yield of the monolithic solution is likely to be no more than 10% in an area of 700~800mm, and the cost price of the chiplet solution based on mature chips will be much lower than that of the monolithic SoC solution. The chip integrated in Chiplet will be a "super" heterogeneous system, which can bring more flexibility and new development opportunities to AI computing.
The chiplet and microsystem technologies that are fully integrated with the two technical approaches of SoC/SiP described by More than Moore have provided basic support for the low power consumption, miniaturization, high reliability, and intelligent development of embedded systems, making embedded systems have higher added value and further promoting the leapfrog and popular development of embedded systems.
2. Applications drive the development of embedded technology
Human beings' endless pursuit of information acquisition, representation, transmission, processing and use has continuously promoted the hot spots of embedded technology. Each era and different stages in the era have different characteristics for embedded systems. In the industrial era, instrument control, industrial equipment and automatic control were the earliest places for embedded systems to be used; in the information age, home appliances, computers, communications and networks have developed rapidly, and each era is inseparable from embedded technology. The hot topics of the times such as virtual reality, big data, cloud computing, Internet of Things, 5G, blockchain, and artificial intelligence have led to the emergence of a large number of applications such as live streaming, face recognition, smart furniture, autonomous driving, and smart cities. There are a wide variety of smart phones, multi-purpose drones, smart assisted cars, robots and other products, and the demand for embedded applications is becoming increasingly rich and diverse. With the rapid implementation of the Internet of Things, big data, and artificial intelligence technologies in the future, embedded systems will enter human life with greater breadth and depth than ever before.
The continuous expansion and innovation of application scenarios have put forward more requirements on the software and hardware ecosystem of embedded systems. In the early days, there were only microcontrollers for industrial control, but soon DSPs (digital signal processors) and GPUs for signal processing and graphics processing were developed. In recent years, artificial intelligence has also been eager to try. After 2010, with the continuous enrichment of application scenarios and service content, the types of embedded system chips have increased rapidly and the complexity has increased exponentially.
Customized heterogeneous, multi-core embedded SoC system chips have appeared in different application fields such as aircraft, automobiles, mobile phones, and watches. The rapidly developing and continuously subdivided application scenarios require embedded systems to be more professional and customized. The gradual implementation of artificial intelligence will intensify the demand for subdivision of application scenarios. Customizing dedicated processors for application scenarios is the development trend of embedded systems in the future. The increasing complexity of processor functions and the diversification of application scenarios also put forward higher requirements for the software ecosystem.
As embedded systems are increasingly used in high-security fields such as finance, aircraft, automobiles, and nuclear power, higher requirements are placed on the security, reliability, and trustworthiness of embedded systems. Various industries have developed various software and hardware development specifications, standards, and process control systems, and developed corresponding processors and operating systems. As the complexity of applications continues to increase and the scale of embedded systems continues to expand, design methods that meet the characteristics of security, reliability, and trustworthiness still need to be further explored. Applications will continue to drive the coordinated and sustainable development of various embedded technologies.
3. Computer technology is the core of building embedded systems
Applications drive the coordinated development of embedded technology, and different computing architectures and corresponding software and hardware technologies support each stage of embedded computing development. The birth of processors in the 1970s solved the control problem, forming a microcontroller with CPU as the core and integrated various IO interfaces, which quickly realized industrial control, home appliances and other applications; the birth of DSP in the 1980s solved the signal processing problem, and also formed a mobile communication control + processing system with CPU and DSP as the processing core, which promoted the development of mobile communication equipment; in the 21st century, the birth of GPU solved the graphic display problem, formed a graphic display system with CPU and GPU as the core, and promoted the widespread application of visual industrial control and electronic instruments. Since 2010, the visual mobile communication embedded system with CPU, DSP and GPU as the core has triggered a boom in smart phones. The birth of GPGPU in 2006 exponentially improved parallel computing capabilities. NVIDIA also took the lead in launching an embedded system for autonomous driving big data processing with CPU and GPGPU as the core. With the rise of deep learning neural networks, NPU (neural network processor) came into being in 2017. Huawei was the first to integrate NPU into smartphone SoCs, adding AI elements to embedded systems and greatly enhancing intelligent applications such as face recognition and smart photo processing.Each generation of computing technology innovation adds new vitality to embedded technology, making embedded systems have rich functions, powerful performance and better implementation efficiency, and promoting the rapid implementation of various applications.
4. Software technology is the soul of embedded systems
The collaboration between software and hardware is a major feature of embedded systems. With the rapid development of embedded systems, embedded software has also developed greatly. The development language has evolved from assembly language and C language in the early days to C++, Python, and JAVA today. The contention of a hundred programming languages has expanded the application space of embedded systems on the one hand, and on the other hand, it has further subdivided the profession, allowing the potential of hardware to be better explored. At the same time, the emergence of embedded operating systems represented by VxWorks, Android, and embedded Linux has added strong impetus to the development of embedded systems. In today's intelligent world, the development of embedded software has created a software ecosystem for various application fields, including consumer mobile phone terminals represented by Android, robot ecosystem represented by ROS, and unmanned driving ecosystem represented by Apollo. These embedded development ecosystems are based on embedded software technology, unified software architecture and user API (application programming interface), and use hardware abstraction layer technology to build an open hardware support architecture. The emergence of embedded ecosystems not only promotes the orderly development of embedded systems for application fields, but also further promotes the rapid integration of application needs and hardware in industrial development. Looking at the development history of embedded system hardware and software, microelectronics technology provides a strong body for embedded systems, while software technology gives embedded systems a flexible brain, vitality and soul. With the increasing complexity of embedded systems, the complexity and scale of the software ecosystem have grown exponentially, and the implementation of artificial intelligence has accelerated the increase in the complexity of the software ecosystem. Software engineering, open source software, and software quality will become the focus of embedded software. Especially for software systems used in high-security fields such as aerospace, automobiles, and finance, the software design and certification of security, reliability, and trustworthiness are particularly important. The country has invested heavily in trusted software, and has achieved remarkable results in the development of application software in the fields of finance, the Internet, and aerospace. However, in the embedded field, software and hardware are deeply integrated, and trustworthiness design is difficult to achieve through a single software or hardware level. It is necessary to combine the processing characteristics of software and hardware, cooperate with each other, complement each other, and design collaboratively to jointly build a safe, reliable, and trustworthy system. The collaborative design of trusted software and hardware ecosystems will become a research hotspot.
5. Embedded talent training is a shortcoming in China
Embedded development is a technology based on knowledge from multiple disciplines, oriented to specific needs, and characterized by applications. It is a comprehensive application of multiple knowledge. At present, the professional division and knowledge transfer in China are too fragmented. Teaching often emphasizes the development of application software and APP based on a certain hardware and software platform, which basically stays at the application level.
How to build the system and software and hardware platforms, as well as the coordinated and integrated development of software and hardware are the key and core. The cultivation of these comprehensive capabilities is the key to the cultivation of innovative talents, but it is obviously insufficient. System thinking, multidisciplinary integration, coordinated development of software and hardware, and innovative capabilities are also difficult to cultivate in a single discipline and professional direction. The embedded discipline construction system that emphasizes the cultivation of innovative technology and comprehensive application capabilities is very inconsistent with the discipline construction and evaluation system guided by SCI papers, resulting in a disconnect between school talent training and the embedded talents urgently needed for corporate innovation and development. It has become the norm for companies to "run universities" to cultivate the talents they need, which is also a helpless move.
The embedded industry needs a large number of talents, and even more so high-level embedded talents with leadership capabilities. However, there are serious shortcomings in the current domestic embedded talent training, which is far from meeting the urgent needs of the industry for embedded talents at all levels. How to build a talent training system with embedded characteristics requires the industry and universities to calm down and think carefully, and come up with practical solutions. Schools and scientific research institutions should base themselves on the needs of development and "write papers on the land of the motherland." Enterprises should provide more practical and training positions and a vast scientific research "land" so that teachers and students can immerse themselves in the scientific research practice of the enterprise.
6. Conclusion
This paper reviews the origin and development of embedded technology, and analyzes and summarizes the development relationship between technology and embedded systems from four aspects: microelectronics technology, application technology, computer technology, and software technology. Through the evolution of technology, it points out the future development trend and future research focus of embedded technology. Finally, it analyzes the shortcomings of domestic embedded system talent training, and proposes that universities and enterprises should develop embedded talents based on their own needs, and develop collaboratively and innovatively.
提升卡
变色卡
千斤顶
京公网安备 11010802033920号
