1.Concept of Embedded Systems
An embedded computer is a dedicated computer system designed for specific applications, based on microprocessors, with customizable hardware and software to meet strict requirements for functionality, reliability, cost, size, and power consumption. In simple terms, an embedded system is software embedded into the Flash memory of application devices, specifically tailored for a particular application domain and scenario. Embedded systems are widely used, from household appliances like washing machines and refrigerators to transportation tools like bicycles and cars, and even in office remote conferencing systems.
2. Technical System of Embedded Systems
Traditional embedded computers can be divided into four layers: hardware layer, driver layer, operating system layer, and application layer. From a practical perspective, embedded systems can be simply divided into five technical systems: microprocessors, DSP processors, operating systems, wireless (or wired) protocols, and application software.
(1) Microprocessors
1. Architecture and Instruction Set of Microprocessors
In software development, taking C/C++ as an example, developers first write source code (like C++) in an IDE, then use a compiler to compile it into a binary file (executable file). During the compilation process, the compiler actually performs two steps: first, it translates high-level programming language into assembly language (programs can also be written directly in assembly language), and then the assembly language is translated into machine code (a series of 0101 codes).A machine code is called an instruction, which can be executed by the CPU. All instructions that the CPU can execute are called the instruction set (ISA). The architecture can be understood as the hardware circuits built to implement the instruction set.
Currently, the commonly used architectures in the industry include Intel’s X86 architecture, AMD, and RISC-V architecture. Both X86 and AMD are closed-source architectures, and their intellectual property must be authorized by Intel and AMD for chip manufacturing, while RISC-V architecture is an open-source architecture that emerged after 2010.
The ARM architecture, formerly known as Advanced RISC Machine, is a 32-bit RISC processor architecture. There are also derivative products based on ARM design, with important products including Marvell’s XScale architecture and Texas Instruments’ OMAP series. ARM processors are widely used in embedded system design, mobile communications, and consumer electronics, such as smartphones, tablets, set-top boxes, digital TVs, routers, and even military facilities like missile onboard computers.
The X86 architecture was specifically developed by Intel for its first 16-bit CPU (i8086), and the CPU used in the world’s first PC launched by IBM in 1981 – the i8088 (a simplified version of i8086) – also used the X86 instruction set. Intel has continuously developed and improved the functionality of X86, which has been used to this day.
The RISC-V architecture refers to the RISC-V instruction set and its derived ecosystem. RISC-V was born in 2007, founded by SiFive. To date, it has accumulated many manufacturers, including Qualcomm, SK Hynix, NXP, Alibaba, Huawei, and others. As an open-source architecture, there are no concerns about technology licensing and political blockades, making it a strong push in China.
2. The Ecological Chain Dilemma of the Architecture
Currently, the main architectures of embedded systems across various industries are still dominated by X86 and AMD, which control a significant portion of the embedded market.
RISC-V, on the other hand, is not controlled by any single company, and its widespread adoption faces the biggest obstacle of the software ecosystem. Many practitioners in the embedded industry start learning from microcontrollers to AMD or X86, while RISC-V has become more of a learning tool for hobbyists.
Furthermore, in terms of software availability and development, RISC-V is still catching up with companies like AMD and X86. For example, many compilers do not prioritize RISC-V, leading to lower code efficiency; major software vendors have not fully embraced RISC-V, resulting in low application rates; and RISC-V has not made significant progress in cloud computing or large data center fields like AMD and X86.
RISC-V emphasizes a completely open-source design, allowing users to add proprietary instruction sets at will, and even choose to keep the architecture closed or maintain it as open-source. This leads to the situation where, although RISC-V architecture has more instruction sets, they cannot be shared. When companies create chips with their unique features, if they choose to keep their proprietary instruction sets confidential, it is likely that years later, the chips will become incompatible.
3. Current Status and Progress in China
Currently, domestic chip companies mainly use processor architectures based on varying degrees of authorization from AMD and X86, and have not fully achieved localization.
(1) Architecture/Instruction Set Authorization: The highest level of authorization allows users to design their own architecture and instruction set (such as extensions or reductions). Typical cases include Haiguang (AMD Zen architecture), Loongson (LoongArch), and Shenwei (SW64).
(2) Core Authorization/Soft Core Authorization: Users can use the processor core but cannot modify the core design, only adjust peripheral interfaces. Representative manufacturers include Kunpeng and Feiteng (based on ARMv8).
(3) Usage Authorization/Hard Core Authorization: Users can only purchase packaged processors, and functional expansion relies on external modules. Zhaoxin (x86 secondary authorization) falls into this category.
In 2019, Alibaba’s T-head released the RISC-V-based Xuantie 910 processor, which performs comparably to ARM A76. However, its specific application and promotion status remain unclear. The rise of open-source instruction sets has forced the X86 camp to rethink its licensing strategy. Intel has quietly initiated the development of the “x86S” pure 64-bit architecture, attempting to simplify the instruction set to meet challenges.
(2) DSP Processors
1. Applications of DSP Processors
DSPs are not always required in embedded systems. For example, in some smart home applications, where the main function is to remotely control and monitor home devices, only simple logical judgments and data transmission are needed, without complex digital signal processing. In industrial control, small motor controllers, temperature controllers, and pressure controllers primarily monitor and control equipment in real-time, using sensors to gather data and adjusting based on preset control algorithms. These control algorithms are usually simple, such as PID control algorithms, which do not require the high-speed computing capabilities of DSPs. In automotive applications, systems like dashboards, window controls, and seat adjustments mainly collect, display, and control various information about the vehicle, typically not requiring complex digital signal processing.
However, in many critical areas, DSPs are essential. For instance, in the 5G field, extensive signal processing is required, such as beamforming, channel coding, and modulation/demodulation, where DSPs can quickly process these complex signals to ensure communication stability and efficiency. In mobile phones, audio processing tasks like noise reduction, voice enhancement, and sound effect processing, as well as video processing tasks like image decoding and video encoding, all require DSPs. In electric vehicle motor control, real-time calculations of motor torque, speed, and other parameters are necessary for precise control, where DSPs can quickly process this data to achieve efficient operation and accurate control of the motor.
2. Challenges of DSP Chips
(1) Dedicated DSPs are suppressed by SoCs and FPGAs, and general-purpose embedded chips have integrated basic DSP instruction sets. For example, DSPs often experience performance degradation due to memory access delays when processing real-time signals. This is especially true in edge computing scenarios, where data cannot reach the computing unit in time, leading to idle multiply-accumulate units, wasted power, and even task delay accumulation. While multi-core DSPs can enhance computing power (such as TI’s TMS320C6474 tri-core chip), integrating more cores requires more complex thermal management and power design, and optimizing parallel programs is challenging, resulting in high development costs. Therefore, dedicated DSPs are currently facing difficulties, with general-purpose processors gradually replacing them. For instance, ARM architecture processors (like Cortex-M/A series) and GPUs are increasingly surpassing DSPs in performance, power consumption, and development convenience. ARM processors can handle 1080P video through soft coding, while GPUs excel in signal processing due to their parallel computing capabilities, leading to DSPs being marginalized in traditional fields like audio and video encoding/decoding. Dedicated IP cores (like H.265 encoders) and FPGAs offer greater flexibility and energy efficiency, gradually replacing DSPs in signal processing within communication systems, with DSPs remaining only for proprietary algorithms or small-batch scenarios.
(2) Technical blockades and severe instability in the supply chain for DSP chips.In high-end DSPs, such as those for automotive applications, there is a long-term reliance on imports (e.g., from TI, ADI in the USA). Military applications face export restrictions, forcing domestic equipment to use lower-grade chips, which limits performance and complicates maintenance. For example, Chinese radar systems once relied on Freescale chips but later developed the Huairui DSP independently. Additionally, domestic DSP-related chips are primarily controlled by companies like TI, which have established a complete industrial ecosystem, making it difficult for domestic companies to achieve breakthroughs.
(3) (Real-time) Operating Systems
1. Functionality of Real-time Operating Systems
A real-time operating system (RTOS) is a specialized operating system capable of precisely and efficiently handling tasks. Unlike general-purpose operating systems (GPOS) that prioritize user interaction and multitasking, RTOS focuses on real-time execution. RTOS emphasizes hard real-time (μs-level response), with interrupt latency typically <10μs. It ensures that critical tasks are completed within strict time constraints, with predictable response times, making it suitable for scenarios with high real-time requirements, such as industrial automation control and aerospace.
In contrast, Linux is a time-sharing operating system and a general-purpose OS, but it can also participate in real-time operations. However, its kernel is inherently insufficient for real-time processing (requiring patches for processing delays), leading to uncertainty in response times when handling high real-time tasks, making it generally suitable for scenarios with less stringent real-time requirements.
For simple functions or extremely resource-limited scenarios, a real-time operating system may not be necessary, and bare-metal programming remains a reasonable choice. For example, extremely simple functions (like LED blinking) with severely limited resources (MCUs with only a few KB of memory). However, due to the rapid development of technology and hardware capabilities, hardware devices must have RTOS to implement increasingly complex functions.
2. Types of Real-time Operating Systems
Currently, most real-time operating systems are open-source systems, generally free, with open-source code, portability, customizability, and flexible scheduling strategies, making them easy to port to various microcontrollers. Examples include RT-Thread, which integrates a real-time operating system kernel, middleware components, and a developer community, featuring a complete and rich component set, high scalability, easy development, ultra-low power consumption, and high security. It has the largest embedded open-source community in China and supports mainstream compilation tools and standard interfaces, with commercial support for all mainstream MCU architectures. It is widely used in various industries, including energy, automotive, medical, and consumer electronics. Other types of real-time operating systems with diverse functionalities include FreeRTOS, μC/OS-II, eCos, Nuttx, SylixOS, VxWorks, and more.
(4) Wireless (or Wired) Protocols and Interfaces
Embedded systems have a wide variety of communication protocols and interfaces, with common types including serial communication protocols and interfaces, such as UART (Universal Asynchronous Receiver-Transmitter), SPI (Serial Peripheral Interface), parallel communication protocols and interfaces, and other communication protocols and interfaces, such as CAN (Controller Area Network), USB (Universal Serial Bus), Ethernet, etc. Others include Bluetooth (2.4GHz band, low power, low cost, high speed), Wi-Fi (2.4GHz or 5GHz band), ZigBee (based on IEEE 802.15.4 standard, low power, low speed), and Z-Wave (a wireless connection protocol designed for embedded systems, operating at frequencies below 1GHz, with a range of up to 30 meters).
(5) Application Software
In the embedded systems industry, microprocessors are often sourced externally, with companies deciding which architecture to adopt based on actual conditions. Most real-time operating systems are open-source and only require customization and porting, making development relatively easy. Therefore, most business activities in the embedded industry are reflected in the application software development process, as well as the adaptation and debugging of hardware and software. This often requires embedded development engineers to possess two core technical capabilities: hardware fundamentals and software programming, along with their integration capabilities. Engineers need to master the hardware fundamentals of embedded systems, including basic circuits and PCB circuit design capabilities; they also need to be proficient in various programming languages; and finally, they must have a thorough understanding of application software and top-level applications, including file I/O, multi-threaded programming, UI, IoT applications, AI applications, etc.
3. Current Status of the Embedded Industry
(1) In the domestic processor (including DSP) architecture, the automotive-grade market remains a challenging area.
The automotive industry’s demand for specialized, efficient, and reliable processors leads to a market that still heavily relies on mature proprietary standards, with limited powerful design tools and support ecosystems. Firstly, automotive-grade chips must meet strict safety and reliability requirements (e.g., ISO 26262), and currently, not all RISC-V solutions can provide these requirements. Additionally, localization has not yet achieved a complete ecosystem, not only for processors but also for corresponding full-stack solutions—software, safety, and long-term support. At the same time, automotive manufacturers are also reluctant to take risks when adopting new technologies, making the transition to RISC-V a slow process.
(2) Increasingly complex embedded applications demand higher requirements for talent with composite technical capabilities.
Firstly, there is a need for understanding hardware. Embedded engineers must not only understand basic circuits and PCB design but also be well-versed in hardware devices like sensors and actuators, fully understanding the specific parameters of hardware devices to guide software development.
Secondly, the complexity of functions is increasing. For example, in complex environments, data collection, image processing, navigation, control, and special industries are becoming more complex, which raises higher demands on the embedded industry. If technical capabilities are too singular, it will be difficult to achieve better development in the industry.
Thirdly, the ability to integrate understanding and analysis is crucial. A complete embedded product, from hardware to software, is interconnected. If there is a problem with the product, is it a software issue or a hardware issue? How to scientifically and accurately judge and debug is also a technical challenge.
(3) Development Trends in the Interactive and Embedded Industry
1. Interactive Technology
Embedded devices ultimately serve human needs, evolving from matrix keyboards and small LCD screens to the now-common touch displays. Human-computer interaction methods continue to improve and become more convenient, with gesture recognition and natural language interaction being hot technologies in recent years. Human-computer interaction is evolving towards more precise and less intrusive methods, such as brain-machine interface technology (BMI, BCI).
2. Edge AI
The integration of embedded systems with AI models is an important trend in current technological development. Through hardware optimization, algorithm adaptation, and development tool support, AI capabilities can be efficiently run on resource-constrained terminal devices. However, this also places higher demands on both hardware and software.
On the hardware side, high-performance processors (such as ARM Cortex-A series) need to be integrated with dedicated AI acceleration modules (such as NPU, TPU) to achieve efficient computing. Additionally, low power consumption is required; for embedded devices with power constraints, hardware-level power management technologies (such as dynamic voltage frequency adjustment) and lightweight AI model designs (such as quantization and pruning) are necessary to extend battery life.
On the software side, models need to be compressed, quantized, and pruned, converting high-precision floating-point models into low-precision integers (such as INT8) to reduce parameter storage and computational load. At the same time, knowledge distillation can be used to simulate the behavior of large models with smaller models, reducing computational complexity suitable for real-time inference on embedded devices. Modular process design should also be considered, such as integrating model preprocessing and post-processing. When deploying generative AI in embedded systems, it is necessary to combine modules like speech recognition (ASR) and text generation (TTS).
4. Current Status of Military Embedded Systems
(1) Special Requirements of Military Embedded Systems Compared to Civilian Fields
Military embedded systems must meet specific reliability and anti-interference capabilities under certain environmental conditions. High reliability and anti-interference capabilities are required. For processors, they need to support wide temperature ranges, resist electromagnetic pulses, be physically reinforced, balance low power consumption with high computing power, and provide long-term technical support. For DSPs, high floating-point computing capabilities and strong radiation resistance are required, along with low latency and deterministic responses from the operating system, ensuring safety and reliability.
(2) Low Reusability of Military Embedded Module Technologies
In military embedded systems, functional modules mainly include interface control, data (signal) acquisition, data processing, image processing, flight control, measurement and control, communication, navigation, positioning, and simulation testing systems based on the above functions. Due to task differences, military branch differences (aircraft, ships, vehicles, missiles, ground, aerospace), functional differences, hardware dependencies (embedded systems cannot operate without customized hardware), and security and confidentiality (military encryption, anti-interference, and limited interface standardization), standardized modules cannot be achieved, necessitating deep custom development. This results in low reusability of various standardized modules, with companies only able to realize technology monetization in civilian fields as much as possible.
(3) Key Hardware of Military Embedded Systems is Transitioning from “Ensuring Existence” to “Complete Domestic Replacement”
Domestic military embedded processors have undergone years of independent development, forming a multi-architecture parallel industrial pattern, such as Phytium, which targets government and military sectors, possessing three core series: FTC8XX (high performance), FTC6XX (balanced), and FTC3XX (low power). Its key technology is based on ARMv8 permanent authorization for independent core design, with the FTC870 series released in 2023 continuously catching up with international standards, supporting military scenarios like individual soldier equipment and vehicle-mounted systems.
However, Phytium’s processors still rely on AMD cores. Some domestically produced products adopt x86-compatible designs, such as Haiguang Hygon (Zen series) and Zhaoxin (KaiXian/KaiSheng series), using technology authorization to achieve compatibility with mainstream ecosystems, serving as transitional solutions in specific military information systems.
(4) Military-Civilian Integration Business Mainly Reflects in Design and Software Layers, Testing Enterprises’ System Design, Algorithm, and Hardware-Software Integration Capabilities
For military supply chain enterprises, specific work involves selecting components and designing schematics at the hardware level; at the driver level, writing hardware peripheral driver programs based on customer functional requirements, allowing applications to access peripherals without knowing the working details; at the operating system level, generally choosing existing domestic or open-source operating systems; and at the application layer, designing and developing corresponding application programs based on user needs to enable products to complete specified functions. The core part of the business lies in writing driver programs and application software independently, forming embedded software with the operating system on the customized hardware layer.
