The core board selection follows a five-step process: "determine the scenario → select the core board → match core parameters → ecosystem and quality → cost and supply chain," balancing current needs with long-term stability.

I. Determine the Scenario and Environment Level First

The scenario determines performance and interfaces.

Industrial Control/Gateway: Choose industrial grade, wide temperature range (40℃~85℃), strong EMC, and RTOS support. Focus on RS485/CAN/multiple network ports.

Edge AI/Machine Vision: Consider NPU/GPU/ISP, such as RK3588 or i.MX 8M Plus. High-bandwidth memory and storage are required.

Low-Power IoT Nodes: Choose MCU (STM32/ESP32) or A7/A53 small cores. Prioritize LPDDR4/LPDDR5 and eMMC. Power consumption <1W.

Automotive/Medical: Automotive-grade AECQ100, medical-grade, long lifecycle (5-10 years) supply.

Environment Level Must Be Correctly Selected: Commercial grade (0℃~70℃), industrial grade (40℃~85℃), automotive grade (40℃~125℃). Select according to actual operating conditions to avoid rework later.

II. Matching Core Hardware Parameters

Main Control and Architecture

Lightweight Control: Cortex-M (STM32) or Cortex-A7, low cost and low power consumption

General-Purpose Computing: Cortex-A53/A55 (e.g., i.MX 8M Mini, RK3568), balancing performance and power consumption

High-Performance/AI: Cortex-A76/A72 + NPU (RK3588, i.MX 8M Plus), supporting complex algorithms and video encoding/decoding

Special Scenarios: FPGA (e.g., Xilinx Zynq) suitable for real-time signal processing; RISC-CV suitable for domestic production and low power consumption

Memory and Storage

Memory: Linux/Android recommended ≥2GB LPDDR4/LPDDR5; Industrial gateways/AI recommended ≥4GB; RTOS/MCU can be as low as 256MB

Storage: System + Application recommended eMMC ≥8GB; for large local storage needs, choose ≥64GB eMMC or NVMe expansion

Interfaces and Expansion

Required Interfaces: Confirm UART/SPI/I2C/USB/Ethernet/CAN/PCIe, etc., according to peripheral requirements, with a 20% margin.

High-Speed Interfaces: AI/video requires PCIe 3.0/4.0, MIPI CSI/DSI, and Gigabit/2.5G Ethernet ports.

Connection Methods: Gold fingers (for mass production reliability), stamp holes (for miniaturization), board-to-board (for high density and high speed). Standardized pin definitions are preferred for easy board replacement.

Power Supply and Consumption: Specify the input voltage (e.g., 12V/5V). The core board must be able to output stable voltages such as 3.3V/1.8V/1.2V. Control power consumption according to the scenario; battery power is preferred (<1W), industrial gateways (<5W).

III. Software Ecosystem and Development Support

System Adaptation: Confirm support for target OS (Linux/Yocto, Android, FreeRTOS, RTLinux), complete BSP, and no missing drivers (e.g., Ethernet, WiFi/Bluetooth, display). Tools and Documentation: Manufacturers should provide SDKs, cross-compilation chains, debugging tools, hardware reference designs, and user manuals. Avoid incomplete documentation.

Community and Technical Support: Prioritize major manufacturers (e.g., Rockchip, NXP, Advantech) or established solution providers. They offer technical support and community resources, making problem-solving easier.

IV. Quality, Supply Chain, and Cost

Reliability and Certification: Industrial/automotive applications require EMC/ESD/vibration testing and CE/FCC/ISO or industry-specific certifications (e.g., AECQ100).

Supply and Lifecycle: Core chips and core boards must have a stable supply throughout the project lifecycle. For industrial products, a lifespan of ≥5 years is recommended. Avoid using niche models with a high risk of discontinuation.

Cost and Mass Production

Small Batch (<100 units): Prioritize mature solutions to reduce development costs.

Large Batch (≥500 units): Customizable BOM for cost optimization; overall cost of core board + baseboard must be controllable, balancing performance and price.

V. Selection Steps and Pitfalls

Four-Step Quick Selection Method

1. List Scenarios: Environment, Power Consumption, OS, Required Interfaces, Performance Indicators (Computing Power/Bandwidth/Storage)

2. Narrow Down: Filter by scenario based on main control architecture, environment level, memory/storage, and interfaces, narrowing down to 23 models.

3. Prototype Testing: Evaluate BSP completeness, driver adaptation, stability, and performance; conduct high/low temperature/EMC baseline testing.

4. Finalize the Solution: Confirm supply cycle, technical support, and cost; finalize the solution.

Common Pitfalls and Countermeasures

Performance Overkill: Select based on actual load to avoid high cost and high power consumption.

Insufficient Interfaces: Reserve 20% interface margin to prevent limited future expansion.

Weak Ecosystem: Prioritize models with mature BSPs and complete documentation to reduce debugging time.

Unstable Supply: Choose mainstream chip solutions, avoid niche models, and confirm long-term supply commitments.

VI. Quick Selection Reference (2025 Mainstream)

Industrial Gateway: i.MX 8M Mini (A53), RK3568, wide temperature range, multiple Ethernet/CAN ports, supports Yocto/RTLinux

Edge AI: RK3588 (A76+NPU), i.MX 8M Plus, 4GB+ RAM, supports 8K codecs and multi-camera support

Low-power IoT: STM32MP1 (A7+M4), ESP32S3, power consumption <1W, supports FreeRTOS/Arduino

Automotive central control: i.MX 8QuadMax, automotive-grade, multi-display, supports Android Automotive