ARMxy BL450 Embedded Industrial Computer in Semiconductor Wafer Defect Inspection

Based on the hardware and software features of the ARMxy BL450 series embedded industrial computer, it can efficiently meet the high-precision, high-throughput, and reliable requirements of semiconductor wafer defect inspection. Below is a detailed analysis of the solution tailored for this application.

Suitability of ARMxy BL450 for Wafer Defect Inspection

The BL450 series, powered by the Rockchip RK3588J/RK3588 processor, offers the following features that make it suitable for wafer defect inspection:

• High-Performance Computing: Equipped with a quad-core ARM Cortex-A76 (up to 2.4GHz) + quad-core Cortex-A55 and a 6TOPS NPU, it supports efficient inference for deep learning models (e.g., CNN, YOLOv8), ideal for defect detection and classification tasks.
• Image Processing Capability: Supports dual ISPs (up to 48MP, 8064×6048@15fps), compatible with high-resolution industrial cameras to meet the imaging needs for nanoscale wafer defects.
• Rich Interfaces: Provides 2×USB 3.1 (5Gbps), 1–3×RJ-45 Ethernet ports, and HDMI 2.1 (supporting 8K@60fps decoding), facilitating connections to cameras, sensors, and displays.
• Wide Temperature Range and Reliability: Operates reliably from -40°C to 85°C, suitable for semiconductor cleanroom environments; passes electromagnetic compatibility (EMC) and environmental adaptability tests (GB/T, IEC standards) for stability in harsh industrial settings.
• Flexible Expansion: Supports X/Y-series IO boards with RS485, RS232, DI/DO, AI/AO interfaces, enabling seamless integration with wafer inspection equipment (e.g., probe stations, CMP tools).

Wafer Defect Inspection Solution Architecture

The BL450-based solution covers the entire process of image acquisition, preprocessing, defect detection, data management, and automation control:

(1) Image Acquisition

• Hardware Configuration:
  • Connects high-resolution industrial cameras (e.g., Basler or Hikvision) via USB 3.1 (5Gbps), supporting 48MP (8064×6048@15fps) or 32MP (6528×4898@30fps) image capture.
  • Optional Mini PCIe supports 4G/5G modules for real-time image data transmission to cloud or local servers.
  • HDMI 2.1 (8K@60fps) enables real-time wafer image display for debugging and monitoring.
• Software Support:
  • Integrates OpenCV (supported by development examples) for image acquisition and preprocessing, optimizing image quality (e.g., HDR, 3DNR noise reduction).
  • Runs camera drivers on pre-installed Ubuntu 20.04 or Debian 11, compatible with mainstream industrial camera protocols (e.g., GigE Vision, USB3 Vision).

(2) Image Preprocessing and Defect Detection

• Preprocessing:
  • Leverages the Mali-G610 MP4 GPU for accelerated image preprocessing (e.g., grayscale conversion, edge detection, image enhancement).
  • Supports ISP hardware acceleration to handle high-resolution images, reducing CPU load.
• Defect Detection:
  • Deep Learning Models: The 6TOPS NPU supports TensorFlow, PyTorch, and other frameworks, running lightweight models (e.g., MobileNet, EfficientNet) for defect classification (scratches, particles, pattern deviations). The NPU supports INT8/FP16 precision for fast inference, ideal for real-time detection.
  • Development Support: BL450 provides NPU and OpenCV development examples, enabling rapid deployment of pretrained or customized models (e.g., YOLOv8 for object detection).
  • Example Model Performance: Using YOLOv8n, BL450 achieves over 99% defect detection accuracy with single-frame inference time <50ms (for 32MP images).
• Software Tools:
  • Uses Qt-5.15.10 to develop graphical interfaces for real-time display of detection results.
  • Node-Red enables rapid development of data processing workflows, facilitating integration of image analysis with downstream processes.

(3) Automation Control and Integration

• Hardware Support:
  • X-series IO boards (e.g., X23: 4×RS485 + 4×DI + 4×DO) enable communication with wafer processing equipment, controlling robotic arms or marking defective wafers.
  • Y-series IO boards (e.g., Y01: 4×DI + 4×DO) support trigger signal acquisition and control for production line automation.
  • DIN35 rail mounting simplifies integration into existing wafer inspection systems.
• Protocol Conversion:
  • BLIoTLink software supports Modbus, OPC UA, MQTT, and other protocols, connecting to SCADA systems or cloud platforms (e.g., AWS IoT Core, Alibaba IoT) for real-time data upload and process monitoring.
• Remote Maintenance:
  • BLRAT tool enables remote access for convenient device debugging and maintenance, reducing manual intervention in cleanrooms.

(4) Data Management and Traceability

• Local Storage: 32/64/128GB eMMC stores inspection images and results, supporting SQLite or MySQL database management.
• Cloud Integration: Uploads data to cloud platforms via 4G/5G or Ethernet (2×10/100/1000M), using Thingsboard or IgnitionSCADA for data analysis and defect traceability.
• Docker Support: Deploys data processing and analysis applications in Docker containers, ensuring modularity and scalability.

Recommended Product Configuration

Based on wafer defect inspection requirements, the following BL450 configuration is recommended:

• Model: BL452B-SOM452-X23-Y63
  • Host: BL452B (2×10/100/1000M + 1×10/100M Ethernet ports, 2×USB 3.1, 1×HDMI 2.1, 2×Y-board slots)
  • SOM Module: SOM452 (RK3588J, 128GB eMMC, 16GB LPDDR4X, -40°C to 85°C)
  • X Board: X23 (4×RS485 + 4×DI + 4×DO, 20PIN)
  • Y Board: Y63 (4×RS485/RS232, compatible with equipment communication)
  • Expansion: BL452BL (adds 4G module and antenna for enhanced data transmission)
• Rationale:
  • High storage and memory (128GB eMMC + 16GB RAM) support large-scale image data processing.
  • Multiple Ethernet ports and RS485 interfaces meet multi-device connectivity and high-speed communication needs.
  • Wide temperature range and EMC compliance ensure stability in cleanroom environments.

Advantages and Optimization Suggestions

Advantages:

• Efficient Edge Computing: The 6TOPS NPU and GPU enable real-time defect detection, reducing reliance on cloud computing.
• Flexible Expansion: X/Y-series IO boards support diverse interfaces, adapting to various inspection equipment and scenarios.
• Cost-Effectiveness: Compared to x86 industrial computers, BL450 offers lower power consumption (9-36VDC) and competitive pricing.
• Ease of Development: Provides NPU, OpenCV, and Qt development examples, lowering the development barrier.

Optimization Suggestions:

• Model Optimization: For high-resolution images, use INT8 quantization or model pruning (e.g., TensorRT optimization) to improve NPU inference speed.
• Thermal Management: In high-temperature cleanroom environments, pair with active cooling fans to ensure stable operation of RK3588J in 2.0GHz high-performance mode.
• Network Optimization: Utilize 2×10/100/1000M Ethernet ports to separate image data and control signals, enhancing system real-time performance.
• Data Augmentation: Deploy generative adversarial networks (GANs) using BL450’s Docker support to generate defect samples, improving model robustness.

Conclusion

The ARMxy BL450 series industrial computer, with its high-performance NPU, robust image processing capabilities, rich interfaces, and industrial-grade reliability, is an ideal choice for semiconductor wafer defect inspection. It supports deep learning, real-time image processing, and automation control, effectively meeting nanoscale defect detection requirements. With flexible IO expansion and a comprehensive software ecosystem (e.g., BLIoTLink, Node-Red), BL450 seamlessly integrates into wafer production lines, enhancing yield and quality.