In the continuous pursuit for near-zero defects per million (DPM), many manufacturers are turning to industrial vision AI to fully automate their inspection processes. Unlike traditional machine vision that relies on rigid, rule-based systems, vision AI has the capability to learn, adapt, and handle day-to-day variability.
Using high-resolution imaging technologies, AI algorithms, and real-time processing, vision AI achieves precise, reliable defect detection for meeting rigorous global standards.
One sector fully embracing and deploying vision AI is semiconductor fabrication. Manufacturing semiconductors comprises multiple steps like deposition, etching, doping, and lithography. During each step, vision AI is used to identify imperfections that could lead to downstream issues. Along with pinpointing scratches, pattern errors, or particles on wafers, vision AI inspects etched features at the nanometer scale, verifies uniformity in material deposition, and confirms perfect metallisation.
For chips requiring soldering or bonding, vision AI precisely monitors alignment in die bonding, wire bonding, and solder joints. Because of this, vision AI has the proven ability to improve yield and minimise costly waste.
For more than a decade, Advantech has driven innovation in vision AI. Its advanced systems are supercharged by AI Edge servers featuring Intel Xeon scalable processors and leveraging multiple NVIDIA GPU cards to accelerate inspection without compromising performance.
In 2023, frame grabber manufacturer BitFlow became part of the Advantech family. Its CoaXPress (CXP) and Camera Link interface frame grabbers are now being deployed in Advantech’s vision AI, taking these industry-leading systems to the next level in image acquisition.
BitFlow frame grabbers are making it possible for Advantech vision AI to capture and analyse ultra high-resolution images without experiencing latency or frame loss, even in multi-camera setups. BitFlow’s diverse range of frame grabbers also enables the configuration of modular and scalable Advantech Vision AI systems customised to a manufacturer’s unique processes.
As an example, let’s explore an Advantech vision AI solution integrated into an optical inspection machine in a semiconductor fabrication plant.
Components include:
- AI AOI Edge Inference system: Advantech HPC-6240 2U 20” Short-Depth Edge Accelerator Server and ASMB-622V3 5th/4th Generation Intel Xeon Scalable Proprietary Board supporting 8 expansion slots
- Frame Grabber Card: BitFlow Axion Camera Link half-size x4 PCI Express Gen 2.0 frame grabber connected to multiple high-resolution Camera Link cameras
- GPU Card: NVIDIA RTX 6000 Ada high-end professional graphics cards handle complex AI computations in visual inspections
- Cameras: Axion-CL frame grabbers support as many as four Camera Link cameras in base, medium, full or 80-bit formats. CL cameras can be synchronised or completely independent, operating at speeds up to 85MHz
Faster throughput is possible by substituting the Camera Link interface components for a BitFlow Aon, Cyton or Claxon CoaXPress (CXP) frame grabber capable of accelerating transmission speeds to 12.5 Gb/s per link, depending on the model. BitFlow CXP frame grabbers work seamlessly with the Advantech AIR-030 AI Inference System Box based on the NVIDIA Jetson AGX Orin, or the Advantech MIC-770 Compact Fanless System, or several other Advantech IPC options.
Industrial vision AI is transforming modern manufacturing by providing a powerful tool for enhancing product quality and operational efficiency. Frame grabbers are a lynchpin in vision AI, especially in real-time applications requiring latency-free, reliable data transfer.
