ASE, Inc. Boosts Yield and Efficiency with Dragonwing Edge AI

Title

ASE, Inc. Boosts Yield and Efficiency with Dragonwing Edge AI

Content

ASE (Advanced Semiconductor Engineering, Inc.), a leading provider of outsourced semiconductor assembly and test (OSAT) services serving IoT, automotive, and AR/VR markets, developed a GenAI-powered maintenance support tool internally called aMA. ASE selected the Qualcomm Dragonwing AI On-Prem Appliance Solution with Qualcomm Cloud AI 100 Ultra cards to deliver scalable, high-performance edge AI across its cleanroom operations. ASE deployed the Dragonwing AI On-Prem Appliance with Qualcomm Cloud AI 100 Ultra cards to run vision and language inference workloads directly at the edge. This cleanroom-optimized appliance integrates with ASE's IT and manufacturing stack to automate inspection, documentation, and onsite safety compliance tasks without relying on the cloud. Each Qualcomm Cloud AI 100 Ultra card packs 64 AI cores and delivers up to 870 TOPS (INT8) and 288 TFLOPS (FP16), enabling high throughput for real-time wafer defect and PPE detection. Technically, the architecture combines on-prem Dragonwing appliances with locally run Large Language Models that power the ASE Maintenance Copilot (aMA), enabling maintenance documentation and inspection report summaries without cloud latency. The solution delivers 150W TDP with passive cooling, achieving 3.8× better performance-per-watt than NVIDIA A100 in factory environments. The system supports 1–2 AI cards per appliance and up to 40 concurrent users, ideal for global factory rollouts. Containerized AI and standard APIs enable integration with ASE's factory systems, including MES and ERP, supporting automated workflows and unified operations across sites. Business outcomes include reduced downtime with faster maintenance and fewer unplanned outages through On-Prem AI diagnostics, predictive maintenance improving uptime and reduced equipment servicing costs per wire bonder, improved defect accuracy with high-throughput CV inference, lowered total cost of ownership by eliminating recurring cloud inference costs, enhanced compliance with cleanroom protocols through real-time safety monitoring, and faster local decision making and root-cause analysis with On-Prem LLMs and vision copilots.