Manufacturing

SoM 520/720 Accelerates Edge AI for Generative Applications

As generative AI continues to reshape industries, edge AI devices are entering a transformative era. Powered by MediaTek’s Genio 520 and 720 chipsets, AMobile Solutions has introduced production-ready system-on-modules (SoM 520 / SoM 720) that deliver up to 10 TOPS of AI performance via an integrated 8th-generation NPU. These modules are fully optimized for both Transformer-based language generation (ideal for voice assistants) and CNN-based vision analytics (such as visual defect detection). With support for running popular large language models (LLMs), the platform provides powerful AI acceleration for a wide range of AI-camera applications.
Featuring multi-camera input support, high-speed image processing, and edge inferencing, AMobile’s SoM 520 / 720 platforms are ready to be deployed in smart home, smart retail, and industrial automation environments, enabling private, efficient, and secure on-device generative AI solutions.

🔹 Smart Home: Enhanced Safety with On-Device Intelligence and Conversational Control
With support for up to 2 internal ISPs and 6 FHD MIPI virtual channels, the SoM platform is ideal for building multi-camera AI-powered home surveillance systems:


•    Real-time Person Detection with Event Summary Generation
Automatically identifies family members or unknown visitors and generates contextual alerts (e.g., "19:42 – Unknown person detected near back door").
•    On-Device Voice Assistant with LLM Support
Run LLM models locally to power smart speakers and hubs with natural, multilingual conversational capabilities—even offline.
•    AI Smart Camera for Home Safety Monitoring
Monitor critical zones like the kitchen. If unattended boiling on the stove is detected, the system can issue immediate fire alerts to prevent accidents.
•    Privacy-Preserving Edge AI Processing
All voice and video data are processed locally, ensuring maximum user privacy without dependence on the cloud.•

🔹 Smart Retail: AI That Sees, Understands, and Responds
With multi-display output support (eDP 4K60, DP 5K60) and extensive I/O interfaces, the SoM 520 / 720 can be embedded into smart shelving, in-store AI camera systems, and interactive signage:
•    Real-Time Customer Profiling with Dynamic Ad Generation
Detects age, gender, and dwell time to dynamically generate promotional content based on current campaigns—automatically displayed on connected digital signage.
•    LLM-Based Virtual Shopping Assistant
AI-powered assistants offer natural-language product recommendations tailored to shopper needs and inventory data.
•    Anomaly Detection and Smart Alerts
Generate real-time alerts like "Customer lingering too long in the refrigerated section" to assist staff in managing behavior and store flow.

🔹 Industrial Automation: Visual Inspection and Smart Reporting Combined
With a wide operating temperature range (-10°C to 50°C) and efficient 6nm processing, SoM 520 / 720 are engineered for stable operation in demanding industrial environments:
•    Defect Detection with Auto-Generated Maintenance Summaries
AI-Camera modules detect visual defects and generate actionable insights (e.g., "Misaligned solder joint in Module B – manual inspection recommended").
•    AI-Driven Inspection Robots with Voice Feedback
Responds to operator voice queries and generates real-time system summaries such as: "Line 3 is running at 10% lower frequency than last week."
•    Simultaneous Multi-Camera Monitoring with Edge Decision-Making
Handle multiple video feeds on a single device, enabling real-time analysis and reducing manual inspection overhead.

AMobile’s SoM modules are backed by a complete hardware validation framework, including board support packages (BSPs), development kits, and software documentation. This allows customer engineering teams to focus on application-level development while reducing system integration risk, shortening validation cycles, and accelerating time-to-market.

The AMobile SoM-SD520 and SoM-SD720 are now available for order.
To learn more or request a sample, please visit the product page for detailed specifications and documentation.

Go Back