Edge AI Compute Modules

Overview

System-on-Module (SoM) platforms combining Cortex-A application processors with high-performance GPU/NPU/TPU accelerators for edge AI workloads beyond MCU class. Common examples include NVIDIA Jetson Orin / AGX, Hailo-8/15 plus carrier modules, Google Coral Dev Board, Qualcomm RB5/RB6, and emerging modules from BeagleBoard.org and SeeedStudio.

Benefits

• TOPS-class inference performance for vision, perception, and multimodal models
• Pre-integrated Linux and driver stacks accelerate prototyping
• Carrier-board ecosystems make custom enclosures tractable
• Long-term roadmaps from major silicon vendors

Limitations & Risks

• Power and thermal envelopes are real (10–50W typical)
• Driver and firmware blobs are still common, complicating long-term support
• Pricing premium versus MCU class is significant
• Vendor SDK lock-in (TensorRT, Hailo Tappas, Coral compiler) is real

Recommended Actions

Choose by deployed-application headroom rather than peak TOPS marketing — a smaller module that runs cool and meets the framerate is usually better than the biggest. Plan for vendor SDK migration cost up front.

Additional Notes

Distinct from AI-Capable / NPU MCUs — this is the higher-power, Linux-running class. Pairs with #11 Edge ML Lifecycle and Embedded Linux SoC Platforms.

References & Links

• https://developer.nvidia.com/embedded-computing
• https://hailo.ai/