In today’s Coredump Session, we dive into the fast-evolving world of Edge AI and its real implications for device makers. From robots that detect humans to welding machines that hear errors, we explore the rise of intelligent features at the hardware level. The conversation spans practical tools, common developer traps, and why on-device AI might be the most underrated revolution in embedded systems today.

Key Takeaways:

• Edge AI means real-time inference on embedded devices, not just “AI at the edge of the network.”
• Privacy, latency, and power efficiency are core reasons to use Edge AI over cloud processing.
• Hardware accelerators like the Cortex-M55 + U55 combo have unlocked GPU-like performance in microcontrollers.
• Battery-powered AI devices are not only possible—they're already shipping.
• Data collection and labeling are major bottlenecks, especially in real-world form factors.
• Start projects with data acquisition firmware and plan ahead for memory, power, and future use cases.
• Edge AI applications are expanding in healthcare, wearables, and consumer robotics.
• Business models are shifting, with AI driving recurring revenue and service-based offerings for hardware products.

Chapters:

00:00 Episode Teasers & Intro02:57 What Is Edge AI Anyway?06:42 Tiny Models, Tiny Devices, Big Impact10:15 The Hardware Leap: From M4 to M55 + U5515:21 Real-World Use Cases: From ECGs to Welding Bots17:47 Spec’ing Your Hardware for AI24:15 Firmware + Inference Frameworks: How It Actually Works26:07 Why Data Is the Hard Part34:21 Where Edge AI Will—and Won’t—Take Off First37:40 Hybrid Edge + Cloud Models40:38 Business Model Shifts: AI as a Service44:20 Live Q&A: Compatibility, Labeling, On-Device Training56:48 Final Advice: Think of AI as Part of the Product

Join the Interrupt Slack

⁠⁠⁠Watch this episode on YouTube⁠

Follow Memfault

• ⁠⁠LinkedIn⁠⁠
• ⁠⁠Bluesky⁠⁠
• ⁠⁠Twitter⁠⁠

Other ways to listen:

⁠⁠Apple Podcasts

iHeartRadio⁠⁠

⁠⁠Amazon Music

GoodPods

Castbox

⁠⁠

⁠⁠Visit our website