Dual-Core Edge AI Sticker Runs ML Models on ARM Cortex-M33 and RISC-V

Edge AI is moving from bulky developer kits to sticker‑thin form factors that can be placed anywhere. The Ghost Edge AI Sticker packs a dual-core ARM Cortex‑M33 and RISC-V processor, a 6‑axis IMU, and a microphone into a flexible FPC board—running machine learning models entirely on-device. This architecture not only demonstrates how to partition high‑frequency sensor acquisition from inference, but also highlights the growing feasibility of tinyML in real‑world environments where latency and privacy are critical.

The emergence of the Ghost Edge AI Sticker marks a notable leap in edge computing form factors, blending paper-thin flexibility with capable on-device machine learning. Built around Nordic Semiconductor's nRF54L15 wireless system-on-chip, this open-source project combines a dual-core processor (ARM Cortex-M33 plus RISC-V FLPR) with a 6-axis IMU (LSM6DSO) and a digital PDM microphone on a flexible printed circuit (FPC) that can conform to curved surfaces. This design directly challenges the traditional bulk of IoT sensor nodes, which often remain constrained by rigid PCBs and battery enclosures, opening up new possibilities for unobtrusive installations in wearables, structural health monitoring, and physical activity tracking.

“

Nordic Semiconductor's nRF54L15, with its ultra-low power radio and processing capabilities, is well-positioned as a platform for such sticker-like sensor nodes.

The sticker's core innovation lies in its ability to run real-time machine learning models entirely on-device, leveraging the dual-core architecture to separate high-frequency sensor acquisition (handled by the RISC-V core) from application processing and ML inference (handled by the Cortex-M33). This split not only optimizes power consumption but also demonstrates a practical template for efficient edge AI workloads—a crucial step as edge AI adoption is projected to grow 25% annually through 2028, driven by demands for lower latency and enhanced data privacy (McKinsey & Company, 2023). By processing data locally, the Ghost Edge AI Sticker sidesteps the bandwidth and latency penalties of cloud-dependent IoT architectures, aligning with industry findings that edge processing can cut data processing times by up to 70% compared to cloud-only systems (IEEE, 2024).

The project's open-source nature—complete with KiCad design files, Gerber outputs, and schematic PDFs—democratizes advanced edge AI hardware, enabling makers, startups, and even small enterprises to prototype and iterate quickly. The inclusion of a digital microphone alongside motion sensing broadens the potential use cases to acoustic event detection, voice activity monitoring, or multimodal sensor fusion, all within a form factor that can be stuck onto pipes, beams, or clothing. However, the Ghost Edge AI Sticker also highlights the persistent hurdles of edge AI adoption: while the sticker itself is cost-effective and flexible, scaling such designs for mass production demands robust supply chains, thorough regulatory approval, and optimization of battery life. The nRF54L15 SoC, though powerful, operates within constraints typical of microcontrollers—limited memory and compute—meaning that the on-device models must be carefully pruned and quantized, a task that still requires significant ML expertise. Additionally, as the broader article from Archynewsy notes, the cost of upgrading existing infrastructure to support on-device processing and the scarcity of edge AI talent remain barriers for many organizations.