TinyML for Edge Analytics in IoT Systems: A Systematic Survey of Hardware, Software and Optimization Techniques

Abstract

The proliferation of Internet of Things (IoT) devices has generated unprecedented volumes of data at the network edge, overwhelming traditional cloud-centric processing models. Tiny Machine Learning (TinyML) has emerged as a transformative paradigm enabling machine learning inference directly on resource-constrained microcontrollers with memory footprints under 1 MB and power consumption below 100 mW. This survey provides a comprehensive review of TinyML-enabled edge analytics for IoT systems, synthesizing the state of the art across hardware platforms, software frameworks, optimization techniques, application domains, and system integration. We systematically review 56 key papers from 2018 to 2026, identifying key trends, design patterns, and open challenges. Our contributions include: (1) a taxonomy of TinyML optimization techniques including quantization, pruning, knowledge distillation, and hardware-aware neural architecture search; (2) a comparative analysis of microcontroller platforms (ARM Cortex-M, ESP32, RP2040) and software frameworks (TensorFlow Lite Micro, Edge Impulse, MicroTVM); (3) a structured review of edge analytics techniques spanning inference strategies, sensor fusion, anomaly detection, time-series analysis, vision, audio, and adaptive learning; (4) an analysis of eight application domains including smart health, industrial IoT, agriculture, and smart buildings; and (5) a forward-looking discussion of open challenges including long-term reliability, distribution shift, energy harvesting, and TinyMLOps. We conclude by identifying critical research gaps and proposing a roadmap for future work toward sustainable, autonomous edge intelligence. This survey serves as a foundational reference for researchers, practitioners, and system designers developing TinyML-IoT systems.