TinyML is the field of deploying machine learning models on ultra-low-power microcontrollers (MCUs) with kilobytes of memory — enabling AI inference on devices that cost under $1, run on coin-cell batteries for years, and are embedded in sensors, wearables, and industrial equipment.

TinyML Constraints

• Memory: 256KB-1MB flash, 64-256KB RAM — models must be extremely small.
• Compute: ARM Cortex-M class processors — no GPU, limited integer/fixed-point arithmetic.
• Power: Microwatt to milliwatt power budgets — must run on batteries for years.
• Frameworks: TensorFlow Lite Micro, microTVM, CMSIS-NN for optimized inference.

Why It Matters

• Ubiquitous AI: TinyML enables AI everywhere — in every sensor, actuator, and embedded device.
• Semiconductor Sensors: Embed ML directly in process sensors for real-time, on-device anomaly detection.
• Always-On: Ultra-low power enables always-on sensing and inference without cloud connectivity.

TinyML is AI on the smallest computers — deploying machine learning on microcontrollers for ubiquitous, always-on, battery-powered intelligence.