IoT & Embedded Systems IP

IoT Patents Span Every Layer of the Stack

IoT patent claims exist at every layer of the technology stack — from the physical sensing element, through the wireless protocol, up through the cloud data processing pipeline. An IoT FTO analysis must consider not just the application software but the complete technology stack below it, each layer potentially carrying independent patent thickets from different industry players.

Application Layer — IoT Protocol IP

IoT communication protocols are themselves patent territories. MQTT (Message Queuing Telemetry Transport, ISO/IEC 20922) was developed at IBM in the late 1990s — IBM's original patents have largely expired, but implementation-specific patents covering QoS level 2 message delivery, session persistence across reconnections, and retained message handling remain active. CoAP (RFC 7252) is covered by IETF contributor patents from Ericsson, Huawei, and others. LwM2M (Lightweight M2M) for device management carries IP from OMA SpecWorks contributors.

The Matter protocol (formerly CHIP — Connected Home over IP) — the smart home interoperability standard backed by Apple, Google, Amazon, and Samsung — uses IP over Thread (802.15.4) or IP over WiFi, with IP contributions from all founding members. As Matter devices ship in hundreds of millions, its patent landscape becomes commercially significant.

Network Layer — Wireless Protocol IP

Wireless IoT protocol patents are dense with foundational IP. LoRaWAN uses Semtech's chirp spread spectrum (CSS) modulation — covered by Semtech's foundational LoRa patents which, while expiring in key jurisdictions, are being supplemented by implementation patents from LoRa Alliance members covering adaptive data rate algorithms, time-division channel access, and gateway diversity combining.

Zigbee (IEEE 802.15.4) and the successor Thread protocol (also 802.15.4 based, with IPv6 mesh networking) share physical layer implementations but differ in network layer architecture. The Thread Group IP framework aims to ensure open access, but individual member company patents covering mesh routing algorithms (DODAG root selection, RPL routing), security bootstrapping (PAKE-based commissioning), and device type implementations remain relevant for FTO analysis.

RTOS and Embedded Software IP

Real-Time Operating System software patents cover scheduling algorithms, interrupt handling architectures, memory management approaches, and deterministic execution guarantees. FreeRTOS (now Amazon FreeRTOS) is MIT licensed, but Amazon holds patents on extensions including OTA update management, AWS IoT Greengrass integration APIs, and fleet management SDKs. Microsoft's ThreadX (acquired from Express Logic) underpins many commercial embedded platforms with patent coverage on priority inheritance mutex implementations and event flag synchronisation primitives.

Zephyr RTOS (Linux Foundation) is Apache 2.0 licensed — but contributor patents from Intel, Nordic Semiconductor, NXP, and others on specific driver implementations, power management subsystems, and Bluetooth stack HCI implementations must be considered in FTO analysis for commercial products.

FOTA — Firmware Over-The-Air Update IP

Firmware OTA update mechanisms have become a major IoT patent area driven by security requirements and device lifecycle management. Key patent areas cover differential update algorithms (only transmitting the binary delta between old and new firmware), A/B (active/inactive) partition schemes enabling rollback on failed update, multi-image update coordination for systems with separate application processor and connectivity module firmware, cryptographic signature verification of update packages, and bandwidth-efficient streaming update protocols for constrained devices.

Edge AI and TinyML

The deployment of machine learning inference directly on microcontrollers — TinyML — has created a rapidly growing patent area. Key innovation clusters include: model compression techniques (post-training quantisation to INT8/INT4, structured and unstructured pruning, knowledge distillation from large teacher models), efficient neural architecture design (MobileNetV3, EfficientNet-Lite, MCUNet, TinyBERT optimised for Arm Cortex-M), hardware-aware neural architecture search (NAS optimising for specific inference hardware), and on-device federated learning frameworks enabling privacy-preserving model improvement without centralising raw sensor data.