Product Name: ECS-DoT

Manufacturer: EMASS, a Nanoveu subsidiary

Product Category: Computing Hardware, Software and Systems

Supporting Documentation (if available)

ECS-DoT is a purpose-built edge AI system-on-chip that delivers continuous always-on intelligence at sub-milliwatt power levels, up to 20 times lower energy than comparable edge AI solutions. It is designed for battery-constrained devices including wearables, IoT sensors, and autonomous systems, integrating a RISC-V MCU, dual deep learning accelerators, multi-sensor interfaces, and 4MB of on-chip memory in an event-driven architecture. This architecture enables sub-10 millisecond inference latency with no cloud dependence. By processing audio, motion, vision, and environmental data entirely on-device, ECS-DoT supports always-on functions such as wake word detection, anomaly detection, and activity recognition without compromising battery life or real-time responsiveness.

Battery-powered edge AI systems must balance continuous sensing with strict energy constraints. Conventional edge processors often rely on duty-cycled operation, external memory, or cloud connectivity to manage workloads. These approaches introduce latency, increase power consumption, create security and privacy risks, and reduce device operating life, making continuous intelligence difficult to sustain in wearables, IoT sensors, and autonomous systems. ECS-DoT addresses these limitations by delivering sub-milliwatt always-on AI inference with latency below 10 milliseconds. Its event-driven system-on-chip architecture allows the device to remain in an ultra-low-power state until meaningful sensor activity is detected, eliminating unnecessary computation, simplifying system-level power management, and reducing overall power draw.

At the core of ECS-DoT is an event-driven processing model that activates compute only when sensor activity occurs. Traditional embedded systems rely on constant polling and clock-driven execution, consuming power even when no relevant data is present. ECS-DoT eliminates this inefficiency by remaining in ultra-low-power states until triggered by sensor input, enabling continuous multimodal sensing while maintaining sub-milliwatt operation and deterministic latency under 10 milliseconds. The platform also integrates 4MB of on-chip memory, allowing AI models, intermediate data, and sensor inputs to remain local to the device. This reduces data movement and eliminates reliance on external DRAM, improving both system efficiency and responsiveness.

A key technical feature is its proprietary on-chip AI weight decompression engine. Neural network weights are compressed offline to as little as two bits per parameter, with typical averages near 1.3 bits per weight. At runtime, these weights are decompressed back to 8-bit precision before execution on the deep learning accelerators. This enables larger models to be stored in compact form while maintaining full inference accuracy during execution. The decompressor operates alongside flash memory to support efficient model streaming without the energy penalties typically associated with external memory access. This compression-based approach reduces memory footprint, lowers access energy, and supports efficient inference within constrained power budgets.

The system is designed for straightforward integration into embedded platforms. It supports standard embedded toolchains, RTOS and bare-metal environments, and existing AI model optimization workflows, enabling both hardware and software teams to deploy and iterate on models without redesigning system architecture. By integrating compute, memory, and sensing into a single device, ECS-DoT reduces bill of materials complexity, minimizes board footprint, and accelerates development timelines through available reference designs and validated use cases.

In real-world applications, ECS-DoT enables continuous edge intelligence across multiple domains. In autonomous drones, the device processes onboard sensor data locally for navigation, environmental awareness, and anomaly detection. In hardware-in-the-loop validation trials, this contributed to up to 80% improvement in flight endurance without changes to batteries or propulsion systems. In industrial predictive maintenance, ECS-DoT processes vibration and acoustic data directly on-device to detect early signs of equipment failure, transmitting only relevant anomaly events to reduce bandwidth usage and extend battery life. In wearable devices such as smart glasses and hearables, it supports continuous voice recognition, gesture detection, and contextual awareness within strict power constraints. It has also been used for bone-conduction voice detection by analyzing inertial measurement data for keyword spotting, enabling private and energy-efficient voice interaction.

Additional applications include acoustic event detection for security monitoring and vision-based smart meter retrofits, where ECS-DoT processes data locally and transmits only relevant updates or alerts. Across these use cases, the system enables continuous sensing and real-time response while maintaining long operating life in battery-powered environments.

ECS-DoT stands out by delivering high compute efficiency and low energy per inference within a compact integrated architecture. It achieves approximately 12 TOPS per watt while operating in an approximately 0.1 to 5 milliwatt power range. Energy per inference is approximately 0.8 to 5.5 microjoules, with inference latency around 5 milliseconds. Compared to alternative solutions that require external memory or operate at higher power levels, ECS-DoT reduces system-level energy consumption and latency by keeping models and data on-chip and activating compute only when required.

The return on investment for users is realized through reduced system cost, improved operational efficiency, and enhanced product capability. Eliminating external memory and reducing reliance on cloud infrastructure lowers bill of materials and system complexity. Ultra-low-power operation extends battery life and reduces maintenance requirements in large-scale deployments. At the same time, ECS-DoT enables always-on AI capabilities that are difficult to achieve within comparable power budgets, allowing developers to build more responsive and energy-efficient products with reduced firmware complexity.

Through its event-driven architecture, integrated memory design, and compression-enabled AI execution, ECS-DoT enables continuous on-device intelligence in environments where power, latency, and reliability are critical constraints. It provides design engineers and system integrators with a practical and scalable solution for deploying always-on edge AI in battery-constrained systems.