Butterfly: μW Level ULP Sensor Nodes with High Task Throughput

Area of Science:

• Computer Science
• Electrical Engineering
• Embedded Systems

Background:

• Internet of Things (IoT) applications require sensor nodes with low power consumption and high task efficiency.
• Existing systems force a trade-off between power efficiency and task execution in IoT sensor nodes.
• Developing lightweight IoT sensor nodes is crucial for widespread adoption and advanced applications.

Purpose of the Study:

• To introduce the Butterfly design paradigm for edge-to-end integration in IoT sensor nodes.
• To optimize both power consumption and task execution efficiency in IoT sensor nodes.
• To eliminate the need for power-consuming Microcontrollers (MCUs) in sensor nodes.

Main Methods:

• Offloading energy-intensive computational tasks from sensor nodes to a high-performance gateway.
• Implementing instruction and data buffering to minimize downlink data transmission.
• Utilizing a novel last-bit transmission and packaging mechanism for efficient uplink data transfer.
• Designing a task prediction mechanism on the gateway for concurrent task scheduling.

Main Results:

• Achieved a 4.91 times increase in task rate compared to benchmarks.
• Reduced power consumption of each sensor node by 94.3%.
• Demonstrated natural security advantages, such as anti-capture, due to offloaded control functions.

Conclusions:

• The Butterfly paradigm effectively addresses the trade-offs in IoT sensor node design.
• Edge-to-end integration offers significant improvements in efficiency and power savings for IoT applications.
• Butterfly enhances IoT sensor node performance, security, and scalability.