Batteryless Sensing

Batteryless sensors collect small amounts of sporadic energy from the surrounding environment and use it to perform complex tasks. At BatterylessLab we advocate modularity and reusable design. In harvesting-based systems, this means decoupling the environment from the application. More precisely, the transducer’s voltage and current are made independent from those in the application circuit. By doing so, it becomes possible to optimize the energy harvesting input and application circuit’s output independently and simultaneously. These design techniques have proven to be powerful enough to process data streams from sensors and broadcast results wirelessly. Thanks to advanced energy management and distribution mechanisms, this device needs less than 10 micro-Watts of input power to operate. What started our as academic research has been applied to commercial, open-source products like the miroCard.

Embedded AI

Embedded Artificial Intelligence is an emerging discipline that covers different approaches to AI on highly memory- and power-constrained devices. One large area of work focuses on machine learning and aims to bridge the gap between complex inference models and their execution on embedded systems. Multiple research projects have implemented multiple machine learning models to process data from (high-bandwidth) sensors and transmit results with low-bandwidth transceivers. At BatterylessLab, we work with different types of sensors, from accelerometers to gas sensors and cameras, developing practical applications like the people-counting one shown here. We also work with novel energy-efficient architectures combining low-power, parallel, and heterogeneous computing.

2_pre_nms

Wireless Networks

selective_flooding

Long-Range communication and low power were, until several years ago, two mutually exclusive topics. In recent years, multiple long-range wireless protocols have been investigated, including LoRa, NB-IoT, and Sigfox. The research community has primarily focused on technology using the unlicensed frequency spectrum thanks, in part, to the ease of operation and availability. Although there are known limitations to the scalability of Aloha-based communication these systems use, they have accelerated the growth of IoT in multiple business markets. We are currently investigating new methods to bring energy-efficient synchronous communication into batteryless sensing networks. This has the potential to bring many new application scenarios such as federated learning and autonomous agents into the batteryless sensing domain.