Abstract
A large number of sensing elements communicating with each other or with central control units is an integral part of the Internet of Things. The associated transmission and processing of an extreme amount of produced data is problematic. One of the solutions is processing of sensory data in close proximity to the sensor (near-sensor computing) or directly in the sensor (in-sensor computing), which radically reduces the requirements for their subsequent transmission and processing. In this project, we focus on the development of resistive gas and temperature sensors implemented into synaptic matrix of a hardware neural network, enabling low-level processing of measured sensory data directly in this network using a hardware algorithm. In the project, we will develop a methodology to calculate this algorithm and to encode it into the sensor matrix. This methodology will be one of the outputs of the project with the potential for wider application in hardware neural networks.