The amount of raw data that the ever growing amount of sensors is producing is growing rapidly. On the other hand, usually, it is the information contained in the sensory data, that we are interested in, not the raw data. This research direction concentrates on extracting meaningful data from the sensory input right next to the sensor. Usually we are targeting real-time operation, or faster, where the speed requirements pose major challenges for the information extraction processor units in either speed, power consumption, or both. Also, ultra-low power sensor-processor architectures are studied with the emphasis on having the sensor-processor loop closed in a manner that the extracted information can be used in enhancing the sensing operation and achieving lower power consumption.
The main focus of the research has been on near-sensor processing in the area of vision chips for visible light spectrum. The developed architectures follow a locally connected pixel-parallel approach with the processing and data storing being optimized using analog, mixed-signal and digital building blocks and operating inasynchronous mode in critical places to gain speed and power advantage.
In vision chips, the main type of sensor has been the photodiode, but the future research is targeted to account also other types of sensors as well as other parts of the electromagnetic spectrum. The research is performed by designing integrated circuits, sending the designs to fabrication and after receiving the chips, by measuring their performance.