The EUPHORIC subproject proposes the development of a bioinspired sensing and processing subsytem. During the project, a novel electronically foveated dynamic vision sensor with an on-chip integrated smart digital controller of the foveation regions will be developed. In its normal default mode, the sensor produces a low-resolution output that is processed by a visual attention system which detects the regions of interest in the visual scene that are configured as high resolution in the sensor in real-time. Each of the high-resolution regions of interest will be processed by a small-size recognition system. Thus, the size requirements and power consumption of the recognition system are reduced.
In order to realize the visual feature extraction in the visual recognition system, the project proposes the interconnection of the sensor with a neuromorphic CMOS-memristive processor with on-line learning capability. The neuromorphic processor proposed in the project will be an advanced multicore processor equipped with reconfiguration and programmability capabilities of the different cores which allows the parallelization of the different regions of interest processing. Additionally, in the proposed prototype we will introduce circuit design and algorithmic techniques for analog learning of the memristive synaptic weights. The sensing and processing neuromorphic modules of the EUPHORIC project will be interconnected to audio sensing modules, stochastic processors, and robotic platforms of the other subprojects to develop demonstrator systems of sensing-processing-actuation in high-speed low-power applications.
Advertising and communication:
Project PID2023-149071NB-C51 funded by MICIU/AEI/10.13039/501100011033 and FEDER, UE