We aim to understand the role of spike-based learning in brain-inspired systems considering various constraints that are not usually taken into account in simulations, such as the effect of variability in the neural network parameters or the impact of bounded weights in the learning/training phase. Our mid-term goal is to develop an innovative cognitive brain-inspired system that can effectively run in parallel with energy efficiency on sequence processing tasks (i.e., a stream of events from sensors), produce intelligent behavior, interact, and adapt to the environment.
Currently, we are researching robust algorithms, architectures, and digital implementations targeting specific applications of spiking neural networks in adaptive control of prosthetic robotic arms and sensory processing applications. For a proof of concept, we prototyped a reliable three-dimensional digital neuromorphic system geared explicitly toward the 3D-ICs biological brain's three-dimensional structure (R-NASH), where information in the network is represented by sparse patterns of spike timing and learning is based on the local spike-timing-dependent plasticity rule. R-NASH enables real-time and low-power solutions targeted at full-custom manycore system-on-chip integration.