System-level integration in neuromorphic co-processors

Abstract

In this chapter we present results on system-level integration of memristive devices with neuromorphic circuits and systems. Specifically we present an overview of the current state-of-the-art hybrid memristive-complementary metal–oxide–semiconductor (CMOS) mixed-signal neuromorphic circuits for learning and plasticity and present perspectives toward integration of memristive devices in neuromorphic spiking neural network architectures. We focus on neuromorphic circuits and architectures that allow for a relatively natural integration of memristive devices, irrespective of the specific characteristics of the specific memristive device technology adopted. We address the cointegration of memristive devices with on-chip learning mechanisms, using both analog and digital CMOS circuits, to build a solid background of the functionality of neuromorphic circuits explaining how memristive devices can be implemented on them. Furthermore we also address the system-level integration of such architectures in multicore and multichip systems, for connecting them to input and output devices, such as sensors, actuators, and conventional CMOS processing devices.