Advances in neuromorphic devices for the hardware implementation of neuromorphic computing systems for future artificial intelligence applications: A critical review

Abstract

Neuromorphic Computing (NC) is considered as the next generation of artificial intelligence (AI). AI can transform the way people live and work, however, the current Neumann computing systems limits the potential of AI applications due to their large energy consumption and limited efficiency in information processing. Therefore, the hardware realization of neuromorphic computing is gaining tremendous interest as one of the most attractive technologies for overcoming the bottleneck of conventional Von-Neumann based computing systems. Neuromorphic devices which are capable of mimicking the functionality of biological synapses and neurons such as memristors and neuromorphic transistors are the fundamental elements of neuromorphic computing systems. Learning ability and learning accuracy are the key requirements of synaptic devices which can be measured in terms of parameters such as STP (Short Term Plasticity), STDP (Spike Time Dependent Plasiticity), LTP (Long Term Plasticity), LTD (Long-Term-Depression), PPF (Paired Pulse Facilitation), and EPSC (Excitatory Post Synaptic Current). In the modern Big Data era, indicated by Internet of things (IoT) & artificial intelligence, the biggest challenge is to process large amount of information at high speed & low power. In this scenario, non-linear & parallel data processing based neuromorphic computing (NC) has emerged as a research topic of huge interest. Therefore, this article, critically reviews the recent advances in materials, synaptic devices such as memristors and neuromorphic transistors for future neuromorphic computing.