2T1M Neuromorphic Synapse with Pt-Hf-Ti Memristor Model

Abstract

AbstractNeuromorphic computing systems are the future of the technological revolution due to their containment of multiple processing layers, low energy consumption, high performance, adaptability, faster data processing and ability to work in parallel. Neuromorphic computing mimics the way information is processed inside the human brain compared to Von Neumann architecture. Von Neumann architecture has several disadvantages like high power consumption, slow speed, low performance, and a lower density of information to be processed. Because of above arguments, neuromorphic computing has become the new architecture for processing. Neural computing is the mainstay of any type of Artificial Neural Networks (ANNs) as it processes information in the same way that biological neural systems such as the brain. This is what makes it unique from any other system. ANNs consists of two main parts, neurons and synapses, just like biological neural networks in the human brain. In the past decade, ANNs depended on transistors in their manufacture, but after the discovery of the memristor and its advantages such as nanoscale, low energy consumption, low emitted heat, non-volatile memory, high performance and high storage space, it became a good alternative to the transistor and this will increase effectiveness of Von Neumann structure. In this paper, an electrical synapse circuit consisting of two transistors of MOSFET type and a Pt-Hf-Ti memristor (2T1M synapse) is used. This work is done on Cadence virtuoso simulation environment using Verilog-A VTEAM model for memristor modeling. This paper shows practical results of using Pt-Hf-Ti memristor in 2T1M neuromorphic synapse, as the Verilog-A VTEAM model is based on experimental data rather than other literature results based on simple simulation model. The simulation results reveal that, symmetric read signals used before in literature can’t be applied to the Pt-Hf-Ti memristor as the device is asymmetric by its nature. However, a small pulse duration read signal can be applied without causing destruction to the ANN weight stored in the memristor.KeywordsNeuromorphic computingPt-Hf-Ti memristorNeuromorphic SynapseHardware acceleration