Memristor-based synapse design and training scheme for neuromorphic computing architecture
A fabricated memristor behavior has been remodel in LT-SPICE and presented in this paper. Memristor SPICE models are important for designers to exhibit memristor behavior since memristor is not yet available in market. Among important parameters in memristor SPICE model is R<sub>ON</sub>, R<sub>OFF</sub> and TiO<sub>2</sub> thickness. This is because different R<sub>ON</sub> and R<sub>OFF</sub> value results in different switching behavior. In this paper, R<sub>ON</sub> and R<sub>OFF</sub> are obtained from fabricated memristor graph gradient. The values are applied as memristor model parameters. The behavior of this model is in agreement with the measurements of fabricated memristor. Next, NAND and NOR circuits are designed using the SPICE model and circuits are working based on the simulation results. The memristor SPICE model parameters are based on fabricated memristor model and it has similar behavior with fabricated memristor model. Thus it is convincing that fabricated memristor will work in real circuit.