Investigation of Non-Linear Selection Effect on RRAM based Neuromorphic Computing Array with Passive Selective Element

This paper presents a comprehensive investigation of exploiting passive RRAM crossbar array with nonlinear selective elements for neuromorphic computing. Design considerations, including weight-mapping method, voltage-applying strategy, and optimization scheme, are discussed. By using these methods, the dysfunction of the non-linear selection characteristic during inference is removed. Moreover, we evaluate the IR-drop effect on the passive crossbar array with non-linear selection characteristic. Finally, this work provides insightful analysis for designing a neuromorphic computing system based on a passive crossbar with non-linear selection characteristics.