Hybrid Write Bias Scheme for Non-Volatile Resistive Crossbar Arrays

Abstract

Crossbar arrays based on non-volatile resistive devices are planned for future memory systems due to the scalability and performance as compared to conventional charge based memory systems. To enhance the feasibility of these resistive memory systems, the energy consumption needs to be reduced. The write operation of a resistive memory based on a one-selector-one-resistor crossbar array consumes significant energy. The energy consumed by a crossbar array is dependent on the device and interconnect characteristics as well as the bias scheme. While the device and circuit parameters are the same for a specific application, the bias scheme of an array can be tuned to improve the energy efficiency. In this paper, an intelligent write scheme is proposed to provide a hybrid bias scheme. The proposed system adaptively sets the bias schemes to enhance energy efficiency. The most energy efficient bias scheme depends upon several parameters such as the size of the array, nonlinearity factor, and number of selected cells. For a specific array size and device characteristics, a power delivery system is described that sets the bias voltages based on the number of selected cells. Energy improvements of more than 2× are demonstrated with this hybrid bias scheme.