Work

Optimization of RRAM Read Performance and Area Efficiency: A Large-Scale and Low-Parasitic Array with Novel Interconnection Schemes