Input-Independent Array Compensation in Memristor-Arrays-Based Neuromorphic Systems for Input Resistance

Neuromorphic systems using memristors as artificial synapses have attracted widespread interest for low-power consumption and low-latency computing applications. However, networks based on the paradox of limited sources and progressively larger array sizes are affected by non-ideal effects, such as input resistance problems due to matrix switches. During vector array multiplication, the input resistance leads to great distortion of the out current and significant degradation of the network performance. Here, an input-independent array compensation (IAC) method for the input resistance effect is proposed, which only require a few known parameters (memristor column average conductance, array size, input resistance) independent of the input information to effectively solves current distortion problems. The simulation and experiment results demonstrate that the current distortion problem can be successfully mitigated and the accuracy of the network computing can be effectively restored by using the proposed IAC method. These results provide a feasible approach for the precise implementation of the neuromorphic system based on memristor arrays.