On the Fine Tuning of RRAM Resistance Under Variability Using Current Pulses at SET

Emerging devices are widely selected for the design of higher density memories and for the exploration of hardware implementations of low-power artificial neural networks. In particular, a Resistive Random Access Memory (RRAM) device is highly immune to cosmic radiation, to have high endurance and retention. Moreover, its internal structure is quite simple which makes it compatible with mature CMOS processes. However, the device-to-device variability and probabilistic switching are its Achilles’ heel. In this paper a circuit for fine-tuning of RRAM’s resistance is proposed based on the injection of current pulses at SET. A compact model which includes the variability effects is selected and calibrated with a real RRAM and is included in a 65 nm CMOS process to evaluate the proposal. Two tuning modes are obtained: discrete and continuous programming. In the discrete mode, predetermined levels of the resistance can be set up, storing up to 4 bits in a single RRAM in the most favorable case. In the continuous mode, the resistance can be gradually modified in finer steps, but the achieved resistance level is unknown and must be measured.