Variation-aware Design Space Exploration of Mott Memristor-based Neuristors

Mott memristor (MM)-based neuristors are promising candidates for artificial neuron implementations due to their scalability, energy efficiency, and CMOS-compatibility. A neuristor exhibits threshold-driven spiking under a voltage input and diverse spiking patterns under constant current. The design principle of the neuristor relies on a close match between two MMs, which is challenging to ensure in practical designs. In this work, we perform a simulation-driven comprehensive analysis to identify the possible effects of parametric mismatch between these MMs, and the overall impact of device/circuit-level variations. We perform sensitivity analysis with individual device parameters to understand their unique roles. We also perform 10,000-point Monte-Carlo simulations considering variations in the device parameters of the MMs (switching thresholds and resistance levels), along with other circuit components. We observe that the current-biased neuristor can withstand ~15% ( ~5%) mismatch in metallic (insulating) state resistance of the MMs. While these tolerance limits are specific to a given set of nominal parameters, their relative values clearly illustrate that the mismatch in the insulating state resistance renders more critical influence. The Monte-Carlo simulations demonstrate that a neuristor, biased with a super-threshold voltage, can plunge into sub-threshold mode due to variations beyond its tolerance limits. Our work provides a pathway towards designing a neuristor-based neuromorphic system considering the impacts of mismatch and variations.