Sub-10 nm HfZrO ferroelectric synapse with multiple layers and different ratios for neuromorphic computing

Abstract

Listen

Translate article

To break the von Neumann bottleneck, emerging non-volatile memories have gained extensive attention in hardware implementing neuromorphic computing. The device scaling with low operating voltage is of great importance for delivering a high-integrating and energy-efficient neuromorphic system. In this paper, we fabricated sub-10 nm ferroelectric capacitors based on HfZrO (HZO) film with varying HfO and ZrO components. Compared to the conventional HZO capacitors (a constant component of 1:1), the varying component ferroelectric capacitors show similar remnant polarization but a lower coercive electric field (Ec). This enables the partial domain switching processed at a lower pulse amplitude and width, which is essential for emulating typical synaptic features. In the MNIST recognition task, the accuracy of sub-10 nm ferroelectric artificial synapse can approach ∼85.83%. Our findings may provide great potential for developing next-generation neuromorphic computing-based ultra-scaled ferroelectric artificial synapses.