Unraveling the Dynamics of HfO2-Based NW-CTT as an Artificial Synapse

In this work, we investigate the synaptic characteristics of HfO2-based nanowire-charge trap transistor (NW-CTT) using a fully calibrated 3-D technology computer-aided design (TCAD) simulation, based on self-consistent solutions of 3-D Poisson’s equation, Boltzmann transport equation, and self-heating equation with interface traps charge correction terms. The results indicate that the Si NW-CTT exhibits excellent nonvolatile memory characteristics with the modulation of threshold voltage around 1 V between program and erase pulses when interface trap charge density ( ${{N}_{\text {it}}}) = {1} \times 10^{17} {\mathrm {cm}}^{-3} $ present in the HfO2 layer. In addition, almost linear conductance modulation in long-term potentiation (LTP) and long-term depression (LTD) with NW-CTT synapse promises the recognition accuracy of around 94.7% and write energy efficiency of around 2.3 mJ in the $784\times 100\times 10$ neural network for handwritten MNIST dataset. Our analysis highlights the critical role of optimal nanowire diameter, thickness of the charge trapping HfO2 layer, gate length, and metal gate work function in enhancing short- and long-term memory characteristics, while concurrently preserving recognition accuracy and energy efficiency. Overall, our results provide valuable insights into the synaptic behavior of conventional NW-CTTs and offer guidance for further harnessing their weight-update capabilities in neuromorphic computing applications.