Recent Progress in Developing Next Generation Neuromorphic Devices Based on Ferroelectric Hafnia by the U.S. Army Research Laboratory

In this presentation, I will present on the research progress by the U.S. Army Research Laboratory (ARL) in creating a framework for maturing emerging non-volatile memory devices based on ferroelectric hafnia for integration into front- and back-end CMOS neuromorphic integrated circuits. Next generation edge applications require a dramatic shift in computing paradigms, which utilize new electronic devices and circuit architectures to address the increasing power demands of enabling cognitive sensing and autonomy in extreme environments. Ferroelectric hafnia exhibits promising attributes, such as scalability, sub-nanosecond switching, back-end-of-the-line (BEOL) and front-end-of-the-line (FEOL) CMOS-compatibility, and good endurance that make it a promising candidate to address these needs.<br/><br/>In this presentation, I will discuss our material and device efforts in developing BEOL 2D ferroelectric field effect transistors. Our research demonstrates that within the CMOS compatible thermal processing envelope the ferroelectric landscape of HZO shows unique characteristic that have significant implications on the neuromorphic device properties of FeFETs. We examine these properties by integrating HZO into BEOL 2D WSe2 based-FeFET device architectures, a prototypical van der Waals system, and verify their robust synaptic plasticity within a 3.5 order of magnitude conductive range. These discoveries highlight a roadmap for material processing, dimensional scaling, and integration of HZO-based FeFETs.<br/><br/>I will further discuss our complementing activities that focus on creating the design and simulation tools required for integration of FeFETs into CMOS. In this effort we seek to develop a framework for multiscale modeling of ferroelectric-based devices. To that end we will focus on our efforts in developing a physics-based compact device model for three-terminal ferroelectric field effect transistors. The model has been designed with explicit support for transient analysis to capture time-dependent polarization switching dynamics down to nanosecond scale with multiple types of ferroelectric domains. In the presentation, we will discuss the model’s calibration against experimental data and use in optimizing the design of a ferroelectric-metal field effect transistor (FeMFET). Finally, we conclude with future direction of the work, including how we intend to couple this model to a Monte Carlo analysis framework to feed into neural network-level hardware simulations of neural accelerator crossbar architectures.