EAGER: Ferroelectric Memristive Devices Emulating Synapses in Subcortical Information Processors

点击翻译按钮获取中文摘要

Neuromorphic computing is an interdisciplinary field that aspires to create physical architecture and design principles based on biological nervous systems for applications in vision systems, auditory systems and autonomous robots. There is an increasing interest in developing electronic analog circuits to mimic neuro-biological architectures present in the nervous system. In the nervous system of the brain, a synapse is a structure that permits a neuron to pass an electrical or chemical signal to another cell. This exploratory research proposes to create a two-terminal memristive device that can emulate the function of a synapse. The resistance of the device will change depending on the amount, direction, and duration of voltage applied. The device has advantage of maintaining its state until another voltage pulse is applied over conventional computer memory, which requires regular charge to maintain its state. The principle behind the proposed approach is to make dipoles in a film switch up or down depending on voltage polarity in ferroelectric materials. If the thickness of the ferroelectric layer is made small enough, it can allow tunneling of electrons that is a function of the relative density of dipoles in up or down position thus preserving a memory similar to that of the synapse, thereby making electronic analog circuits to mimic brain. The proposed novel synapse circuits will enable integration of complex systems with power-constrained devices. The research on hafnium oxide (HfO2) based FTJ will also open the door for further scaling of FE capacitor based random access memory and enable the fabrication of FE field effect transistor (FE-FET) based memory. The graduate students would have a significant opportunity in advancing their interdisciplinary skills in semiconductors device design and fabrication, integrated circuit design, machine learning, and neuroscience. This proposal explores a two-terminal memristive device based on newly discovered ferroelectricity in CMOS compatible high permittivity dielectric, HfO2 doped with silicon or aluminum. The switching mechanism in FTJ is driven by polarization that is relatively immune from stochastic variations observed in other resistive memory devices. Fabrication of a high permittivity HfO2 based ferroelectric device will allow thinner films which can be scaled. The goal of the proposed research is to explore the design and fabrication of HfO2 based ferroelectric tunnel junction (FTJ) memristive device and its characterization to generate models. In addition, neuron circuits with multiple signaling types for behavioral emulation of biological neurons will be designed, synaptic circuits based on the proposed memristor device models will be trained, and the feasibility of incorporating the neuron and synapse circuits into subcortex-inspired information processing (SIIP) system will be evaluated. The outcome of the proposed exploratory research will result in a novel device that can mimic synaptic behavior and can be integrated with conventional CMOS electronics thereby forming the basis for the next generation of intelligent computing.