Periodic Reporting for period 1 - MIETMAN (Modeling of Ionic and Electronic Transport in 2D Materials Toward Memristive Applications in Neuromorphic Computing)
Okres sprawozdawczy: 2022-03-01 do 2024-02-29
The digital transformation of the last half a century has revolutionized human societies, enabling unthinkable information and communication services. However, the stringent computational requirements that these technologies demand, results in a (long-term) unaffordable energy consumption and environmental stress. This energy voracity is rooted in the von Neumann computational architecture, which physically separates the information storage and processing modules in present electronic systems.
To confront this challenging scenario, the last decades have witnessed a strong scientific push toward the exploration of neuromorphic computing architectures taking inspiration from the power-efficiency of the biological brain. The memristor, with added functionality provided by two-dimensional materials (2DMs), has shown the capability of achieving the innate high density of the biological networks, with efficient hardware realization of both neurons and synapses. Moreover, memristors-based neuromorphic systems are not restricted to solve the energy consumption of the existing technology, but will also enable much-advanced functionality through the realization of artificial intelligent (AI) systems.
This field, although promising, is in its infancy and needs strong theoretical support to guide the experimental work in order to push forward the state of the art. In this respect, MIETMAN sought the development of a multi-scale modelling and simulation framework for 2DM-based memristors, combined with the fabrication of working prototypes for their application in brain-inspired computation. The overall aim of the proposal was to demonstrate the feasibility of the 2DMs to implement novel neuromorphic applications able to lead the forthcoming revolution in the semiconductor industry. The project work was carried out at two institutions: i) the University of Granada (UGR), Granada Spain, where a comprehensive computational study of the main properties of these materials was realized; and ii) the Gesellschaft fur Angewandte Mikro- und Optoelektronik (AMO GmbH), Aachen, Germany, where the 2DMs-based memristive devices were fabricated and characterized.
At the end of the project we were able to achieve most of the originally proposed objectives. We studied the 2DM-based memristors from different abstraction levels generating and forwarding critical information to build a bottom-up understanding of the device. Along with the fabricated prototypes, we were able to show the feasibility of 2DMs in realizing important learning features of the biological synapse as well as emulating the neurons behaviour. The knowledge pool generated is currently being carried forward to advance the 2DM-based memristive systems toward a common goal of reducing energy consumption in computing.
To confront this challenging scenario, the last decades have witnessed a strong scientific push toward the exploration of neuromorphic computing architectures taking inspiration from the power-efficiency of the biological brain. The memristor, with added functionality provided by two-dimensional materials (2DMs), has shown the capability of achieving the innate high density of the biological networks, with efficient hardware realization of both neurons and synapses. Moreover, memristors-based neuromorphic systems are not restricted to solve the energy consumption of the existing technology, but will also enable much-advanced functionality through the realization of artificial intelligent (AI) systems.
This field, although promising, is in its infancy and needs strong theoretical support to guide the experimental work in order to push forward the state of the art. In this respect, MIETMAN sought the development of a multi-scale modelling and simulation framework for 2DM-based memristors, combined with the fabrication of working prototypes for their application in brain-inspired computation. The overall aim of the proposal was to demonstrate the feasibility of the 2DMs to implement novel neuromorphic applications able to lead the forthcoming revolution in the semiconductor industry. The project work was carried out at two institutions: i) the University of Granada (UGR), Granada Spain, where a comprehensive computational study of the main properties of these materials was realized; and ii) the Gesellschaft fur Angewandte Mikro- und Optoelektronik (AMO GmbH), Aachen, Germany, where the 2DMs-based memristive devices were fabricated and characterized.
At the end of the project we were able to achieve most of the originally proposed objectives. We studied the 2DM-based memristors from different abstraction levels generating and forwarding critical information to build a bottom-up understanding of the device. Along with the fabricated prototypes, we were able to show the feasibility of 2DMs in realizing important learning features of the biological synapse as well as emulating the neurons behaviour. The knowledge pool generated is currently being carried forward to advance the 2DM-based memristive systems toward a common goal of reducing energy consumption in computing.
We started by studying the 2DMs at an atomic level, focusing on imperfections within the material and the presence of metal atoms coming from the contact that are thought to cause memristance (broadly speaking the ability of the material to evidence memory effect encoded in its resistivity). Through careful analysis, we were able to understand the role of point defects and metal atoms in the experimental behaviour. We are currently investigating more complex defect structures, such as grain boundaries.
Then, we studied various memristive devices that use 2DMs and obtained parameters for further study. We consider both two-terminal and the three-terminal mem-transistors from a macroscopic device level. To achieve this goal we developed a numerical simulator capable of self-consistently solving the coupled equations that describe the physical phenomena ruling the behaviour of the memristors. By doing this, we were able to quantify different figures of merit of the devices and connect them to the physical parameters. As a result, we generated a guiding principle for their application-specific experimental realization.
Next, in close collaboration with partners in MIETMAN, we developed three different prototypes of memristors based on 2DMs. These prototypes included Ag/MoS2/Pd lateral volatile memristors, Ni/PtSe2/Pd vertical non-volatile memristors, and laser-induced graphene (LIG) flexible memristors (volatile and non-volatile). We performed comprehensive electrical characterizations to explain the underlying physical mechanisms responsible for the memristance and demonstrated their ability to perform synaptic functions such as short and long-term plasticity.
Finally, making use of the knowledge gained from performing the simulations and experiments, we developed compact models to describe the memristor experimental behaviour. We encoded the models in a Verilog-A code and used a commercial simulator to designs and analyse circuits comprised of memristors. The simulation showed that it is feasible to use the fabricated memristors to realize the Leak-Integrate and Fire (LIF) model of biological neurons. Furthermore, we implemented a memristive crossbar using another compact model, calibrated with the experimental data of the LIG flexible memristors to emulate the weights of an Artificial Neural Network (ANN). We then used the ANN to perform an image recognition task and calculated the energy consumption, thus completing our end-to-end study of the memristors.
The findings of MIETMAN and the knowledge gained from it were shared with non-scientific audiences, such as high school students and the general public, through events like European Researchers Night, Science Week and Engineering Fair. The scientific community was also informed about these results through conferences. Detailed articles explaining the numerous findings carried out in MIETMAN are in different stages of publication in peer-reviewed journals.
Then, we studied various memristive devices that use 2DMs and obtained parameters for further study. We consider both two-terminal and the three-terminal mem-transistors from a macroscopic device level. To achieve this goal we developed a numerical simulator capable of self-consistently solving the coupled equations that describe the physical phenomena ruling the behaviour of the memristors. By doing this, we were able to quantify different figures of merit of the devices and connect them to the physical parameters. As a result, we generated a guiding principle for their application-specific experimental realization.
Next, in close collaboration with partners in MIETMAN, we developed three different prototypes of memristors based on 2DMs. These prototypes included Ag/MoS2/Pd lateral volatile memristors, Ni/PtSe2/Pd vertical non-volatile memristors, and laser-induced graphene (LIG) flexible memristors (volatile and non-volatile). We performed comprehensive electrical characterizations to explain the underlying physical mechanisms responsible for the memristance and demonstrated their ability to perform synaptic functions such as short and long-term plasticity.
Finally, making use of the knowledge gained from performing the simulations and experiments, we developed compact models to describe the memristor experimental behaviour. We encoded the models in a Verilog-A code and used a commercial simulator to designs and analyse circuits comprised of memristors. The simulation showed that it is feasible to use the fabricated memristors to realize the Leak-Integrate and Fire (LIF) model of biological neurons. Furthermore, we implemented a memristive crossbar using another compact model, calibrated with the experimental data of the LIG flexible memristors to emulate the weights of an Artificial Neural Network (ANN). We then used the ANN to perform an image recognition task and calculated the energy consumption, thus completing our end-to-end study of the memristors.
The findings of MIETMAN and the knowledge gained from it were shared with non-scientific audiences, such as high school students and the general public, through events like European Researchers Night, Science Week and Engineering Fair. The scientific community was also informed about these results through conferences. Detailed articles explaining the numerous findings carried out in MIETMAN are in different stages of publication in peer-reviewed journals.
MIETMAN has led us to advance the current understanding of the 2DM-based memristors. We believe that the knowledge accumulated during the course of this project will greatly support the industry and the research institutes to address the challenging issue of the relentless growing energy consumption in computing. In close collaboration with our partner AMO GmbH, whose focus is on bridging the gap between academia and industry, the prototypes already fabricated have demonstrated some of these capabilities. We have also written proposals for the upcoming European Innovation Council Pathfinder Challenges 2024, which includes a specific challenge on "Nanoelectronics for Energy-Efficient Smart Edge Device", where the results from the MIETMAN project serve as a perfect precursor. Further, through the multidisciplinary nature of the project and the exchange of knowledge, we generated interest among graduate and undergraduate students, and some of them actively participated in MIETMAN performing shared tasks as part of their thesis projects.