Author ORCID Identifier

https://orcid.org/0000-0002-3986-942X

Defense Date

2024

Document Type

Dissertation

Degree Name

Doctor of Philosophy

Department

Mechanical and Nuclear Engineering

First Advisor

Jayasimha Atulasimha

Abstract

The data generated annually has reached unprecedented levels, necessitating the integration of energy-efficient computing systems to achieve high performance and sustain our society’s computing needs. Specifically, energy-efficient computing, once considered advantageous, has now become essential in managing this exponential data growth. For example, the energy used solely by Google was equivalent to the energy consumed by a country with population >20 million in 2019 and this energy need is increasing exponentially [1]. Therefore, high-performance computing systems with energy-efficient, faster operability, and scalable computing units are necessary to address present and future computing challenges.

Apart from traditionally used conventional von Neumann computing systems, several unconventional computing systems have been proposed to achieve high-performance computing. The currently used complementary metal-oxide-semiconductor (CMOS) memory units in both conventional and unconventional computing systems are volatile in nature and therefore require standby power to retain an information. Additionally, they suffer from excessive heat generation when downscaled. The leading non-volatile alternative to CMOS-based volatile memory is nanomagnetic memory, known as magneto-resistive random-access memory (MRAM), where no standby power is required after writing information. The fundamental building block of MRAM is a magnetic tunnel junction (MTJ), where the resistance changes depending on the magnetization orientation and can be used for storing '0' and '1' bits of binary memory system. Since MRAM offers the added advantage of non-volatility and scalability, while potentially retaining the benefits of CMOS-based memory units, it can become the preferred choice for high-performance computing. Unfortunately, commercial MRAM based on spin-transfer torque (STT-MRAM) requires ~100 fJ/bit, which is 1000 times more energy to write a bit than to switch a CMOS device. This factor has prevented its large-scale commercial adoption. Therefore, various manipulation methods, such as mechanical strain, spin-orbit torque (SOT), and electric field, are being studied to control the magnetization of the MTJs for energy-efficient memory solutions.

One energy-efficient method of controlling magnetization in a nanomagnet for implementing memory and neuromorphic computing, is the use of the voltage-control of interfacial magnetism. To bring about energy-efficient memory and neuromorphic computing by leveraging the advantages of voltage control of interfacial magnetism, we have:

i) Studied the feasibility of scaling of dynamic skyrmion-mediated switching in perpendicular magnetic tunnel junctions (p-MTJs) to ~20 nm lateral dimension in the presence of room-temperature thermal noise and defects [2].

ii) Theoretically shown that three states can be written in an MTJ using the voltage-control method in an energy-efficient way, as opposed to the two-state memory systems traditionally used in computing systems, to achieve high-density memory units.

iii) Theoretically and experimentally demonstrated that the voltage-control method can be used for energy-efficient neuromorphic computing, such as reservoir computing. We showed that a skyrmion-based physical reservoir can be utilized for classifying and predicting temporal data in an energy-efficient manner. We also showed that a magneto-ionic heterostructure can potentially be used to achieve energy-efficient reservoir computing.

iv) Experimentally studied the magneto-ionic control of skyrmion (a voltage-control method) for energy-efficient spintronic devices.

Future research can focus on experimentally demonstrating skyrmion-mediated memory, both binary and ternary, in an MTJ and skyrmion-based reservoir computing using the voltage-controlled method. Additionally, efforts can be directed toward implementing faster and scalable magneto-ionic reservoir computing. These advancements could enable the realization of energy-efficient, scalable, high-performance computing devices.

Reference:

[1] Bryce, R., 2020. How Google Powers Its ‘Monopoly’ With Enough Electricity For Entire Countries. https://www.forbes.com/sites/robertbryce/2020/10/21/googles-dominance-is-fueled-by-zambia-size-amounts-of-electricity/?sh=255a251968c9

[2] Rajib, M.M., 2022. Electric field control of confined magnetic skyrmions for energy efficient scalable nanomagnetic memory.

Rights

© The Author

Is Part Of

VCU University Archives

Is Part Of

VCU Theses and Dissertations

Date of Submission

8-6-2024

Share

COinS