xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim
leader nam 22 Ka 4500
controlfield tag 007 cr-bnu---uuuuu
008 s2010 flu s 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0004786
Field-coupled nano-magnetic logic systems
h [electronic resource] /
by Javier Pulecio.
[Tampa, Fla] :
b University of South Florida,
Title from PDF of title page.
Document formatted into pages; contains X pages.
Dissertation (PHD)--University of South Florida, 2010.
Includes bibliographical references.
Text (Electronic thesis) in PDF format.
Mode of access: World Wide Web.
System requirements: World Wide Web browser and PDF reader.
ABSTRACT: The following dissertation addresses the study of nano-magnetic devices configured to produce logic machines through magnetostatic coupling interactions. The ability for single domain magnets to reliably couple through magnetostatic interactions is essential to the proper functionality of Magnetic Cellular Automata (MCA) devices (p. 36). It was significant to explore how fabrication defects affected the coupling reliability of MCA architectures. Both ferromagnetic and anti-ferromagnetic coupling architectures were found to be robust to common fabrication defects. Experiments also verified the functionality of the previously reported MCA majority gate  and a novel implementation of a ferromagnetic MCA majority gate is reported. From these results, the study of clocking Magnetic Cellular Automata (MCA) interconnect architectures was investigated (p. 54). The wire architectures were saturated under distinct directions of an external magnetic field. The experimental results suggested ferromagnetic coupled wires were able to mitigate magnetic frustrations better than anti-ferromagnetic coupled wires. Simulations were also implemented supporting the experimental results. Ferromagnetic wires were found to operate more reliably and will likely be the primary interconnects for MCA. The first design and implementation of a coplanar cross wire system for MCA was constructed which consisted of orthogonal ferromagnetic coupled wires (p. 68). Simulations were implemented of a simple crossing wire junction to analyze micro-magnetic dynamics, data propagation, and associated energy states. Furthermore, two systems were physically realized; the first system consisted of two coplanar crossing wires and the second was a more complex system consisting of over 120 nano-magnetic cells. By demonstrating the combination of all the possible logic states of the first system and the low ground state achieved by the second system, the data suggested coplanar cross wire systems would indeed be a viable architecture in MCA technology. Finally, ongoing research of an unconventional method for image processing using nano-magnetic field-based computation is presented (p. 79). In magnetic field-based computing (MFC), nano-disks were mapped to low level segments of an image, and the magnetostatic coupling of magnetic dipole moments was directly related to the saliency of a low level segment for grouping. A proof of concept model for two MFC systems was implemented. Details such as the importance of fabricating circular nano-magnetic cells to mitigate shape anisotropy, experimental coupling analysis via Magnetic Force Microscopy, and current results from a complex MFC system is outlined.
Advisor: Sanjukta Bhanja, Ph.D.
Magnetic Cellular Automata
Quantum Cellular Automata
Magnetic Force Microscopy
x Electrical Engineering
t USF Electronic Theses and Dissertations.
Field Coupled Nano Magnetic Logic Systems by Javier F Pulecio A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy Department of Electrical Engineering College of Engineering University of South Florida Major Professor: Sanjukta Bhanja, Ph D Shekar Bhansali, Ph D Sudeep Sarkar, Ph D Rudy Schlaf, Ph D Aaron Stein, Ph D Date of Approval: September 30 2010 Keywords: Magnetic Cellular Automata, Q uantum Cellular Automata, Nano F a brication, Magnetic Force Microscopy, Micromagnetics Copyright 2010, Javier F. Pulecio
Dedication T o my late grandparents Elfa and Tiberio Lemos.
Acknowledgements Thi s work was supported in part by, NSF Florida Georgia LSAMP Bri dge to the Doctorate Aw ard HRD #0217675, NSF S STEM DUE #0807023, NSF C areer CCF #0639624, NSF EMT # 0829838 the Alfred P. Sloan fellowship and the FEF McKnight fellowship.
i Table of Contents List of Tables ................................ ................................ ................................ .......... iii List of Figures ................................ ................................ ................................ ......... iv Abstract ................................ ................................ ................................ ................ viii Introduction ................................ ................................ ............................ 1 Chapter 1 Motivation ................................ ................................ .............................. 1 1.1 Significance ................................ ................................ ............................ 4 1.2 Contributions ................................ ................................ .......................... 6 1.3 Outline of the Dissertation ................................ ................................ ....... 8 1.4 Nano Magnetic Systems ................................ ................................ .......... 10 Chapter 2 Introduction ................................ ................................ .......................... 10 2.1 Magnetic Cellular Automata ................................ ................................ .... 10 2.2 2.2.1 Single Domain Nano Dots ................................ ....................... 12 2.2. 2 Field Coupled Nano Pillars ................................ ....................... 14 2.2.3 Biaxial Magnetocrystalline Anisotropy ................................ ...... 17 Magnetic Nano Rings ................................ ................................ ............. 19 2.3 Magnetic Domain Wall Logic ................................ ................................ ... 20 2.4 Fabrication of Nano Structures via Electron Beam Lithography .................... 23 Chapter 3 Introduction ................................ ................................ .......................... 23 3.1 Overview of Electron Beam Lithography Process ................................ ....... 23 3.2 Electron Beam Resists ................................ ................................ ............ 24 3.3 Preparation and Application of PMMA Resist ................................ ............. 25 3.4 Scanning Electron Beam Exposure ................................ ........................... 28 3.5 Resist Development ................................ ................................ ............... 31 3.6 Film Deposition ................................ ................................ ...................... 32 3.7 Liftoff ................................ ................................ ................................ .... 35 3.8 Introduction to Single Domain Nano Magnets for Cellular Automata ............ 36 Chapter 4 Introduction ................................ ................................ .......................... 36 4.1 Magnetism and Mag netic Materials Overture ................................ ............ 36 4.2 Magnetic Domains ................................ ................................ .................. 40 4.3 Fabrication Process ................................ ................................ ................ 43 4.4 Magnetic Coupling of Single Domain Nano M agnets ................................ .. 45 4.5 Magnetic Cellular Automata Interconnects ................................ ................ 54 Chapter 5 Intr oduction ................................ ................................ .......................... 54 5.1 Wire Architectures ................................ ................................ .................. 55 5.2
ii Fabrication Process ................................ ................................ ................ 57 5.3 Experimental Setup ................................ ................................ ................ 58 5. 4 Results ................................ ................................ ................................ .. 60 5.5 Conclusion ................................ ................................ ............................. 66 5.6 Magnetic Cellular Automata Coplanar Cross Wire Systems .......................... 68 Chapter 6 Introduction ................................ ................................ .......................... 68 6.1 Design ................................ ................................ ................................ .. 70 6.2 Fabrication Process ................................ ................................ ................ 71 6.3 Experimental Setup ................................ ................................ ................ 72 6.4 Results ................................ ................................ ................................ .. 74 6.5 Conclusion ................................ ................................ ............................. 77 6.6 Magnetic Field Based Computing for Image Processing .............................. 79 Chapter 7 Introduction ................................ ................................ .......................... 79 7.1 Proof of Concept ................................ ................................ .................... 82 7.2 Fabrication Process ................................ ................................ ................ 83 7.3 Results ................................ ................................ ................................ .. 85 7.4 Conclusion ................................ ................................ ............................. 94 7.5 Conclusion ................................ ................................ ............................. 98 Chapter 8 Synopsis ................................ ................................ ............................... 98 8.1 Future Endeavors ................................ ................................ ................. 101 8.2 List of References ................................ ................................ ................................ 103
iii List of Tables Table 3 1 outlines and describes the process steps used for the RCA cleaning procedure. ................................ ................................ ............................. 26 Table 3 2 outlines the resist spinner reci pe used for 495K and 950K PMMA ranging from 30 120 nm of thickness. ................................ ...................... 27 Table 3 3 outlines the development process used for a single layer of 950K PMMA and for s bilayer 495K 950K PMMA (bottom top) resist. ................... 32 Table 5 1 provides various energies as sociated with a two nano magnet ferromagnetic and anti ferromagnetic system.. ................................ ......... 63
iv List of Figures Figure 2 1 (A) is a diagram of transistors in series which depicts various external energies and interconnecting wires.. ................................ ........................ 11 Figure 2 2 et al. circular MCA chain. ................................ 12 Figure 2 3 shows the operation of a chain of circular nano dots propagating information. ................................ ................................ ........................... 13 Figure 2 4 (A) shows the fabrication dimensions of a permalloy nano pillar. ................ 15 Figure 2 5 (A) depicts a m agnetic QCA wire. ................................ ............................. 16 Figure 2 6 (A) is a BCC crystal of iron which exhibits magnetic anisotropy along different crystal planes. ................................ ................................ .......... 18 Figure 2 7 (A) and (B) show t he two stable configurations of the onion state in a magnetic nano ring. ................................ ................................ ............... 19 Figure 2 8 (A) shows two different nano wires with a contiguous magnetic domain in opposite directions. ................................ ................................ 20 Figure 2 9 (A) shows the basic domain wall logic structure of an AND gate. ............... 22 Figure 3 1 shows a chemical reaction induced by the exposu re of the positive resist polymethylmethacrylate (PMMA).. ................................ ................... 25 Figure 3 2 is an abstraction of a scanning electron microscope column. ..................... 28 Figure 3 3 represents two different electron beam energies entering an electron sensitive resist and substrate. ................................ ..................... 29 Figure 3 4 (A) depicts the sequential development process (top t o bottom) for a positive resist such as PMMA. ................................ ................................ .. 31 Figure 3 5 is a schematic of an electron beam evaporator. ................................ ........ 33 Figure 3 6 is cros s sectional view of the liftoff process. ................................ ............. 35 Figure 4 1 (A) depicts the anti ferromagnetic ordering of a Chromium BCC crystal. ................................ ................................ ................................ .. 38 Figure 4 2 (A) is a two dimensional slice of a ferrimagnetic crystal lattice. .................. 39
v Figure 4 3 is half of a magnetic hysteresis curve. ................................ ...................... 41 Figure 4 4 (A) is an AFM height amplitude plot (left) and a MFM phase plot (right).. ................................ ................................ ................................ 43 Figure 4 5 is four different sets of data of the same MCA interconnect. ....................... 45 Figure 4 6 is another four sets of data of a different MCA wire. ................................ .. 46 Figure 4 7 is another four different sets of data of an ideal MCA wire. ........................ 47 Figure 4 8 is a n AFM amplitude height plot of a ferromagnetic chain which shows the highly irregular shape and surface roughness. ........................... 49 Figure 4 9 is the AFM height amplitude plot of five ferromagnetic chains (left). ......... 49 Figure 4 10 (A) is an AFM height amplitude plot of a MCA majority gate. .................... 5 1 Figure 4 11 (A) is an AFM height amplitude plot of a ferromagnetic MCA majority gate.. ................................ ................................ ....................... 52 Figure 4 12 is a MFM phase plot of two ferromagnetic MCA m ajority gates feeding into a third majority gate. ................................ ............................ 53 Figure 5 1 depicts correct data propagation for ferromagnetic and anti ferromagnetic wires, which are the lowest ground state energy. ................ 55 Figure 5 2 depicts incorrect data propagation for ferromagnetic and anti ferromagnetic wires. ................................ ................................ ............... 56 Figure 5 3 (A) is an SEM image of the ferromagnetic wire and (B) is of the anti ferromagnetic wire. ................................ ................................ ................ 58 Figure 5 4 (Spin magnetic field) (A) is a MFM image of the ferromagnetic wire and (B) is the anti ferromagnetic wire ................................ ..................... 59 Figure 5 5 (Z axis magnetic field) (A) is the ferromagnetic wire and (B) is anti ferromagnetic wire. ................................ ................................ ................ 59 Figure 5 6 (Easy axis magnetic field) (A) is the ferromagnetic wire and (B) is anti ferromagnet ic wire. ................................ ................................ .......... 59 Figure 5 7 (Hard axis magnetic field) (A) is the ferromagnetic wire and (B) is anti ferromagnetic wire. ................................ ................................ .......... 59 Figure 5 8 is an MFM image w h ere a magnetic field was provide to ferro magnetic wires along the hard axis. ................................ ......................... 60 Figure 5 9 is a graph which summarizes the perce ntage of frustrations per wire based on the direction of the magnetic field. ................................ ............ 60
vi Figure 5 10 shows the possible states for the ferromagnetic and anti ferromagnetic wires and over the duration of the experiments. .................. 65 Figure 5 ....................... 66 Figure 5 ....................... 66 Figure 6 1 is a standard abstract representation of MCA an d EQCA cells. .................... 69 Figure 6 2 (A) is an abstra ct representation of two coplanar crossing wires. ................ 69 Figure 6 3 is an OOMMF simulation of our MCA cross wire system design. .................. 70 Figure 6 4 is a SEM image of the two coplanar cross wire system. .............................. 72 Figure 6 5 is an AFM image of the two coplanar cross wire system. ............................ 73 Figure 6 6 is a MFM image of all possible combinations for a simple cross wire system. ................................ ................................ ................................ 74 Figure 6 7 is a SEM image of the complex cross wire system. ................................ .... 75 Figure 6 8 is a MFM image of the complex cross wire system.. ................................ 77 Figure 7 1 (A) is a grayscale image of a group of buildings taken from above. ............. 81 Figure 7 2 (A) is a layout of numbered magnetic nano disk which correspond to the ed ges of an image. ................................ ................................ ........... 81 Figure 7 3 (A) is a CAD design of nano disk in microns.. ................................ ............ 84 Figure 7 4 (A) shows the phase graph and magnetic image of a single domain magnet taken via a MFM. ................................ ................................ ........ 85 Figure 7 5 ( A) is an AFM height amplitude image and corresponding graph. ............... 87 F igure 7 6 is an AFM height amplitude image and corresponding graph of the second simple MFC system. ................................ ................................ ..... 88 Figure 7 7 (A) is a MFM scan of the second MFC system at a 0 degree scan angle. ................................ ................................ ................................ ... 89 Figure 7 8 is a AFM amplitude image of magnet 4 of the second MFC system (right). ................................ ................................ ................................ .. 90 Figure 7 9 is a AFM amplitude image of magnet 5 of t he second MFC system (right). ................................ ................................ ................................ .. 91
vii Figure 7 10 (A) is a AFM height amplitude force plot of MFC system 3. ....................... 96 Figure 7 11 (A) is a AF M height amplitude force plot of MFC system 4. ....................... 97
viii Abstract T he following dissertation addresses the study of nano magnetic devices configured to produce logic machines through magnetostatic coupling interactions. The abil ity for single domain magnets to reliably couple through magnetostatic interactions is essential to the proper functionality of Magnetic Cellular Automata (MCA) devices (p. 36 ). I t was significant to explore how fabrication de fects affect ed the coupling reliability of MCA architectures. Both f erromagnetic and anti ferromagnetic coupling architectures were found to be robust to common fabrication defects. Experiments also verified the functionality of the previously reported MCA majority gate [ 1 ] and a novel implementation of a ferromagnetic MCA majority gate is reported From these results, the study of clocking Magnetic Cellular Automata (MCA) i nterconnect architectures was investigated (p. 54 ) The wire architectures were saturated under distinct directions of an external magnetic field. The experimental results suggested ferromagnetic coupled wires were able to mitigate magnetic frustrations better than anti f erromagnetic coupled wires. Simulations were also implemented supporting the experimental results. F erromagnetic wires were found to operate more reliably and will likely be the primary interconnects for MCA. The first design and implementation of a coplan ar cross wire system for MCA was constructed which consisted of orthogonal ferromagnetic coupled wires (p. 68 ). Simulations were implemented of a simple crossing wire junction to analyze micro magnetic dynamics, data propagati on, and associated energy states Furthermore, two systems were physically realized; the first system consisted of two coplanar crossing
ix wires and the second was a more complex system consisting of over 120 nano magnetic cells. By demonstrating the combina tion of all the possible logic states of the first system and the low ground state achieved by the second system, the data suggest ed coplanar cross wire systems would indeed be a viable architecture in MCA technology. Finally, ongoing research of an unconv entional method for image processing using nano magn etic field based computation is presented (p. 79 ). In magnetic field based computing (MFC), nano disks were mapped to low level segments of an image, and the magnetostatic co upling of magnetic dipole moments was directly related to the saliency of a low level segment for grouping. A proof of concept model for two MFC systems was implemented Details such as t he importance of fabricating circular nano magnetic cells to mitigate shape anisotropy, experimental coupling analysis via Magnetic F orce Microscopy, and current results from a complex MFC system is outlined
1 Chapter 1 Introduction Motivation 1.1 As the t stream, at times it would seem the neoteric phenom enology of nano science has been whittled into a chic prefix which is often misused. The physical constraints of nano dimensions create fundamentally unique behaviors of materials and particles thought to be thoroughly understood. For example, a material such as gold at macroscopic scales is chemically inert but when reduced to nano particles can be used as an energetic catalyst, or carbon, when found in a nanotube structure exhibits ballistic electron transportation properties mimicking conductivity only found in superconducting materials. These are just two natural examples of nano science that substantiate s unique opportunities for the technological advancement of humanity and should not be lost in translation. The study and understanding of nano phenome na presents a defining opportunity to fundamentally The United States of America formally acknowledged the importance of Nanotechnology in 2000 with the creation of The National Nanotechnology Initiative (NNI), whi dimensions between approximately 1 and 100 nanometers, where unique phenomena
2 interested in nano pheno mena as it is directly related to th e profit model of the industry. In 2003, the CEO of Intel Craig Barnett stated, "The importance of Moore's Law, and its promise of smaller, more powerful products built at lower cost, remains the focus of continuing inv estments by those in our industry who are determined to remain competitive. For Intel, the path is clear. The fundamental value of silicon innovation and integration is certain for future product generations ahead, as it continues to create customer benefi ts and serve as the foundation for outstanding new products." since the prediction in 1965 [ 2 ] but it is nearing the end of its life due to quantum effects at critical scaling dimensions. This presents dilemmas for current CMOS technology, yet it also actualizes new opportunities to explore alternative novel technologies. Dr. Paola Gargini, the director of strategy at Intel, believes around 2015 hybrid chips will emerge with other technologies such as grafted nanotubes and novel materials [ 3 ] Nano electro mechanical field effect transistors (NEMFET) and Carbon Nanot ube Tunnel field effect transistors (CNTFET) are excellent examples of nanotechnology creating interesting alternatives to resolve issues of mainstream devices, such as CMOS. In the following paragraphs NEMFETs and CNTFETs are briefly discussed to emphasiz e how nanotechnology can provide creative solutions to issues with current technology. As the dimensions of CMOS transistors continue to scale parasitic leakage currents have become more problematic and are non ideal for low power applications. A non trad itional approach to resolve sub threshold leakage in CMOS transistors is the Nano Electro Mechanical Field Effect Transistor (NEMFET) [ 4 6 ] A NEMFET uses a nano
3 elec tro mechanical transducer to convert electrical energy into mechanical displacement G =0), the pu ll s in the gate electrode into direct contact with the dielectric. This causes the channel to fully deplete, cuts current flow from the source to the drain, and becomes the off state of the NEMFET. To turn the device on the gate voltage is increased, whic h reduces the depth of the depletion region and allows for current to flow from the source to drain. Once a threshold is reached, the gate voltage is sufficient to overcome the electrostatic forces and the gate breaks contact with the dielectric above the channel. This device demonstrates a sub threshold swing smaller than 60mV/decade at room temperature, which is the fundamental limit of MOSFETs, and is ideal for low power applications. It is interesting to observe the likeness of the first 1947 transisto r, which was a mechanically suspended point contact switch, and how its contemporary state of the art counterpart NEMFET returns to those roots. Similarly, Carbon Nanotube Tunnel Field Effect Transistors (Tunnel CNFET) provide solutions to issues such a s thermally limited sub threshold swing, source drain leakage current, and higher device density when compared to traditional CMOS devices [ 7 9 ] Carbon nanot ubes have been heavily researched due to their unique properties, such as high strength, ballistic transportation, and the ability to mimic semiconductors. When designing a FET the semiconducting single wall carbon nanotube (CN) is particularly interesting By altering the doping profile throughout the length of the nanotube such that a P I N (p doping, intrinsic doping, n doping) junction is formed, a transistor can be created using the carbon nanotube as the channel. The CN is connected to a source, drain and a gate similar to MOSFET device. By changing the voltage of the gate, the
4 field effect is able to ben d the conduction / valence bands and reduce the band gap. Carriers are then able to use the quantum tunneling effect to tunnel from the con duction to valence band or vice versa. The Schottky barriers between the contacts at the source/drain and the carbon nanotube are typically minimal and the energy barrier need ed to be overcome is of the band gap in the carbon nanotube. Both NEMFETs and CNTF ETs provide alternatives to overcome physical design issues with field effect transistors, which are enabled by nanotechnology, and were specifically discussed to demonstrate solution s offered by nano science. The scope and breath of nanotechnology is quit e considerable and affects many aspects of humanity, from enhancing the structural properties of cement mortar through the use of nano particles for construction [ 10 ] to the use of magnetic nano particles for drug targeting and cancer therapy [ 11 ] Using the unique properties of nanotechnology, the focus of this work is to demonstrate a logic paradigm that utilizes a ferromagnetic nano structure as a logical switch in a fundamentally different magnet ic computing machine. Significance 1.2 In the past, there has always existed an alternative technology to continue the advancement of computational technology. From punch cards to vacuum tube s to bipolar transistors to current complementary metal oxide semico nductor field effect transistors ; scientists and engineers have made great advancements in the computational power and physical minimization of logic devices. At the initial stage of many successor technologies, history has shown them to be inferior to the ir respective predecessor and often impractical to implement ; but as the devices c ontinued to be developed the successor technology eventually provided greater benefits to industry and ultimately the consumer. Currently, conventional CMOS technology is pr edicted to
5 reach its full potential by 2015 after which there is no clear front runner to continue the rapid improvement of technology. The impact of CMOS technology is cle arly evident by the public consumption of electronic devices from the late 1980s t o the present day. The research presented in this dissertation explores a new computing paradigm where the single domain magnetic dipole moment of a ferromagnetic nano structure is used as a logical switch The nano magnets are arranged to couple with one another to create larger complex networks of a magnetic computing machine and has been labeled by the industry as an emergent technology Every year an international consortium of experts from the technology sector, called the International Technology Road map for Semiconductor ( ITRS ), gather to provide a roadmap for the future with the sole purpose of identifying sig nificant technologies to advance the industry Companies such as Intel Corporation, Hewlett Packard, Texas Instru ments, and IBM Corporation, re ly on this roadmap as a guide to help determine what significant technologies are materializing. In 2009 the ITRS Emerging Research Devices identified Magnetic Logic as an emergin g technology and stated, Nano magnetic logic has potential advantages rela tive to CMOS of being non volatile, dense, low power, and radiation hard. Such magnetic elements are compatible with MRAM technology, which can provide input output interfaces. Compatibility with MRAM also promises a natural integration of memory and logic Nano magnetic logic also appears to be scalable to the ultimate limit of using [ 12 ] Detailed within is the study of that fundamental shift from the displacement of the electron as a basic sw itch, as found in transistor technology, to using the intrinsic spin of the electron to create a switch for logic. By fabricating ferromagnetic structures at nano
6 dimens ions a unique phenomenon occurs, the formation of a single magnetic domain (as opposed to multiple domain structures found at larger scales). This distinct magnetic state, which occurs only at the nano scale, presents unique opportunities to create a new type of computation machine based on the interaction of magnetic dipole moments Magneti c logic exhibit s extraordinary innovation possibilities, such as the potential to scale t he size of a magnetic switch down to an atom, and could conceivably be the next significant advancement in computational logic. Contribution s 1.3 The use of nano phe nomena at the device level is relatively modern, and the use of nano magnetism as a logical switch in computation is even earlier in its development. Therefore, the intention of this study was to simulate, implement, and test fundamental aspects of differe nt magnetic logic systems. In particular, the focus was on the interaction between neighboring magnetic cells and how several factors, such as surface roughness or coupl ing architectures, affected nano magnetic systems. Within this study, not only were pre viously reported pioneering results verified [ 13 14 ] but the design, simulation, fabr ication, and testing of no vel nano magnetic logic systems un der various conditions were explored Specifically the following contributions are detailed : The effect of size, surface, and shape irregularities of rectangular nano magnet s and the associated magnetostatic coupling reli ability in Magnetic Cellular Automata (MCA) was analyzed (p. 36 ) T he first implementation of a ferromagnetic majority gate as well as two ferromagnetic majority gates feeding into a third ferromagnetic majority gate was pres ented This was the large st known logic circuit demonstrated in MCA. The investigations demonstrate d a highly predictable interaction of ferromagnetic and anti
7 ferromagnetic coupling and a robustness to nominal fabrication defects associated with bottom up fabrication techniques. T he first direct comparison of magnetic frustrations for ferromagnetic and anti ferromagnetic MCA wires under external magnetic clocking fields in the in plane hard axis, out of plane hard axis, in plane easy axis, and spinning mag netizing field was experientially investigated (p. 54 ) Simulations were also implemented to attain the magnetic energy associated with magnetic frustrations and correctly coupled ground states. The results show ed that ferroma gnetic wires were able to mitigate magnetic frustrations better than anti ferromagnetic wires and suggest out of plane magnetic fields as a possible clocking solution. T he design, implementation, and testing of a novel ferromagnetic coplanar crossing wire systems was also presented (p. 68 ) The simulation of two ferromagnetic wires intersecting at a 90 degree angle has been discussed as well as the experimental implementation of two coplanar crosswire systems was reported. The novel crosswire system was fou nd to be functional and offered a unique architecture for Magnetic Cellular Automata. The analysis of a unique Magnetic Field Based Computing (MFC) system consisting of nan o magnetic disks to implement perceptual grouping of segmented edges through magnetic energy minimization for image processing was explored (p. 79 ) A proof of concept was explained and fabricated. The results show ed that the circular disk can indeed be used to establish a sali ency between edges of an image by exhibiting coupling along a particular magnetization direction. This can reduce a computational
8 bottleneck found in the perceptual organization process of grouping segments in computer vision. The se contributions helped to bolster the foundation of magnetic computational devices and provided perspective to the feasibility o f realizing magnetic logic Potential benefits of using magnetism for comput ation range from the prospect of ultra low power devices for mobile applicati ons to radiation tolerant circuitry for space exploration and other extreme environments. Outline of the Dissertation 1.4 The outline of this dissertation is as follows : Chapter 2 N ano M agnetic Systems review ed current state of the a rt nano magnetic systems. Chapter 3 Fabrication of Nano Structures via Electron Beam Lithography provided an overview and details of the Electron Beam Lithography fabrication process used in the physical realization of the magneti c nano structures. Chapter 4 Introduction to Single Domain Nano Magnets fo r Cellular Automata investigate d rectangular single domain magnetic structures and discusse d how physical parameters of fabrication such as defects, affect the coupling interaction between neighboring cells. MCA majority gates structures are also discussed. Chapter 5 Magnetic Cellular Automata Interconnects discusse d ferromagnetic and anti ferromagnetic wire architectures for MCA u nder various external
9 magnetic clocking fields Det ails on the ability of the wire architectures ability to mitigate frustrations were discussed Chapter 6 Magnetic Cellular Automata Coplanar Cross Wire Systems detail ed the desig n and implementation of a novel cross wire system that could intersect on the same device plane without undesired cross talk Chapter 7 M agnetic Field Based Computing for Image Processi ng explored the proof of concept and physical implementation of mapping the quadratic energy minimization of coupled dipole moments to the maximization of the quadratic edge affinity in the grouping proc ess of perceptual organization.
10 Chapter 2 N ano M agnetic Systems Introduction 2.1 Nano magnetic systems have r ecently shown great promise in the advancement of logic and memory. Novel emerging magnetic systems have been reviewed in the following sections. In particular Magnetic Cellular Automata was discussed in great detail as it demonstrated very promising resul ts such as scalability, room temperature operation, and functioning logic components. Magnetic Cellular Automata 2.2 In Cellular A utomata (CA) a basic cell has distinct enumerated states which are determined by its current state and the state of its neighbo rs. These cells are positioned such example of a [ 15 16 ] Quantum Cellular Automata, or QCA, is a flavor of CA in which the intrinsic and extrinsic properties of the cell(s) are based on the quantum effects of an elect ron(s). An interesting fact of CA is that it is not limited to a specific technology ; meani ng that there are several phy sical implementations of a CA cell which embody the electron differently and make use of distinct quantum effects. Currently, there are three prominent variants of CA; Magnetic, Electronic, and Molecular [ 17 42 ] The beauty of CA is that each cell is a self contained element, needing no external energy to maintain a state. The desired state will always be one of the possible energy minimum configurations of the cell.
11 There is no need for cons tant power to operate a CA system; external energy is only provided during switching of states, m aking CA an inherently low power system. Another excellent characteristic of CA is that it is a lead less system, meaning there is no need for physical in terco nnects between all of the CA cells. Cells interact with neighbors via coupling forces which depend on the particular CA schema. This significantly reduces the already stringent requirements to fabricate any real nano scale device. Figure 2 1 depicts a simple comparison of transistor and QCA technologies. Thus far, of the various CA schema, only Magnetic QCA has been able to demonstrate successful operation at room temperature [ 43 ] [ 44 ] In Magnetic Cellular Automat a (MCA), a basic cell is a nano magnet; these magnetic cells are arranged in various grid like fashions to accomplis h computing. Cells in M CA Figure 2 1 (A) is a diagram of transistors in series which depicts various external energies and interconnecting wires T ransistors require a co nstant energy supply t o operate which is not necessary for QCA. (B) is an abstract representation of a QCA wire. external source forced the leftmost cell to change its state to a which forced the neighboring center cell to a logical eventually change to a logic enced from the neighboring cell.
12 are enumerated based on their single domain magnetic dipole moments and are inherently energy minimums. The single domain phenomenon only occurs i n nano scale magnets and if engineered properly can reduce ma gnetic fields to alter the mag netic moment of a cell and provides desired switching characteristics. Two distinct and significant advancements in M CA ha ve been made and are summarized in following subsections. 2.2.1 S ingle Domain Nano Dots In this scheme a computing network is created by placing an elongated elliptical input dot followed by circular dots to form a wire as shown in Figure 2 2 [ 43 45 52 ] The elliptical input dot requires a greater magnetic field to change states than the circular dots due to the shape anisotropy. Once the state of the input dot has settled, it is propagated down the circular M CA wire. The input dot Oe (returning to 0 Oe) or 300 Oe (returning to 0 Oe), res pectively. To ensure that the magnetic pulse used to set the input dot did not set the circular dots states as well, the same magnetic pulse of 300 Oe was sent along the wire without the presence of the input dot The Figure 2 2 is a depiction of et al. circular M CA chain. The leftmost magnetic d can be assumed to be a ots have coupled Assuming information is being driven from left to right t flip to a logical to minimize local magnetic energies.
13 dipole moments of the circular n ano dots were not uniformly aligned to the external field, which ensured the propagation of information was due to the ma gnetostatic coupling forces and not the external field applied to set the input dot. A weak oscillating magnetic field of 25 Oe at a 30Hz frequency, combined with a 10 Oe bias magnetic field was also applied to the entire network of dots to propagate th e state of the input dot down the wire of circular dots. To propagate or reverse the state of the network a soliton was created [ 43 ] Once created, usually near the end of the QCA wire, the soliton can move back and forth along the chain depending on the magnetic oscillating field. Figure 2 3 depicts how the propaga tion of states could occur. Notice at T2 when the soliton (dark blue bar) reaches the input dot it will stay between the input dot and the first circular nano dot. Once the magnetic oscillating field begins to rise the soliton will begin to move forward along the chain, as depicted at T3. Had the input dot been set to 0 (magnetic dipole moment pointing to the left) and the soliton moving backward, Figure 2 3 shows the operation of a chain of circular nano dots propagating information. The graph depicts an oscillating external magnetic field with a 10 Oe bias which is used to drive information along the wire At time T1 the majority of the circular wire has followed suit with the larger elliptical input dot. At T2 the negative phase of oscillation combined with the bias has sufficient energy to reverse the state of the circular wire by propagating a soliton At time T3 the soliton ha s begun to move forwa rd along the chain during the positive phase of oscillation of the applied field
14 as depicted at time T2, the soliton would be lost and another would need to be created. In the network described above the circular dots were fabricated to a diameter of 110 nm, at a thickness of 10 nm, on a pitch of 135nm as shown in Fi gure 2 2. The size and shape ensured that the ferromagnetic circular dots made of a Supermalloy ( Ni 80 Fe 14 Mo 5 X 1 where X was other metals) had single domain magnetic moments. This is important considering that the enumeration of states is based on the orie ntation of magnetic dipole moments and switching dynamics. It was found that the circular dots with a diameter of 100 nm and thickness of 10 nm exhibited ideal switching characteristic [ 53 ] The minute amount of energy required to saturate the dot s starting at about 5 Oe, and the high residual magnetization, about 80%, ar e prime characteristics produced by these single domain nano dots [ 53 ] T he magnetostatic energy (PE in the magnetic field) between two dots in this network was about 200 kb T, ( where kb is the nd T is the temperature) and should be at lea st 40 kbT to keep thermal errors below one per year [ 53 ] 2.2.2 Field Coupled Nano Pillars In a nano pillar network a basic cell consists of a shape engineered single domain magnet as shown in Figure 2 4 (A) [ 1 13 22 25 29 30 44 54 60 ] Due to the shape anisotropy, the pill ars develop easy and hard axis of magnetization A magnetic dipole moment experiences lower demagnetization energy when aligned along the easy axis of magnetization and hence is in an energetic ally favorable state. Conversely when aligned along the hard axis, the magnetic dipole moment is unstable and will eventually align itself along the easy axis to reduce the d ema gnetization energy. This allows for the enumeration of states along the easy axis of magnetization and creates a digital mode of operation for Magnetic Cellular Automata.
15 A M CA wire is construct ed by placing nano p illars side by side as shown in Figure 2 5 (A). Proper data propagation occurs when all the magnetic dipole moments of each cell in the M CA wire are in an anti parallel arrangement. A s initial state in Figure 2 5 (A) magnetic dipole moment s pointing along the hard axis of the magnets. Csaba et al. suggest ed [ 55 ] where the energy barrier required for switching is reduced by first applying a magnetic field along the hard axis This c auses the n ano magnetic dipole moment to align itself along the hard axis which is the initial state befo re an input is given This state is energetically unfavorable, making it possible to supply a weaker external magnetic input field to the system. Once an input is given magnetostatic coupling energies between the magnets causes data to propagate down the wire. Several shapes shown in Figure 2 4 (B ) were also studied under an adiabatic and rotating demagnetization fields to investigate the anti ferromagnetic coupling order [ 23 ] It was determi ned that symmetric shapes demonstrate longer data propagation lengths with an adiabatic demagnetization field, while the asymmetric shapes performed better under the rotating demagnetization field. Figure 2 4 (A) shows the fabric ation dimensions of a p ermalloy nano pillar Due to the shape anisotropy an easy and hard axis of magnetization were created. (B) are various shapes that were studied under different demagnetization fields to determine coupling order length s
16 Figure 2 5 (B) is a MCA majority g ate, which was the first logic g ate fabricated in this type of CA [ 1 ] The shape of the rect angular nano pillar was chosen because of the bi stable natur e. The shape anisotropy of the n ano pillars introduced a high energy barrier between the two stable states thereby making them les s error prone. The high er energy barrier also led to more power dissipation but was mitigated by an adiabatic switching/pum ping mentioned previously. The n ano pillars dimensions were 35nm thick had a width o f 70nm, height of 135nm, and were made of p ermalloy The n ano pillars were fabricated via an el ectron beam litho graphy process The MCA majority gate was shown to have performed all possible logic combina tions of a three input gate and operated at room temperature. This was significant since a majority gate can implement the essential logic gate functions necessary to implement any Boolean function. Research studies have also been conducted to investigat e the low power operation and defect characterization of MCA devices. MCA devices will need a clock implementation for synchronous operation and many have proposed using an external magnetic field as a clocking mechanism [ 44 61 62 ] The clock energy along with the Figure 2 5 (A) depicts a m agnetic QCA wire. An external inpu t is given to the wire which ha d sufficient strength to coerce the rest of the wire to change its state accordingly. (B) is a Magnetic QCA Majority Gate. The gate has three inputs (external energies) ; t he output will depend on the dipole moments of the inputs. In this example the topmost and leftmost inputs have the positive end of the dipole moment pointing downward a nd the bottom input has its mo ment pointing upward. The majority of the inputs are poin ting downward so the center pillar is the result of the majority.
17 energy necessary to provide an input is the total energy of an MCA system. The energy loss was calculated for a 32 bit MCA ripple carry adder and was found to be 0.004 0.285 pJ at a switching frequency of 10 8 which is in the worst case scenario on par with larger low power CMOS devices [ 61 ] MCA devices have also been found to be robust to common fabrication defects and have shown long ra nge ordering [ 23 63 ] This particular implementation of Magnetic Cellular Automata, which uses rectangular na no magnets as the basic cell, was of primary interest in this dissertation due to their bi stable nature and room temperature operation [ 44 ] The formation and reliability of a single domain magnetic moment of a ferromagnetic nano structure, and the magnet ic coupling between neighbors are discussed in detail Chapter 4 Chapter 5 explores how ferromagnetic and anti ferromagnetic wire architectures perform under different clocking fields. A novel ferromagnetic cross wire syste m, investigated in Chapter 6 is also presented which utilizes rectangular ferromagnetic nanostructure as a basic cell for MCA. The research presented within this dissertation, coupled with the pioneering work previously reported, continues to move MCA to the forefront of magnetic lo gic devices. 2.2.3 Biaxial M agnetocrystalline Anisotropy MCA will most likely need a clock ing structure to align magnetic cells into a hard axis of magnetization Alignment into the hard axis allows for previous states to be erased and lower s the work need ed t o overcome the energy barrier to flip nano magnet. Once a nano magnet is aligned in a hard axis of magnetization for example via an external magnetic field, the external energy is removed to allow the magnetic cells to settle into a lower energy state along the easy axis of magnetization During this relaxation time the nano magn ets are very eas il y influenced by the external forces of a
18 neighboring cell which is the very essence of cellular automata. If an input is given to one nan o magnet it will drive its neighbor which in turn drives its neighbor and so on. Recently Carlton et al. has suggested introducing a magnetocrystalline biaxial anisotropy energy into magnetic simulations to increase the relaxation time of a nano magnet [ 64 ] Many ferromagnetic materials exhibit magnetic anisotropy such th at magnetic properties differ along crystal planes as shown in Figure 2 6 If a magnetic cell is engine ered such that the clock is provide along the hard axis, the magnetocrystalline anisotropy will provide energy to increase the relaxation time. The benefit of the biaxial anisotropy energy is it will allo w time for an input to flow throughout a system For example if the propagation time for a magnetic cell is 100 picoseconds as estimated by Carlton et al. [ 64 ] the relaxation time for a clock domain of a wire will have to be less than that propagation time to insure signal integrity. The extra energy provided by t he biaxial anisotropy will increase the relaxation time and therefore allow for proper data propagation within clocking boundaries and mitigate magnetic frustrations Figure 2 6 (A) is a BCC crystal of iron which exhibits magnetic anisotropy along different crystal planes. Image (B) is a FCC crystal of cobalt. Note for cobalt the  crystal direction is the easy axis of magnetization.
19 Magnetic Nano Rings 2.3 Magnetic nano rings present an interesting architecture, particularly for storage elements [ 54 65 69 ] Magnetic Random Access Memory (MRAM) h as many of the same benefits as magnetic logic, such as non volatility extreme longevity, radiation hardness, etc ., but one of th e current dilemmas in the technology is the neighbor neighbor magnetostatic interaction Magnetic nano rings have shown a propensity to reduce this interaction as well as controllability via external magnetic fields and more recently through spin torque tr ansfer [ 68 70 71 ] Magnetic nano rings consist of circular soft magnetic disk with the center removed as shown in Figure 2 7 An intriguing feature of the magnetic nano rings design is the possible onion or vortex states. Figure 2 7 (A) and (B) show the magnetic onio n state s What is striking about t hese configurations is the fact that similar magnetic poles are in close proximity and the presence s of ma gnetic flux where the poles coincide The other two sta ble configurations are given in Figure 2 7 (C) and (D). These closed flux states are in a vortex configuration. The benefit of a vortex state for memory is the zero Figure 2 7 (A) and (B) show the two sta ble configurations of the onion state in a magnetic nano ring. The arrows represent the magnetic moment of the nano rings. The onion states are similar to single domain magnets due to the production of stray fields. Image (C) and (D) are the stable vortex configurations of the nano rings. This configuration has zero stray fields.
20 stray field of the closed flux configurations This reduce s any significant magnetostatic interaction with neighboring bits allowing for higher bit densities. Magnetic nano rings can also act similar to single domain nano disks. The flux created by an onion state is similar to the stray fields created by single doma in magnets By placing nano rings in close proximity they can interact through their magnetostatic field s Experiments have also demonstrated a structure where the na no rin gs are linked together and magnetic information is coupled between rings [ 72 ] Magnetic Domain Wall Logic 2.4 Another type of nano magnetic system is driven by the motion of a domain wall. A domain wall is the boundary where a magnetic dipole moment gradually transitions from one direction to the opposite direction, such as a Bloch wall By constructing a magnetic nanowire magnetic dipole moments are formed parallel to the length of the wire due to magnetic shape anisotropy as shown in Figure 2 8 (A) In a sufficiently long nanowire, multiple domain s form along the length of the wire and in between these magnetic Figure 2 8 (A) shows two different nano wires with a contiguous m agnetic domain in opposite directions. (B) depicts two different types of domain wall configurations, one with a head to head configuration and the other with a tail to tail. Whenever there is a domain wall present the magnetization opposing the motion of the domain wall is NOT gate. For this abstraction an external magnetic field (H) w as applied to move the domain wall into and through the kink which performs an inver sion of the logical state.
21 domains are domain walls as s hown in Figure 2 8 (B) Researchers have demonstrated the manipulation of domain walls through the use of external magnetic fields and spin torque moment transfer [ 73 77 ] A pioneering experiment in domain wall logic was the dem onstration of a NOT gate [ 78 ] By introducing a kink in the nanowire as shown in Figure 2 8 (C) the inverse of the logical state can be attained. The ma nipulati on of the domain wall was done through a rotating externa l magnetic field (H) which moved the domain wall through the kink as shown in Figure 2 8 (C) (E). To enumerate the logic the flow of the domain wall must be taken into acco unt. In the example given in Figure 2 8 (C) the domain wall was moving from left to right. To the left of the domain wall was opposite magnetization on the right side of the domain wall was a log domain wall was then propagated up and through the corner and then down the wire via an external rotating field (H). As can be seen in Figure 2 8 (D) once the domain wall passed through the corner the state of the w ire was inverted and could be shifted along the rest of the wire. Using this structure a magnetic shift register was also demonstrated [ 73 78 ] S hift registers are used in many logical components and are an essential component of a central processing unit. The magnetic shift register was successfully operated at 27 MHz and inverted the same domain wall approxim ately 100,000 times. This demonstrated the stability of the domain wall. Subsequently a logic al AND gate was experimentally proven and is shown in Figure 2 9 (A) [ 75 ] The AND gate consist ed of two input wire s which taper down to a junction point This junction point was connected to a thicker outp ut wire where the domain wall was propagated along via an external magnetic field. The junction point acts as a valve
22 providing an energy barrier. Using th e same logical nomenclature as the NOT gate, a magnetic moment that oppose s the flow of the domain wall was Figure 2 9 right). Once the domain wall reached the junction, the external magnetic field could not overcome the energy barrier. Therefore the output remained Figure 2 9 (C) has both Input 1 and Input 2 set to a into both inputs and the external magnetic field there now was enough energy to overcome the barrier of the junction point and Figure 2 9 (D) proceeds. Domain walls were introduced in the referenced work through the use of injection pads It is also possible to induce domain walls through the use of short cu rrent pulse or via magnetic fields from current carrying wires [ 73 76 ] With the demonstration of a n AND and NOT gate all the necessary requirements to produce any logic circuit had been achieved. More recently domain walls have shown the propensity to be controlled via spin torque momentum transfer [ 76 77 79 80 ] This could lead to more practical implementation of controllers and integration with other te chnologies such as CMOS and MRAM. Figure 2 9 (A) shows the basic domain wall logic structure of an AND gate. It is made of two tapered input wires which meet at a junction point. Assuming that (A) was the o both Input 1 and Input 2. D ue to the energy barrier at the junction point (D) can only be reached from (C).
23 Chapter 3 Fabrication of Nano Structures via Electron Beam Lithography Introduction 3.1 Electron beam lithography (EBL) is a high resolution fabrication method often used to repair/create masks for photolithography and n ano imprint lit hography The shorter wavelength of an electron and the lack of diffraction limitations found in tradition al photolithography light sources allows for featur es down to 4 nanometers to be realized using EBL [ 81 ] Althoug h the throughput of an EBL direct write pr ocess is much lower than photolithography research continues to create a commercially viable EBL system. For the purpose of investigating ferromagnetic nano structures EBL provides the flexibili ty of easily proto typing several different structures with the high resolution necessary for the fabrication of nano magnetic logic systems. In the following sections the EBL process used through out this research is detailed. Overview of Electron Beam Lithography Process 3.2 Th e bottom up EBL approach began by coating a silicon substrate with an electron sensitive positive resist namely polymethylmethacrylate ( PMMA ) The resist was then exposed using a J E O L 840 thermionic Scanning Electron Microscope (SEM) retrofitted with a Na bity NPGS lithography system to create the desired mask patterns [ 82 ] The exposed areas were removed via a chemical developer using a solven t of m ethyl isobutyl ketone ( MIBK ): isopropanol 1: 3 A soft ferro magnetic material was then deposited via a Varian Model 980 2462 Electron Beam Evaporator A bond was formed between the
24 ferromagnetic material on the resist in the non developed areas, as w ell as on the substrate where the developed resist h ad been removed. The final liftoff step involved using a solvent (acetone) which removed the resist from the substrate leaving behind only the ferromagnetic metal on the exposed substrate areas. Several v ariations of the EBL process outline above were used in the fabrication of the nano magnetic structures such as molecular weight s of electron sensitive resists, development times, and metal thickness Specific details pertaining to each phase of the resear ch can be found in the respective chapt ers. In the proceeding sections background information on the theory for each step of the process and the rational e for specific methods used are discussed Electron Beam Resist s 3.3 Electron sensitive resist s may be cl assified into two cat egories, positive and negative When exposed to an electron beam, a chemical reaction is induced in a negative electron beam resist which causes crosslinking between polymer chains. The crosslinked polymer chai ns become insoluble whil e the unexposed polymer chains are soluble creating selectivity. A positive electron beam resist also has chemical reaction s induced by the electron beam energy which degrades the polymer molecule s into smaller chain scission molecules. The chain scission molecules are more soluble than the unexposed polymer and can be developed using a n organic solvent. For the purpose of fabricating nano structures with thickness ranges and lateral dimensions in the nano regime, polymethylmethacrylate (PMMA) is the positi ve resist standard for high resolution lithography Although the low sensitivity of PMMA reduces the commercial usability of the resist, the high contrast and low swelling during development are significant factors contributing to the high lateral resolut ion. The
25 dissolution rate of PMMA can be understood via the exposure resist model given in equation 3 1 where is the dissolution rate for the exposed resist, is the dissolution rate for the unexposed resist is the initial molecular weight, is the molecular weight post exposure, is the developers kinetic solubility dissolution parameter, and is the scission efficiency [ 83 ] ( 3 1 ) From the model, it is given that the dissolution rate is inversely proportional to the molecular weight. As the electron beam induces chain scissions in the positive resist, the molecular weight of PMMA is reduced and therefo re more soluble than the unexposed higher molecular weight PMMA. Figure 3 1 shows a main chain scission for polymethylmethacrylate which lowers the molecular weight of the resist creating the desired selectivity between exposed an d unexposed areas. Preparation and Application of PMMA Resist 3.4 In order to fabricate a device a substrate is needed. N type 25 0mm thick silicon wafers were used throughout the fabrication process as the substrate due primarily to Figure 3 1 shows a chemical reaction ind uced by the exposure of the positive resist polymethylmethacrylate (PMMA) The main molecular chain is given on the left and the scission of the main chain is shown on the right. The lower molecular weight of the chain on the right makes the resist more so luble in a developing solvent.
26 the atomically smooth surface and lack of magnetic response. The surface was prepared for resist coating using the RCA clea ning procedure outlined in Table 3 1 The purpose of the RCA cleaning procedure was to remove organic, ionic, and heavy metal co ntaminants in order to prepare the Si wafer surface for the application of the electron sensitive resist. A spray on solvent clean of acetone, followed by isopropanol, and dri ed Table 3 1 outlines and describes the process steps used for the RCA cleaning procedure. Process Description Rinse wafers with DI water for two rinsing cycles Rinses any superficial d ebris Dip in HF (50:1) for 20 seconds Removes native oxide Rinse with DI water for two rinsing cycles Rinses off any residue Dip in SC1 for 10 minutes @ 60C 2000ml of DI water; 330ml of Hydrogen Peroxide; 330ml of Ammonium Hydroxide (6:1:1) Removes insoluble organic contaminants Rinse with DI water for two rinsing cycles Rinses off residue Dip in HF (50:1) for 20 seconds Removes chemical/native oxide Rinse with DI water Rinses off residue Dip in SC2 for 10 minutes @ 60C 2000ml of water; 330ml of Hydrogen Peroxide; 330ml of Hydrochloric acid (6:1:1) Removes ionic and heavy metal contaminants Rinse with DI water for two rinsing cycles Rinses off residue Dip in HF (50:1) for 20 seconds Removes chemical/native oxide Rinse with DI water for two r insing cycles Rinses off residue Dry with Nitrogen Gas Removes possible water spots
27 using nitrogen gas was also done prior to the resist application to remove an y possible contaminates from handling The substrate s w ere spun dried at 5000 rpm for 2 minutes to promote the evaporation of any residual solvents without altering the temperature of the substrate. The application of the PMMA onto the wafer w as a ccomplished using a Laurell Technologies WS 400A 8NPP/Lite Spin Processor Due to the nature of the research, most of the fabricated devices were less than 40 nm thick and therefore thin layers of resist were necessary. A Teflon pipette was used to deposit a 25mm circle of PMMA at room temperature. The recipe used to apply a uniform coating of PMMA is given in Table 3 2 After the resist was spun a prebake was done in a convection oven at 170C for 30 minutes to remove any excess s olvent. The spin recipe was successfully used for applying resist layers ranging from 30 nm to 120 nm with no adhesion issues The t wo molecular weights of PMMA used were 495,000 and 950,000 in anisole, which were commercially a vailable from MicroChem [ 84 ] The percentage of PMMA to anisole w as diluted to less than 2% for the desired film thickness by adding anisole. For h igh resolution electron beam lithography, a single layer of 950k MW PMMA was able to Table 3 2 outlines the resist spinner recipe used for 495K and 950K PMMA ranging from 30 120 nm of thickness. Process Ste p Description 1 Dispensed a 25mm diameter drop of PMMA resist 2 Pre ramp up: 0 500 rpm 5 seconds 3 Ramp up: 500 5000 rpm 45 seconds 4 Ramp down: 5000 0 rpm 5 seconds 5 Prebake: 170C for 30 minutes
28 successfully meet the critical dimensions of 20 nm between each nano magnetic device. A bilayer resist process was also used where the lower molecular weight PMMA was spun as bottom layer, prebaked, followed by the same procedure for the higher molecular weight for the top layer The higher molecular weight PMMA provided the better contrast while the more sensitive bottom laye r created an undercutting profile ideal for a cl ean liftoff. Scanning El ectron Beam Exposure 3.5 Traditional exposure methods use light to develop photoresist down to just below a micron but for high resolution lithography electron beam exposure systems continue to offer researchers low cost rapid prototyping of submicron featu re sizes. Figure 3 2 show s a typical Scanning Electron Microscope (SEM) column The beam is focused via magnetic lenses and is scanned across the surface of a sample via deflectors. SEMs have several possible electron gun sources, such as ther mal field emission cathodes, cold field Figure 3 2 is an abstraction of a scanning electron microscope column. For electron beam lithography thermionic emission filaments are often used due to the stable emission current
29 emission cathodes, and thermionic emission sources which are preferred electron beam lithography Field emission sources are excellent for imaging via secondary electron excitation due to the high brightn ess of the source, but are prone to contamination from gases in the vacuum system which can cause a drift i n t he emission current [ 85 ] The stability of a emission current is critical to the proper exposure of resists and is not an issue with thermionic sources. A tungsten thermionic emission source emits electrons by providing thermal energy to overco me the work function of the metal ; break ing the electron free from the material into the vacuum. A uniform thermionic emission current density can be model by the Richardson Dushman ( equation 3 2 ) where is emission current density is the Richardson constant (1.20 x 10 6 Am 2 K 2 ), is temperature is and is the Boltzmann constant ( 10 5 ) [ 86 ] ( 3 2 ) Figure 3 3 represents two different electron beam energies entering an electron sensitive resist and substrate The figure on the left is at a lower acceleration voltage and the backscattered electrons expose the resist as well as the forward scattered electrons. The figure on the right is at a higher acceleration energy and the resist is exposed primarily by forward scattered electrons.
30 The exposure mechanisms of electron sensi tive resists depend on several factors such as the atomic number of the substrate, beam energy, and resist thickness. Figure 3 3 depicts how electrons scatter as they traverse the resist and substrate mediums. As electrons move th rough the resist energy is deposited into the medium through interactions between the primary beam and the resist molecules. When the incident beam of accelerated electrons ( where acceleration voltages range from 20 50kV) enters the resist, interactions wi th a thin resist ( < 100nm) are small and only m inute amount of initial energy is deposited into the resist. The transparent resist allows most of the energy to enter the substrate which induces several types of reactions such as x rays, auger electrons, and secondary electrons. As the incident electrons interact with the sub strate several scatter back into the resist know n as backward scattered electrons Higher energy beams penetration deeper into the substrate and have less resist exposure due to backsc attered electrons while lower energy beams have a broader lateral profile in the resist and can suffer from t he proximity effect. H igher energy electron beams offer improved resist exposure profiles but require higher doses of energy to expose the resist. T he energy deposited by energetic electrons into resist can be model ed usin g the Bethe energy model, where is the electron energy, is the distance of the electron trajectory, is the electron density of the atoms, is the charge, and is the mean excitation energy [ 87 ] ( 3 3 ) From this equation i t can be understood that as electron beam energy is increased the amount of energy imparted to the resist per unit length decreases. As higher acceleration energies are used higher doses are needed to expose the resist.
31 Resist Development 3.6 In order to selectively remove resist a developer solution normally removes a resist of lower molecular weight more effectively than the same resist of a higher molecular weight The developer solution is specific to the particular chemistry of a given resist and the necessary activation energy need ed to dissolve it Figure 3 4 (A ) depicts the developing process of a sample that has both exposed and unexposed areas of a positive resi st place d into a developer solution. T he developer penetrate s the lower molecular weight resist removing it from the substrate while the higher molecular resist areas swell as a resistance mechanism to dissolution Swelling is normally an undesired condition of d eveloping and can cause negative effects such as bridging [ 87 ] Once the developer is remo ved and the resist is dried a mask is created. In order for the dissolution of the lower molecular weight resist to occur a solvent must provide the necessary activation energy to uncoil the resist molecules. T he solubility parameter and resist contrast f or various developer s of PMMA have been studied and are used to Figure 3 4 (A) depicts the sequential development process (top to bottom) for a positive resist such as PMMA. Figure (B) shows the interaction between the developer solution and resist. As the developer penetrates the resist a gel layer is formed. The activation energy necessary for the PMMA to uncoil is provided by the penetrating developer molecules.
32 calculate the disso lution rate in equati on 3 1 [ 83 ] A poor developer with low activation energy will provide insuff icient energy to efficiently break intermolecular and intramolecular forces of the resist This increases the dissolution time which in turn increa ses swelling deforming the desired mask pattern. The combination of a high molecular weight PMMA resist an d a developer solution of MIBK:Isopropanol 1:3 was used in this research to attain the high resolution lithography necessary for the creation of magnetic logic devices. The same development recipe was used for a single layer of 950K PMMA and the bilayer 49 5K 950K (bottom top layer) of PMMA and i s given in Table 3 3 Since the top 950K PMMA layer of the bilayer resist provides the desired mask dimension it dictates the development time and is the same when using only single layer of 950K PMMA. The lower molecular weight bottom layer of 450K PMMA dissolv es at a fast rate which created the desired undercutting profile and ensured proper liftoff. Film Deposition 3.7 For the fabrication of nano magnetic logic devices a soft ferromagnetic metalli c material demonstrates ideal characteristics for logical switching [ 14 49 ] There are several m ethods available for the deposition of thin ferromagnetic film s such as electroplating and sputtering. Both aforementioned techniques are ideal for specific Table 3 3 outlines the development process used for a single layer of 950K PMMA and for s bilayer 495K 950K PMMA (bottom top) resist. Process Step Description Development MIBK:Isopropanol 1:3 for 60 seconds constant stirring Rinse Isopropanol 20 seconds c onstant stirring Dry Nitrogen gas
33 fabrication situations, particularly when the conformal coating of a surface is desired. In this research e lectron beam evaporation was used for thin film deposition ; primarily due to line of sight deposition which aids in the liftoff process a nd the preservation of the ferromagnetic alloy stoichiometry [ 88 ] An e lectron beam evaporation uses a n electron filament (as known as an electron gun) to vaporize a solid material source shown schematically in Figure 3 5 As the source material is heated via the thermal energy provided by the electron beam it under goes a matter phase transitions from a solid to a gas The phase transitions of a so urce are dependent on the physical material properties and can be characterized by the vapor pressure. If the vapor pressure for a material is above 10 2 Torr it will typical ly go through a standard matter transition (solid to liquid to gas), and are readi ly evaporated. Other elements such as Chromium (Cr) have vapor pressure below 10 2 Torr and transition directly from a solid to a gas in a process known as sublimation. These materials are harder to evaporate and have lower evaporation rates. Most of the d evices Figure 3 5 is a schematic of an electron beam evaporator The electron beam from the filament provides thermal energy to the solid source which causes the matter to transition into a gas phase. The evaporated source proceeds to grow a thin film of the material as it interacts with the surface of the target.
34 reported in this research were made of commercially available p ermalloy and n ickel pellets. Both materials were readily evaporated in a vacuum of 10 6 Torr using a Varian Model 980 2462 Electron Beam Evaporator at rates of 1 8 sec 1 Once the sour ce material is evaporated into a gas using a vacuum deposition system such as an electron beam evaporator, the gas can be modeled using the ideal gas law An important parameter to consider during deposition is the pressure at which the material is evapor ated The vacuum pressure at the time of the evaporation determines the mean free path of a gas mole cule and is given in equation 3 4 where is the pressure of the vacuum system, and is the diameter of the gas m olecule. ( 3 4 ) Assuming the source material travels in a straight line from the evaporation source to the intended target the mean free path of a material d ictates how far a molecule can travel before it interacts with another molecule in its environment. For example a molecule with a diameter of 4 at a vacuum pressure of 10 7 Torr has a mean free path of 60 m eters [ 89 ] It was determined that for the physical setup of the Varian evaporator used in this research, a vacuum pressure of 10 6 Torr would ensure minimal contamination during deposition. The position of the target in relation to the evaporation source was also considered and the growth rate can be predicted using equation 3 5 where is the density, is the mass evaporation rate, is the angle normal to the source plane, and is th e angle normal to the target plane as shown in Figure 3 5 ( 3 5 )
35 Liftoff 3.8 The final step in the additive electron beam lithography process involved selectively dissolving the remaining resist mask. Similar dissolution kinetics for the chemical development of a resist applies to the liftoff process. In stead of having dissimilar dissolution rates for a resist of different molecular weights as with the development step in the liftoff process a solvent has to select between an organic resist material and, in this case, an inorganic ferromagnetic metal. The p rocess used for the dissolution of the PMMA resist consisted of an ultrasonic agitation of the sample in a warm bath of acetone. The sample was agitated for approximately 10 minutes followed by an isopropanol rinse and dried using nitrogen gas. Once the resis t was dissolved the only remaining metal was the metal which formed a bond directly with the substrate and a cross section al diagram of the liftoff process is given in Figure 3 6 Figure 3 6 is cross se ctional view of the liftoff process. After the ferromagnetic metal layer has been evaporated, the sample is placed in a solvent which dissolves the resist and removes the unwanted metal layers. The final metal device layer is left in place bonded to the su bstrate surface.
36 Chapter 4 Introduction to Single Domain Nano Magnets fo r Cellular Automata Introduction 4.1 The interactions of orderly fabricated nano magnets and the viability of nano magnetic structures as logic al building blocks is an area of great interest [ 12 ] With few examples of order ly coupled nano magnetic logic systems published, in this chapter the study of coupling between single domain dipole moments is discussed. A transitory introduction to magnetism is given leading into a discussion on the formation of single domain dipole mo ments. The coupling dependences of physical parameters such as size, surface roughness, and shape irregularities are also discussed In particular nano magnets with an aspect ratio near 2:1 were of interest due to their bi stable nature and functionali ty [ 1 44 ] The bi stability created by the shape anisotropy (i.e. the 2:1 aspect ratio of the cell) significantly reduces meta stable states, which in turn allows proper coupling between neighboring nano magnetic cells. The objective of this chapter was to alter physical parameters of the cells and examine the resultant magnetostatic coupling interaction of single do main rectangular nano magnetic Magnetism and Magnetic Materials Overture 4.2 The phenomenon known as magnetism arises from the angular momentum of a charge d particle, generally the electron. Magnetism from an atom can originate from the orbital angular moment of an electron (L) the intrinsic angular momentum (S), or a combination of the two. The ene rgies of both quantities are defined in quantum
37 mechanics by the quantum numbers l and s The other two quantum number s associated with magnetism m l and m s ; corre spond to the angular momentum component along an external magnetic field (B). xclusion principle and n atom to exhibit a net magnet ic resultant force known as a magnetic dipole moment. Based on the elect ron configurations of elements, materials can be classified into five magnetic categories: diamagnetic, paramagnetic, anti ferromagnetic, ferrimagnetic, and ferromagnetic. Diamagnetism is defined by materials that have a generally weak negative magnetic s usceptibility ( X m = 10 6 ). Magnetic susceptibility is defined in equation (4 1), where ( 4 1 ) is the magnetization vector, is the magnetic susceptibility, and is an external magnetizing field. From this relationship it is clear if the magnetic susceptibility is negative, the magnetization of the material will be in the opposite direction of the applied magnetizing field Materials made of elements with closed shells and subshells, as well as many organic materials, are diamagnetic. Paramagnetic materials have a small positive magnetic susceptibilities (10 5 ~10 4 ) which are comprise d of atoms with unpaired electrons. These unpaired electrons have a net resultant magnetic moment but are weakly coupled to neighboring atoms. Each atomic magnetic moment can be considered independent such that in the absence of a magnetic field the magnetic moments are aligned in random directions and the material has no net magnetization. Once an external magnetic field is applied, the magnetic moments experience a torque which causes them to align in the direction of the field causing a net magnetization. When the external field is removed, the alignment of the
38 moments are randomized again and the net magnetization is lost. Anti ferromagnetic, ferrimagnetic, and ferromagn etic material also exhibit paramagnetic behavior above a critical temperature, were the atomic magnetic ordering energy is dominated by thermal energy. Anti ferromagnetic materials also have small positive magnetic susceptibilities similar to paramagnetic materials but demonstrate atomic magneti c ordering as shown in Figure 4 1 (A) B elow the Neel temperature T N a Body Centered Crystal (BCC) unit cell of Chromium (T N ~ 37.5C), exhibits anti ferromagnetic ordering such that the total net magnetic moment of the crystal is z ero. Above the Neel temperature anti ferromagnetic materials become paramagnetic. Different types of anti ferromagnetic ordering exist for compounds such as cubic perovskites, which are show n in Figure 4 1 (B) For perovskites, the G type ordering is due to an indirect superexchange interaction of the oxygen atoms which force the magnetic moments of neighboring atoms into an anti ferromagnetic arrangement. Figure 4 1 (A) depicts the anti ferromagnetic ordering of a Chromium BCC crystal The (+) and ( ) signs represent t he two possible spin states. The center atom contributes one atomic magnetic moment and the corner atoms combine to contribute a total of one opposing atomic magnetic moment to the unit cell. The magnetic moment for the unit cell is zero, wh ich is the defi nition of an anti ferromagnetic material. (B) shows several different anti ferromagnetic atomic magnetic orderings for a simple cubic lattice.
39 Ferrimagnetism is very similar in nature to anti ferromagnetism with regards to magnetic atomic ordering but the magnetic sublattices are not equivalent as shown in Figure 4 2 (A) This causes a net magnetization difference b etween sublattices and, unlike anti ferromagnetism where there is no net magnetization, a ferrimagnetic materi al exhibits a net magnetization ( M >0) below the Curie temperature (T c ) Ferrimagnetic materials have very large, positive magnetic susceptibility and become paramagnetic above the Curie temperature Ferrimagnetic materials are often electrical insulators and do not suffer from eddy currents caused by rapidly changing magnetic fields This makes ferrimagnetism very useful for automated applications and is used in many electronic devices. If all the atomic magnetic dipole moments of a material are ali gned in the same direction as sh own in Figure 4 2 (B), this material is defined as ferromagnetic. Ferromagnetic materials have very large positive values of magnetic susceptibility and vary based on the strength of the magnetizin g field The origin of ferromagnetic and Figure 4 2 (A) is a two dimensional slice of a ferrimagnetic crystal lattice. The larger magnetization of the green atoms creates a total net magnetization along that particular direction. (B) is a BCC unit cell of an iron crystal. A ferromagnetic material is defined by each atomic magnetic moment having the same spin. The (+) signs indicates all the atoms are of the same spin.
40 anti ferromagnetic ordering between nearest neighbor atoms is actually not so classically intuitive. If two bar magnets were placed near each other it understood that they would couple in an anti ferromagnetic or fe rromagnetic manner, but this direct dipole dipole interaction is actually not the origin of magnetic atomic ordering between nearest neighbors. Indeed, within a ferromagnetic material a quantum mechanical exchange interaction (direct and/or indirect) is th e basis for magnetic ordering. The exchange interaction is governed by the lowering of electrostatic columbic energies direct Heisenberg exchange int eraction is given in equation (4 2) whe re and are two interacting ( 4 2 ) spins separated by distance and is the ex c hange integral with units of energy. When the exchange integral is positive the resulting coupling of the spins is ferromagnetic; conversely when the exchange constant is negative the spins are coupled in an anti ferromagnetic fashion. The Heisenberg exch ange interaction is the dominant effect for spins and that have electron clouds with considerable overlapping and are atomically separated on the order of nearest neighbor (NN) or next to nearest neighbor (next NN ). Ferromagnetic materials are used in various electronic applications and were the primarily materials used in this study Magnetic Domains 4.3 Although f erromagnetic t ransitional metals such as Ni, Fe, and Co are permanent magnets they can also be found to exert no net magnetic force due to the formation of domains within a material When a ferromagnetic material is sufficiently large, in general gr eater than the mesoscopic range, multiple magnetic domains are created inside of the
41 m agnet w hich arrange themselves to reduce the overall potential energy These domains are commonly found in opposition to one another by 180 degrees an d separated by a domain wall as depicted in the initial state (I) of Figure 4 3 of the hysteresis curve Following the different points throughout the hysteresis curve provides a good example of magnetic domains and domain wall motion through the stimulus of an external field. It also presents the co ncepts of multiple domain structures and single domain magnetic dipole moments. At point (I) i n Figure 4 3 a magnet was found with a total magnetization of zero due to the minimization of magnetostatic energy through the creation of multiple domains As an external magneti zing field (H) was applied the total magnetization of the magnet increase d T he magnetic domain with a larger component in the direction o f the applied field expanded and is shown on the subsequent point on the curve Eventually all the magnetic domains became align ed in the same direction as the applied field. The absence of domain walls is known a s a single domain magnetic Figure 4 3 is half of a magnetic the magnet has a net magnetization of 0. As an external magnetic field is applied (H) the appropriate magnet domains expand and contract as shown throughout the curve.
42 dipole moment. As can be seen at the next point the magnet was still not completely aligned in the direction of the applied magnetic fi eld. This was due to an easy axis of magnetization which in this case was due to the shape anisotropy of the magnet Magnetic saturation (M s ) is defined as the point where the magnet has all possible magnetic moments aligned in the direction of the external field and is indicated by the image at the following point on the curve Any external field applied greater than (M s ) will not change the magnitude or directi on of magnetization. As (H) was ramped down until there was no external field present, the curve shows that the magnet still had a net magnetization. This is known as the remanent magnetization (M r ) Finally another field was applied known as the coercive field. The energy ne cessary to demagneti ze the remanent magnetization is defined as the coercivity The final point on the curve demonstrates the formation of closure domains which remove d all magnetic flux. The formation of multiple domains is a complic ated process and for the purpose of MCA is undesirable. The objective was to produce switches that couple with one another through magnetostatic field interactions and therefore the intent was to create magnetic cells that were single domain in nature. Figure 4 4 is of a nickel anti ferromagnetic MCA wire. In Figure 4 4 (A) each magnetic cell was approximately 400 ( l ) x 200 ( w ) x 30 ( th ) nm 3 The image on the left was a n atomic force microscopy ( AFM ) height image which represented the physical dimensions of the system, and on the right was the magnetic force microscopy ( MFM ) image illustrating the magnetic moments. The complicated magnetization found in the MFM image was due to the mul tiple domains formed wi thin each magnetic cell in the wire. The formation of multiple domains i s dependent on energy minimizations
43 factors such as crystalline anisotropy and shape anisotropy [ 90 ] Figure 4 4 (B) is another chain with the sam e aspect ratio of 2:1, where each cell was approximately 200(l) x 100(w) x 30(th) nm 3 It was evident that the each magnet had a more regular single domain magnetic dipole moment Furthermore, the magnetic dipole moments appear to be coupled to one another via magnetostatic fields. T he reduction of size caused single domain moments of each cell to couple in a digital fashion between neighbors and is the essence of MCA systems A single domain magnetic system affords the opportunity for a new logic and memory paradigm to be explored. Fabrication Process 4.4 The electron beam lithography process began by coating a Si wafer with a resist, n amely PMMA. This was accomplished by using a Laurell Technologies WS 400A 8NPP/Lite Spin Processor which spun the wafer at high speeds causing the PMMA resist (A) (B) Figure 4 4 (A) is an AFM height amplitude plot (left) and a MFM phase plot (right). The dimensions of each Nickel rectangle in (A) were 200(w) x 400(l) x 30(th) nm 3 The complicated multiple domain structures arose from the physical dimensions and material properties of n ic kel. (B) is a n ickel system with dimensions of 100(w) x 200(l) x 30(th) nm 3 The MFM image clearly shows a single domain magnetization as well as an anti ferromagnetic coupling order.
44 to spread evenly over the entire wafer. A bilayer PMMA recipe was chosen using 495K molecular we ight as the bottom layer and 950K molecular weight as the top layer. The 495K bottom layer was spun first and, then baked, followed by the 950K top layer being spun and baked. The purpose of the bilayer PMMA was to ensure a clean liftoff, which was aided b y the undercutting profile created in the low molecular weight PMMA. A pattern was designed using DesignCAD2000 NT. The most effective line spacing exposure doses points, and focus were determined by using diagnostic designs, such as the wheel pattern a nd spot burn process. A sample of the Si wafer coated with bi layer PM MA was then loaded into the JEO L 840m retrofitted with the NPGS lithography system and a beam blanker. The SEM was then focused, and the stigmation was adjusted for optimal resolution T he pattern was selected and written via the NPGS system. Afterwards, the sample was unloaded and prepared for chemical development. It was placed in a bath of MIBK and Isopropanol (1:3) for approximately seventy seconds, followed by a twenty second Isoprop anol bath, and finally nitrogen dried. A ferromagnetic mat erial was chosen for deposition in this case n ickel and p ermalloy (Ni80Fe20), and was evaporated via a Varian Model 980 2462 Electron Beam Evaporator A vacuum of approximately 2 Torr was achieve d and the mat erial was evaporated at a rate of 5 sec 1 Once the desired thickness of the material was deposited, the next stage was the liftoff step. The sample coated with the ferromagnetic material was placed in a heated ultrasonic acetone bath for ap proximately 15 minutes, which selectively lifted off the unwanted PMMA and left the fabricated magnetic structure intact.
45 Magnetic Coupling of Single Domain Nano M agnets 4.5 For magnetic logic system s to be considered a viable technology, reliabl e interaction between magnetic switches must be realized Interaction s between magnetic switches a re accomplished through the coupling of spins which create magnetic dipole moments. There are different ways for spins to interact but here the interaction was based on magnetostatic field interactions. The f ield interaction between magnetic cells will intuitively be affected by the physical dimensions of each magnetic cell. Particularly, how factors such as surface roughness and shape irregularities affect coupling interaction is important since the surface a rea begins to dominate volume at the nano scale. Figure 4 5 Figure 4 6 and Figure 4 7 are a set of images of three different chains of nano magnetic cells and have the follow ing images from top to bottom: a SEM surface image, an AFM 3 D surface plot, an AFM 2 D surface image, and a MFM image. Each figure is of the same chain of nano magnetic cells and each cell has the Figure 4 5 is four different s ets of data of the same MCA interconnect. The top figure is a SEM image, followed by a 3D representation of an AFM plot. Next is an image of a 2D AFM plot and the bottom image is a MFM phase plot. From these images s ignificantly irregular ities in the cell structure and surfaces are visible The fourth cell from the right has no in plane moment detected due to the surface imperfection.
46 dimensionality of approximately 160nm height x 80nm width x 40nm thickness. The rational w as to compare the magnetic moments of each cell and the resulting magnetic coupling of neighbors to : surface roughness, shape regularity, cell spacing, and cell size. These results show ed that n ickel h ad single dom ain characteristics parallel to the substrate (in plane) at dimensions mentioned above as evident in Figure 4 5 Figure 4 6 and Figure 4 7 It can also be noted that strong magnetic coupling is e x hib ited due to the anti parallel alignment of the cells In Figure 4 5 there are several n ano magnetic cells with very irregular shapes, irregularly spaced and with uneven surfaces. The AFM images clearly show the surface roughness where the SEM images present the shape irregularities, cell spacing, and cell spatial orientation. Here it can be seen that even if the magnet d id not have a uniform shape and neighbors are not completely parallel to one another that single domain magnetic dipole moments still formed and data propagation still occurred, evident via the anti parallel coupling of cells 5 through 10 (left to right). C ells 1 and 2 had an Figure 4 6 is another four sets of data of a different MCA wire. The top figure is a SEM image followed by a 3D representation of an AFM plot. Next is an image of a 2D AFM plot and the bottom image is a MFM phase plot. In this chain, the cells with a smoother surface have better defined symmetrical dipole moments.
47 unwanted coupling state which can be attributed to the surface irregularities. It is also noted that where there are large protrusions on the surface of a cell, the magnetic moment seems to be void ed as evident in cell 4. Figure 4 6 is a different chain of nano magnets. Overall the various cells ha d a smoother surface and more regular shape than the chain presented in Figure 4 5 Cells 2 and 4 had large bulges on the s urfaces and the MFM image depicted magnet ic voids in corresponding areas. Cells 2 and 3 were in an undesired configuration, with like orientations of their magnetic moments, which was attribute d to the irregular surfaces. It was also note d that even though cell 1 had a very irregular shape (almost square) it still managed to form a desired single domain magnetic moment. Figure 4 7 was an ideally fabricated chain of nano magnetic cells. It was see n that the shape, surface, spacing, and orientation were better than that of the previous chains mentioned above. In this fabricated chain of nano magnet ic cells, there was no configuration errors, all magnets were in a n anti parallel orientation and thus Figure 4 7 is another four different sets of data of an ideal MCA wire. The top figure is a SEM image followed by a 3D representation of an AFM plot. Next is an image of a 2D AFM plot and the bottom image is a MFM phase plot. This is an ideal chain, where the shape of the cells was regular and the surface roughness had been reduced. As a result the single domain dipole moments are were strongly coupled with no magnetic frustration s.
48 propagating information correctly. T his could mainly be attributed to the surface evenness of the nano magnetic cells. These experiments demonstrate d that for an anti ferromagnetic coupling scheme, shape uniformity was a much more lenient requirement than the surface roughness for reliable data propagation. S everal irregularly shaped rectangular nan o cells were able to establish single domain magnetic moments. The irregularly shaped cells were also able to propagate da ta to neighboring cells which was exhibit ed via an a nti parallel magnetic coupling. It was note d that configuration errors, where neighbors have similar magnetic dipole moment orientations, occur red near cells that had irregular surfaces. The reliability of anti ferromagnetic M CA wire s therefore relied on the fact the surface mu st be uniformly smooth in order to propagate data properly. The arrangement of how nano magnets couple with one another and the robustness to fabrication defects was a more subtle aspect of MCA. As mention ed previously an anti ferromagnetic arrangement wa s susceptible to significant fabrication defects. In Figure 4 8 a ferromagnetic chain was shown with similar dimensions to the anti ferromagnetic wires mentioned earlier in this section. The top image was an AFM image and the bot tom was the corresponding MFM image. The AFM image clearly showed substantial irregularities in both shape irregularities and surface roughness. Similar types of defects show ed magnetic frustrations in an anti ferromagnetic wire but the ferromagnetic wire demonstrated the robustness to such defects in the MFM image of Figure 4 8 The reproducibility of ferromagnetic coupling in wires with surface irregularities was confirmed in Figure 4 9 Overall ferro magnetic coupled wires were found to be more robust to fabrication defects when compared to anti ferromagnetic MCA wires due to coupling energies and is explored in Chapter 5
49 Figure 4 8 is an AFM amplitude height plot of a ferromagnetic chain which shows the highly irregular shape and surface roughness. The image on the bottom is the MFM phase plot of the chain and shows the single domain magnetic moments. In general ferroma gnetic coupling was found to be more robust to large fabrication defects. The sizes of the nano magnetic cells were 100(l) x 50(w) x 30(th) nm 3 Figure 4 9 is the AFM height amplitude plot of five ferromagn etic chains (left). The image to the right is the corresponding MFM phase plot. Although the AFM image shows irregularities in the structures the MFM image shows the highly ordered coupling of the chains.
50 Although fabrication defects such as surface roughn ess, shape irregularities, and spacing irregularities may cause erroneous data propagation ; these defects could be significantly reduced through the perfection of the fabrication process. The levels of irregularities, at these dimensions, needed to serio usly affect the coupling dynamics was very significant and most fabrication defects could be avoided through the fine tuning of the fabrication process such as having a constant evaporation rate without significant evaporation spikes Therefore it could be concluded that magnetic cells for MCA had the capability to mitigate erroneous coupling due to common fabr ication defects. I f the m agnetic coupling of MCA cells were reliable it follows that more complicated arrangements of MCA should be able to perform c omplex logical operations. Imre et al. demonstrated the first MCA logic gate to function at room temperature [ 14 ] The gate was a majority voter which can also be consider ed to have the functionality of both a two bit OR gate or a two bit AND gate by fixing an input to a All the possible combinations of the 3 bit majority gate were demonstrated and Figure 4 10 is an expansion of the maj ority gate. T he majority gate presented in Figure 4 10 had a total of 9 extra magnetic cells added to the logic gate inputs and output then the previ ously reported gate As can be seen in Figure 4 10 the logic operation performed a majority vote of the inputs given. Figure 4 11 is a ferromagnetic majority gate which was the first to be proposed and demonstrated The diff erence between this layout and the previous ly reported MCA majority gate was that the coupling at the input cells as well as the output cells was ferromagnetic, unlike Figure 4 10 that had both ferromagnetic and anti ferromagnetic coupling. As mentioned earlier in the section, ferromagnetic coupling seemed to be more
51 robust to fabrication defects. The AFM image of the ferromagnetic majority gate shown in Figure 4 11 (A) clearly showed some surface rough ness presen t The logic operation of a ferromagnetic majority gate also differed from the previously demonstrated mixed coupled gate. The lo gical state of a wire was defined by the direction of the informational flow as shown in the MFM image Figure 4 11 The white arrows represent ed the inputs to the vot er cell which was outlined by a white rectangle. Based on the given inputs to the voter cell the correct output was propagated along the black arrow. What was particularly interesting about the configuration in this MFM image was the opposing result of the voter cell to the vertical input wire. The fact that the voter cell was opposed to the vertical input wire was a direct result of the dipole moments provided by the horizontal in puts This demonstrated the functionality of the voter cell responding to the inputs provided by the gate and shows the possibility of a ferromagnetic MCA majority gate structure. Figure 4 10 (A) is an AFM height amplitude plot of a MCA majority gate. Notice that there is anti ferromagnetic coupl ing as well as ferromagnetic coupling in the design of the structure. (B) is a MFM phase plot. The poles have been overlaid with identifiers to help interpret the image. The rectangle in the center outlines the voter cell which The inputs were the vertical cells at the bottom, left and right of the image.
52 Figure 4 12 was a set of ferromagnetic majority gates feeding into a third majority gate. This was one of the largest systems of MCA logic to be demonstrated. There was one magnetic frustration found in the entire system Through experimentation it was found that one frustration was predicta bly present in a three majority gate system. It was also important to note that there was no t a frustration at the voter cell; all magnetic frustrations had been found at the periphera ls of the gates Since ferromagnetic coupling was determined to be reliable, as shown in several examples in this chapter, it was r easoned the issue arose from the ability to control informational flow. How the magnetic system was driven fro m input to ou tput was a non trivial issue It was also r easoned that propagation errors, such as the one shown in Figure 4 11 were directly related to the fact that there was no control of where the data started and ended Without a means o f synchronous computation via a clocking mechanism, magnetic frustrations and the orderly operation of logic would not be possible. The following Figure 4 11 (A) is an AF M height amplitude plot of a ferromagnetic MCA majority gate. Notice that all the coupling between the input chains and the output chains were ferromagnetic. (B) is the corresponding MFM phase plot. The poles have been overlaid with identifiers to help in terpret the image. The rectangle in the center outlines the voter cell has voted based on the two horizontal input chains and opposed the vertical input chain
53 chapter explores how different types of external fields interact ed with ferromagnetic and anti ferromagnetic MCA wires along different mag netization axes By studying th e interaction of MCA wires with an external clocking field a more realistic clocking mechanism may be realized. A thorough understanding of signal propagation, switch delay, and system clocking will be a necessity if MCA logi c systems are to materialize as a viable technology. Figure 4 12 is a MFM phase plot of two ferromagneti c MCA majority gates feeding into a third majority gate. All voter cells in the design had no magnetic frustrations.
54 Chapter 5 Magnetic Cellular Automata Interconnects Introduction 5.1 In order for any logic machine to work, there must be communication between the various logic components throughout a design. One can observe this a t several different levels throughout the design automation process w ith transistor technology. From individual transistor level, to inter gate signaling, or the signaling between logical black boxes, the interconnecting wire is fundamental to every aspect of any logic system. Without a means of communicating results from one logic structure to another, computation is inconsequential. Hence, the reliability of an interconnecting wire providing accurate information is crucial to the functionality of a system In MCA, there have been wires prototyped demonstrating a high order of coupling [ 23 ] ; encouraged by these results, similar work was produced and the results indeed verified successful operation [ 91 ] To progress towards implementing more complex circuitry, it was necessary to control information propagation of an interconnecting wire via a clock. Csaba et al proposed using an adiabatic clocking method that used an e xternal magnetic field [ 13 ] The external magnetic clocking field saturated the nano magnets, causing the overall magnetic dipole moment to align itself along the in plane hard axis of the nano structure. The clocking field would then be reduced as an input was provided. The study
55 of wire archit ectures under such clocking fields was significant to the advancement MCA systems and presented here. Wire Architectures 5.2 In MCA there were two types of interconnects that had been proposed, ferromagnetic coupled wires and anti ferromagnetic coupled wires Th e different types of interconnects were created through fabricating the wires in the different arrangements shown in Figure 5 1 For both wire architectures the desired state was always the lowest possible local energy minimum. Since magnetic systems work ed to lower their overall magnetization energy MCA has taken advantage of the physics. Anti ferromagnetic wires had several intriguing factors associated with them such as a built in inverter. When information was properly pr opagat ed down a wire each magnet had the inverted state of its immediate neighbor as shown in 1 This occurred due to the repulsion forces experienced by similar magnetic poles. A logical Figure 5 1 depicts correct data propagation for ferromagnetic and anti ferromagnetic wires, which are the lowest ground state energy. For the ferromagnetic wire the mag netic dipole moments are oriented in the same direction, where anti ferromagnetic wires are in opposing directions.
56 error occurred when neigh boring magnets had like states shown in Figure 5 2 This is known as a magnetic frustration because of the high demagnetization energy experienced by pinning like poles next to each other. Ferromagnetic wires were arranged in a head to tail fashion as shown in 1 Data propagation occurred through magnetic field interactions of each cell. Proper logical exchange occurred when all magnetic dipole moments were arranged in the same direction as shown in 1 Magnet frustrations occurred when neighboring cells had dissimilar magnetic dipole moments as depicted in Figure 5 2 In the proceeding sectio ns the first experimental results were presented comparing shape engineered ferromagnetic and anti ferromagnetic wires under various magnetic clocking fields. The experimental observations show that ferromagnetic wires were less likely to have magnetic fru strations Simulations were also implemented which bolster the conclusion that ferromagnetic wires operate in a more reliable manner than anti ferromagnetic wires. Figure 5 2 depicts incorrect data propagation for ferromagnetic and anti ferromagnetic wi res The presence of like poles in close proximi ty makes this an undesired high energy state This is called a magnetic frustration.
57 Fabrication Process 5.3 The electron beam lithography process began by coating a Si wafer with a resist, namely PMMA. This was accomplished by using a Laurell Technologies WS 400A 8NPP/Lite Spin Processor which spun the wafer at high speeds causing the PMMA resist to spread evenly over the entire wafer. A bi layer PMMA recipe using a 4 95K molecular wei ght as the bottom layer and 950K molecular weight as the top layer was chosen The bottom layer was spun, then baked, followed by the top layer being spun then baked. The purpose of the bilayer PMMA was to ensure a clean liftoff which was aided by the undercutting profile created in the low molecular weight PMMA. A pattern was designed using DesignCAD2000 NT. The most effective line spacing, exposure doses, points, and focus were determined by using diagnostic designs such as the wheel pattern. A sampl e of the Si wafer coated with bi layer PMMA was then loaded into the JEOL 840m retrofitted with the NPGS lithography system and a beam blanker. The SEM was then focused and the stigmation was adjusted for optimal resolution. The pattern was selected and wr itten via the NPGS system. Afterwards, the sample was unload ed and prepared for chemical development. It was placed in a bath of MIBK and Isopropanol ( 1:3 ) for approximately seventy seconds, followed by a twenty second Isopropanol bath, and finally nitroge n dried. Permalloy (Ni80Fe20), was then deposited via a Varian Model 980 2462 Electron Beam Evaporator in a vacuum of approximately 5 Torr at a rate of 2 sec 1 The sample coated with the ferromagnetic material was placed in a heated ultrasonic acetone bath for approximately 15 minutes which selectively lifted off the unwanted PMMA and left t he fabricated magnetic structure intact
58 Experimental Setup 5.4 The specific ferromagnetic and anti ferromagnetic wires used for the experiment are shown in Figure 5 3 A Veeco DI 3100 was used to collect all Magnetic Force and Atomic Force Microscopy data. SEM metrology was accomplished using a Hitachi S 800. The dimension each nano magnetic cell was approximately 100 (l) x 50 (w) x 30 (th ) nm. The spacing between each cell was approximately 20 nm and the nano magnets were not annealed. An electromagnet powered by a 300W regulated dc power supply was used to provide uniform magnetic fields of approximately 200mT (2000 Gauss). A NIST traceab le Gauss meter from AlphaLab Inc. was used to measure the fields provided by the electromagnet. The remanent magnetization was recorded via Magnetic Force Microscopy and the binary states of a nano magnetic cell were enumerated based on the alignment of th e magnetic dipole moment along the easy axis. The following procedure was used to provide the different directions of external magnetic excitations to the MCA wires. The sample was placed inside of an electromagnet which had an approximate re manent field of 0T. It was followed by a (A) (B) Figure 5 3 (A) is an SEM image of the ferromagnetic wire and ( B) is of the anti ferromagnetic wire. The dimensions of the magnets were approximately 100 (l) x 50 (w) x 30 (th) nm 3
59 (A) (B) Figure 5 4 ( Spin magnetic field ) (A) is a MFM image of the ferromagnetic wire and (B) is the anti ferromagnetic wire. There were very few frustrations in (B) and the ferromagnetic wire had no frustrations. (A) (B) Figure 5 5 (Z axis magnetic field ) (A) is the ferromagnetic wire and (B) is anti ferromagnetic wire. The number of incorrect alignments increas ed in (B). The ferromagnetic wire had no frustrations. (A) (B) Figure 5 6 (Easy axis magnetic field) (A) is the ferromagnetic wire and (B) is anti ferromagnetic wire. In (B) there was a significant amount of frustrations. For anti ferromagnetic wires, this type of clocking field is was undesirable. The magnetic moments must overcome the shape anisotropy energy to align in an anti ferromagnetic fashion. The ferromagnetic wire had no frustrations. (A) (B ) Figure 5 7 (Hard axis magnetic field) (A) is the ferromagnetic wire and (B) is anti ferromagnetic wire. In (A) an interesting configuration occurred as pointed out by the moment appeared to be void. It can could be seen that the cell is was surrounded by like poles, hence forcing the peculiar state.
60 ramp up period of approximately 2 seconds to 200mT. The sample was left in the saturating field for 5 seconds, followed by a ramp down over 2 seconds. Simulations were calculated using the NIST provided OOMMF tool. The following parameters were used to simulate Permalloy systems via OOMMF; magnetic saturation (M s ) 860e3 A/m, exchange stiffness (A) 13e 12 J/m, and a damping co efficient 0.5. Results 5.5 In Figure 5 4 through Figure 5 7 the MFM images depict the remanent magnetization of both the ferromagnetic (top) and anti ferromagnetic (bottom) MCA wires. Each individual mag net has been overlaid with an outline to help interpret results. A blue outline was used to point out an undesired condition, i.e. a frustration. In Figure 5 4 the wires were rotated while inside of the magnetic field, similar t o the method used by Bernstein et al. [ 24 ] The magnetic field was provided to both wir e arc hitectures using the procedure outlined above. Afterwards the nano magnets were allowed to settle in to the lowest possible ground state. As shown in Figure 5 4 there Figure 5 8 is an MFM image where a magnetic field was provide to ferromagnetic wires along the hard axis. The bottom most wire was the same wire shown in Fig 5 7. Although that particular wire had frustrations, the three neighboring wires above it did not. Figure 5 9 is a graph which s ummarizes the percentage of frustrations per wire based on the direction of the magnetic field. The field strength for all fields was approximately 200mT (2000 Gauss).
61 w ere a few frustrations present in the anti ferromagnetic wire architecture. Interestingly, the ferromagnetic wire was able to reach the lowest possible energy state with no frustrations present. In Figure 5 5 the magnetic field was provided along the z axis of the wires (out of plane hard axis) This aligned the magnetic moments in a high energy state due to the shape anisotropy of the nano magnets. The field was removed and the wires were allowed to settle. It was evident that the number of frustrations present in the anti ferromagnetic wi re architecture increased, whereas the ferromagnetic MCA wire architecture was able to settle with no frustration s Similarly, in Figure 5 6 a magnetic field was provided along the easy axis of the magnets for both wire architect ures. In this instance, the differences in behavior between the architectures became clearly evident. For the anti ferromagnetic wires this was the least desirable type of field to provide. It forced the magnetic moments of each magnet to arrange itself in a high energetically unfavorable state, by forcing repelling poles to exist next to each other. Once the field was removed, the repelling forces of like poles must overcome the shape anisotropy energy to align in a more energetically favorable anti ferrom agnetic state. It was clear that the number of frustrations in Figure 5 6 for the anti ferromagnetic wire architecture, was significantly worse. In the case of the ferromagnetic wire architecture though, this would clearly be the preferred type of field to provide. Since the magnetic field lies along the easy axis of the nano magnets, this would lead to the least amount of frustrations for the ferromagnetic wire architecture. In Figure 5 7 the magnetic f ield was provided along the in plane hard axis of the nano magnets. This forced the magnetic moments to align themselves in an unfavorable
62 state due to the shape anisotropy of the nano magnets. Once the magnetic field was removed, the magnetic moments atte mpted to align themselves along the easy axis of the nano magnetic cells. In the anti ferromagnetic wire, there were several frustrations present. The ferromagnetic wire also showed some susceptibility to magnetic frustrations, with a few undesired states. A very interesting configuration is pointed out by an arrow in Figure 5 7 There were two similar poles on either side of the nano magnet. The magnetic moment for the trapped nano magnetic cell appears to absent. It is possible t hat the stray field of the out of plane magnetic moment was too weak for the MFM to detect. In Figure 5 8 the ferromagnetic wire from Figure 5 7 was the bottom most wire. As can be seen in the image, th ere were three neighboring wires above it. The MFM image shows all three neighboring wires were able to reach a ground state with no magnetic frustrations. So although the ferromagnetic wire in Figure 5 7 did have some frustration s present when providing a field along the in plane hard axis, neighboring wires did not. The graph in Figure 5 9 summarizes the number of frustrations for both architectures for a given type of magnetic field. This experimental data suggested that the ferromagnetic MCA wire architecture could be more robust to fabrication defects than the anti ferromagnetic wire and was more likely to reach a lowest possib le ground state than the later. OOMMF solved a given problem by integrati ng the Landau L ifshitz equation (5 1) where M was the point wise magnetization, H eff was the effective magnetic field, was the gyro magnetic ratio, and was the damping co efficient H eff was calculated in Eq.
63 (5 2 ) where E was the total energy of the system. The total energy E calculated in Eq. (5 3 ) described the total average energy density of all the magnetically active elements where; E z was the Zeeman energy, E d was the demagnetization energy, E ex was the exchange energy, and E A was the anisotropy energy [ 92 ] ( 5 1 ) ( 5 2 ) ( 5 3 ) The total energy density of all magnetic elements ( E ) from the simulation is highlighted in dark blue in Table 5 1 for both th e desired ground state (G) and a meta stable state (M eta ) for anti ferromagnetic and ferromagnetic architectures. Interestingly, it was worth noting that the total energy of both possible states for the anti ferromagnetic scheme was lower than the total en ergy of the equivalent ferromagnetic state. At first glance this would lead one to believe anti ferromagnetic wires would act more reliably, but this does not capture the observed results. In the experiments above, Table 5 1 provides various energies associated with a two nano magnet ferromagnetic and anti ferromagnetic system. It captured the two possible states of both schemes, namely the ground state and meta stable state. Architecture States (G=ground, M S =meta stable) Exchange Energy (J /m^3) Demag Energy (J /m^3) Total Energy (J /m^3) (M eta ) 3255.46 56798.92 60054.38 ( G ) 3918.31 43112.37 47030.69 ( G ) 1365.99 50932.46 52298.45 (M eta ) 2687.185 67491.31 70178.49
64 the ferromagnetic scheme was able to mini mize the total number of frustrations and therefore operate in a more reliable manner. These results are attributed but not limit too the ferromagnetic wires ability to mitigate frustrations more so than the anti ferromagnetic wires due to the energy dif ference between the two possible states for the local architectures, also known as the kink energy. The kink energy was defined in Eq. (5 4 ) where; U MS was the energy associated with the meta stable (frustrated) state, and U G was the energy associate d wi th the desired ground state. ( 5 4 ) Figure 5 10 shows an abstraction of the experiment carried out above where the external magnetic field was given along the z axis. When the system was forced to align itself along the z axis, the total energy was very high, and once the external field was removed both architectures attempt to minimize their local energies. As mentioned above, the total ener gy of the desired state for the anti ferromagnetic wire was lower than that of the ferromagnetic wire. Nevertheless, this was not considered by the locally implemented architectures, only the kink energy or the energy between the possible ground and meta s table states for the particular architecture was present at the time the external field was removed. As shown in Figure 5 10 the kink energy for the ferromagnetic wire was much greater than the anti ferromagnetic wire, and theref ore, the ferromagnetic architecture had a higher propensity to mitigate undesired frustrations. The nucleation of frustrations in the experiments for both wire architectures were primarily attributed to a multiple driver scenario. Without an input driving the system, it was highly probable for a wire to have multiple drivers. This could be explained by
65 picturing a wire in a clocked state along the hard axis. Once the magnetic clocking field was removed, any nano magnetic cell in the chain could become the driver to its neighbor. In fact, multiple drivers could occur throughout a wire, in opposing directions. This lead to magnetic frustrations and was not representative of the ideal operation of a MCA system. A more ideal ferromagnetic wire s ituation was implemented in a simulation and is shown in Figure 5 11 and Figure 5 12 A driver was present as well as a clocking field. In Figure 5 11 and Figure 5 12 the leftmost magnet was used as the driving input cell by providing an external input field to this particular magnet only. The rest of the wire was in a clocked state, where a saturating magnetic field was provided along the in plane hard axis. The clock was released instantaneously and the wire was allowed to settle. As shown in Figure 5 11 and Figure 5 12 the ferromagnetic wire settled according to the driver. For the experiment above, providing the clocking field along the Figure 5 10 shows the po ssible states for the ferromagnetic and anti ferromagnetic wires and over the duration of the experiments. The higher kink energy for the ferromagnetic wire provided an explanation of its greater ability to mitigate unwanted frustrations.
66 in plane hard axis lead to only a few frustrations, but in simulation with an input present the wire operated as desired. Conclusion 5.6 T wo types of MCA wire architectures were fabricated ; ferromagnetic and anti fer romagnetic. By providing clocking fields to the wires, and investigating the remanent magnetic dipole moment s the results revealed ferromagnetic wires to exhibit exceptional data propagation. T frustrati ons over the anti ferromagnetic wires was attributed to the higher kink energy of the ferromagnetic scheme. Providing a clocking field along the z axis of the nano magnets, where experimentally there were no frustrations found in the ferromagnetic architec ture, would eliminate possible clocking complexities encountered in dense layouts that contain in plane wires at different angles. A practical and feasibly simplistic clocking method could then be implemented for ferromagnetic wire architectures by providi ng the clocking field along the thickness of the magnetic cells. Figure 5 11 shows three snapshots of a (a) the ferromagnetic wire was in an initial As can be seen at time step (c) the chain completely switched to match the driver cell. Figure 5 12 shows three snapshots of a (a) the ferromagnetic wire was in an initial As can be seen at time step (c) the chain completely switched to match the driver cell.
67 Magnetic Cellular Automata does not have to be limited to just one type of wire architecture per design though Indeed, the ferromagnetic wires could be mixed with the anti ferromagnetic wi res in the same design and there could be some circumstances where one would be more desirable than the other. The data suggest ed that for relatively long data propagation, the ferromagnetic wire architecture would be the preferred implementation for MCA s ystems.
68 Chapter 6 Magnetic Cellular Automata Coplanar Cross Wire Systems Introduction 6.1 The novelty of Cellular Automata ( CA ) and the associated fabrication difficulties created a great opportunity for experimental research to be conducted. An Electronic Quantum Cell ular Automata (EQCA) proof of concept cell was demonstrated to work, but the experiment was conducted at a temperature of 70 mK [ 93 ] Due to the energies involved, the dimensions necessary for EQCA to work at room temperature were approximately a few nanometers. This created a challenge for current fabrication technology. This difficulty allowed for designs such as crossing wires to be theoretically proposed and critiqued but required a physical actualization to make such architectures viable. Tougaw et al. proposed the idea of coplanar wire crossings for EQCA, and explained that by physically rotating the electron sites by 45, as shown in Figure 6 1 (C) two coplanar crossing wires could propagate information successfully [ 94 ] Figure 6 2 (B) shows the crossing wire system propagating information. Wire 1 was propagating a logical 1, while wire 2 was propagating a logic al 0. The cells in wire 2 had a 45 rotation in order to minimize any influence that it cou ld have had on wire 1. Walus et al. explained that the 45 rotated junction cell had a null effect on neighboring cells of wire 1 regardless if it was propagating a 0 or a 1 [ 95 ] Concerns are also expressed over coplan ar crossing wires. Walus et al. states that 45 cross wires breaks the non rotated
69 wire into several weakly coupled segments [ 95 ] This is due to the energies associated with the columbic inte ractions of EQCA. Bhanja et al. further characterized various cross wire architectures with triple modular redundancy, double triple modular redundanc y and thick crossing wires in terms of polarization loss and thermal stability by modeling the cross wire system as probabilistic Bayesian network model [ 38 ] The coplanar cross Figure 6 1 is a standard abstract representation of MCA and EQCA cells. The top row can could be interpreted as a logical zero and the bottom a logical one. ( A ) e numerates the nano magnetic cells based on their magnetic dipole moments. ( B) d epicts t he standard QCA c ell. Two electrons are represented as red dots. The white dots are the alternate electron sites which the electrons tunnel to in order to reduce the cells overall energy in accordance with its neighbors. ( C ) shows the same EQCA cell only the sites are rota ted by 45. This is necessary for a cross wire layout in EQCA. Figure 6 2 (A) is a n abstract representation of two coplanar crossing wires. The area where the two wires cross is called the junction. Inform ation propag ation is shown via the arrows. (B) is a representation of QCA coplanar crossing wires. For traditional current carrying wires, the junction would create a source of error and this type of structure would not be viable.
70 wire systems were probabilistically inferred under various thermal conditions and the robustness was characterized under single missing cell defects. T he architecture was ported into a MCA construct by designing a ferromagnetic coplanar cross wire system. Figure 6 3 shows an abstract representation of the ferromagnetic cross wire system that was designed. It was composed of two perpendicularly intersecting ferromagnetic coupled wires. A s depicted by the green arrows, information was propagated along the two wires without a loss of information. T he MCA cross wire systems were experimentally studied under various magnetic fields and the results are presented in the following sections Desi gn 6.2 The design of the ferromagnetic coupled cross wire system is depicted in Figure 6 3 The orthogonal wires allow ed for minimal interference between the crossing wires. The shape anisotropy created a preferred axis of magnetizat ion which was also the ground states of the nano magnet. The energy necessary to align the magnetic moment a long Figure 6 3 is a n OOMMF simulation of our MCA cross wire system design. There are two ferromagnetic coupled wires outlined by the green arrows. The magnetic dipole moments of each nano magnet ic cell is represented by the grey arro ws. The magnet moments lie along the easy axis of the nano magnets.
71 the hard axis was very high therefore the wires were designed perpendicular to one another. The magnetic fields produce d by wire 1 interacted with the hard axis of the cells in wire 2, which mitigated undesired interference coupling. The following parame ters were used to simulate the P ermalloy systems via OOMMF; a magnetic saturation of (Ms) 860e3 A/m, an exchange stiffness of (A) 13e 12 J/m, and a damping co efficient of the magnets. It was then allowed to settle into the ground state depicted in. It is worth noting that the interaction of the intersecting wire s was neutral du ring the operation of the cross wire system due to the nature of ferromagnetic coupled wires. As can be seen in Figure 6 2 at the junction both north and south poles were presented to the other wire, which in essence produced a n et cancelling effect. Thus, the wires were able to cross in the same plane with virtually no interference. Fabrication Process 6.3 A Si wafer was coated with PMMA via a Laurell Technologies WS 400A 8NPP/Lite Spin Processor. It spun the wafer at high speeds ca using the PMMA resist to spread evenly over the entire wafer. A single thin layer of 950K molecular weight PMMA was used Afterwards it was baked in a n oven at 170C to provide even heat over the entire wafer. This evaporate d and removed any solvent left in the resist to ensure resi st purity, which was critical to the nano meter electron beam lithography process The cross wire pattern s were designed using DesignCAD2000 NT. The most effective line spacing, exposure doses, points, and focus were determined b y using the diagnostic wheel pattern. A sample of the Si coated with 950 K molecular weight PMMA was then loaded into the JEOL 840m retrofitted with the NPGS lithography system and a beam blanker. The SEM was then focus ed and the stigmation was adjusted for optimal results.
72 The pattern was selected and written via the NPGS system. Afterwards, the sample was unload e d and was ready to be developed. For this particular process the developer, MIBK: Isopropanol 1:3 was cooled via liquid nitrogen. The sample was then developed at the sub zero temperatures. This allowed for glassing to occur during the molecular chain scission process [ 81 ] This result ed in sharper resist edges after development which translated to better defin ed edges for the nano magnets. Permalloy was then deposited, via the Varian Model 980 2462 Electron Beam Evaporator. A vacuum of approximately was achieved and an evaporation rate of 2 sec 1 was used. Once the desired thickness of the material w as deposited the liftoff process was done The sample coated with Permalloy was placed in a heated ultrasonic acetone ba th for approximately 15 minutes. Experimental Setup 6.4 T wo cross wire s similar to the design shown in Figure 6 2 were fabricated A Veeco DI 3100 was used to collect all Magnetic Force and Atomic Force Microscopy data. SEM Figure 6 4 is a SEM image of the two coplanar cross wire system. Each cell was made of permalloy rectangles with the dimensions of 200(l) x 100(w) x 30(th) nm 3
73 metrology was accomplished using a Hitachi S 800. Figure 6 4 shows a topological SEM image of a simple cross wire syst em where each wire was composed of ten nano magnets. Figure 6 5 is a 3D representation of an AFM image taken of the same cross wire structure. As can be seen in the SEM image, there were a few lateral spatial irregularities as we ll as some irregularly shaped cells. On average, most cells were approximately 100 x 50 nm 2 with a spacing of 20 nm between each cell. This was a non ideal physical implementation of a ferromagnetic crosswire system, but if successful, w ould demonstrate th e robustness of the system with respect to fabrication defects. The following procedure was used to provide the various external excitations to the magnetic system via an electromagnet. The sample was placed inside the electromagnet followed by a ramp up p eriod of approximately 2 seconds to 200mT along the thickness of the magnets. The sample was left in the saturating field for 5 seconds, followed by a ramp down over 2 seconds. Figure 6 5 is a n AFM image of the two coplanar cross wire system. The AFM height image captured the surface irregularities of the crosswire system.
74 Simulations were calculated using the NIST provided OOMMF tool. The following pa rameters were used to simulate p ermalloy systems; magnetic saturation (M s ) 860e3 A/m, exchange stiffness (A) 13e 12 J/m, and a damping co efficient 0.5. Results 6.5 A cross wire similar to the one shown in Figure 6 3 was fabri cated Figure 6 4 shows a topological SEM image of the cross wire system where wire 1 and wire 2 were each composed of ten nano magnets. Figure 6 5 is a 3D representation of an AFM image taken of the same cross wire structure. By combining these types of metrology, a more accurate representation of the cross wire structure was obtained I n the SEM image, there were a few lateral spatial irregularities, as well as, some irregularly shaped cells. On average, most cells were approximately 100 x 50 nm 2 with a spacing of 20 nm between each cell. The AFM data allowed the surface roughness and thickness of the nano magnets to be determined I n the 3D AFM image, the surfaces of the nano magnetic cells were visibly non uniform, with several peaks covering the surfaces. Previously, it was reported that surface roughness led to faulty data propagation for anti ferromagnetic wires, here, the intent was to determine if surface roughness had a Figure 6 6 is a MFM image of all possible combinations for a simple cross wire system White ellipses have been overlaid to outline each nano magnet. The arrows repr esent the flow of data propagation for each wire.
7 5 significant role in the cou pling of ferromagnetic wires as well [ 91 ] Bryan et al. also noted that edge roughness increased the coercivity of rectangular nano magnetic structures [ 96 ] The thicknesses of the cells are approximately 35 nm with a maximum peak height of 104 nm. We have determined, via bearing analysis, that approximately 55% of the surface is covered with peaks greater than 10 nm above the thickness. This presents a less than ideal case for the physical implementation of a ferromagnetic cross wire system, but if successful, could d emonstrate the robustness to fabrication defects of the system. Figure 6 6 shows the four possible combinations for data propagation in a two coplanar cross wire system. An external magnetic field was provided for stimulus and the n removed. Afterward s the system was allowed to settle in an energy minimum. The blue arrows represent the orientation of the magnetic dipole moments along the wires. As can be seen, there were no frustrations present in any four of the combinations and Figure 6 7 is a SEM image of the complex cross wire system. The system consisted of six wires with nine junctions. Each wire was 20 nano magnets long.
76 t he system reacted as expected. It is worth stating the particular system present ed here was a non ideal sample, due to the irregularities mentioned above. also important to mention that in all of the experiments with coplanar cross wires, frustrations at t he most critical area of the system, namely, at the junction were never observed The junction area can be considered as the four nano magnetic cells where the two wires intersect each other, as shown in Figure 6 3 Due to the nat ure of magnets, if the junction performed in a reliable manner, any subsequent nano magnetic cell in the wire would not experience signal loss. Meaning, for MCA coplanar cross wire systems, the wire was not segmented into smaller slices. The magnetization of the cell participating in the junction, once settled, would provide a self gaining effect. This is due to the nano magnet attempting to minimize its internal magnetic energy. For a nano magnet neighboring the junction, the magneto static energy it exper iences would be similar to neighborin g a cell in a traditional wire. In an effort to determine if indeed the hypothesis of wires being self gaining after the junction was true, in other words wires not becoming segmented into smaller sections at each junct ion, a more complex cross wire system was fabricated The MCA system consisted of six wires, each 20 nano magnet s long, as shown in the SEM image Figure 6 7 Each magnet was approximately 100 (l) x 50 (w) x 35 (th) nm 3 and the syste m had a total of nine junctions. As mentioned before, an external field was provided for stimulus, removed, and the n the system was allowed to reach a ground state. The arrows in Figure 6 8 depict the orientation of the magneti c d ipole moment along the wires and there were no frustrations at any of the junctions. In fact, there were no frustrations in the entire system except in one location. This is highlighted in with a blue circle. T his scenario was familiar and only occurred on ce in large complex
77 system. Due to the complex nature of providing inputs and clocking fields at the nano level, currently, the system was left to attain an energy minimum after the external field was removed. During this occurrence, a multiple driver sce nario was created in the wire. As depicted in Figure 6 8 this occurred most frequently near the terminal locations of the wire. T his result was found to bolster the via bility of MCA as a technology. If a complex system such as th is was able to attain an energy minimum on its own without the help of a driving input, it would seem that the system will be very capable of minimizing its total energy. Conclusion 6.6 Unlike EQCA, where there was a rotated cell in the cent er of the junction ( Figure 6 2 (B) ) the design of an MCA cross wire system has a free area of space shown in Figure 6 3 The four possible logic combinations of a two coplanar cross wire system were Figure 6 8 is a MFM image of the complex cross wire system. The system consisted of six wires with nine junctions. Each wire was 20 nano magnets long. The blue circle shows the only magnetic frustration in th e system.
78 experimentally demonstr ated even though, physically, the system was less than ideal. This demonstrated the robustness of ferromagnetic coupled cross wires systems towards common fabrication defects Furthermore, a complex cross wire system consisting of 120 nano magnetic cells with nine junctions was fabricated and was able to reach an agreeable minimum energy. The concerns of segmentation of wires due to junctions, as in EQCA, did not seem to manifest themselves in these experimental investigations. The results lead to the conc lusion that MCA cross wire systems were inherently able to reach energy minimums and that ferromagnetic cross wire architectures were likely to be realized in MCA. This could possibly enable the technology to increase the density of switches, develop new l ayout algorithms to optimize space in the design automation process, as well as simplify the fabrication process by removing several multilayer alignment steps for signal propagation thus increase yield.
79 Chapter 7 M agnetic Field Based Computing for Image Processi ng Introduction 7.1 The detection of significant features in an image is a computationally expensive process. The term perceptual organization is used to describe the act of rec ognizing important features of an image. It is generically broken down into three s teps: Segmentation: extraction of features from an image such as edges, pixel intensity, etc. Grouping: relating low level segments of an image into larger perceptual groups such as surface s background, foreground, etc. Recognition: matching groupings ide ntified as significant to a known model such as a building, human, etc. Although each step has its own associated complex ities, grouping is specifically important when reducin g computational intensity. P articularly useful is the fact that grouping does not require exact computational accuracy, meaning that solutions that are near the optimal result are acceptable. The grouping of low level segmen ts can be accomplished through a quadratic energy maximization process [ 40 ] Traditional Boolean architectures found in supercomputers normally reduce problems such as the optimization of quadratic problems into finding exact solutions to arithmetic problems and logical operations This is very demanding on a Boolea n system and is not
80 necessarily required for the grouping of visual objects and magnetic nano systems present a unique way to accomplish such a task [ 40 ] The magnetic energy minimization of a nano syste m can provide sufficiently accurate solutions when variable s are carefully mapped to nano s tructures The correlation between a variable from the segmentation of an image, in this case an edge, can be translated to a nano magnetic disk. The magnetization of th e nano disk along a particular direction can then be related to the saliency of that particular variable (edge). The energy minimization occurs when n eighboring magnets influence magnetization vector through magnetic dipole field interactions. The magnetic dipole moments arrange themselves such that similar poles are far apart reducing the magneto static energy of the system. The reduced form of the Hamiltonian for magneti c energy is given in equation (6 1) where is the magnetization vector of nano disk i is the magnetization of a neighboring nano disk j and is the quadratic energy term in terms of the distance between neighboring magnets given in equation (6 2) [ 40 ] ( 7 1 ) ( 7 2 ) In vision processing edge groupi ng can be accomplished through assigning an affinity between edges based on factors such as the angle between the edges, distance between edge s, and continuity. Therefore when grouping edges based on assigned affinities, the desire is to maximize the a ffinity energy The affinity energy used here is
81 Figure 7 1 (A) is a grayscale image of a group of building s taken from above (B) is the same image post low level segmentation. (C) is the magnetic field based model where the white circles are nano magnetic disk. The spacing between each magnet is carefully placed so that the quadratic dipolar interaction is proportional to the quadratic affinity between edges. Figure 7 2 (A) is a layout of numbered magnetic nano disk which correspond to the e dges of an image. (B) depicts the magnetic initialization of the MFC system, where the red represents a positive pole and yellow represents a negative pole. (C) repr esent a possible ground state of the system after the external field is removed. If the magnetization is taken along axis M, the magnetization component of nano disk 1, 2, and 3, will be larger than those of magnets 4 and 5.
82 given in equation (6 3), where is an edge, is another edge, and is the quadratic affinity term given in equation (6 4) [ 40 ] ( 7 3 ) ( 7 4 ) Proof of Concept 7.2 A M agnetic F ield based C omputing (MFC) system is able to solve the edge grouping quadratic problem through the natural minimization of magnetic energy. By carefully spacing the magnetic cells such that the dipolar interaction energy i s inversely propor tional to edge affinity energy the magnetization vector of the cells c an then be related to the sal iency of that edge [ 40 ] If an image is given for feature detection such as the one in Figure 7 1 (A) Figure 7 1 (B) is th e low level segmentation of that image. The goal is to define a magnetic field based computing (MFC) model, such as the one presented in Figure 7 1 (C), to determine the significant features of the segmented image and group them t ogether for the final recognition process. A small MFC system consisting of five nano magnets very similar to a subsection of Figure 7 1 (C) given in Figure 7 2 was the starting point of the study The wh ite circles represent magnetic nano disk s enumerated from left to right. The disk s were made of p ermalloy with the thickness being much less than the lateral dimensions to force in plane single domain magnetic dipole moments. The experiment would proceed b y applying an external magnetizing field along a particular direction as shown Figure 7 2 (B). Afterwards the field would be removed and the magnetic cells would be allowed to
83 settle into a ground state as shown in Figure 7 2 (C). The natural tendency for magnetic energy to be minimized, in this case primarily due to magneto static coupling, would lead magnets 1 3 to arrange themselves in a ferromagnetic fashion. It was also possible for magnet 4 to exper ience a degree of c oupling with magnets 1 3 which c ould be demonstrated via an anti ferromagnetic coupling. The coupling exhibited by magnets 1 3 would be intentional since the inter spacing distant was smaller than that of magnets 4 and 5, and would be part of the computational linear placement algorithm [ 40 ] If the magnetization vector of each magnet is taken along axis M as shown in Figure 7 2 (C), the magnetization of magne ts 1 3 would be greater than those of magnets 4 and 5. As mentioned previously, this magnetic interaction can be modeled via a quadratic term and is computational intensive but occurs naturally in the physical world at least on the order of nano seconds [ 97 ] The magneto static quadratic term was inversely proportional to the qu adratic edge affinity which determines the salient feature s of the image. So by setting appropriate affinity thresholds the edges corresponding to magnets 1 3 would be regarded as a significant grouping while magnets 4 and 5 would not. F abrication Process 7.3 A Si wafer was coated with PMMA via a Laurell Technologies WS 400A 8NPP/Lite Spin Processor. It spun the wafer at high speeds causing the PMMA resist to spread evenly over the entire wafer. A single thin lay er of 950K molecular weight PMMA was used Afterwards it was baked in a co nvection oven at 170C to provide even heat over t he entire wafer. This evaporated any solvent left in the resist to ensure resi st purity, critical to the nano meter electron beam lithography process.
84 The MFC systems were designed using DesignCAD2000 NT. The most effective line spaci ng exposure doses points, and focus were determined by using diagnostic wheel pattern. A s ample of the Si coated with 950K molecular weight PMMA was then loaded into the JEOL 840m retrofitted with the NPGS lithography system and a beam blanker. The SEM w as then focus ed and the stigmation was adjusted for optimal results. The pattern was selected and written via the NPGS system. Afterwards, the sample was unload ed and was ready to be developed. For this particular process the developer, MIBK: Isopropanol 1 :3 Permalloy was then deposited, via the Varian Model 980 2462 Electron Beam a 2 sec 1 Once the desired thickness of the material was deposited; the liftoff step was n ext The sample coated with Permalloy was placed in a heated ultrasonic acetone bath for approximately 15 minutes. Figure 7 3 (A) is a CAD design of nano disk in microns. Critical dimensions are given as 36nm in between 100nm disks. (B) depicts the corresponding fabricated pattern.
85 Results 7.4 As mentioned above, the size, spacing, and placement all correlated to features of an image ; therefore proper fabrication of a design was imperative. Figure 7 3 (A) and (B) is a CAD design and SEM image, respectively. The purpose of this figure is to demonstrate the degree of accuracy in which the desired layout was fabricated On a larger scale though there were instances where some common ir reg ularities associated Figure 7 4 (A) shows the phase graph and magnetic image of a single domain magnet taken vi a a MFM. (B) is the same for the vortex state which is identified by the eye in the center of the magnet. Notice that the repulsive force at the eye of the vortex was much weaker than the attractive force experienced around the eye. In fact the force exper ienced at the eye was on the same order of magnitude as the background noise.
86 with the electron beam fabrication process occurred It was important to explore how all these para meters affect the ability of a MFC system to couple and therefore extract salient f eatures. In this section the possible states of a nano magnet in the system were discussed and how physical dimensions of a nano magnetic affect ed the overall coupling. The magnetic nano disks used throughout the experiments were made of p ermalloy which had uniaxial magnetocrystalline anisotr opy. The circular shape combined with the magnetocrystalline anisotropy allow ed for a nano disk to have an easy axis of magnetization of 360 degrees. For the proof of concept the thickness was reduced to try to induce an in plane single domain magnetic mo ment. The in plane moments would couple with neighbors through magneto static interactions. Figure 7 4 are M FM phase graphs with the corresponding image of the possible states. There was an interesting dynamic between the formatio n of a vortex or single domain moment in nano disks which was a formulation of ma terial, diameter, and thickness Indeed, a complete implementation of magnetic field computation could benefit from the ability to switch between single domain and v orte x states particularly because the vortex state had no stray field. This would essentially remove any participation of a nano disk ( in the vortex state ) from the computational process F or the purpose of these experiments the goal was to create in plane dipole interactions through single domain magnetic moments A simple MFC system was chosen for the first experiment and shown in Figure 7 5 The p ermalloy nano disks had a thickness of approximately 32.527 nm and a diameter of a 156.56 nm. The purpose of this experiment was to see if the three leftmost magnets would exhibit strong coupling while the two rightmost magnets exhibit a
87 weaker coupling. The stronger coupling would be due to the closer spacing and Figure 7 5 (B) shows the magnet ic image of the system. In the MFM image certain magnets were in a vortex state while other s w ere not It was clear that the desire coupling between the leftmost magnets was not evident. T he magnets were expected to be in a similarly oriented single domain magnet moment which demonstrat ed the coupling between the states. The fact that there were vortex states in the system in t his situation was undesired and was most likely due to the thickness to radius ratio. Although there ha s been research in the formation of vortex and single domain moment s most of this work was studying the nano disk as decoupled entities Indeed most work had spaced the nano disk far enough apart such that magneto stat ic interaction from neighbors c a n be neglected. T he interest of the MFC systems presented in this study was to have nano disk couple though single domain magneto static interactions. Investigating how the potential e nergy of a neighboring magnet affect ed the state of a nano disk and the f ormation of vortices was necessary. Figure 7 5 (A) is an AFM height amplitude image and corresponding graph. (B) is the MFM image of the first simple MFC system. It was difficult to interpret if any coupling had taken place.
88 When considering the fabrication layout and dimensions of the next system the size of the nano disk was reduced to remove the possible formation of a vortex state. Also, an elongated structure was introduced (magnet 1) to be used as a driver of the coupled magnets (magnets 2 and 3) The shape anisotropy energ y of the elongated magnet force s a magnetization d irection along the long axis and therefore if neighboring magnets were coupled they should respond to the driver cell. Figure 7 6 shows a force plot of the AFM amplitude image to the right. The average dimensions of the nano disk s were approximately 130 nm in diameter and 20 nm thick. The reduction of dimensions should reduce or eliminate vortex states in the system and force all the magnetic moments to be single domain As expect ed, all the nano disks were in single domain config urations with the magnets 1 3 exhibiting strong coupling Figure 7 7 are MFM images of the second MFC system scanned in various directions sequentially. Figure 7 7 (A) is a typical MFM scan of the sys tem with the slow scan direction from bottom to top and the fast scan direction from left to right as depicted by Figure 7 6 is an AFM height amplitude image and corresponding graph of the second simple MFC system. As can be seen in the graph the diameter of the measured nano disk i s 131.75 nm and the thickness is 21.658 nm. This was a 15.8% reduction in diameter and 33.4% reduction of thickness when compared to the first system.
89 the arrow. As desired, the three leftmost magnets seemed to be in a single domain state. Moreover, the general angle of the magnetization for magnets 1 3 was in the same direction. This demonstrated ferromagnetic coupling order a nd therefore magnets 1 3 were conside red to be strongly coupled. I f a magnetic analysis of the state of the entire system were made these three magnets would be considered strongly coupled along axis M of Figure 7 2 (C) The high degree of magnetization along the M axis represent s a salient relationship. To ensure that the magnetization was not due to the scanning of the MFM tip, meaning that the tip was not significantly altering the state of coupling between neighbors along the scan direction, Figure 7 7 (B) and (C) are images with the scan direction at 90 degrees and 180 degrees respectively. From these sets of images it was clear that the set of three magnets were strongly coupled and were not Figure 7 7 (A) is a MFM scan of the second MFC system at a 0 degree scan angle. The imag es have been rotated to ease interpretation and the arrows represent the scan paths. The slow scan direction was at the tail end and the fast scan direction is depicted by the head of the arrow. The magnets are referred to as magnets 1 5 from left to right (B) is the same system scanned at 90 degrees. Notice the magnets 1 3 retained their magnetization from (A) while magnets 4 and 5 were altered by the scan. (C) is a scan at 180 degrees. Magnets 1 3 retained their magnetization due to the ferromagnetic cou pling between each other.
90 being aligned by the scanning of the tip. This demonstrated the ability of nano disks to establish single domain moment s and couple based on placement. When analyzing the two right most magnets (magnets 4 and 5) wh ich should be in a decoupled single domain state due to the distance in between their nearest neighbors, some abnormal magnetic states are depicted in Figure 7 7 M agnet 4 of the two decoupled nano disk s in Figure 7 7 (A) exhibited a pinw heel type magnetization At first glance it did not appear to be either of the possible magnetic configurations discussed earlier in the chapter, namely single domain or a vortex. This was due to sam ple tip interaction whi ch caused the nano disk to flip its magnetization during data acquisition Even though the tip alter ed the magnetic state of the nano disk the single domain moment could still be extrapolated via the presence of the strong dipoles. By retracing the scanni ng process of the tip, as represented by the arrow, the single domain moment could be followed as it flips throughout the scan from bottom to top. The stray field from the scanning tip was sufficiently strong to flip the magnetization of the magnet Figure 7 8 is a AFM amplitude image of magnet 4 of the second MFC system (right). As can be seen in the graph, the magnet is in fact elliptical. This creates an easy axis of magnetization in the horizontal direction which is along the blue line.
91 approximately 3 4 times during the scan. Even t hough the magnet was in a single domain configuration, because it was decoupled from neighboring elements, the magnetic stray field from the tip was sufficiently strong to alter the state of the nano disk. When the leftmost nano disk was examined (magnet 5) it appeared as if there were two smaller magnets inside of the nano disk aligned from bottom to top in Figure 7 7 (A) Again the interpretatio n of this magnet was not obvious. Figure 7 7 (B) can help deduce what type of magnetic state that magnet 5 was in By following the scan direction in Figure 7 7 (B) a single domain state could be resolved by using similar logic mentioned in the paragraph above for magnet 4 Although this did explain the situation in Figure 7 7 (B) it did not explain the magnetic state depicted in Figure 7 7 (A) or (C) fo r the same magnet. It also did not explain the magnetic configuration shown in Figure 7 7 (B) for magnet 4. In order to interpret the magnetization in the aforementioned example the shape of the magnets must be account ed for as w ell as the scan direction. Figure 7 8 and Figure 7 9 are AFM amplitude scan s of magnets 4 and 5 respectively. In Figure 7 9 is a AFM amplitude image of magnet 5 of the second MFC system (right). As can be seen in the graph the magnet is in fact elliptical. This created an easy axis of magnetization in a vertical dire ction which is along the blue line.
92 those images the p urple line represented the shorter axis and the light blue line was draw n across the longer axis. After closer inspection the shape was found to be more elliptical than circular. Figure 7 8 shows magnet 4 having a longer horizontal axis by approximately 24 nm, while Figure 7 9 shows magnet 5 to have a longer vertical axis by 19 nm. The elliptical shape created an easy axis of magnetization along the longer axis due to shape anisotropy. When the tip was scanned horizontally along magnet 4 in Figure 7 7 (A) and (C), the single domain moment could be seen before flipping but when scanned along the vertical hard axis, as in Figure 7 7 (B), only one scan of the brighter pole was seen before the magnetization flips. The tip was sca nned over one pole several times (the left half of the magnet) and showed the darker profile Once it reached the opposite pole (the brighter right side) the tip sample interaction forced the like poles of the tip and nano disk to flip the magnetization w hich made the right half of the disk represent the attractive force (darker) as well A similar explanation can be used for the magnetization of magnet 5 in Figure 7 7 The reason that more of the repulsive force (bright pole) cou ld be extracted for magnet 5 in Figure 7 7 (A) than magnet 4 in Figure 7 7 (B) can be explained by the smaller difference between the easy and hard axis. The shape of magnet 5 was closer to a circle than magnet 4 by 5 nm. This reduced the shape anisotropy energy which allow ed for more of the moment to align in the harder axis before completely flipping the magnetization. With this understanding of the shape anisotropy and scan directions it was determined that the decoupled magnets were also single domain in nature W hen interpreting the results from the second MFC system some conclusions were drawn. T he smaller dimension and ratio of diameter to thickness had created the desired single do main magnetizatio ns. The magnets that were more closely spaced display ed
93 ferromagnetic cou pling while the magnets that were further apart were more strongly influenced by sample tip field interactions The distinction of flipping or not flipping the magnets via tip scanni ng was in fact a satisfying way to qualify whether or not magnets were strongly coupled and therefore whether the magnets represent ed a salient relationship. The successful implementation and new understanding of the coupling and scanning dynamics encoura ged the fabrication of a larger more complex system. A third MFC system was implemented and the layout is given in Figure 7 10 (B). The feature sizes of the system were altered, as shown in Figure 7 10 (A ), to investigate changes in the magnetization of the nano disk. It was evident by the layout that this syste m was much more complex and more neighbor neighbor interaction would take place. Figure 7 10 (C) is the corresponding mag netic image of the system where the outline s of the nano disk s have been overlaid in white During the evaluation of the system it was evident that as the thickness of the system was reduced it became much more susceptible to the stray field of the tip. Thi s interaction made it more difficult to interpret results. Some important features to point out were that many of the horizontally coupled magnets display ed a ferromagnetic ordering. It also appeared that a row of ferromagnetic coupled nano disks were coup led to a neighboring row through an anti ferromagnetic arrangement. There were also several instances of vortex states throughout the system as well as many magnets displaying very weak magnetizations. The current hypothesis is that these weaker magnetizat ion s are in anti vortex configurations Another instance of the pattern was fabricated and shown in Figure 7 11 The graph in Figure 7 11 (A) shows a difference of 32 nm in diameter between the third an d fourth system. Interestingly, the sample appeared to be more heavily influenced by the tip s
94 stray field which was demonstrated by the flipping of the nano disk. Also when comparing the magnetic data from this system to the previous one, more magnets were in a single domain configuration. There were also more magnetic moments that were not resolvable in the previous system with the larger nano disk. Currently steps to reduce the tip sample interaction are being taken such as exploring alternative magneti c microscopy methods and fabricating custom magnetic tips to reduce the sample tip interaction. Conclusion 7.5 Magnetic Field based Computing (MFC) has the potential to provide solutions towards the grouping of low level segments into relevant features of an i mage in a unique manner A proof of concept consisting of 5 nano magnetic disks that related to low level edge segments of an image has been provided The proof of concept was fabricated and strictly adhered to a placement algorithm that correlated magneto static interaction s to the edge affinity energies of an image. The significance of creating nano disk to specific dimensions was displayed by the undesired formation of preferred axes of magnetization and the formation of magnetic vortices. The strong mag neto static interaction revealed by nano disks 1 3 was established via the ferromagnetic coupling exhibited by the nano magnets. Nano disks 4 5 were strongly influenced by the MFM tip and caused the magnet moments of the magnets to flip during data acquisi tion. The switching of nano disk 4 5, and the lack of variation in magnetic dipole moments of nano disk 1 3 due to the magneto static coupling, was determined to be a satisfying way of qualifying strong field interactions of magnets. The strong coupling re solved the ed ges that mapped to nano disk 1 3 as significant and could be grouped into a salient feature of an image. The time necessary for the magnetic systems to settle into an
95 energy minimum was the computational time necessary to accomplish the group ing of low level segments. Two larger systems have also been fabricated and are currently being evaluated to investigate the effectiveness of MFC for perceptual organization in computer vision
96 (C) Figure 7 10 (A) is a AFM height amplitude f orce plot of MFC system 3. (B) shows the entire arrangement of nano disk. (C) is the MFM data gathered from the system. White circles were overlaid on the image to relate the physical placement of the nano disk.
97 (C) Figure 7 11 (A) is a AFM height amplitude f orce plot of MFC system 4. (B) shows the entire arrangement of nano disk. (C) is the MFM data gathered from the system.
98 Chapter 8 Conclusion Synopsis 8.1 The use of nano m agnetic switches for computation is a novel application that offers several potential benefits such as a high density of switches, ultra low power devices, integration with MRAM, non volatile operation, and radiation hardness. C orroborative data has been p resented regarding the reliability of magnetic coupling through field interactions and it has been found that nano magnetic cells were robust to common fabrication defects associated to the electron beam lithography process. The data experimental ly verifi ed the functionality of a previously reported Magnetic Cellular Automata (MCA) majority gate and a larger majority gate system was shown to be operational. Furthermore, a new ferromagnetic coupled majority gate based on the experience of ferromagnetic arc hitectures being less prone to magnetic frustrations was introduced A working ferromagnetic majority gate was demonstrated and a logic network of two ferromagnetic majority gates feeding into another ferromagnetic majority gate was also presented Though the results have shown that the ferromagnetic majority gate network always had one magnetic frustration, this could be attributed to the absence of computational synchronization for Magnetic Cellular Automata. The first comparative study of ferromagnetic a nd anti ferromagnetic MCA wire architectures under external clocking fields was also introduced. F our different external
99 magnetic fields to magnetize the system were provided: an in plane easy axis, an in plane hard axis, an out of plane hard axis, and a s pinning field to both a ferromagn etic and anti ferromagnetic wire with magnetic cells of approximately 100 (l) x 50 (w) x 30 (th) nm 3 The external magnetic energy was removed and the wires were allowed to settle into local magnetic energy minimums and ana lyzed for magnetic frustrations. The e xperimental results showed ferromagnetic wires were less susceptible to magnetic frustrations than anti ferromagnetic wires by up to 67 % when magnetized along the in plane easy axis. T he energy associated with two magn etic cells arranged in an anti ferromagnetic and ferromagnetic coupling scenario was simulated and the high energy meta stable state (magnetic frustration) and the low energy ground state were compared The kink energy, which is the energy difference betwe en the meta stable state and the ground state, was 4,856.35 J/m 3 larger for ferromagnetic wire than the anti ferromagnetic w ire architecture The improved mitigation of ferromagnetic wires over anti ferromagnetic wires could be attributed to the larger ene rgy gap between the meta stable state and ground state of ferromagnetic wires. From these results it was concluded that ferromagnetic wires will likely be the main interconnect for MCA. It was determined that providing an out of plane clocking field could eliminate complexities that could arise from having both horizontal and vertical in plane magnetic interactions for computation. The out of plane clocking f ield could allow for magnetic clocking domains to be established and be insensitive to the in plane orientation of data propagation in a MCA circuit. The capacity of having wires cross on the same device plane had been a theorized feature of Quantum Cellular Automata (QCA) yet to be physically realized T he first design and implementation of a Magnetic Cellular Automata coplanar crosswire system
100 was presented here The system was design ed to use the ferromagnetic wire architecture and intersect two wires at a 90 degree angle. The rectangular shape of the magnetic cells created an easy and hard axis of ma gnetization and the 90 degree angle utilized the in plane hard magnetization axis to abate any cross talk interference between intersecting wires. Simulations were implemented to verify functionality and analyze coupling dynamics. The coplanar crossing wir es were then fabricated in two systems; a simple two crosswire system with a total of 20 magnetic cells and a larger complex system consisting of 6 ferromagnetic coupled wires with 120 nano magnetic cells. T he four possible logic combinations of the smalle r system were demonstrated via Magnetic Force Microscopy and results were shown from the larger complex system. The ability for the complex system to settle into a nominal ground state energy and the validation of all the possible logic combinations of th e smaller system established the feasibility of coplanar crossing wires in Magnetic Cellular Automata. A novel unconventional mapping of a vision algorithm to the energy minimization of magnetic dipolar interaction was also presented. Magnetic Field Bas ed Computing (MFC) using circular magnetic nano disk s, were used and correlated to edges of a low level segmented image. The purpose was to use the magnetic nano disk to solve the quadratic problem of grouping edge segments together. Computing edge affinity through accepted Boolean methods is computationally intensive and is often the bottle neck in the perceptual organization of image processing in the area of computer vision. By placing nano disk s at specific locations the quadratic edge affinity term of t he edge grouping equation was effectively mapped to the quadratic distance term of magneto static interactions for nano disk s T he solution of grouping salient edges together was accomplished in the time it takes for the magnetic system to settle into a gr ound state.
101 A simple 5 nano disk system was fabricated and explored how factors such as disk thickness affected the formation of single domain and vortex states. Examples were provided demonstrating that a minimal value of 19 nanome ters caused an elliptica l shaped nano disk and formed an easy and hard axis of magnetization due to shape anisotropy energy, which alter ed the behavior of the MFC system. The correlation of the magnetization vector along a particular axis to the significance of an edge was establ ished and proven to be feasible. T he experimental results of larger more complex system s were also presented and are currently in the process of being analyzed. Future Endeavors 8.2 The synchronization of a larger Magnetic Cellular Automata (MCA) system via a clo cking structure has yet to be realized. This essential feature of MCA is an ongoing area of investigation and we are exploring the use of nanowires and multiple metallization planes to create clocking domains. This will allow for the control of data flow f rom input to output and move the technology closer to a functioning logic device. Also, the ability to provide inputs in an efficient manner is of pronounced importance. Currently MRAM has employed differe nt methods of writing to a magnetic storage cell. W e are presently exploring MRAM writing methods such as spin torque transfer as a means of providing inputs to magnetic cells in MCA. In addition, w e are also exploring alternative magnetic microscopy methods to reduce the interaction between the measurin g apparatus and the sample, such as Lorentz Micro scopy, Electron Holography, and X ray M agnetometry Magnetic Force Microscopy can have strong sample tip interactions with a magnetically soft sample and can alter the state of the nano magnet such as descr ibed when imaging the Magnetic Field based Compu ting systems in Chapter 7. The alternative microscopy methods
102 mentioned above could potentially provide quantitative analysis and yield real time dynamics of a system.
103 List of References 1. Imre, A., et al., Majority logic gate for magnetic quantum dot cellular automata. Science, 2006. 311(5758): p. 205 208. 2. Moore, G.E., Cramming More Components Onto Integrated Circuits. Proceedings of the IEEE, 1998. 86(1): p. 82 85. 3. Kanellos, M. (2005 ) New life for Moore's Law CNET news. 4. Hei, K., et al. A new nano electro mechanical field effect transistor (NEMFET) design for low power electronics in Electron Devices Meeting, 2005. IEDM Technical Digest. IEEE International 2005. 5. Jang, W.W., et al., NEMS switch with 30 nm thick beam and 20 nm thick air gap for high density non volatile memory applications. Solid State Electronics, 2008. 52(10): p. 1578 1583. 6. Pruvost, B., H. Mizuta, and S. Oda, 3 D Design and Analysis of Functional NEMS gate M OSFETs and SETs. Nanotechnology, IEEE Transactions on, 2007. 6(2): p. 218 224. 7. Appenzeller, J., et al., Band to Band Tunneling in Carbon Nanotube Field Effect Transistors. Physical Review Letters, 2004. 93(19): p. 196805. 8. Guangyu, Z., et al. Carbon N anotubes: From Growth, Placement and Assembly Control to 60mV/decade and Sub 60 mV/decade Tunnel Transistors in Electron Devices Meeting, 2006. IEDM '06. International 2006. 9. Knoch, J. and J. Appenzeller, Tunneling phenomena in carbon nanotube field ef fect transistors. physica status solidi (a), 2008. 205(4): p. 679 694. 10. Li, H., et al., Microstructure of cement mortar with nano particles. Composites Part B: Engineering, 2004. 35(2): p. 185 189. 11. Ito, A., et al., Medical application of functionali zed magnetic nanoparticles. Journal of Bioscience and Bioengineering, 2005. 100(1): p. 1 11. 12. Semiconductors, I.T.R.f. (2009) Emerging Research Devices 13. Csaba, G., W. Porod, and A.I. Csurgay, A computing architecture composed of field coupled single domain nanomagnets clocked by magnetic field. International Journal of Circuit Theory and Applications, 2003. 31(1): p. 67 82.
104 14. Imre, A., et al., Majority Logic Gate for Magnetic. Science, 2006. 311: p. 205 208. 15. Gardner, M., The fantastic combinati ons of John Conway's new solitaire game "life". Scientific American, 1970. 223: p. 120 123. 16. Conway, J.H., The game of life. Scientific American, 1970. 17. Lent, C.S. and P.D. Tougaw, Device architecture for computing with quantum dots. Proceedings of t he Ieee, 1997. 85(4): p. 541 557. 18. Tougaw, P.D. and C.S. Lent, Effect of Stray Charge on Quantum Cellular Automata. Japanese Journal of Applied Physics Part 1 Regular Papers Short Notes & Review Papers, 1995. 34(8B): p. 4373 4375. 19. Tougaw, P.D. and C .S. Lent, Quantum Cellular Automata Computing with Quantum Dot Molecules. Compound Semiconductors 1994, 1995(141): p. 781 786. 20. Bernstein, G.H., et al., Practical issues in the realization of quantum dot cellular automata. Superlattices and Microstruc tures, 1996. 20(4): p. 447 459. 21. Gin, A., et al., Hierarchical of design of quantum dot cellular automata devices. Journal of Applied Physics, 1999. 85(7): p. 3713 3720. 22. Csaba, G., A. Csurgay, and W. Porod, Computing architecture composed of next ne ighbour coupled optically pumped nanodevices. International Journal of Circuit Theory and Applications, 2001. 29(1): p. 73 91. 23. Imre, A., et al., Investigation of shape dependent switching of coupled nanomagnets. Superlattices and Microstructures, 2003. 34(3 6): p. 513 518. 24. Bernstein, G.H., et al., Magnetic QCA systems. Microelectronics Journal, 2005. 36(7): p. 619 624. 25. Orlov, A., et al., Magnetic Quantum dot Cellular Automata: Recent developments and prospects. Journal of Nanoelectronics and Opt oelectronics, 2008. 3(1): p. 55 68. 26. Hennessy, K. and C.S. Lent, Clocking of molecular quantum dot cellular automata. Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 2001. 19(5): p. 1752 1755. 27. Momenzadeh, M., M. Ottavi, and F. Lombardi. Modeling QCA defects at molecular level in combinational circuits in Defect and Fault Tolerance in VLSI Systems, 2005. DFT 2005. 20th IEEE International Symposium on 2005. 28. Walus, K. and G.A. Jullien, Design tools for an emerg ng SoC technology: Quantum dot cellular automata. Proceedings of the Ieee, 2006. 94(6): p. 1225 1244.
105 29. Alam, M.T., et al. On Chip Clocking Scheme for Nanomagnet QCA in Device Research Conference, 2007 65th Annual 2007. 30. Alam, M.T., et al. Clocking scheme for nanomagnet QCA in Nanotechnology, 2007. IEEE NANO 2007. 7th IEEE Conference on 2007. 31. Karim, F., et al., Modeling and Evaluating Errors Due to Random Clock Shifts in Quantum Dot Cellular Automata Circuits. Journal of Electronic Testing Theo ry and Applications, 2009. 25(1): p. 55 66. 32. Parish, M.C.B. and M. Forshaw, Physical constraints on magnetic quantum cellular automata. Applied Physics Letters, 2003. 83(10): p. 2046 2048. 33. Parish, M.C.B. and M. Forshaw, Magnetic cellular automata (M CA) systems. Circuits, Devices and Systems, IEE Proceedings 2004. 151(5): p. 480 485. 34. Bhanja, S. and N. Ranganathan. Modeling switching activity using cascaded Bayesian networks for correlated input streams in Computer Design: VLSI in Computers and Processors, 2002. Proceedings. 2002 IEEE International Conference on 2002. 35. Bhanja, S. and S. Sarkar. Switching Error Modes of QCA Circuits in Nanotechnology, 2006. IEEE NANO 2006. Sixth IEEE Conference on 2006. 36. Bhanja, S. and S. Sarkar, Probabi listic Modeling of QCA Circuits Using Bayesian Networks in Nanotechnology, IEEE Transactions on 2006. p. 657 670. 37. Saket, S. and B. Sanjukta, Hierarchical Probabilistic Macromodeling for QCA Circuits in Computers, IEEE Transactions on 2007. p. 174 1 90. 38. Bhanja, S., et al., QCA Circuits for Robust Coplanar Crossing. Journal of Electronic Testing, 2007. 23(2): p. 193 210. 39. Bhanja, S. and S. Sarkar, Thermal Switching Error Versus Delay Tradeoffs in Clocked QCA Circuits. Very Large Scale Integratio n (VLSI) Systems, IEEE Transactions on, 2008. 16(5): p. 528 541. 40. Sarkar, S. and S. Bhanja. Direct Quadratic Minimization Using Magnetic Field Based Computing in Design and Test of Nano Devices, Circuits and Systems, 2008 IEEE International Workshop on 2008. 41. Srivastava, S., S. Sarkar, and S. Bhanja. Error Power Tradeoffs in QCA Design in Nanotechnology, 2008. NANO '08. 8th IEEE Conference on 2008. 42. Srivastava, S., S. Sarkar, and S. Bhanja, Estimation of Upper Bound of Power Dissipation in QCA Circuits. Nanotechnology, IEEE Transactions on, 2009. 8(1): p. 116 127. 43. Cowburn, R.P. and M.E. Welland, Room Temperature Magnetic Quantum Cellular Automata. Science, 2000. 287: p. 1466 1468.
106 44. Csaba, G., et al., Nanocomputing by field coupled nanomag nets. Ieee Transactions on Nanotechnology, 2002. 1(4): p. 209 213. 45. Cowburn, R.P., A.O. Adeyeye, and M.E. Welland, Configurational anisotropy in nanomagnets. Physical Review Letters, 1998. 81(24): p. 5414 5417. 46. Cowburn, R.P. and M.E. Welland, Microm agnetics of the single domain state of square ferromagnetic nanostructures. Physical Review B, 1998. 58(14): p. 9217 9226. 47. Cowburn, R.P., et al., Designing nanostructured magnetic materials by symmetry. Europhysics Letters, 1999. 48(2): p. 221 227. 48. Cowburn, R.P. and M.E. Welland, Analytical micromagnetics of near single domain particles. Journal of Applied Physics, 1999. 86(2): p. 1035 1040. 49. Cowburn, R.P., et al., Single Domain Circular Nanomagnets. The American Physical Society, 1999. 83(5): p. 1042 1045. 50. Cowburn, R.P., et al., Sensing magnetic fields using superparamagnetic nanomagnets. Journal of Applied Physics, 2000. 87(9): p. 7082 7084. 51. Cowburn, R.P., et al., Lateral interface anisotropy in nanomagnets. Journal of Applied Physics, 2 000. 87(9): p. 7067 7069. 52. Cowburn, R.P., Where Have All the Transistors Gone? Science, 2006. 311: p. 183 184. 53. Cowburn, R.P., et al., Single domain circular nanomagnets. Physical Review Letters, 1999. 83(5): p. 1042 1045. 54. Imre, A., et al., Contr olled domain wall motion in micron scale permalloy square rings. Physica E Low Dimensional Systems & Nanostructures, 2003. 19(1 2): p. 240 245. 55. Csaba, G., P. Lugli, and W. Porod. Power dissipation in nanomagnetic logic devices in Nanotechnology, 2004. 4th IEEE Conference on 2004. 56. Imre, A., et al., Flux closure magnetic states in triangular cobalt ring elements. Ieee Transactions on Magnetics, 2006. 42(11): p. 3641 3644. 57. Csaba, G., et al., Activity in field coupled nanomagnet arrays. Internatio nal Journal of Circuit Theory and Applications, 2007. 35(3): p. 281 293. 58. Becherer, M., et al., Magnetic ordering of focused ion beam structured cobalt platinum dots for field coupled computing. Ieee Transactions on Nanotechnology, 2008. 7(3): p. 316 32 0. 59. Tougaw, P.D., C.S. Lent, and W. Porod, Bistable Saturation in Coupled Quantum Dot Cells. Journal of Applied Physics, 1993. 74(5): p. 3558 3566.
107 60. Alam, M.T., et al. Magnetic Logic Based on Coupled Nanomagnets: Clocking Structures and Power Analysi s in Nanotechnology, 2008. NANO '08. 8th IEEE Conference on 2008. 61. Niemier, M., et al., Clocking structures and power analysis for nanomagnet based logic devices in Proceedings of the 2007 international symposium on Low power electronics and design 2007, ACM: Portland, OR, USA. p. 26 31. 62. Kumari, A. and S. Bhanja, Landauer Clocking for Magnetic Cellular Automata (MCA) Arrays. Very Large Scale Integration (VLSI) Systems, IEEE Transactions on, 2010. PP(99): p. 1 4. 63. Niemier, M., M. Crocker, and X .S. Hu. Fabrication Variations and Defect Tolerance for Nanomagnet Based QCA in Defect and Fault Tolerance of VLSI Systems, 2008. DFTVS '08. IEEE International Symposium on 2008. 64. Carlton, D.B., et al., Simulation Studies of Nanomagnet Based Logic Arc hitecture. Nano Letters, 2008. 8(12): p. 4173 4178. 65. Casta, et al., Metastable states in magnetic nanorings. Physical Review B, 2003. 67(18): p. 184425. 66. Landeros, P., et al., Stability of magnetic configurations in nanorings. Journal of Applied Phys ics, 2006. 100(4): p. 044311 6. 67. Yang, T., et al., Magnetic characterization and switching of Co nanorings in current perpendicular to plane configuration. Applied Physics Letters, 2007. 90(2): p. 022504 3. 68. Zhu, F.Q., et al., Magnetic Bistability an d Controllable Reversal of Asymmetric Ferromagnetic Nanorings. Physical Review Letters, 2006. 96(2): p. 027205. 69. Klaui, M., et al., Vortex circulation control in mesoscopic ring magnets. Applied Physics Letters, 2001. 78(21): p. 3268 3270. 70. Klaui, M. et al., Domain wall coupling and collective switching in interacting mesoscopic ring magnet arrays. Applied Physics Letters, 2005. 86(3): p. 032504. 71. Zhu, X., Magnetic force microscopy studies of submicron and nanoscale magnet arrays. Ph.D. dissertati on 2002, McGill University (Canada). 72. Imre, A., EXPERIMENTAL STUDY OF NANOMAGNETS FOR MAGNETIC QUANTUMDOT CELLULAR AUTOMATA (MQCA) LOGIC APPLICATIONS 2005, University of Notre Dame. 73. Allwood, D.A., et al., Magnetic Domain Wall Logic. Science, 2005. 309(5741): p. 1688 1692. 74. Atkinson, D., et al., Magnetic domain wall dynamics in a submicrometre ferromagnetic structure. Nat Mater, 2003. 2(2): p. 85 87.
108 75. Faulkner, C.C., et al., Controlled switching of ferromagnetic wire junctions by domain wall i njection. Magnetics, IEEE Transactions on, 2003. 39(5): p. 2860 2862. 76. Hayashi, M., et al., Current Controlled Magnetic Domain Wall Nanowire Shift Register. Science, 2008. 320(5873): p. 209 211. 77. Xu, P., et al., An all metallic logic gate based on cu rrent driven domain wall motion. Nat Nano, 2008. 3(2): p. 97 100. 78. Allwood, D.A., et al., Submicrometer Ferromagnetic NOT Gate and Shift Register. Science, 2002. 296(5575): p. 2003 2006. 79. Vernier, N. and et al., Domain wall propagation in magnetic na nowires by spin polarized current injection. EPL (Europhysics Letters), 2004. 65(4): p. 526. 80. Thiaville, A., et al., Domain wall motion by spin polarized current: a micromagnetic study. Journal of Applied Physics, 2004. 95(11): p. 7049 7051. 81. Hu, W., et al., Sub 10 nm electron beam lithography using cold development of poly(methylmethacrylate). Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, 2004. 22(4): p. 1711 1716. 82. [cited 2010 11/11]; Nanometer Pattern Gener ation System, a direct write lithography system for Electron or Ion Beam Microscopes]. Available from: http://www.jcnabity.com/ 83. Moreau, W.M., Semiconductor Lithography Principles, Practices, and Materials Micr odevices Physics and Fabrication Technologies, ed. J.J.M.a.I. Brodie. 1988, New York and London: Plenum Press. 84. MicroChem PMMA datasheet 85. Zheng, C., Micro Nanofabrication: Technologies and Applications 1 ed. 2006: Springer. 300. 86. Herring, C. and M.H. Nichols, Thermionic Emission. Reviews of Modern Physics, 1949. 21(2): p. 185. 87. Nonogaki, S., Microlithography Fundementals in Semiconductor Devices and Fabrication Technology 1998, New York: Dekker. 88. Leavitt, J.A., et al., Permalloy stoichiome try by nuclear backscattering. Nuclear Instruments and Methods in Physics Research Section B: Beam Interactions with Materials and Atoms, 1985. 10 11(Part 2): p. 719 721. 89. Jeager, R.C., Introduction to Microelectronic Fabrication Modular Series on Soli d State Devices, ed. Second. Vol. V. 2002, Upper Saddle River Prentice Hall. 90. Kasap, S.O., Principles of Electronic Materials and Devices 2002: McGraw Hill.
109 91. Pulecio, J.F. and S. Bhanja. Reliability of bi stable single domain nano magnets for Cellul ar Automata in Nanotechnology, 2007. IEEE NANO 2007. 7th IEEE Conference on 2007. 92. Gilbert, T.L., A phenomenological theory of damping in ferromagnetic materials. Magnetics, IEEE Transactions on, 2004. 40(6): p. 3443 3449. 93. Amlani, I., et al., Expe rimental demonstration of a leadless quantum dot cellular automata cell. Applied Physics Letters, 2000. 77(5): p. 738 740. 94. Tougaw, P.D. and C.S. Lent, Logical Devices Implemented Using Quantum Cellular Automata. Journal of Applied Physics, 1994. 75(3): p. 1818 1825. 95. Walus, K., G. Schulhof, and G.A. Jullien. High Level Exploration of Quantum Dot Cellular Automata (QCA) in Signals, Systems and Computers, 2004. Conference Record of the Thirty Eighth Asilomar Conference on 2004. 96. Bryan, M.T., D. At kinson, and R.P. Cowburn, Experimental study of the influence of edge roughness on magnetization switching in Permalloy nanostructures. Applied Physics Letters, 2004. 85(16): p. 3510 3512. 97. Kikuchi, R., On the Minimum of Magnetization Reversal Time. Jou rnal of Applied Physics, 1956. 27(11): p. 1352 1357.