Baatar / Roska / Porod Cellular Nanoscale Sensory Wave Computing

1. Auflage 2010
ISBN: 978-1-4419-1011-0
Verlag: Springer
Format: PDF
Kopierschutz: 1 - PDF Watermark

E-Book, Englisch, 249 Seiten

This book is loosely based on a Multidisciplinary University Research Initiative (MURI) project and a few supplemental projects sponsored by the Of?ce of Naval Research (ONR) during the time frame of 2004-2009. The initial technical scope and vision of the MURI project was formulated by Drs. Larry Cooper and Joel Davis, both program of?cers at ONR at the time. The unifying theme of this MURI project and its companionefforts is the concept of cellular nonlinear/neuralnetwork (CNN) technology and its various extensions and chip implementations, including nanoscale sensors and the broadening ?eld of cellular wave computing. In recent years, CNN-based vision system drew much attention from vision scientists to device technologists and computer architects. Due to its early - plementation in a two-dimensional (2D) topography, it found success in early vision technologyapplications, such as focal-plane arrays, locally adaptable sensor/ processor integration, resulting in extremely high frame rates of 10,000 frames per second. More recently it drew increasing attention from computer architects, due to its intrinsic local interconnect architecture and parallel processing paradigm. As a result, a few spin-off companies have already been successful in bringing cel- lar wave computing and CNN technology to the market. This book aims to capture some of the recent advances in the ?eld of CNN research and a few select areas of applications.

Autoren/Hrsg.

Roska, Tamas

Porod, Wolfgang

Baatar, Chagaan

Inhaltsverzeichnis

1;Preface;5
2;Contents;7
3;1 A Brief History of CNN and ONR;9
4;2 Cellular Wave Computing in Nanoscale via Million Processor Chips;13
4.1;2.1 Introduction;13
4.2;2.2 From Standard CNN Dynamicsto the Cellular Wave Computer;16
4.3;2.3 Various physical implementationsof the Cellular Wave Computer;19
4.4;2.4 Virtual Cellular Machine;20
4.4.1;2.4.1 Notations and Definitions;20
4.4.1.1;2.4.1.1 Core=Cell;20
4.4.1.2;2.4.1.2 Elementary Array Instructions;21
4.4.2;2.4.2 Physical Implementation Types of Elementary Core/Cell Array Instructions (A, B, C);21
4.4.3;2.4.3 Physical Parameters of Array Processor Units (Typically a Chip or a Part of a Chip) and Interconnections;22
4.4.4;2.4.4 Virtual and Physical Cellular Machine Architectures and Their Building Blocks;22
4.4.5;2.4.5 The Design Scenario;24
4.4.6;2.4.6 The Dynamic Operational Graph and its Use for Acyclic UMF Diagrams;25
4.5;2.5 Recent, Non-Standard Architecture Combining Spatial-Temporal Algorithms with Physical Effects;26
4.6;2.6 Hints for Architectural Principles for Non-CMOS Nano-Scale Implementations;29
4.7;2.7 Biological Relevance;30
4.8;References;32
5;3 Nanoantenna Infrared Detectors;34
5.1;3.1 Introduction;35
5.1.1;3.1.1 Project Overview;36
5.1.2;3.1.2 Infrared Detectors;37
5.1.2.1;3.1.2.1 Thermal Infrared Detectors;38
5.1.2.2;3.1.2.2 Quantum Infrared Detectors;40
5.1.2.3;3.1.2.3 Radiation-Field Infrared Detectors;41
5.1.3;3.1.3 Detector Characterization;41
5.1.3.1;3.1.3.1 Figures of Merit;42
5.1.3.2;3.1.3.2 Electrical Noise Considerations;43
5.1.3.3;3.1.3.3 Detector Comparison;45
5.2;3.2 Antenna-Coupled MOM Diodes;46
5.2.1;3.2.1 Dipole Antenna;47
5.2.2;3.2.2 MOM Diodes;48
5.2.2.1;3.2.2.1 MOM Diode Design;51
5.2.2.2;3.2.2.2 Point Contact MOM Diodes;52
5.2.2.3;3.2.2.3 Thin-Film MOM Diodes;53
5.2.3;3.2.3 Conduction Mechanisms;54
5.2.4;3.2.4 Substrate and Antenna Effects;59
5.3;3.3 Fabrication;61
5.3.1;3.3.1 Substrate;61
5.3.2;3.3.2 Bonding Pad Fabrication;61
5.3.3;3.3.3 ACMOMD Fabrication;62
5.3.3.1;3.3.3.1 Electron Beam Lithography;62
5.3.3.2;3.3.3.2 Metal Deposition;66
5.3.3.3;3.3.3.3 Packaging;70
5.4;3.4 Detector Characterization;70
5.4.1;3.4.1 Current-Voltage Characteristics;71
5.4.1.1;3.4.1.1 Air-Oxidation;72
5.4.1.2;3.4.1.2 Controlled Oxidation;75
5.4.2;3.4.2 Infrared Response Characteristics;76
5.4.2.1;3.4.2.1 IR Detector Characterization;78
5.4.2.2;3.4.2.2 Polarization Dependence;81
5.4.2.3;3.4.2.3 Antenna Length Dependence;83
5.5;3.5 Comparison to Current Technologies;85
5.5.1;3.5.1 Comparison with Currently Available IR Detectors;86
5.5.2;3.5.2 Integration with CMOS Imaging Chips;86
5.6;References;89
6;4 Memristors: A New Nanoscale CNN Cell;94
6.1;4.1 Introduction;94
6.2;4.2 Background Information on Memristors;95
6.2.1;4.2.1 HP Memristor;100
6.2.2;4.2.2 How to Read and Write Memory States;102
6.3;4.3 Memristive Devices and Systems;106
6.3.1;4.3.1 Potassium Memristor;111
6.3.2;4.3.2 Sodium Memristor;111
6.4;4.4 Lossless Nonvolatile Memory Circuit Elements;113
6.4.1;4.4.1 Memory Capacitor;113
6.4.2;4.4.2 Memory Inductor;117
6.5;References;121
7;5 Circuit Models of Nanoscale Devices;123
7.1;5.1 Introduction;123
7.2;5.2 Vacuum Fluctuations in Nanocircuits;126
7.3;5.3 Mixed Quantum Classical Electromechanical Models;127
7.4;5.4 Circuit Model of a Double-Band Infrared Sensor;130
7.5;References;132
8;6 A CMOS Vision System On-Chip with Multi-Core, Cellular Sensory-Processing Front-End;134
8.1;6.1 Introduction;134
8.2;6.2 Architectural Concept of the Eye-RIS System;136
8.3;6.3 The Eye-RIS Chip;140
8.4;6.4 The Eye-RIS' Front-End: The Q-Eye;141
8.5;6.5 The Eye-RIS Chip in Operation;143
8.6;6.6 Discussion;147
8.7;References;150
9;7 Cellular Multi-core Processor Carrier Chip for Nanoantenna Integration and Experiments;152
9.1;7.1 Introduction;152
9.2;7.2 Algorithmic Considerations;154
9.2.1;7.2.1 Numeric Precision;155
9.3;7.3 Architecture of the Nanoantenna Carrier Chip;155
9.3.1;7.3.1 High-Gain Sensor Readout Channel;156
9.3.2;7.3.2 Digital Processor Architecture;159
9.3.3;7.3.3 Partitioning;159
9.3.4;7.3.4 Control Processor;161
9.3.5;7.3.5 SIMD Processor Array;162
9.4;7.4 Nanoantenna Integration;163
9.4.1;7.4.1 Antenna Coupled Nanodiode Interfacing;163
9.4.2;7.4.2 Physical Integration of the Nanoantenna Array;165
9.5;7.5 Measurement Environment;166
9.6;7.6 Concluding Remarks;166
9.7;References;167
10;8 Circuitry Underlying Visual Processing in the Retina;168
10.1;8.1 Introduction;168
10.1.1;8.1.1 Background Circuit Organization;169
10.1.2;8.1.2 Extreme Complexity of Amacrine Cell Interactions;171
10.1.3;8.1.3 A Dozen Different Representations;171
10.1.4;8.1.4 Each of the Ganglion Cell Outputs Extends over a Specific and Different Space--Time Domain;173
10.1.5;8.1.5 Crossover Circuitry of Vertical Amacrine Cells Affects Bipolar Amacrine and Ganglion Cells;174
10.1.6;8.1.6 The Visual Functional Roles of Crossover Circuitry;176
10.1.6.1;8.1.6.1 Active Surround Mediated by Crossover Inhibition in Ganglion Cells;177
10.1.7;8.1.7 Crossover Inhibition Helps to Distinguish Brightness from Contrast (Molnar et al. 2008);177
10.1.7.1;8.1.7.1 Crossover Inhibition Allows Neurons to Linearly Add Intensities Distributed Across the Receptive Field Center for Ganglion Cells (Molnar and Werblin 2007b);177
10.1.7.2;8.1.7.2 In-layer Interactions are Mediated by GABAergic Pathways;179
10.1.8;8.1.8 Specific Ganglion Cell Circuitries;180
10.1.8.1;8.1.8.1 Directionally Selective Ganglion Cells;180
10.1.8.2;8.1.8.2 Alpha Ganglion Cells;181
10.1.8.3;8.1.8.3 Local Edge Detectors;182
10.1.8.4;8.1.8.4 ON Beta Cells;183
10.2;References;184
11;9 Elastic Grid-Based Multi-Fovea Algorithm for Real-Time Object-Motion Detection in Airborne Surveillance;186
11.1;9.1 Introduction;186
11.1.1;9.1.1 Unmanned Aerial Vehicles;186
11.1.2;9.1.2 Multi-Fovea Approach;187
11.1.3;9.1.3 Airborne Motion Detection;188
11.2;9.2 Independent Motion Analysis;189
11.2.1;9.2.1 Images and Video Frames;189
11.2.2;9.2.2 Background and Objects;189
11.2.3;9.2.3 Global Image Motion Model;190
11.2.4;9.2.4 Motion Detection, Object Extraction, and Global Background Mosaic;191
11.3;9.3 Multi-Fovea Framework: Abstract Hardware Model;192
11.4;9.4 Algorithms;194
11.5;9.5 Corner Pairing Algorithm;197
11.5.1;9.5.1 Block Matching Algorithms;197
11.5.2;9.5.2 KLT Algorithm;198
11.5.3;9.5.3 SIFT Algorithm;199
11.5.4;9.5.4 Global Registration-Based Detection;200
11.5.5;9.5.5 Elastic Grid Multi-Fovea Detector;202
11.6;9.6 Performance of Methods;204
11.6.1;9.6.1 Metrics for Quality;204
11.7;9.7 Comparison;206
11.8;9.8 Summary;208
11.9;References;217
12;10 Low-Power Processor Array Design Strategy for Solving Computationally Intensive 2D Topographic Problems;219
12.1;10.1 Introduction;219
12.2;10.2 Architecture Descriptions;220
12.2.1;10.2.1 Classic DSP-Memory Architecture;221
12.2.2;10.2.2 Pipe-Line Architectures;223
12.2.3;10.2.3 Coarse-Grain Cellular Parallel Architectures;225
12.2.4;10.2.4 Fine-Grain Fully Parallel Cellular Architectures with Discrete Time Processing;226
12.2.5;10.2.5 Fine-Grain Fully Parallel Cellular Architecture with Continuous Time Processing;227
12.3;10.3 Implementation and Efficiency Analysis of Various Operators;228
12.3.1;10.3.1 Categorization of 2D Operators;229
12.3.1.1;10.3.1.1 Execution-Sequence-Variant VersusExecution-Sequence-Invariant Operators;231
12.3.2;10.3.2 Processor Utilization Efficiency of the Various Operation Classes;233
12.3.2.1;10.3.2.1 Execution-Sequence-Invariant Content-DependentFront-Active Operators;233
12.3.2.2;10.3.2.2 Execution-Sequence-Variant Content-Dependent Front Active Operators;235
12.3.2.3;10.3.2.3 1D Content-Independent Front Active Operators (1D Scan);236
12.3.2.4;10.3.2.4 2D Content-Independent Front Active Operators (2D Scan);238
12.3.2.5;10.3.2.5 Area Active Operators;239
12.3.3;10.3.3 Multiscale Processing;239
12.4;10.4 Comparison of the Architectures;240
12.5;10.5 Optimal Architecture Selection;242
12.6;10.6 Conclusions;247
12.7;References;248
13;Index;250