book

Event-Based Neuromorphic Systems

by Shih-Chii Liu, Tobi Delbruck, Giacomo Indiveri, Adrian Whatley, Rodney Douglas

February 2015

Intermediate to advanced

440 pages

15h 37m

English

Wiley

Read now

Unlock full access

1.1 Origins and Historical Context1.2 Building Useful Neuromorphic SystemsReferences

2.1 Introduction2.2 Address-Event Representation2.2.1 AER Encoders2.2.2 Arbitration Mechanisms2.2.3 Encoding Mechanisms2.2.4 Multiple AER Endpoints2.2.5 Address Mapping2.2.6 Routing2.3 Considerations for AER Link Design2.3.1 Trade-off: Dynamic or Static Allocation2.3.2 Trade-off: Arbitered Access or Collisions?2.3.3 Trade-off: Queueing versus Dropping Spikes2.3.4 Predicting Throughput Requirements2.3.5 Design Trade-offs2.4 The Evolution of AER Links2.4.1 Single Sender, Single Receiver2.4.2 Multiple Senders, Multiple Receivers2.4.3 Parallel Signal Protocol2.4.4 Word-Serial Addressing2.4.5 Serial Differential Signaling2.5 DiscussionReferences
3.1 Introduction3.2 Biological Retinas3.3 Silicon Retinas with Serial Analog Output3.4 Asynchronous Event-Based Pixel Output Versus Synchronous Frames3.5 AER Retinas3.5.1 Dynamic Vision Sensor3.5.2 Asynchronous Time-Based Image Sensor3.5.3 Asynchronous Parvo–Magno Retina Model3.5.4 Event-Based Intensity-Coding Imagers (Octopus and TTFS)3.5.5 Spatial Contrast and Orientation Vision Sensor (VISe)3.6 Silicon Retina Pixels3.6.1 DVS Pixel3.6.2 ATIS Pixel3.6.3 VISe Pixel3.6.4 Octopus Pixel3.7 New Specifications for Silicon Retinas3.7.1 DVS Response Uniformity3.7.2 DVS Background Activity3.7.3 DVS Dynamic Range3.7.4 DVS Latency and Jitter3.8 DiscussionReferences
4.1 Introduction4.2 Cochlea Architectures4.2.1 Cascaded 1D4.2.2 Basic 1D Silicon Cochlea4.2.3 2D Architecture4.2.4 The Resistive (Conductive) Network4.2.5 The BM Resonators4.2.6 The 2D Silicon Cochlea Model4.2.7 Adding the Active Nonlinear Behavior of the OHCs4.3 Spike-Based Cochleas4.3.1 Q-control of AEREAR2 Filters4.3.2 Applications: Spike-Based Auditory Processing4.4 Tree Diagram4.5 DiscussionReferences
5.1 Introduction5.1.1 Determining Functional Biological Elements5.1.2 Rhythmic Motor Patterns5.2 Modeling Neural Circuits in Locomotor Control5.2.1 Describing Locomotor Behavior5.2.2 Fictive Analysis5.2.3 Connection Models5.2.4 Basic CPG Construction5.2.5 Neuromorphic Architectures5.3 Neuromorphic CPGs at Work5.3.1 A Neuroprosthesis: Control of Locomotion in Vivo5.3.2 Walking Robots5.3.3 Modeling Intersegmental Coordination5.4 DiscussionReferences
6.1 Introduction: Synaptic Connections, Memory, and Learning6.2 Retaining Memories in Neuromorphic Hardware6.2.1 The Problem of Memory Maintenance: Intuition6.2.2 The Problem of Memory Maintenance: Quantitative Analysis6.2.3 Solving the Problem of Memory Maintenance6.3 Storing Memories in Neuromorphic Hardware6.3.1 Synaptic Models for Learning6.3.2 Implementing a Synaptic Model in Neuromorphic Hardware6.4 Toward Associative Memories in Neuromorphic Hardware6.4.1 Memory Retrieval in Attractor Neural Networks6.4.2 Issues6.5 Attractor States in a Neuromorphic Chip6.5.1 Memory Retrieval6.5.2 Learning Visual Stimuli in Real Time6.6 DiscussionReferences
7.1 Introduction7.2 Silicon Neuron Circuit Blocks7.2.1 Conductance Dynamics7.2.2 Spike-Event Generation7.2.3 Spiking Thresholds and Refractory Periods7.2.4 Spike-Frequency Adaptation and Adaptive Thresholds7.2.5 Axons and Dendritic Trees7.2.6 Additional Useful Building Blocks7.3 Silicon Neuron Implementations7.3.1 Subthreshold Biophysically Realistic Models7.3.2 Compact I&F Circuits for Event-Based Systems7.3.3 Generalized I&F Neuron Circuits7.3.4 Above Threshold, Accelerated-Time, and Switched-Capacitor Designs7.4 DiscussionReferences
8.1 Introduction8.2 Silicon Synapse Implementations8.2.1 Non Conductance-Based Circuits8.2.2 Conductance-Based Circuits8.2.3 NMDA Synapse8.3 Dynamic Plastic Synapses8.3.1 Short-Term Plasticity8.3.2 Long-Term Plasticity8.4 DiscussionReferences
9.1 Introduction9.2 Voltage-Domain Second-Order Filter9.2.1 Transconductance Amplifier9.2.2 Second-Order Low-Pass Filter9.2.3 Stability of the Filter9.2.4 Stabilised Second-Order Low-Pass Filter9.2.5 Differentiation9.3 Current-Domain Second-Order Filter9.3.1 The Translinear Loop9.3.2 Second-Order Tau Cell Log-Domain Filter9.4 Exponential Bias Generation9.5 The Inner Hair Cell Model9.6 DiscussionReferences
10.1 Introduction10.2 Floating-Gate Circuit Basics10.3 Floating-Gate Circuits Enabling Capacitive Circuits10.4 Modifying Floating-Gate Charge10.4.1 Electron Tunneling10.4.2 pFET Hot-Electron Injection10.5 Accurate Programming of Programmable Analog Devices10.6 Scaling of Programmable Analog Approaches10.7 Low-Power Analog Signal Processing10.8 Low-Power Comparisons to Digital Approaches: Analog Computing in Memory10.9 Analog Programming at Digital Complexity: Large-Scale Field Programmable Analog Arrays10.10 Applications of Complex Analog Signal Processing10.10.1 Analog Transform Imagers10.10.2 Adaptive Filters and Classifiers10.11 DiscussionReferences
11.1 Introduction11.2 Bias Generator Circuits11.2.1 Bootstrapped Current Mirror Master Bias Current Reference11.2.2 Master Bias Power Supply Rejection Ratio (PSRR)11.2.3 Stability of the Master Bias11.2.4 Master Bias Startup and Power Control11.2.5 Current Splitters: Obtaining a Digitally Controlled Fraction of the Master Current11.2.6 Achieving Fine Monotonic Resolution of Bias Currents11.2.7 Using Coarse–Fine Range Selection11.2.8 Shifted-Source Biasing for Small Currents11.2.9 Buffering and Bypass Decoupling of Individual Biases11.2.10 A General Purpose Bias Buffer Circuit11.2.11 Protecting Bias Splitter Currents from Parasitic Photocurrents11.3 Overall Bias Generator Architecture Including External Controller11.4 Typical Characteristics11.5 Design Kits11.6 DiscussionReferences
12.1 Introduction12.1.1 Communication Cycle12.1.2 Speedup in Communication12.2 AER Transmitter Blocks12.2.1 AER Circuits within a Pixel12.2.2 Arbiter12.2.3 Other AER Blocks12.2.4 Combined Operation12.3 AER Receiver Blocks12.3.1 Chip-Level Handshaking Block12.3.2 Decoder12.3.3 Handshaking Circuits in Receiver Pixel12.3.4 Pulse Extender Circuits12.3.5 Receiver Array Peripheral Handshaking Circuits12.4 DiscussionReferences
13.1 Introduction13.1.1 Monitoring AER Events13.1.2 Sequencing AER Events13.1.3 Mapping AER Events13.2 Hardware Infrastructure Boards for Small Systems13.2.1 Silicon Cortex13.2.2 Centralized Communication13.2.3 Composable Architecture Solution13.2.4 Daisy-Chain Architecture13.2.5 Interfacing Boards using Serial AER13.2.6 Reconfigurable Mesh-Grid Architecture13.3 Medium-Scale Multichip Systems13.3.1 Octopus Retina + IFAT13.3.2 Multichip Orientation System13.3.3 CAVIAR13.4 FPGAs13.5 DiscussionReferences
14.1 Introduction14.1.1 Importance of Cross-Community Commonality14.2 Chip and System Description Software14.2.1 Extensible Markup Language14.2.2 NeuroML14.3 Configuration Software14.4 Address Event Stream Handling Software14.4.1 Field-Programmable Gate Arrays14.4.2 Structure of AE Stream Handling Software14.4.3 Bandwidth and Latency14.4.4 Optimization14.4.5 Application Programming Interface14.4.6 Network Transport of AE Streams14.5 Mapping Software14.6 Software Examples14.6.1 ChipDatabase – A System for Tuning Neuromorphic aVLSI Chips14.6.2 Spike Toolbox14.6.3 jAER14.6.4 Python and PyNN14.7 DiscussionReferences
15.1 Introduction15.2 Requirements for Software Infrastructure15.2.1 Processing Latency15.3 Embedded Implementations15.4 Examples of Algorithms15.4.1 Noise Reduction Filters15.4.2 Time-Stamp Maps and Subsampling by Bit-Shifting Addresses15.4.3 Event Labelers as Low-Level Feature Detectors15.4.4 Visual Trackers15.4.5 Event-Based Audio Processing15.5 DiscussionReferences
16.1 Introduction16.2 Large-Scale System Examples16.2.1 Spiking Neural Network Architecture16.2.2 Hierarchical AER16.2.3 Neurogrid16.2.4 High Input Count Analog Neural Network System16.3 DiscussionReferences
17.1 Introduction17.2 The Nature of Neuronal Computation: Principles of Brain Technology17.3 Approaches to Understanding Brains17.4 Some Principles of Brain Construction and Function17.5 An Example Model of Neural Circuit Processing17.6 Toward Neuromorphic CognitionReferences

Content preview from Event-Based Neuromorphic Systems

5 Locomotion Motor Control

images

This chapter turns from the sensors discussed in Chapters 3 and 4 to the motor output side of neuromorphic systems. Understanding how to engineer biological motor systems has far reaching implications, particularly in the fields of medicine and robotics. However, there are many complexities when it comes to understanding biological design. This chapter describes ways in which scientists and engineers have uncovered truths about the control circuitry behind animal locomotor skills (the motor circuits responsible for an organism’s locomotion) and discusses practical design solutions used for implementing these circuits. Neuromorphic design of locomotor systems in particular aims to replicate the efficiency and efficacy of biological design for its remarkable performance under real-world conditions. The end result of such efforts helps spawn innovative solutions for individuals with impaired motor function, and in the field of robotics, neuromorphic design will continue to redefine the level of robotic interactions within nature, as well as situations and environments adapted to human ergonomics. (Sections of this chapter have been adapted from the thesis work of Vogelstein (2007).)

5.1 Introduction

Evolution has produced a variety of animals that have developed interesting ways of moving in their environment. As scientists and engineers, we sometimes ...