| Introduction | p. 1 |
| The Diminishing Returns of Instruction-Level Parallelism | p. 1 |
| The Dawn of the Communication-Centric Revolution | p. 2 |
| The Global Wiring Challenge | p. 2 |
| The Network-on-Chip (NoC) Solution | p. 4 |
| Overview of Research | p. 6 |
| Legend for Figs. 1.4 and 1.5 | p. 7 |
| A Baseline NoC Architecture | p. 13 |
| MICRO-Architectural Exploration | |
| ViChaR: A Dynamic Virtual Channel Regulator for NoC Routers [39] | p. 19 |
| Importance of Buffer Size and Organization | p. 19 |
| Related Work in Buffer Design | p. 22 |
| The Proposed Dynamic Virtual Channel Regulator (ViChaR) | p. 24 |
| Variable Number of Virtual Channels | p. 27 |
| ViChaR Component Analysis | p. 30 |
| Simulation Results | p. 35 |
| Simulation Platform | p. 35 |
| Analysis of Results | p. 35 |
| Chapter Summary | p. 39 |
| RoCo: The Row-Column Decoupled Router - A Gracefully Degrading and Energy-Efficient Modular Router Architecture for On-Chip Networks [40] | p. 41 |
| Introduction and Motivation | p. 41 |
| Related Work in Partitioned Router Architectures | p. 43 |
| The Proposed Row-Column (RoCo) Decoupled Router | p. 44 |
| Row-Column Switch | p. 44 |
| Blocking Delay | p. 48 |
| Concurrency Control for High-Contention Environments | p. 50 |
| Flexible and Reusable On-Chip Communication | p. 51 |
| Fault-Tolerance Through Hardware Recycling | p. 51 |
| Performance Evaluation | p. 56 |
| Simulation Platform | p. 56 |
| Energy Model | p. 57 |
| A Performance, Energy, and Fault-Tolerance (PEF) Metric | p. 57 |
| Performance Results | p. 58 |
| Chapter Summary | p. 62 |
| Exploring Faulto Tolerant Network-on-Chip Architectures [37] | p. 65 |
| Introduction and Motivation | p. 65 |
| Simulation Platform Preliminaries | p. 67 |
| Handling Link Soft Faults | p. 68 |
| Flit-Based HBH Retransmission Scheme | p. 69 |
| Deadlock Recovery | p. 72 |
| Handling Soft Errors in Intra-Router Logic | p. 77 |
| Virtual Channel Arbiter Errors | p. 78 |
| Routing Computation Unit Errors | p. 80 |
| Switch Allocator Errors | p. 81 |
| Crossbar Errors | p. 83 |
| Retransmission Buffer Errors | p. 83 |
| Handshaking Signal Errors | p. 83 |
| Handling Hard Faults | p. 83 |
| Proximity-Aware (PA) Fault-Tolerant Routing Algorithm | p. 84 |
| Extension of PA Routing for Hot-Spot Avoidance | p. 86 |
| Service-Oriented Networking (SON) | p. 88 |
| SON - Direction Lookup Table (DLT) and Service Information Provider (SIP) | p. 89 |
| Chapter Summary | p. 91 |
| On the Effects of Process Variation in Network-on-Chip Architectures [45] | p. 93 |
| Introduction and Motivation | p. 93 |
| Related Work in Process Variation (PV) | p. 94 |
| The Impact of PV on NoC Architectures | p. 95 |
| Evaluation Platform | p. 95 |
| PV Effects on Router Components | p. 98 |
| The Proposed SturdiSwitch Architecture | p. 105 |
| IntelliBuffer: A Leakage-Aware Elastic Buffer Structure | p. 106 |
| VA Compaction Mechanism | p. 108 |
| SA Folding Mechanism | p. 111 |
| Chapter Summary | p. 115 |
| Macro-Architectural Exploration | |
| The Quest for Scalable On-Chip Interconnection Networks: Bus/NoC Hybridization [15] | p. 119 |
| Introduction and Motivation | p. 119 |
| Exploration of Existing On-Chip Bus Architectures | p. 121 |
| Traditional Bus Architectures | p. 121 |
| TDMA Buses and Hybrid Interconnects | p. 123 |
| Constraints of Traditional Buses | p. 124 |
| CDMA Interconnects | p. 125 |
| The Dynamic Time-Division Multiple-Access (dTDMA) Bus | p. 127 |
| Operation of the Dynamic Timeslot Allocation | p. 128 |
| Implementation of the dTDMA Bus | p. 129 |
| Comparison with a Traditional Bus Architecture | p. 130 |
| dTDMA Bus Performance | p. 132 |
| Comparison with Networks-on-Chip | p. 138 |
| Experimental Setup | p. 138 |
| Results | p. 139 |
| Interconnect Hybridization | p. 141 |
| Affinity Grouping | p. 142 |
| Simulation Methodology | p. 143 |
| Hybridization Results | p. 144 |
| Chapter Summary | p. 146 |
| Design and Management of 3D Chip Multiprocessors Using Network-In-Memory (NetInMem) [43] | p. 147 |
| Introduction and Motivation | p. 149 |
| Background | p. 149 |
| NUCA Architectures | p. 150 |
| Network-In-Memory (NetInMem) | p. 150 |
| Three-Dimensional (3D) Design and Architectures | p. 151 |
| A 3D NetInMem Architecture | p. 153 |
| The dTDMA Bus as a Communication Pillar | p. 155 |
| CPU Placement | p. 157 |
| 3D L2 Cache Management | p. 162 |
| Processors and L2 Cache Organization | p. 162 |
| Cache Management Policies | p. 163 |
| Experimental Evaluation | p. 164 |
| Methodology | p. 164 |
| Results | p. 166 |
| Chapter Summary | p. 169 |
| A Novel Dimensionally-Decomposed Router for On-Chip Communication in 3D Architectures [44] | p. 171 |
| Introduction and Motivation | p. 173 |
| Three-Dimensional Network-on-Chip Architectures | p. 176 |
| A 3D Symmetric NoC Architecture | p. 176 |
| The 3D NoC-Bus Hybrid Architecture | p. 177 |
| A True 3D NoC Router | p. 179 |
| A Partially-Connected 3D NoC Router Architecture | p. 182 |
| The Proposed 3D Dimensionally-Decomposed (DimDe) NoC Router Architecture | p. 182 |
| Performance Evaluation | p. 190 |
| Simulation Platform | p. 190 |
| Energy Model | p. 191 |
| Performance Results | p. 191 |
| Chapter Summary | p. 196 |
| Digest of Additional NoC MACRO-Architectural Research | p. 199 |
| A Distributed Multi-Point Network Interface for Low-Latency, Deadlock-Free On-Chip Interconnects [42] | p. 199 |
| Design of a Dynamic Priority-Based Fast Path Architecture for On-Chip Interconnects [46] | p. 201 |
| Exploring the Effects of Data Compression in NoC Architectures [47] | p. 203 |
| Conclusions and Future Work | p. 207 |
| References | p. 211 |
| Table of Contents provided by Ingram. All Rights Reserved. |