xv
BookID 186671_ChapID FM1_Proof# 1 - 28/08/2009
List of Figures
Fig. 6.11 The VA Compaction Architecture................................................... 109
Fig. 6.12 The Staggered Arbitration Mechanism Employed by
the VA Compaction Architecture .................................................... 110
Fig. 6.13 Effect of No-Grants on Generic and VA Compaction Modules ..... 111
Fig. 6.14 The SA Folding Concept ................................................................ 112
Fig. 6.15 Detailed View of the SA Folding Architecture ............................... 113
Fig. 6.16 Effect of No-Grants on Generic and SA Folding Modules ............. 114
Fig. 7.1 Bus Segmentation ........................................................................... 122
Fig. 7.2 Unfairness Caused by Timeslice Reassignment ............................. 123
Fig. 7.3 Normalized CDMA Throughput Using Walsh
Orthogonal Codes ........................................................................... 126
Fig. 7.4 Example of Dynamic Timeslot Allocation ..................................... 128
Fig. 7.5 Architecture of the dTDMA Receiver and Transmitter .................. 129
Fig. 7.6 A Simple 2-PE Transfer on dTDMA and AMBA .......................... 131
Fig. 7.7 dTDMA Bus Throughput (Uniform Injection) ............................... 133
Fig. 7.8 Unicast Latencies for Several Bus Sizes ......................................... 134
Fig. 7.9 Latency Using Three Different Arbitration Policies ....................... 136
Fig. 7.10 Power Consumption of the dTDMA Bus ........................................ 137
Fig. 7.11 dTDMA Bus vs. NoC Latencies for Uniform (a-d)
and Multimedia (e-h) Traffic Injection. A Comparison
of Average Power Consumption is Shown in (i). ............................ 140
Fig. 7.12 Hybridization of the AG Mesh Using the dTDMA Bus ................. 143
Fig. 7.13 Hybrid dTDMA Bus/NoC vs. Pure NoC (AG Mesh)
Latencies for Uniform (a-c) and Multimedia (d-f)
Traffic Injection ............................................................................... 145
Fig. 7.14 Hybrid dTDMA Bus/NoC vs. Pure NoC (AG Mesh)
Average Power Consumption (Uniform Traffic Injection) ............. 145
Fig. 8.1 The Progression from 2D Hybridization in Chapter 7 to
3D Hybridization in Chapter 8. ....................................................... 148
Fig. 8.2 Wiring Scales in Length as the Square Root of the
Number of Layers in Three Dimensions ......................................... 151
Fig. 8.3 Face-to-Face and Face-to-Back Wafer-Bonding in
3D Technology ................................................................................ 152
Fig. 8.4 The Proposed 3D Network-In-Memory (NetInMem)
Architecture ..................................................................................... 154
Fig. 8.5 Transceiver Module of a dTDMA Bus ........................................... 155
Fig. 8.6 Side View of the 3D Chip with the dTDMA Bus (Pillar) ............... 156
Fig. 8.7 A High-Level Overview of the Modified Router of the
Pillar Nodes ..................................................................................... 158
Fig. 8.8 A CPU has More Cache Banks in its Vicinity in the
3D Architecture ............................................................................... 158
Fig. 8.9 Hotspots can be Avoided by Offsetting CPUs in All
Three Dimensions ........................................................................... 159
Fig. 8.10 Placement Pattern of CPUs Around the Pillars ............................... 160
Fig. 8.11 An Example of the CPU Placement Algorithm .............................. 160