i A PHYSICAL IMPLEMENTATION WITH CUSTOM LOW POWER ...

More documents

Recommendations

Info

2.1 ARCHITECTURE OF THE SUPERCISC RHF The architecture of the SuperCISC RHF comprises an array of ALUs with interconnects between the rows of the ALU array. Each row in the reconfigurable fabric is called a stripe. The stripe of ALUs is called the Computational Stripe and the stripe of multiplexers which form the interconnect logic is called the Interconnect Stripe. The multiplexers in the Interconnect Stripe can be configured to select the appropriate operands to the ALU [3]. Figure 2-1 shows the architecture of the reconfigurable hardware fabric with computational and interconnect stripes alternating. Figure 2-1: Architecture of SuperCISC Reconfigurable Hardware Fabric 8
The specifications of the hardware fabric are the number of ALUs per stripe and the number of Computational (ALU) stripes. A 20X18 configuration represents a RHF with 20 ALUs per stripe and 18 ALU stripes. The number of multiplexer stripes is one less than the number of the ALU stripes, as the ALU and MUX stripes alternate each other. The final ALU stripe in the design is followed by a special interconnect stripe called the FINALMUX stripe. The FINALMUX stripe has its inputs from the last ALU stripe and from specialized ALU stripes called “early exit” ALU stripes. The FINALMUX stripe multiplexes these inputs to the get the final output of the hardware fabric. 2.2 ARITHMETIC AND LOGIC UNIT (ALU) The ALU performs conventional arithmetic and logical operations. In addition to the normal operations, a specialized hardware predication function is implemented within the ALU for implementing SDFGs [3] [4]. This operation requires a third single-bit operand called predicator input to be included in the ALU. The predicator input acts as a selector to choose one of the two operands INP1 and INP2 [3]. Figure 2-2 below shows the logical diagram of the ALU used in the hardware fabric. The control pins of the ALU select the logical operation performed by the ALU. 9
Page 1 and 2: A PHYSICAL IMPLEMENTATION WITH CUST
Page 3 and 4: A PHYSICAL IMPLEMENTATION WITH CUST
Page 5 and 6: 4.0 ASIC DESIGN FLOW ..............
Page 7 and 8: 6.5.5 Characterization Results for
Page 9 and 10: LIST OF TABLES Table 5-1: ALU Modul
Page 11 and 12: LIST OF FIGURES Figure 1-1: SuperCI
Page 13 and 14: Figure 5-20: BIGFABRIC Logical Diag
Page 15 and 16: Figure 7-4: EEPROM Erase Physical O
Page 17 and 18: ACKNOWLEDGEMENTS “Many, O Jehovah
Page 19 and 20: 1.0 INTRODUCTION Technological adva
Page 21 and 22: Property) block. The physical desig
Page 23 and 24: the thesis proposes a power gated m
Page 25: 2.0 SUPERCISC RECONFIGURABLE HARDWA
Page 29 and 30: INP2 selects from ALU1, ALU2, ALU3
Page 31 and 32: 3.0 POWER ESTIMATION The need for e
Page 33 and 34: 3.1.1 Static Power Consumption Stat
Page 35 and 36: Internal Power Internal Power is th
Page 37 and 38: Switching activity Generation Switc
Page 39 and 40: Figure 3-4: Leakage Power definitio
Page 41 and 42: The entire array contains informati
Page 43 and 44: 3.3.2 Calculation of Fall Power Fig
Page 45 and 46: Figure 3-10: Off-state leakage Powe
Page 47 and 48: Figure 4-1 : Typical ASIC Design Fl
Page 49 and 50: Figure 4-2: ASIC Physical Design Fl
Page 51 and 52: 4.1.2 Powerplanning Power planning
Page 53 and 54: created in the design. Routing feed
Page 55 and 56: 4.1.7 Routing Once the standard cel
Page 57 and 58: Table 5-1: ALU Module Specification
Page 59 and 60: Table 5-2 shows the important speci
Page 61 and 62: the basic parameters are specified,
Page 63 and 64: direction have to be aligned on the
Page 65 and 66: ALU Stripe Pin Assignment The “ib
Page 67 and 68: 5.2.1 MUX Module Specifications As
Page 69 and 70: Figure 5-12: MUX Stripe Logical Dia
Page 71 and 72: MUX initialization routine The MUX
Page 73 and 74: Table 5-10: MUX Stripe Pin Placemen
Page 75 and 76: Figure 5-15: FINALMUX Module Logica
Page 77 and 78:
Table 5-12: Final MUX Stripe Specif
Page 79 and 80:
pins in the FINALMUX stripe is plac
Page 81 and 82:
larger system. Figure 5-22 shows th
Page 83 and 84:
Figure 5-20: BIGFABRIC Logical Diag
Page 85 and 86:
Figure 5-22: Place and Routed BIGFA
Page 87 and 88:
BIGFABRIC Stripe Pin Assignment As
Page 89 and 90:
The SPEF file can be generated usin
Page 91 and 92:
5.5.1 Power Results for ADPCM Encod
Page 93 and 94:
5.5.3 Power Results for IDCT Row Be
Page 95 and 96:
5.5.5 Power Results for Sobel Bench
Page 97 and 98:
6.0 DELAY ELEMENTS FOR LOW POWER FA
Page 99 and 100:
Transmission gate with Schmitt Trig
Page 101 and 102:
PMOS transistors have their gate in
Page 103 and 104:
output normally. The signal integri
Page 105 and 106:
m-Transistor Cascaded Inverter This
Page 107 and 108:
6.2 LOW POWER FABRIC USING DELAY EL
Page 109 and 110:
The use of delay elements in the ha
Page 111 and 112:
6.3.2 Dynamic Triggering Scheme A d
Page 113 and 114:
Figure 6-16 shows the shunt current
Page 115 and 116:
td1 is the delay in discharging nod
Page 117 and 118:
δt is the regeneration time of the
Page 119 and 120:
transistors, they are active low si
Page 121 and 122:
delay elements are enabled on a log
Page 123 and 124:
charge sharing happens between the
Page 125 and 126:
Figure 6-23: AND gate to generate D
Page 127 and 128:
uffer with a drive capacity of 640f
Page 129 and 130:
Cell Rise Delay The Cell Rise Delay
Page 131 and 132:
Fall Transition Time The time taken
Page 133 and 134:
6.5.6 Characterization Results for
Page 135 and 136:
7.0 EEPROM CIRCUIT DESIGN EEPROM (E
Page 137 and 138:
7.1.1 Erase Operation Figure 7-3: I
Page 139 and 140:
7.1.2 Write Operation Figure 7-6: I
Page 141 and 142:
Figure 7-9: IV Characteristics of a
Page 143 and 144:
The current source IFN, resistor RT
Page 145 and 146:
Figure 7-11: HSPICE Description of
Page 147 and 148:
7.2.1 Ramp Generator The ramp gener
Page 149 and 150:
7.2.4 Column Latch for Bitlines inp
Page 151 and 152:
7.2.5 Power Multiplexer Figure 7-16
Page 153 and 154:
The inverter pair connected to the
Page 155 and 156:
7.2.7 Memory Bank Architecture Figu
Page 157 and 158:
7.2.8 Memory Bank Simulation Figure
Page 159 and 160:
(Vpp) and a normal voltage (Vdd), t
Page 161 and 162:
8.1.2 Dynamic Decoder The address d
Page 163 and 164:
8.3 POWER-ON RESET One problem with
Page 165 and 166:
Figure 8-6 shows the modified charg
Page 167 and 168:
To study the impact of the power co
Page 169 and 170:
9.0 CONCLUSION For this thesis, I h
Page 171 and 172:
definitions, metal layer resistance
Page 173 and 174:
APPENDIX B element standard cell ch
Page 175 and 176:
$k=-1; $j=-1; $i=-1; foreach $ load
Page 177 and 178:
$FALL_START_8 = $OFFSET8 + $ONTIME;
Page 179 and 180:
if ($#rise_power_6 == -1) { # $modi
Page 181 and 182:
$modified_spice_array[$i] = $temp2;
Page 183 and 184:
@fall_power_9 = grep(/fall_power9/,
Page 185 and 186:
print MEASHANDLE $_; print ENERGYHA
Page 187 and 188:
} print MEASHANDLE " "; print ENERG
Page 189 and 190:
set VAL [expr "$NUMBER_OF_MODULES_P
Page 191 and 192:
set DIE_WIDTH_STRIPE [expr {($MODUL
Page 193 and 194:
#*************************PLACING i
Page 195 and 196:
preassignPin stripe $PIN_NAME -loc
Page 197 and 198:
set X_LOC [expr {($X _LOC -( 31*$OU
Page 199 and 200:
#Module related details set NUMBER_
Page 201 and 202:
set X_LOC 0 set Y_LOC 41.2 for {set
Page 203 and 204:
Table D 1 (Continued) 506.232fF 0.0
Page 205 and 206:
121.344fF 0.076ns 0.4634 0.2277 570
Page 207 and 208:
Table D 5 (Continued) 9.48 0. 516 0
Page 209 and 210:
Table D 7: Characterization data fo
Page 211 and 212:
Transi tion Table D 9: Characteriza
Page 213 and 214:
[13] M.F. Aburdene, J. Zheng, and R
show all

i A PHYSICAL IMPLEMENTATION WITH CUSTOM LOW POWER ...

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?