Download slide

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

History of electric power transmission wikipedia , lookup

Control system wikipedia , lookup

Power over Ethernet wikipedia , lookup

Electrification wikipedia , lookup

Electric power system wikipedia , lookup

Transistor wikipedia , lookup

Voltage optimisation wikipedia , lookup

Decibel wikipedia , lookup

Standby power wikipedia , lookup

Opto-isolator wikipedia , lookup

Mains electricity wikipedia , lookup

Islanding wikipedia , lookup

Variable-frequency drive wikipedia , lookup

Alternating current wikipedia , lookup

Power engineering wikipedia , lookup

Audio power wikipedia , lookup

Buck converter wikipedia , lookup

TRIAC wikipedia , lookup

Switched-mode power supply wikipedia , lookup

Power MOSFET wikipedia , lookup

Power electronics wikipedia , lookup

Power inverter wikipedia , lookup

Solar micro-inverter wikipedia , lookup

CMOS wikipedia , lookup

Transcript
Dual Threshold Voltage Domino
Logic Synthesis for High
Performance with Noise and
Power Constraint
Seong-Ook Jung, Ki-Wook Kim and
Sung-Mo (Steve) Kang
DATE’02
1
Outline





Introduction
Simulation Results on Four Feasible
Configuration
Dual Vt Domino Logic Synthesis Flow
Experimental Results
Conclusions
2
Introduction
Domino logic
Clock=0 : precharge
Clock=1 : discharge(evaluate)
Advantages : high performance
Drawback : noise sensitive
3
Introduction
noise
Solution : increase keeper size
Drawbacks : 1. increase power
consumption
2. increase evaluation time
4
Introduction
Vt (threshold voltage)
Vt
performance
But exponential increase in subthreshold leakage
current(Isub) of transistors.
5
Introduction

Tradeoff



Vt   evaluation time   Isub 
keeper size   power consumption 
 evaluation time 
Goal

use low vt devices to speedup evaluation
while maintaining power consumption and
noise constraint.
6
Four Feasible Configuations
low vt assignment to NMOS
tree and/or PMOS transistor
of output inverter.
7
Keeper Sizing for Feasible
Configurations


Increase keeper size (width) to satisfy
noise constraint
Noise Constraint



Topmost NMOS transistors directly connected to
20% of Vdd
Other NMOS transistors turn on with Vdd
Fail if the output voltage of inverter is greater than
10% of Vdd
8
Keeper Sizing for Feasible
Configurations
9
Evaluation Time(tE) and Power
normalized by HH type of each gate
with 4:1 P:N ratio output inverter.
10
Power consumption in active
mode
1. analyze the simulation result of domino logic gate with 4:1 P:N ratio
output inverter.
(1) LL type : fastest speed, highest power consumption
(2) OR gate : LH slower than HL; Pact of LH greater than HL
(3) AND gate : HL slower than LH; Pact is almost the same
2. analyze the effect of up-sizing PMOS transistor from 4:1 to 8:1.
tE is improved by increasing the P:N ratio of output inverter
(1) OR gate : tE of LL are almost the same
(2) OR gate : Pact increases
(3) AND gate : Pact almost the same
11
Leakage current in standby
mode
normalized by HH type of each gate
with 4:1 P:N ratio output inverter.
Ileak is masured in standby mode
by making all logic gates evaluated
to reduce leakage
Ileak is determined by precharge PMOS
and NMOS of output inverter
(the same size and Vt)
Ileak is almost the same for
all simulation case
12
Dual Vt Domino Logic
Synthesis Flow
13
Dual Vt Domino Logic
Synthesis Flow
Type selection for
unmarked logic gates
with power constraint
(LL, LH and HL type)
For each gate in the critical paths, a proper type is selected
for delay minimization with power constraint.
Example:
OR gate : LL, HL, LH, HH
AND gate : LL,LH,HL, HH
14
Dual Vt Domino Logic
Synthesis Flow
Gate selection based on
performance sensitivity
A gate is chosen such that performance of the gate is
maximum.
15
Experimental Results
tE- denotes critical path speed-up.(with respect to the initial circuits HH)
Pact+denotes total active power overhead.(with respect to the initial circuits HH)
16
Experimental Results



Bold numbers : maximum speed-up for each
benchmark circuit(maximum speed up range
from 15.91% to 18.62% with 0.38% to
7.01% active power increase)
Half of benchmark circuits achieve around
18% speed-up with less than 1% active
power increase
The average maximum speed up is 17.43%
with 1.84% average active power increase.
17
Conclusions


Tradeoffs need to be made among
noise , power, and performance.
Propose a dual Vt synthesis method for
high performance with noise (keeper
sizing) and power constraint.
18