22
SPEED-INDEPENDENT FLOATING POINT COPROCESSOR Stepchenkov Yuri, Zakharov Victor, Rogdestvenski Yuri, Diachenko Yuri , Morozov Nickolai, Stepchenkov Dmitri Institute of Informatics Problems, Federal Research Center "Computer Science and Control" of the Russian Academy of Sciences, IPI RAS, Moscow, Russian Federation IPI RAS EWDTS-2015 1

Stepchenkov Yuri, Zakharov Victor, Rogdestvenski Yuri, Diachenko Yuri, Morozov Nickolai, Stepchenkov Dmitri Institute of Informatics Problems, Federal

Embed Size (px)

Citation preview

SPEED-INDEPENDENT FLOATING POINT COPROCESSOR

Stepchenkov Yuri, Zakharov Victor, Rogdestvenski Yuri, Diachenko Yuri, Morozov Nickolai, Stepchenkov

Dmitri

Institute of Informatics Problems, Federal Research Center "Computer Science and Control" of the Russian Academy of Sciences, IPI RAS, Moscow, Russian Federation

IPI RAS EWDTS-2015 1

Contents

Why Speed-Independent circuits? Structure chart of SI-Coprocessor Ternary ST-coding Simplified indication SI-coprocessor’s pipeline SI-coprocessor’s features Conclusions

IPI RAS EWDTS-2015 2

Why Speed-Independent circuits?

IPI RAS EWDTS-2015 3

They reduce power consumption due to removing both clock generator, and "clock tree" out of a circuit

They have wide workability range on power supply and temperature

Divider & Square Rooter, 0.18

IPI RAS EWDTS-2015 4

MOPS

Die, mm2

ΔUdd*ΔT

Consumption, %Synchron, “SRT-4” Synchron, “Newton” Quasi-SI, “SRT-2”

ParameterVariant

Maximum performance at any environment

Extended workability range

Constant faults detection

Average power consumption

Minimum noise level

IPI RAS EWDTS-2015 5

Projected part of SI-logic (ITRS)

Usa

ge o

f S

I-ne

ts in

a d

esig

n,

%

Years

Structure chart of SI-FPC

IPI RAS EWDTS-2015 6

Booth multiplier

IPI RAS EWDTS-2015 7

Ternary ST-coding

IPI RAS EWDTS-2015 8

Synchronous redundant code

Self-Timed redundant (ternary) code

State Binary code

State ST ternary code

A B Ap Am An

+1 1 0 +1 1 0 0

0 0 0 0 0 0 1

–1 0 1 –1 0 1 0

N/A 1 1 spacer 0 0 0

Ternary SI-adder

IPI RAS EWDTS-2015 9

Ternary SI-adder

IPI RAS EWDTS-2015 10

Indication subcircuit

Dual-rail SI-adder

IPI RAS EWDTS-2015 11

Dual-rail SI-adder

IPI RAS EWDTS-2015 12

Dual-rail Wallace tree

IPI RAS EWDTS-2015 13

1700 transistors, 7 stages

Ternary Wallace tree

IPI RAS EWDTS-2015 14

2190 transistors, 4 stages

Simplified indication

IPI RAS EWDTS-2015 15

Bitwise indicators

Full indication only in spacer phase

Pipeline organization: Top view

IPI RAS EWDTS-2015 16

Pipeline organization:One stage

IPI RAS EWDTS-2015 17

One bit of input register

InpR

SI FPC’s parameters

IPI RAS EWDTS-2015 18

Parameter ValueComplexity, transistors 315 000

Die size, mm2 0.47

Performance, Gflops 0.54

Latency, ns 1.9

Power consumption, mW/Gflops

450

Testing SI-Coprocessor

IPI RAS EWDTS-2015 19

SI FPC

First result

Compa-rator

OK

Data

Conclusions At first time in the world, really SI-

FPC unit performing FMA operation was implemented

Usage of ternary ST-code provided best performance of the Wallace tree

Simplified indication allowed for reducing both the complexity and work phase time

Two-stage pipeline sufficiently decreased hardware cost at minimal drop of performance

IPI RAS EWDTS-2015 20

Thanks!

IPI RAS EWDTS-2015 21

Contacts Director: academician Sokolov I.A. Address: Institute of Informatics Problems

of the Federal Research Center "Computer Science and Control" of the Russian Academy of Sciences (IPI RAS), Moscow, Russian Federation, 119333, Vavilova str., 44, b.2

Tel: +7 (495) 137 34 94 Fax: +7 (495) 930 45 05 E-mail: [email protected] Stepchenkov Y.A., tel. +7 (495) 671 15 20,

[email protected] RAS EWDTS-2015 22