14
15- 398 lecture 5 © 2004 Seth Copen Goldstein 1 lecture 8 15-398 © 2004 Seth Copen Goldstein 1 Nanotechnology And Computer Architecture Seth Copen Goldstein [email protected] CMU 15- 398 Introduction to Nanotechnology lecture 8 15-398 © 2004 Seth Copen Goldstein 2 Today Device requirements Manufacturing requirements Fabrication Micro- nano scale matching Reconfigurable computing Architectures lecture 8 15-398 © 2004 Seth Copen Goldstein 3 Computing Devices Completeness Logic Restoration Fan- out Input/Output Isolation Power Gain Inexpensive Low power (zero static power) Dense packing Reliable lecture 8 15-398 © 2004 Seth Copen Goldstein 4 Switching Devices must have some non- linear characteristic that supports logical computation V out V in V dd V out V ih V il V oh V ol V in

Computing Devices Switching

  • Upload
    others

  • View
    4

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Computing Devices Switching

15-398 lecture 5

© 2004 Seth Copen Goldstein 1

lecture 8 15-398 © 2004 Seth Copen Goldstein 1

NanotechnologyAnd

Computer Architecture

Seth Copen [email protected]

CMU

15-398 Introduction to Nanotechnology

lecture 8 15-398 © 2004 Seth Copen Goldstein 2

Today

Device requirementsManufacturing requirementsFabricationMicro-nano scale matchingReconfigurable computingArchitectures

lecture 8 15-398 © 2004 Seth Copen Goldstein 3

Computing Devices

CompletenessLogic RestorationFan-outInput/Output IsolationPower GainInexpensiveLow power (zero static power)Dense packingReliable

lecture 8 15-398 © 2004 Seth Copen Goldstein 4

Switching

Devices must have some non-linear characteristic that supports logical computation

VoutVin

Vdd

Vout

VihVil

Voh

Vol

Vin

Page 2: Computing Devices Switching

15-398 lecture 5

© 2004 Seth Copen Goldstein 2

lecture 8 15-398 © 2004 Seth Copen Goldstein 5

Completeness

Must be able to generate all functions

01

10

aa

a x

ax

NOT AND OR NAND XOR

010

001

1

0

b

11

00

a ba

110

101

1

0

b

11

00

a ba

110

101

1

0

b

01

10

(a b)

a

ab a b

ab

a b

110

101

1

0

b

01

00

a ba

ab

lecture 8 15-398 © 2004 Seth Copen Goldstein 6

Is AND sufficient?

AND

010

001

1

0

b

11

00

a ba

lecture 8 15-398 © 2004 Seth Copen Goldstein 7

Logic Restoration

Why?

VoutVin

Vdd

Vout

VihVil

Voh

Vol

Vin

lecture 8 15-398 © 2004 Seth Copen Goldstein 8

Fan-out

The ability to drive more than one output.Why?

NOT

halfaddersum

carry

a

b

a

b

sum

Page 3: Computing Devices Switching

15-398 lecture 5

© 2004 Seth Copen Goldstein 3

lecture 8 15-398 © 2004 Seth Copen Goldstein 9

I/O Isolation

A change in the output does not affect the value of the input.Why?

lecture 8 15-398 © 2004 Seth Copen Goldstein 10

Power gain

Is it strictly necessary?

SirM modelSwitchingIsolationRestorationMemory

lecture 8 15-398 © 2004 Seth Copen Goldstein 11

Necessary Device Abstraction

Transistor provides it allSwitchFan-outI/O isolationSignal restoration

Replace single abstraction with multiple abstractions: SIRMEnables an incremental path to direct nano-circuits

Necessary for logic

Necessary for large designs

lecture 8 15-398 © 2004 Seth Copen Goldstein 12

SIRM

SIRM decomposes the abstraction into its fundumental components.

SwitchI solationRestorationMemory

SIRM already in useAllows circuit designers to explicitly manage the different components of a scalable circuit technology

Necessary for logic

Necessary for large designs

Page 4: Computing Devices Switching

15-398 lecture 5

© 2004 Seth Copen Goldstein 4

lecture 8 15-398 © 2004 Seth Copen Goldstein 13

SIRM Example

Breaks up transistor abstraction into its components

Allows creation of library of composable components

Increases freedom of design

A

A

B

B

A&B

Old Way

A

B

A&B

New Way

Nano-switchNano-restorer

Nano-isolator

lecture 8 15-398 © 2004 Seth Copen Goldstein 14

Manufacturing Requirements

InexpensiveReliableHigh yieldDense packingScalable

Top-Down: Photolithography-Combines manufacturing and assembly into one step!Bottom-up: ?

lecture 8 15-398 © 2004 Seth Copen Goldstein 15

(scalable) Bottom-up assembly

Many different techniques:Langmuir-Blodgett filesFlow-based alignmentEM alignmentSelf-assembled monolayersChemical synthesisDNA-based assemblyNano-imprinting

Separates manufacturing and assembly!Often, no precise placement control

lecture 8 15-398 © 2004 Seth Copen Goldstein 16

Fabrication Is Different

Devices & wires alone are not usefulKey to nanoscale computing is bottom-up assembly

Con

trol

, con

figu

rati

on d

ecom

pres

sion

, &de

fect

map

ping

see

d C

ontr

ol, c

onfi

gura

tion

dec

ompr

essi

on, &

defe

ct m

appi

ng s

eed

Con

trol

, con

figu

rati

on d

ecom

pres

sion

, &de

fect

map

ping

see

d C

ontr

ol, c

onfi

gura

tion

dec

ompr

essi

on, &

defe

ct m

appi

ng s

eed

Page 5: Computing Devices Switching

15-398 lecture 5

© 2004 Seth Copen Goldstein 5

lecture 8 15-398 © 2004 Seth Copen Goldstein 17

Fabrication PrimitivesParallel wires2-D grids

lecture 8 15-398 © 2004 Seth Copen Goldstein 18

Fabrication PrimitivesParallel wires2-D gridsSelf-assembled monolayers

lecture 8 15-398 © 2004 Seth Copen Goldstein 19

Fabrication PrimitivesParallel wires2-D gridsSelf-assembled monolayersConnections between orthogonal wires

lecture 8 15-398 © 2004 Seth Copen Goldstein 20

Fabrication PrimitivesParallel wires2-D gridsSelf-assembled monolayersConnections between orthogonal wiresConnections to CMOS

Page 6: Computing Devices Switching

15-398 lecture 5

© 2004 Seth Copen Goldstein 6

lecture 8 15-398 © 2004 Seth Copen Goldstein 21

Fabrication PrimitivesParallel wires2-D gridsSelf-assembled monolayersConnections betweenorthogonal wiresConnections to CMOSNon-deterministic placement

lecture 8 15-398 © 2004 Seth Copen Goldstein 22

Fabrication PrimitivesParallel wires2-D gridsSelf-assembled monolayersConnections betweenorthogonal wiresConnections to CMOSNon-deterministic placementSelective variability on a lithographic scale

lecture 8 15-398 © 2004 Seth Copen Goldstein 23

Building a Computing Crystal

lecture 8 15-398 © 2004 Seth Copen Goldstein 24

Building a Computing Crystal

Page 7: Computing Devices Switching

15-398 lecture 5

© 2004 Seth Copen Goldstein 7

lecture 8 15-398 © 2004 Seth Copen Goldstein 25

Building a Computing Crystal

Assembly Computing CrystalsWith defects

lecture 8 15-398 © 2004 Seth Copen Goldstein 26

Scale Matching

1.2nm >40nm40x differenceOverhead?

Input

A2

A2A2

A1

A1A1

A0

A0A0

O0O1O2O3O4O5O6O7

lecture 8 15-398 © 2004 Seth Copen Goldstein 27

Harnessing RandomnessFab primitives prefer low information content structures

Can we design circuits that can exploit this?

Example: Scale Matching

Address 1 nanoscale wire with 5log(N) micronscalewires

Based on random placement of molecules

Address lines

Nanoscale wireslecture 8 15-398 © 2004 Seth Copen Goldstein 28

Reconfigurable!

Implications

Prefer devices at cross-points(only 2 terminal devices?)Only regular structuresDefect-tolerance requiredFunctionality must be added post-fabricationReduce nano-micro overhead

Page 8: Computing Devices Switching

15-398 lecture 5

© 2004 Seth Copen Goldstein 8

lecture 8 15-398 © 2004 Seth Copen Goldstein 29

FPGA

Universal gates and/or

storage elements

Interconnectionnetwork

Programmable Switches

lecture 8 15-398 © 2004 Seth Copen Goldstein 30

Reconfigurable Computing

Compiler

General-Purpose Custom Hardwareint reverse(int x){

int k,r=0;for (k=0; k<64; k++)

r |= x&1;x = x >> 1;r = r << 1;

}}int func(int* a,int *b){

int j,sum=0;for (j=0; *a>0; j++)

sum+=reverse(*b

General-PurposeCustom Hardware

Logic Blocks

Routing Resources

lecture 8 15-398 © 2004 Seth Copen Goldstein 31

Aides defect tolerance

Reconfigurability & DFT

FPGA computing fabricRegularperiodic Fine-grainedHomogenous

programs circuits

Place& Route

Place& Route

lecture 8 15-398 © 2004 Seth Copen Goldstein 32

Sample Proposals

Can be evaluated on determinism required at fab time

Nanocell: purely randomArray-based: quasi-regularNanoscale silicon: determinisitc

Page 9: Computing Devices Switching

15-398 lecture 5

© 2004 Seth Copen Goldstein 9

lecture 8 15-398 © 2004 Seth Copen Goldstein 33

Nanocell

Summer

lecture 8 15-398 © 2004 Seth Copen Goldstein 34

Using the Nanocell

Summer

lecture 8 15-398 © 2004 Seth Copen Goldstein 35

Array-Based approaches

This is the general trendCreate meshes, glue them togetherEach Mesh is reconfigurable

Dehon

+Vdd

Gro

und

C

CGnd

+Vdd

Gro

und

C

CGnd

lecture 8 15-398 © 2004 Seth Copen Goldstein 36

NanoBlock

+Vdd

Gro

und

C

CGnd

Stripped regions indicate connections from the CMOS layer.

Diode-matrix Molecular-latches

Page 10: Computing Devices Switching

15-398 lecture 5

© 2004 Seth Copen Goldstein 10

lecture 8 15-398 © 2004 Seth Copen Goldstein 37

Diode-resistor Logic

A * B

V AND

BA

A ^ B

VVV AND

A

BA

B

A * B

Nano-implementation

Configured ON

Configured OFF

lecture 8 15-398 © 2004 Seth Copen Goldstein 38

Diode-resistor Logic

VDD

A * BA

B

A * B

Nano-implementation Electrical equivalent

Configured ON

Configured OFFV AND

BA

A ^ B

VVV AND

A

BA

B

A * B

Problems?

lecture 8 15-398 © 2004 Seth Copen Goldstein 39

V

BAA

B

V V V

S = A B

V V

CS

C = A ^ B

Half adder

lecture 8 15-398 © 2004 Seth Copen Goldstein 40

Abstract

+Vdd

Gro

und

C

CGnd

SE

Page 11: Computing Devices Switching

15-398 lecture 5

© 2004 Seth Copen Goldstein 11

lecture 8 15-398 © 2004 Seth Copen Goldstein 41

SE

SE

Nearest Neighbor Routing

lecture 8 15-398 © 2004 Seth Copen Goldstein 42

SE

SE

NW

NW

Nearest Neighbor Routing

lecture 8 15-398 © 2004 Seth Copen Goldstein 43

SE

SE

NW

NW

Nearest Neighbor Routing

Switch block

lecture 8 15-398 © 2004 Seth Copen Goldstein 44

2-D Mesh

Page 12: Computing Devices Switching

15-398 lecture 5

© 2004 Seth Copen Goldstein 12

lecture 8 15-398 © 2004 Seth Copen Goldstein 45

2-D Mesh

lecture 8 15-398 © 2004 Seth Copen Goldstein 46

2-D Mesh

lecture 8 15-398 © 2004 Seth Copen Goldstein 47

Cluster of NanoBlocks

lecture 8 15-398 © 2004 Seth Copen Goldstein 48

The NanoFabric

Con

trol

, con

figu

ratio

n de

com

pres

sion

, &de

fect

map

ping

see

d C

ontr

ol, c

onfi

gura

tion

deco

mpr

essi

on, &

defe

ct m

appi

ng s

eed

Nanoscale layer put deterministically ontop of CMOS

Highly regular

~108 long lines

~106 clusters

Cluster has 128 blocks

Page 13: Computing Devices Switching

15-398 lecture 5

© 2004 Seth Copen Goldstein 13

lecture 8 15-398 © 2004 Seth Copen Goldstein 49

NanoFabric Attributes

Hierarchical fabrication1. Manufacture devices2. Non-deterministically align wires3. Self-assemble monolayers onto wires

4. Create meshes5. Deterministically align w/CMOS

ReconfigurableDefect tolerant

lecture 8 15-398 © 2004 Seth Copen Goldstein 50

Trade-offs

Ease of Manufacturing Vs. ?Device capabilities Vs. ?Device reliability Vs ?Information content in design Vs ?

lecture 8 15-398 © 2004 Seth Copen Goldstein 51

Next Time

Take a closer look at nano-devicesRead chapter 5Recommended Ratner 1973

Page 14: Computing Devices Switching

This document was created with Win2PDF available at http://www.daneprairie.com.The unregistered version of Win2PDF is for evaluation or non-commercial use only.