CMOS Integration of field effect plasmonic modulators

(1)

HAL Id: tel-00848107

https://tel.archives-ouvertes.fr/tel-00848107

Submitted on 25 Jul 2013

HAL is a multi-disciplinary open access

archive for the deposit and dissemination of

sci-entific research documents, whether they are

pub-lished or not. The documents may come from

teaching and research institutions in France or

L’archive ouverte pluridisciplinaire HAL, est

destinée au dépôt et à la diffusion de documents

scientifiques de niveau recherche, publiés ou non,

émanant des établissements d’enseignement et de

recherche français ou étrangers, des laboratoires

CMOS Integration of field effect plasmonic modulators

Alexandros Emboras

To cite this version:

Alexandros Emboras. CMOS Integration of field effect plasmonic modulators. Other. Université de

Grenoble, 2012. English. �NNT : 2012GRENT094�. �tel-00848107�

(2)

THÈSE

Pour obtenir le grade de

DOCTEUR DE L’UNIVERSITÉ DE GRENOBLE

Spécialité : Micro et Nano Électronique

Arrêté ministérial :

Présentée par

Alexandros EMBORAS

Thèse dirigée par Dr. Barbara DE SALVO

et codirigée par Dr. Roch ESPIAU DE LAMAESTRE

préparée au sein

Laboratoire d’ Électronique et des Technologies

de l’Information du Commissariat à l’Énergie Atomique

(CEA-LETI-MINATEC)

et de l

′

École Doctorale Électronique, Électrotechnique Automatique et

Traitement du Signal

Intégration en technologie CMOS

d’un modulateur plasmonique à

ef-fet de champ

Thèse soutenue publiquement le 10 Mai 2012,

devant le jury composé de :

Dr., Gérard GHIBAUDO

Professeur, Président

Dr., Jean-Claude WEEBER

Professeur, Rapporteur

Dr, Laurent VIVIEN

Docteur, Rapporteur

Dr, Constantinos CHRISTOFIDES

Professeur, Examinateur

Dr, Anatoly ZAYATS

(3)

I would like to express my gratitude to Dr. Barbara de Salvo and Ro h Espiau de la

Maestrewhopatiently guided methroughout myPhD.Manythanksfor your trust and

your en ouragement, your riti al feedba k and the time spent in numerous and long

meetings.Theywerealways enjoyable. Thank you forall!

As regard to my work, I benet really a lot from the assistan e of many peoples. I

wouldliketothankCharlesLerouxforhelpingintheoreti alsimulationsandexperiments

ofMOS.Adelwasoneofthe keypersonwhospenthistimetorealizetheplasmostorand

Iamgratefultohaveworkedwithhim.Iwouldliketogratitudealsoseveralotherpeoples

who involved on this work like Emmanuel Augendre for hiswork inthe dire t bonding

(big step in the realization of the plasmostor), Jean-Paul Barnes for his assistan e in

SIMSmeasurementsandAlainforhelpingandguidingmefordoingsimulation usingthe

luster.

I am warmly grateful to Dr. Philippe Grosse for always being so prompt on

en ou-raging, supporting, and helping me throughout this pro ess. I would like to extend my

appre iation to Prof. George Pananakakis whose advises were always very helpful and

played a leading role to myprogress. He hasbeen there for mefrom thevery beginning

ofmy PhD.His valuable friendship and expertisehasbeen agreat asset.

Thank youto SiddarthandClement,the besto ematesI ouldeverhave.Lu and

Salimhaveverymu h ontributed inmakingthe timeunforgettable withrefreshing and

produ tive breaks. Seriously,thank youfor all your helpand kindness.

I would also like to express my gratitude to my family parti ularly to my mother

Dora, my father Kikisand to mysiblingsNeophyta and Emilios. Thank you for all the

patien e, the love andsupport ineverystep thatImade.

Finally,I would like to thank from the bottom of my heart mygirlfriend Danae for

her real, in onvenient love, patien e, and en ouragement throughout the ompletion of

(4)

(5)

Compa t and low energy onsumption integrated opti al modulator is urgently

re-quiredfor en odinginformation into opti alsignals.Tothatrespe t,theuseofplasmon

modes to modulate light is of parti ular interest when ompared to the numerous

re-feren es des ribing sili on based opti al modulators. Indeed, the high eld onnement

propertiesofthose modesandthein reasedsensitivityto smallrefra tive index hanges

of the diele tri lose to the metal an help de reasethe hara teristi length s ales of

the devi es,towards to that of mi roele troni s. Thisthesis investigates the realization

of Si eld-ee t plasmoni modulator integrated with a sili on-on insulator waveguide

(SOI-WG) using thestandardCMOS te hnology.

The materialaspe ts and also thete hnologi al steps required inorder to realize an

integratedplasmoni modulator ompatiblewithrequirementsofCMOSte hnologywere

investigated.First,wedemonstrate aMetal-Nitride-Oxide-Semi ondu tor(MNOS)sta k

forappli ationsinele tro-opti alplasmoni devi es,sothataverylowopti allossesand

reliable operation is a hieved. This obje tive is met thanks to a areful hoi e of

mate-rials : (i) opper asa metal for supporting theplasmoni mode and (ii) stoe hiometri

sili on nitrideasan ultrathinlow opti allossdiusionbarrier to the opper. Final

ele -tri alreliabilityisabove95

%

fora3nmthi kSi

3

N

4

layer,leakage urrent densitybelow 10

−

8

A. m

−

2

and opti al losses aslowas 0.4dB.

µ

m

−

1

for a 13 nmthi k insulator

bar-rier, inagreement withthe losses of the fundamental plasmoni mode estimatedby3D

FDTD al ulations, using the opti al onstant ofCu measuredfrom ellipsometry.After

demonstratingtheMNOSasanappropriatestru ture forele tro-opti alCMOS

plasmo-ni s,wefabri ateaverti alMetal-Insulator-Si-Metal(MISM)waveguide integratedwith

an SOI-WG, where theba kmetalwasfabri atedbyipping and mole ular bonding of

High-k diele tri s aredemonstrated aspromising solution to redu eboth theMISM

absorptionlossandtheoperationvoltage.Given thatinterest, weexperimental

demons-trate an ele tri al reliable high-k sta k for future appli ations to the MOS plasmoni

modulators.

A few

µ

m long plasmoni modulator is experimentally investigated. Devi es show leakage urrentbelow10fA throughthe opperele trodesbasedMOS apa itan e. The

a umulation apa itan e (few fF) was foundto s ale with thesurfa e of thedevi e, in

onsistentwiththe expe tedequivalentoxidethi knessoftheMOSsta kofour

modula-tor.Alowele tro-absorption(EA)modulationshowing apa itivebehaviourwas

experi-mentally demonstrated inagreement withsimulations. Finally,lowenergy onsumption

(7)

1 Introdu tion 2

1.1 Sili onTe hnology . . . 2

1.2 CMOSSili on Photoni s . . . 6

1.3 Opti al Modulators . . . 9

1.3.1 Photoni modulators . . . 9

1.3.2 Plasmoni modulators . . . 12

1.4 Integrated PlasMOStor. . . 17

1.5 Outline ofthis thesis . . . 19

2 Experimental and numeri al methods 22 2.1 Ele tri al hara terization methods . . . 23

2.1.1 C-V andG-V measurement . . . 24

2.1.2 Non idealC-V hara teristi s . . . 28

2.1.3 Linear rampvoltage stress measurement . . . 29

2.1.4 Modelling andExtra tion ofMOSparameters . . . 31

2.2 Opti al methods . . . 34

2.2.1 Opti al index ofthematerials usedfor thesimulations . . . 34

2.2.2 Mode solverfor planar MOSplasmonwaveguides . . . 35

2.2.3 Finite Dieren eTimeDomain(FDTD) . . . 37

2.3 Ele tri aland opti al hara terization methods . . . 38

(8)

2.3.3 Ele tro-opti al hara terization . . . 40

3 CMOS Integration of MOS Plasmoni Modulator 45 3.1 Choi eof materials . . . 46

3.1.1 Choi eof Cufor CMOS plasmoni s. . . 46

3.1.2 Choi eof diusionbarrier . . . 47

3.2 Improved ele tri al reliabilityof CubasedMNOSsta k. . . 50

3.3 Low opti allossCubased MNOSsta k . . . 56

3.4 Fabri ation ofan integratedplasMOStor . . . 59

3.4.1 Grating opti al ouplers . . . 62

3.4.2 Siwaveguide . . . 62

3.4.3 Gate metal . . . 64

3.4.4 Ba kmetal . . . 66

3.4.5 Ele tri al onta t pads . . . 67

3.5 Con lusion. . . 67

4 E ient light oupling fromSi-WG tointegrated plasMOStor 69 4.1 Introdu tion . . . 69

4.2 Review ofplasmoni oupler stru ture . . . 71

4.3 Novel oupling stru ture . . . 74

4.4 Dis ussion . . . 80

4.5 Comment on the planar stru tureand theCMOS integrated devi e . . . . 85

4.6 Con lusion. . . 88

5 Ele tro-Opti alInvestigationofIntegratedMOSplasmoni Modulator 89 5.1 Insertionlossof aMISMsta k . . . 90

5.2 Interest ofhigh-kdiele tri s to optimize the insertion lossof aMISMsta k 90 5.3 Ele tri alreliable high-kgate sta ks . . . 96

(9)

5.6 Ele tro-opti al hara terization . . . 113

5.7 Con lusion. . . 115

6 General on lusion and perspe tives 118 7 Dispersion relation for Metal-Insulator-Sili on-Substratewaveguide 123 8 Interest of high-kdiele trisfor Metal-Insulator-Sili on-Insulatora tive plasmoni devi es 127 9 Résumé de THÈSE 1 9.1 abstra t . . . 1

9.2 Introdu tion . . . 2

9.3 Méthodesexpérimentalesetnumériques . . . 6

9.4 Integration CMOS d'unmodulateur plasmonique MOS . . . 8

9.5 Couplage e a e de la lumière d'un guide sili ium vers le plasMOStor intégré . . . 11

(10)

1.1 Transistor per hip for Intel entral pro essing unit (CPU) . . . 3

1.2 The history of the transistor : (a) Firstexperimental demonstration of a

transistor;(b)MOSFETwithgate lengthof35nm;( )Tri-gatetransistor

with22nm gate length . . . 4

1.3 Delay time for using dierent inter onne ts te hnology for dierent

te h-nology nodes [1℄ . . . 5

1.4 The basi opti al omponents of integrated ir uit [2℄ . . . 7

1.5 Opti al mi ros ope view of the sili on-based opti al modulator

demonstra-ted experimentally using standard CMOSpro essing te hniques [3℄. . . 11

1.6 Transverse dependen e of the E eldof SPPs on a Cu/SiO

2

interfa e. . . 13

1.7 Heldmodeprolesofthetwowaveguided SPPmodes:Thehighly onne

modeinsidetheMetal-Insulator-Metal(MIM)sta k(red)andtheless onne

metal/air mode(blue). Figure taken from[4℄. . . 14

1.8 A ording to [5℄, plasmoni s oers the sizeof ele troni s and the speed of

photoni s. . . 15

1.9 Cross se tion (s hemati ) of the plasMOStormodulator [6℄ . . . 18

1.10 Mode proles, refra tive indi es and losses supported by the plasMOStor

sta k (a); Opti al intensity as fun tion of the sour e-drain separation for

the on state ando-state(a) . . . 19

(11)

ture. . . 24

2.2 C-V hara teristi of MOS apa itor onsists of Cu/Si

3

N

4

(3nm)/SiO

2

(10nm)/p-doped Si (10

18

)at 100 kHz voltage sweep frequen y. Blue line:

experimental results, red line : simulations (see se tion 2.1.4.1). The

a - umulation and depletion modeare also presented. . . 25

2.3 Experimental results of ultra low apa itan e measuring using the

impe-dan e analyser meter, HP 4284. The measurement was performed to the

plasmostor devi e of a tive length = 1

µ

m . . . 26 2.4 (a)AnexampleofC-V

g

andG-V

g

measurementat50kHzand100kHz.(b)

Extra tionof interfa e defe t density usingthe ondu tan e method

deve-loped by Ni ollian et al. [7℄. . . 27

2.5 (a) Measured apa itan e at dierent frequen ies illustrating the

tremen-dous ee t of the unwanted loss diele tri layer taken from [8℄; (b) The

equivalent ir uit model of the real MOS apa itor in luding the ee t of

lossydiele tri layer;( )The equivalent ir uit model of (b) but in strong

a umulation. . . 29

2.6 Typi alWeibuls aleplotofthemaximum voltagetobreakdownasobtained

on a population of 70 MOS apa itors using LSRV . . . 30

2.7 (a)S hemati pi ture of a MOSstru ture;(b)Ele tri al equivalent model

for the apa itan e. . . 31

2.8 Diele tri fun tion (

ǫ = ǫ

′

+ iǫ

′′

) for dierent wavelenghts (

λ

) for the fabri ated Cu metalmeasured by ellipsometry, with diele tri fun tionsof

Altaken from[9℄ . . . 34

2.9 Generalizeds hemati ofthesemi-infnitemetal-oxide-sili on-substrate

wa-veguide. . . 35

2.10 Numeri al zero of the determinant estimated by MATLAB. . . 36

(12)

away fromthe metal interfa e of the plasmoni waveguide. . . 38

2.13 Pi ture of the developed EO experimental set-up. . . 39

2.14 Experimentalset up for measuring the transmission of the devi e. . . 40

2.15 Experimental set up used for the wafer s ale EO measurements at low

frequen y (below 10kHz) . . . 41

2.16 (a) A non leaky devi e of 1

µ

m

2

surfa e is measured with its modulation

depth and transmission; (b) A leaky devi e of 5

µ

m

2

surfa e is measured

withitsmodulationdepthandtransmission.Thisdevi e willex luded from

the analysis of the data. . . 42

2.17 Experimentalset upused formeasuring thebandwidthofthe devi e (upto

200MHz) . . . 44

3.1 Plasmostor design [10℄ with silver as a metal ele trode and without any

onsideration about interfa ial layer between the metal and insulator or

semi ondu tor. . . 46

3.2 Plasmonpropagationlossesindu ed by Cu andAl alonga metal/diele tri

interfa e for dierent wavelengths. . . 48

3.3 Fabri ation pro ess of MOS apa itors using standard CMOS te hnology. . 51

3.4 C-V hara teristi of MOS apa itor at100 kHz voltage sweep frequen y,

forasingeoxide(MOS,bla ksymbol)andanoxide/nitridebarrier(MNOS,

blue symbol). The orresponding quantum simulation of the urve is also

shown (Bla k and blueline), followingthe model presented in the se tion 2. 52

3.5 (a)Experimentalreliabilityplot(inWeibuls ale)ofthebreakdowneld;(b)

SIMSmeasurementofdepthproleofCopperintensityforMOSandMNOS

(13)

quen y,fora singleoxide (MOS,bla ksymbol) andan oxide/nitridesta k

(MNOS,bluesymbol);(b)Defe tdensity,D

it

atdierentenergiesintheSi bandgap for a singe oxide (bla k symbol) and an oxide/nitride sta k (blue

symbol). . . 54

3.7 Example of leakage urrents as a fun tion of the applied voltage for (a)

MOS and (b) MNOS multilayer insulator, showing the two dierent

po-pulationwithjust ohmi ondu tionandthe enhan ed ondu tionstarting

from-3.5 V . . . 55

3.8 Leakage urrent in a (a) FNplot and (b)FP plot.. . . 55

3.9 Normalized experimental transmissiondata asa fun tionof the L

GAT E

. . 57 3.10 (a)Crossse tion inthe propagation dire tion indi ating the main

ompo-nents of the hybrid plasmoni WG;(b) The E and H eld intensity of the

fundamental plasmoni mode isdepi ted for Cu-Si3N4 (barrier)-SiO

2

-Si-SiO

2

stru ture taken from 3D mode analysis using FDTD;( ) Cal ulated transission through the devi e by using 3DFDTD.. . . 58

3.11 (a) E and H eld of the fundamental plasmoni mode; (b) E and H eld

of the mode supported inthe side of the metal . . . 59

3.12 S hemati top view of the integrated plasmoni modulator (a) with

dif-ferent ross a ordingly to dierent planes; ross se tion in the dire tion

of propagation (b) rossse tion in perpendi ular dire tion ( ). . . 60

3.13 S hemati representation of the ele tro-opti al plasmoni modulator after

the 5 mainfabri ation phases. . . 61

3.14 (a)S hemati diagramoftheopti al ouplers;(b)SEM rossse tion

Fpi -tureofthe ouplersindi atingthemaindimensionsofthe riti al omponents 62

3.15 Si-WG oupled withthe Si-PWG. . . 63

3.16 (a- ) Fabri ation progress of sili on WG, using fully CMOS ompatible

(14)

andthe gate metal. . . 64

3.18 (a- )Dierentfabri ationphaseofMNOSplasmoni WG;(d)SEMimage

of the nalPWG. . . 65

3.19 (a) S hemati and (b) SEM ross se tion of the nal stru ture; ( ) Top

view of 200 mm wafer in luding more than 30000 devi es;(d) bino ular

top viewof single devi e indi ating the important omponents; . . . 66

4.1 (a)Abruptjun tion between the Si WGanda metalbased waveguide; (b)

The EandH eldintensity of thefundamental plasmoni modesupported

in Si WG and MISM plasmoni WG taken from3D mode analysis using

FDTD . . . 70

4.2 Hybrid opti al hip showing dire tional oupling between the Si-WG and

the metalslot waveguide [11℄. . . 72

4.3 Ee tive index of thethe fundamental modesupported by the MISMsta k

for dierent sili on andinsulator thi kness . . . 73

4.4 (a)Lateral taper[12℄;(b)verti altaper[13℄ onne ting diele tri WGwith

metalWG.. . . 73

4.5 (a) Hybrid plasmoni waveguide showing the oupling stru ture (MISI)

onne ting the SOI-waveguide and MISM; (b) SEM image of the ross

se tion of the MISM stru ture . . . 75

4.6 Cal ulated ree tion for dierent MISMlength for two ases : L

c

= 0

µ

m andL

c

= 0.7

µ

m. . . 76 4.7 Simulated transmissionfor dierent oupling lengths L

c

and L

M ISM

=1

µ

µ

m. . . 78

(15)

4.9 Cal ulated transmission as a fun tion of the MISM length (L

M ISM

) ou-pling lengths equal to for L

c

=0.7

µ

m. By an exponential t (red line) we extra t the propagation andthe insertion loss . . . 79

4.10 EandHeldintensityofthefundamentalmodes supported inthedierent

se tions (SOI-WG,MISI-WGand MISM-WG). . . 81

4.11 (a)Crossse tionoftheinvestigatedstru tureinthepropagationdire tion;

(b) Field intensity through the hole devi e (dire tion x) from a monitor

m,L

c

=0.7

µ

m). The results obtained from the dire t oupling theory (blue symbols) are also presented. . . 86

4.15 Modeproperties of the plasmostor [10℄ . . . 86

4.16 (a)Devi e s hemati of the integrated plasmostor; (b)Mode propertiesof

the integrated plasmostor . . . 87

5.1 S hemati rossse tion of the a tive integrated plasmostor . . . 91

5.2 S hemati of the 2DMISMplasmoni waveguide with innite width . . . . 91

5.3 Plasmoni losses as a fun tion of the gate oxide physi al thi kness with

SiO

2

and Al

2

µ

m . . . 102 5.11 Carrier on entration ina umulation onditionsusingsemi lassi al and

Shrodinger-Poissonmodel . . . 103

5.12 Carrier on entration ina umulation onditionsfor dierent applied

vol-tages . . . 104

5.13 Capa itan e voltage hara teristi s for an Ohmi and S hottky ba k onta t106

5.14 (a)Modulation depthestimated by 3DFDTDsimulations;(b) Mode

ana-lysis of the on state for the fundamental modes supported by MISI and

MISMse tion . . . 108

5.15 Theoreti al investigation of the DC modulation depth for dierent

opera-tionmodes of the MOS apa itor (a umulation, depletion, inversion). . . 109

5.16 (a)Capa itan e-voltage hara teristi fortheintegratedplasmostor;(b)DC

leakage urrent at dierent voltages . . . 111

5.17 Oxide apa itan e at -5V asa fun tionof thesurfa e . . . 112

5.18 Condu tan e-voltage hara teristi for the integrated plasmostor . . . 113

5.19 (a) Experimental modulation depth measured at 1 KHz; (b) Theoreti al

andexperimental data of modulation depthat voltage of -3 V. . . 116

5.20 (a)Smallsignalopti alresponse(datatakenfromthelo kinamplier) as

(17)

veguide. . . 123

8.1 Per entagegaininpropagationlength(%) omparedtotheSiO

2

asa fun -tionof EOT and Si thi kness when Al

2

O

3

or HFO

2

are used. . . 129

9.1 Les omposants debase des ir uits intégrés optiques [2℄ . . . 3

9.2 plasMOStor intégré surplateforme SOI . . . 5

9.3 Imagedu montage de mesure de la modulation. . . 7

9.4 Experimentalset upused formeasuring thebandwidthofthe devi e (upto 200MHz) . . . 9

9.5 (a)la abilitépartielleexpérimentale (àl'é helledeWeibull)du laquage; (b)la mesureSIMS dela profondeurdeproldel'intensitéde uivre pour des empilements MOS et MNOS . . . 11

9.6 Transmission optique normalisée de données expérimentales en fon tion deL

GAT E

. . . 12

9.7 S héma vue de dessus du modulateur intégré plasmonique (a) ave oupe transverse àdes plansdiérents; et(b)se tion transversale dansla dire -tiondepropagation( )se tion transversale dansla dire tion perpendi ulaire. 13 9.8 (a) S héma et (b) se tion transversale MEB de la stru ture nale; ( ) Vue de dessus de wafer de 200 mm, in luant plus de 30000 dispositifs; (d)vision bino ulaire supérieure d'un dispositif en indiquant les éléments importants; . . . 14

9.9 (a) S héma de guide d'ondes hybride plasmonique montrant la stru ture de ouplage (MISI)reliant leguided'ondeSOIetMISM.(b)l'imageSEM dela se tiontransversaledela stru ture MISM.( )imagesMEBdes ou-pleurs envue dedessus . . . 15

9.10 Transmissionsimulépourle ouplagedeslongueursL

c

diérentesetL

M ISM

aga-tion; (b) l'intensitédu hamp(b) par l'intermédiaire dudispositif detrou

(dire tion x) à partir d'un moniteur pla é dans l'isolant degrillede

l'em-pilement ( )intensitéde puissan e dans la dire tion depropagation (Px),

sont utilisés. Pour une tension de fon tionnement faible, il est intéressant de

travailler ave une épaisseur variant de 1 EOT nm

<

EOT

<

10 nm. A ette distan e, enutilisant Al

2

O

3

omme une grande "K" isolant,il est observéau ungainsigni atif(moinsde10

%

)enlongueurdepropagation du mode fondamental plasmonique par rapport au as de l'utilisation de

SiO

2

de l'EOT même. D'autre part, HfO

2

est identié omme un isolant prometteurpour laplasmoniquea tifsdepuis sonutilisationva augmenter

la longueur depropagation jusqu'à 60

%

pour la plage pré itée de l'EOT. . 19 9.14 (a)Taux de modulation estimée par des simulations FDTD 3D,l'analyse

du mode (b) de l'état sur les modes fondamentaux soutenus par l'arti le

MISI et MISM . . . 20

9.15 Réponse optique de petites signaux enfon tion dela tension appliquée et

(19)

1.1 Performan es of opti almodulators re ently demonstrated expirementaly. 12

3.1 Plasmoni MOSmodelossesfordierentdiusionbarriersatawavelength

of 1.55

µ

m. The operation voltage for onstant ele tri al performan e is alsopresented. . . 48

3.2 EOT andV

f b

ofgivenMOSsta klayer,extra ted bytting to C-V mea-surements. Values are averaged out of 10 dierent devi es, for ea h gate

sta k. . . 51

4.1 Experimentalandtheoreti alpropagationlossfortheinvestigated

geome-tries. . . 85

5.1 EOT of dierent gate sta ks investigated in this thesis extra ted from

C-V measurements with the orresponding a umulation apa itan e at

-3MV/ m. . . 99

(20)

Introdu tion

Sili on (Si) te hnology is onsidered as the ideal platform to merge photoni s with

ele troni s owning to its promise of low ost,high yield and devi e integration. In this

platformtherealizationofphotoni -ele troni s ir uitsisstronglydependent ofthe

avai-labilityofSibasedopti almodulator thatenabletheintera tionofele tri alandopti al

signal.Inthis hapter wewill give ageneralintrodu tionaboutthestate oftheartofSi

te hnology andSiphotoni s.Thenwe willfo usonopti almodulatorsandinparti ular

plasmoni modulators basedon MOSte hnology. Finallythe outlineof this manus ript

ispresented.

1.1 Sili on Te hnology

Itisgloballya eptedthattheSili onte hnologyisthedrivingfor eofthe ontinuous

te hnologi al progress. The tremendous su ess of this te hnology is attributed to the

fa t that by s aling the length of the metal oxide semi ondu tor (MOS) transistor we

an a hieve improvement of performan e and redu tionof thepri e pertransistor. This

su ess was predi ted by the famous Moores law [1℄ whi h denes that it is possible

to pa k more and more transistors in the same area as a onsequen e of s aling down

the size of ea h transistor. At the same time better performan e an be a hieved (Fig.

(21)

initial dimension. This thus results in a MOS transistor with smaller, faster and more

energy e ient hara teristi s[14 ℄.

Figure 1.1 Transistor per hip for Intel entral pro essing unit (CPU)

Fig. 1.2 shows the evolution of the transistor, starting from the rst Metal Oxide

Semi ondu torFieldEe tTransistor(MOSFET)withalengthof100

µm

,tothepresent te hnology ofthe 22nmlengthtri-gatetransistor,originallyinventedbyIntel.Finally,it

isseen that moretransistors an be pa kaged ina given area. For massprodu tion this

an be translated into huge e onomies of s ale.

The imminent drawba k of s aling is often referred to as the "Inter onne t

bottle-ne k". As the dimensions of MOS a transistor are s aled down, more inter onne ting

linesarerequiredinorderto onne tea h transistor,resulting inlessspa ebetween the

inter onne ts. This in reases the apa itan e between the inter onne t lines.

Additio-nally,informationtransferthroughinter onne tsresultsinanin reaseofheatdissipation

due to resistivity. Thesetrends are responsible for the in reased inter onne t

Resistive-Capa itive (RC)delayleading toloweroperation bandwidth ofthedevi e. Thesolution

(22)

123 ₁₄₃

₁₅₃

Figure 1.2 The history of the transistor : (a) First experimental demonstration of a

transistor; (b) MOSFET with gate length of 35nm; ( ) Tri-gate transistor with 22 nm

gate length

to onventional usedaluminium inter onne ts due to thefa tthattheformer hasa low

resistivitythan the latterbyafa toroftwo.Theissueofin reased apa itan eissolved

throughthe useof a lowdiele tri onstant material (low"k").

Choosing opper as the new material of the ele tri al inter onne ts reates several

issues to the manufa turing pro ess. In the inexisten e et hing method of opper, new

te hniques for deposition are implemented namely "Damas ene" te hnique. This

te h-nique is based on hemi al-me hani al-planarizatio n (CMP) [15 ℄. Furthermore, Copper

isa fast diusionmaterial inthe Sili on andinsulators and reatesshort ir uits inthe

transistors.Hen e,severaldiusionbarriersfor opperareproposed[16℄.Thedeposition

andpatterningof opperisthereforewell masteredinaCMOSenvironment,andthanks

to that, s alingisallowed to ontinue.

Using opper with very low k insulators as an inter onne t stru ture may extend

the ontinued use of ele tri al intere onne ts for a while (Fig. 1.3). AsMooresaid "No

exponential isforever:but "Forever" anbedelayed"[17 ℄.However,asthespeedofthe

travelling information be omes higher and higher, these solutions will be inappropriate

be ause they will reate too mu h heat. To that respe t new long term solutions are

(23)

nology nodes [1℄

Opti al inter onne ts suggestedasthemost promising innovative path for repla ing

the ele troni data bus thereby solving the issue of "inter onne t bottlene k". In this

prototype inter onne t s heme, the information transmitted from one point to another

uses photons rather than ele trons and opper wires,with expe ted data transferto be

in the range of 50 gigabits per se ond and also with mu h less power dissipation [18℄.

Thebreakthrough ofa light-based te hnology requiresanumberofbuildingblo ks su h

as lasers, opti al waveguide, modulators and photodete tors. Si photoni s has ome to

(24)

Sili on photoni s refers to the te hnology that reates high performan e Si based

opti al omponentstakingadvantageofCMOS ompatiblematerialandte hniquesused

in ele troni te hnology. For example, the long-standing (more than 40 years) use of

Sili onfordevi efabri ationinele troni industry,enablesnotonlysoundunderstanding

of the physi s of Sili on but also manufa turing tools to be well mastered. Thanks to

that,smallele tri al omponents(fewnanometers) whi hareavailable today.Bytaking

advantage of that and also with the development of the sili on-on-insulator te hnology

we are able now to transfer the knowledge and the te hnology of Si based ele troni s

in order to fabri ate small and low ost opti al omponents thus making feasible the

development of sili on-based photoni integrated ir uits.

As it is already observed in long-distan e data propagation (ex eeding 10 meters),

ele tri al "buses" are su essfully repla ed by opti al ones oering more than 100,000

timeshigher datatransfer.Therefore the limitationsof ele tri alinter onne ts are

over- ome.Thesametrendnowappliestoon- hipdata ommuni ationwiththe opper wires

to be repla ed with the so alled opti al inter onne t system. The onstru tion of this

systemrequireselementary opti al omponents su h aslasers,modulator, opti al

wave-guides and photo-dete tors to be integrated on a Sili on wafer (Fig. 9.1). An Si based

photoni integrated ir uit[2℄transfersopti alsignalfromonepointtoanotheroneusing

(a) an opti al sour e to produ e ontinuous light, (b)a modulator to en ode the

opti- aldata, ( )multiplexer to oupledierent wavelengths to thewaveguide,(d) an opti al

waveguide as the data opti al 'bus' and (e) a photodete tor in order to transform the

opti al signal to ele tri al. Several paradigms of Sibased passive devi es su h as

wave-guide and multiplexer arealready proposed in the literature. During thelast few years

most of the eorts have been dire ted towards a tive photoni devi es su h as lasers,

photodete tors and modulators. The riti al issues involving these elements for

intra- hipappli ationsarepower onsumptionand devi efootprint.Someofthesedevi esare

(25)

As an opti alsour e, semi ondu tor diodelaser is an ex ellent andidate be ause it

emits mono hromati and dire tional light. The prin iple of operation is based on

en-han ement of radiation bystimulatingemission. The areful hoi e of materialis again

riti alinthis ase.Using sili onasthea tivematerial for laseris notagood hoi e

be- auseit hasan indire t bandgap.This meansthatwhen the ondu tion bandele trons

areex ited,the re ombination pro ess ofele tronswillgivelightonly ifthereis

momen-tum(absorption oremissionofphonon)andenergy onservation,makingitlesse ient

for radiative emission. A sili on based laser operated by Raman s attering instead of

photon emission is reported [19℄. However, it is hard to implement for pra ti al

appli- ation as the ex itation omes from another pump laser. Better materials for emitting

light arethoseofdire tband-gap,like theIII-Velementsbut arenot CMOS ompatible

asthereis a rystallatti e mismat h ompared to the sili on.Although, theintegration

remains a non-trivial problem it an be solved by the new te hnologi al a hievements

su haswaferbondingte hnology.Giventhat, several typesoflaserswerereported su h

asmi rodisks[20 ,21 ,22℄,Fabry-Perotlasers[23 ,24℄,ra etra kresonatorslasers[25,26℄,

Distributed feedba k lasers [27 ℄ and Distributed Bragg ree tor (DBR) lasers [28 ℄.

Mi- rodisklasersareofparti ular importan eforintra- hipopti alinter onne ts,thanksto

(26)

Campenhoutetal.[20℄,threshold urrentsaslowas0.5mAforsmallfootprint (few

µm

) mi rodisks anbe a hieved.

A photo-dete toris adevi e that onverts in ident light into an ele tri urrent and

isusually a omplished byusing aphotodiodein photovoltai mode. Whena photon of

energy higherthan theband gapofthea tive materialisstrike thediode,itisabsorbed

andanele tron-hole pairisgeneratedinthejun tion.Theinternal potential ofthediode

auses the harge arriers towardstheele trodes anda photo urrent isprodu ed. Using

sili on for photodete tion in the wavelength range of tele ommuni ation (1.4-1.55

µm

) is not a good hoise sin e it is transparent in the infrared (below 1.1

µm

). Therefore materials with lower band gap su h as germanium [29 , 30℄ or InGaAs [31, 32 ℄ grown

on a Si substrate have demonstrated as more promising solutions. Both te hnologies

have approximately the same sensitivity (1 A/W) but the speed of operation of the

InGaAsdete tors remains relatively low(smaller than 10 GHz[31 ℄ ompared to that of

germaniumphoto-dete tors(higherthan40GHz[33 ℄).Thelatterhavealsoami rometer

s ale footprint and operateat 1.5Vwhi h makesitideal for intra hip appli ations.

Modulators are devi es that transform ontinuous opti al signal oming from the

laser sour e to dis rete pie e of light. Two main me hanisms are used to a hieve light

modulation, namely ele tro-absorption and ele tro-refra tion. The former modulate the

amplitudeofthetransmittedpowerandthelatter hangesthephaseofthewave.

Sili on-based modulators have gained mu h attention, with several devi es reported with high

performan es (brieydes ribed tothese tion1.3). Thesedevi esrelyonthefree arrier

plasmadispersionee t [34℄.Astheunderlyingphysi alee tisnotstrong,theseeorts

have led to large stru tures ( ompared to that of laser and photodete tor) with high

driving powersfor obtaining asigni antmodulation depth.

Modulatorsare the largest(regarding the size)and themost demanding in termsof

energy among theopti al omponents of thephotoni intgrated ir uit.Tothat respe t

aparti ularattentionisgiventothosedevi es.Thegoalofthisthesisistoinvestigatean

opti almodulator using theSiphotoni splatform existinginCEA LETI.The following

(27)

Intherealization of hip-based hybrid ele troni -photoni networks, integrated

opti- al modulators areurgently required for any opti al building blo k. The easiest wayto

perform lightmodulationisto swit hon andotheopti alsour everyfast(10GHz)by

modulating the inje ting urrent (dire t modulation). However, this approa h involves

onstant hangesinthetemperatureleadingtoperturbationoftheoperationwavelength

(â hirpâ)and,inturn,distortstheopti aldata.In ontrastanexternalmodulation

oerstheadvantageofhighdatatransfer(10-40GHz)andalso wavelength stabilization.

Inthis s hemethe opti alsour e isswit h tothe ontinuousmodeand theexternal

mo-dulator work su h as shutter on-o for the ontinuous light. Therefore external opti al

modulators expe ted to be the key omponent to push towards higher bandwidths and

lowenergy onsumption of datatransfer.

1.3.1 Photoni modulators

Today, ommer iallyavailableopti almodulatorsat10Gbpsarebasedonlithium

nio-bateandIII-V ompoundsemi ondu tors.Whiletheformer requirestheusenon-CMOS

ele tro-opti al materials, the latter works on the prin iple of ele tro absorption.

Howe-ver both materials are ontaminants within the CMOS integrated photoni -ele troni s

ir uit. Using a tive materials that are ompatible with the urrent mi ro-ele troni s

pro esses,we areableto ir umvent theinherentlimitationsoftheproposedapproa hes

listedabove.Sili onitselfisaCMOS ompatiblematerialandthereforemodulatinglight

through sili on will be the essential solutionfor thefuturephotoni te hnology.Several

demonstrations of integrated Si based modulators have been published/patented sin e

2000 and arebrieydes ribed below.

E ientlightmodulatorsusingsili onasa tivematerialarealreadyproposed.Using

strain we an break the rystal symmetry of Sili on, leading to a linear ele tro-opti

ee t useful for light modulation [35℄. SiGe/Ge quantum wells is also proposed for low

(28)

V materials to sili on may enable to use their strong ele tro-opti al properties su h as

thermo-opti plasmadispersion ee tsaswell asplasmadispersionee t.

Although sili on does not in normal ir umstan es exhibit an ele tro-opti ee t,

other me hanisms are available for modulation and in parti ular the plasma dispersion

ee t has been proven as a promising approa h of a hieving high speed modulation

in sili on devi es. In this ee t the real and imaginary parts of the refra tive index

are hanged due to the on entration of free harges in sili on. Soref et all [34℄ have

experimentallydemonstratedthatindu edrealrefra tiveindexandabsorption oe ient

variation (due to thefree harges) at a wavelength of 1.55

µ

mis given bythe following formula :

∆n = ∆n

e

+ ∆n

h

= −[8.8 × 10

−

22 ∆N

e

+ 8.5 × 10

−

18 (∆N

h

)

0.8 ]

(1.1)

∆a = ∆a

e

+ ∆a

h

= −[8.5 × 10

−

18 ∆N

e

+ 6 × 10

−

18 (∆N

h

∆a

h

are the hanges in absorption resulting from hanges inthefree-ele tron and free-hole arrier on entrations,respe tively.

The required hange inthe free arrier on entration isa hievable througha variety

of dierent me hanism su h as arrier inje tion, depletion or a umulation. Carrier

de-pletionand a umulation, unlike arrierinje tion, arenot limitedbytherelatively long

arrierlifetimeinsili on and onsequentlythefasterreporteddevi eshaveutilised these

me hanisms.However,thefree arrierdispersionee tinsili onisinsu ient,leadingto

longdevi es[3℄(LisoftheorderofmillimetresforthemodulatorshowninFig.1.5)with

highenergy demands and thereforeinappropriate for intra- hipopti alinter onne t.

Two mainee ts, namely Ele tro-Absorption (EA) andEle tro-Refra tion (ER)are

used to a hieve light modulation. EA indu e absorption (by hanging the opti al

(29)

L=1.8 mm

Figure1.5Opti almi ros opeviewofthesili on-based opti almodulatordemonstrated

experimentally using standard CMOSpro essing te hniques [3℄.

ERindu e phaseshiftof thepropagatingwaves(by hanging therefra tiveindex ofthe

material) and an be exploit for modulation in two typi al geometri al onguration,

namely Ma hâZehnder interferometer (MZI) and ring resonator. In the ase of ring

resonators [37 , 38℄, the indu e phase shift, hanges the resonant ondition allowing to

swit hbetweenon-ando-resonantstates.UsingMZI,we onverttheindu ed hangesof

therelativephaseofthetwopropagatingmodes,tointensitymodulation.Althoughmore

ompa t devi es are a hievable using resonant based stru tures, the MZI onguration

remainsmore attra tive.Thisisbe ause theresonator opti albandwidthislimitedto 1

nmand thus makesit ultrasensitive to externalparameters su hasthetemperature.

The ideal Sili on opti al modulators will be hara terized by CMOS ompatibility,

ultralow power onsumption, small devi e size, high modulation speed as well as large

modulationdepthandopti albandwidths.Giventhefa tthattheserequirements

ontra-di tea h other, innovation isneeded inorder to ombineall theserequirementswithan

optimal tradeo. Re ently,resear hers proposedthatbyusing theproperties of surfa e

plasmons [39℄,su h ashigheld onnement it ispossible to reate amodulator whi h

fulllseveral ofthe mentioned requirements. Su hkindof modulatorsisreviewed tothe

(30)

Modulation Speed Footprint Modulation Energy Working

Prin iple A hieved (

µm

) Voltage(V) (fJ/bit) Spe trum

(Gbit/s) (nm) Intel MZ 20/s Depletion Gbit/s 6000 6.5 28000 30 pn jun tion IBM-Si 10 inje tion Gbit/s 200 7.6 5000 -pin Cornell 12.5 inje tion Gbit/s 33 8 300 1 mi roring MITBAE 1.2 GeSi Gbit/s 30 3 50 14 EA Kotura 12.5 Ge Gbit/s 45 4 100 30 EA pin 1.3.2 Plasmoni modulators

Surfa e plasmons polaritons (SPP) are ele tromagneti waves strongly oupled to

thefreeele trons ofthemetal.Theyarepropagatingalongthemetal-diele tri interfa e

(dire tion x) and de ays exponentially asone moves away from theinterfa e (Fig. 1.6).

The eld penetration (D) into the diele tri an be subwavelength and it depends on

[44 ℄.

Surfa e Plasmon eld onnement is expe ted to be very helpful in the ontext of

modulators. Theenvironment sensitivityof metal/diele tri interfa e SPPmodesiswell

known and it is already used for dierent appli ations su h as bio-sensing and

opti- al a tive omponents. The physi al origin of this sensitivity is basi ally related with

the eld onnement of the SPP mode at the metal surfa e. However, when

onsi-dering stru tures more sophisti ated than just a metal/diele tri interfa e, su h

me-tal/semi ondu tor/metal systems,this onnement an be in reased to a levelthat an

be only hardly rea hed with standard index-guided modes (Fig. 1.7). Thus, the eld

onnement properties ofSP modes andthe in reasedsensitivityto small refra tive

in-dex hanges ofthediele tri in onta twith themetalmake thesemodes rather unique

for appli ationstargetinghigh-frequen y modulation ofan opti alsignal. Fig.1.8shows

the performan es of dierent te hnologies in termsof speed and size. Based on this

(32)

mode inside the Metal-Insulator-Metal (MIM) sta k(red) and the less onne metal/air

mode(blue).Figure taken from[4℄.

and alsoto travel datawiththe speed ofphotoni s. The drawba khere isthatthehigh

onnement isasso iatedwithpropagation losses.Thereforebydesigningour devi e we

have to be areful to have high modulation depth for ompa t devi es (resulting from

high onnement)while keepingthe opti allossesin a eptablevalues.

Taking in advantage of the key assets of use of metals for both ele troni s

(ele -trodes) andphotoni spurposes (ele tromagneti on entration) plasmoni s an fuelthe

prospe ts ofs ale downthe dimension of opti almodulators.

Dierent s hemes of SPP-based modulators were proposed and reported so far.

Ba-si ally, one an identify three types of SPP modulators;(i) all-opti al SPP modulators

(ii)thermo-opti s modulators, and nally(iii) ele tro-opti smodulators.

Dierent ongurations of all-opti al SPP modulator were des ribed and reported

re ently. First, a rather spe i onguration using the properties of gallium was

sug-gested and demonstrated byKrasavin etal. [45 ℄ Thissystem relies of thephase hange

of gallium (due to its very low melting point) when illuminated by a pump signal at

(33)

photoni s.

interferometer onguration . In this onguration, the modulation of a SPP mode is

performed by hanging theopti al properties ofCdSe quantum dots. Whenpumped at

a frequen y within their maximum absorption band (around 515 nm), the Q-dots are

pla edinan ex ited statepromoting theabsorptionof theprobebeam(1415nm). This

light-indu ed absorption modies theintera tion between thein ident probe beam and

the probe SPP leading to the modulation of the signal at the probe frequen y. While

these results are impressive, it is lear that all opti al SPP modulators do not have a

pra ti al futurebe ausethey require materialsthat arenot CMOS ompatible.

Thermo-plasmoni modulators [46, 47℄ are already proposed based on metal

wave-guide supporting SPP modes and heat dissipation within the metal to indu e index

modulation. Thejusti ationfor theuseofmetal waveguide is thattheheatsour e an

belo alizedatthepla ewhereitisthemoste ient.Giventhat,Gos iniaketal[46℄

de-monstrate thermo-opti s plasmon modulators relying on Ma h-Zehnder interferometers

omprisedofgoldlong-rangesurfa eplasmonpolariton(LR-SPP)waveguides.Thistype

ofmodulatorsexhibitsstrongoutputmodulation,lowpower onsumption(1.4mW),but

(34)

lution for a hieving higherbandwidth modulation. A proposition was madebyJannson

et al. [48℄ and is based on mode oupling between light propagating in a diele tri

wa-veguide to a nearby metal lm pla ed a ertaindistan e away : the oupling oe ient

between the opti al mode in the waveguide and the plasmon polariton mode is made

variablethrougha e.g. voltage-dependant property(forexample therefra tive index)of

an ele tro-opti al materialin onta t withthe metal. Su h a material is oftentaken as

LiNbO

3

for example. However the ele tro-opti al onstants of those materials are weak and applied voltage remains highand systemdimensions large.Moreover, thelatter

as-pe tpre ludestheuseofplasmonmodesin ongurationwhereele tromagneti eldsare

very onnedbutlossy.The onguration proposedbyBreukelaaretal. [49 ℄,relying on

theele tri allyindu edlongrangeplasmonpolaritonmode ut o, ouldnot ir umvent

this drawba k.

More re ently, Liu and Xiao [50 ℄ suggested in a theoreti al work an ele tro-opti al

swit h(modulator)relyingon theuseofametallmsupporting surfa eplasmonmodes

embedded within a ferroele tri medium. By adjusting a bias voltage at the surfa e of

theele tro-opti al material(su hasBa0.5Sr0.5TiO3), theyshow numeri ally that for a

givenpropagation distan e, interferen ebetween thesymmetri andtheanti-symmetri

surfa eplasmonmodes ouldbeswit hedfrom onstru tivetodestru tive leadingtothe

lo alization of the SPP eld at one or the other metal thin lm interfa e. Following a

very similar strategy, this on ept was extended to the ase of LR-SPP and

demons-trated experimentally by Berini et al. [51℄. In this system, a very thin lm supporting

a long-range plasmon mode is embedded within two layers of lithium niobate with

Z-axisorientation pointinginoppositedire tions.Byapplyingavoltage,arefra tive index

ontrast between the two sidesof thegold lm is reated leading to an in rease of the

LR-SPPpropagationlosses.Amu h more ompa t modulator relyingalso ontheuseof

BaTiO3wasproposedandinvestigatedbyDi kenetal.[52 ℄.Thismodulatorisa tuallya

double-slitsinterferometerexploitingtheintera tionbetween adire tlytransmittedeld

(35)

CMOS platform.

Amongstthenumerouspatentsandpubli ationsdes ribingopti alplasmoni s

modu-lators,veryfewhavethepotentialtointegrateinCMOSintegratedphotoni s.Inare ent

arti le of Melikyan et all [53℄ they demonstrate an EA modulator operates at 1.55

µ

m. Inthis devi e, the absorption hange is aresult ofvoltage-indu ed arriers inITO,

pro-viding thus a modulation speed whi h is free fromany slow arrier-related phenomena.

Furthermore, Zhu et al. [54 ℄ they demonstrate a Si nanoplasmoni ele tro-absorption

modulator using opper as a metal ele trode. This novel devi e is fabri ated within a

CMOS environment. Theysuggestthatthis modulator relieson a umulation of harge

arriers inside a MOS, indu ed by an applied voltage. However, su h devi e dissipate

an una eptable power due to the high leakage urrent that is observed (0.1 mA for 2

µ

m

2

surfa e).Furthermore,we spe ulatethatthethermalee t playsanimportant role

to themeasured modulation depth due to the high urrent passing from the insulator.

Finally, the so- alled PlasMOStor devi e proposedbyDionne et al. [6℄,use Siasa tive

material in order to modulate light in a verti al Ag/SiO

2

/Si/Ag onguration. In the following se tion we present a detailed des ription of this devi e, sin e it is the subje t

ofthis work.

1.4 Integrated PlasMOStor

PlasMOStor it is an interferometri onguration used for performing modulation.

Thedevi eemploysafour-layerofAg-oxide-sili on-Ag(Fig.1.9)thatsupportsboth

pho-toni and plasmoni modes(Fig. 1.10a).Inthisele tro-opti al devi e,themodulation is

a hieved by hangingtheindex ofann-dopedSili onlayerbymeans of harges

a umu-lation in a MOS stru ture. Su h an index hange is at the origin of the disappearan e

of a hybrid photoni mode ex ited at a frequen y lose to its ut-o. The perturbation

of the beating reated bythe interferen e of the photoni mode and a plasmoni mode

(36)

to 4 dBina devi e2.6

µ

mlong (Fig.1.10b). Theoperating voltage of plasMOStor was estimated to be 1V whi h is ompatible with the low energy onsumption demands of

integratedphotoni ste hnology.Itisalsosuggestedthatthebandwidthofthedevi e,is

fundamental limitedbythespeed oftheformationof theMOSa umulation layer(asit

is ina onventional small-geometry MOS eld-ee ttransistor) and therefore gigahertz

operation is expe ted. However, this stru ture is not ompletely ompatible with the

CMOSenvironment be ause theyuse ontaminants metalssu h assilver.Thisissue an

be solved by swit hing to CMOSmetals su h as opper or aluminum.

Theoutstandingdevi eperforman esprovidedbytheinnovativemodulation on ept

oftheplasMOStor, originatestheinterestto integrate this on ept inaCMOS

environ-ment available on LETI platform in olaboration also with CALTECH. To do that we

take advantageofthekeyassetsofmi roele troni sandSiphotoni ste hnology,namely

maturity and the low-losslong-range opti al ir uitry (developed for CMOS photoni s)

andthewellknownfabri ationte hnologyofthemetals(developedformi roele troni s)

forbothele troni s(ele trodes)andphotoni spurposes(ele tromagneti on entration).

The experimental proof of su h approa h would bring a new paradigm in the design

of ele tro-opti al modulators. Fig. 9.2 show s hemati ally the integrated plasMOStor

(37)

123 ₁₄₃

Figure 1.10 Mode proles, refra tive indi es and lossessupported by the plasMOStor

sta k (a); Opti al intensity as fun tion of the sour e-drain separation for the on state

ando-state(a)

waveguide (plasMOStor).

Figure 1.11SOI based integrated plasMOStor

1.5 Outline of this thesis

Thisthesispresentsexperimentalandtheoreti alwork,investigatingthepotential of

CMOS integrationof plasmoni modulator. The hapters areorganizedasfollows :

Chapter 2 des ribes the experimental and numeri al methods used to investigate

(38)

theuse ofa metal insteadof highlydoped poly rystalline sili on onta t and thesingle

mi ron-s ale dimensions of our devi es. Finally an automated and reliable method is

des ribed for large-s aleele tro-opti al hara terization system.

Chapter3 investigatesthe materialaspe tsand alsothete hnologi al steps required

in order to realize an integrated plasMOStor [10℄ within the CMOS environment. The

main onstraintsrelatedwithmaterialaspe ts omefromtheneedfor ele tri al onta ts

withmetalswithhighele tri al ondu tivityinordertoensurelowopti allosses.Tothat

respe twe hooseCuasthemostpromising metalfortheintegrationofplasmondevi es.

Afterthat hoi e,weshowthattheinsertionofanultrathinstoi hiometri

Si

3 N

4

layerin aMOSsta ka tsasane ient diusion barrierofCu, whilepreservingthelowopti al

lossesoperationofthedevi e.Finally,wedes ribethemainte hnologi alstepsinorderto

fabri atethe plasMOStor sta kintegrated withamono-mode Siwaveguide.Fabri ation

ofour modulator requirethe useofbothfront-end andCuba k-end te hnologies, whi h

areproperly ombined only a few nanometers one from ea h otherthanks to theuse of

adiusion barrierlayer.

Chapter 4presents theexperimental realizationof an e ient light oupling s heme

betweentheSiWGandtheintegratedplasmostorontheSOIplatform.Threedimensional

FDTDis alsoexploited inorderto simulate ourplasmoni oupler stru ture. The

expe-rimental and simulation results arein agreement illustrating thatthe highest-e ien y

geometry orrespondsto a ompa t ouplerof0.5

µ

mlengthhaving oupling lossofonly 2.5dBper oupler (experimental value).

Chapter 5 dis usses the ele tro-opti al behaviour of the integrated plasmostor. We

demonstrate rstthatbyusinghighdiele tri onstant (high-k)materialssu hasAl

2

O

3

and HfO

2

we an a hieve an optimized MOS stru ture with low operation voltage and lowopti allosses.Giventhatinterest,wede idedtoexperimentalvalidatetheCu/high-k

asanele tri alreliable sta kfor futureappli ationsto theMOSplasmostorsta k.After

that,wepresentadetailedele tro-opti alinvestigationoftheintegratedplasmostor.The

(39)

wasfollowed bythe ele tri al aswell asele tro-opti al (EO)experimentalinvestigation.

Devi esshowa umulation apa itan eoffewfFandleakage urrentbelow10fA.Finally,

an opti al modulationhaving a apa itive signature it was demonstrated, inagreement

withsimulations.

Finally, we give the nal on lusions of this thesis and we dis uss future resear h

(40)

Experimental and numeri al

methods

In this hapter we dis uss the ele tri al and opti al hara terization methods whi h

areused to investigate the physi al me hanisms underlying the operation ofour opper

based plasmon MOS modulators. The use of opper for the onta t ele trodes is an

important feature of this study. From a plasmoni point of view, thelow DC ele tri al

resistivityof opper an leadtoa lowpropagationloss[55℄.However,withouttheuseof

a metal-diele tri interfa e layer, during annealing, opper may rapidly diuse into the

diele tri layers[56 ℄, degrading signi antly thedevi eperforman e.

Inordertoele tri ally hara terisetheMOSsta kandinparti ularthe opper

diu-sionee t andtheperforman e ofaninterlayerdiusionbarrier, awide rangeof

ele tri- alte hniqueswereemployed.Theunderstanding oftheele tri albehaviourofourMOS

devi es was enhan ed through the use of numeri al simulations obtained using either

SILVACO, a ommer ial software pa kage, or an in-house pa kage, 'Le simulateur' [57℄

(se tion 2.1).

Following ele tri al hara terisation, opti al investigations were performed. Results

of experimental transmission measurements were interpreted in onjun tion with 3D

(41)

devi es (se tion2.2).

Finally, we des ribe a reliable automatedlarge-s ale ele tro-opti al hara terisation

system whi h was developed to test the opti al transmission, urrent-voltage

hara te-risti s and ele tri ally indu ed modulation depth for a large number of devi es. This

system was spe i ally adapted for the single mi ron-s ale dimensions of our devi es

(se tion 2.3).

2.1 Ele tri al hara terization methods

In this se tion, we detail the ele tri al hara terization methods that were used to

investigate the ele tri al performan e of dierent diele tri layers and also to study the

integrity of a diusion barrier designed to mitigate the penetration of opper into the

underlying diele tri layers. To that respe t, Capa itan e-Voltage (C-V),

Condu tan e-Voltage (G-V) and Linear RampVoltage Stress (LRVS)wasperformed and are briey

des ribed below.

The apa itive response C-V, an give insight about keys parameters su h as the

equivalent SiO

2

thi kness of the diele tri (EOT) and the at band voltage V

f b

. The extra tionoftheseparametersis ru ialinthisthesisinordertoestimatethebreakdown

eld a ross the insulator. Moreover by estimating the V

f b

we an extra t information about the defe ts inside the insulator. Finally, C-V measurements an give the harge

on entration in a wide range of voltages. The extra tion of this quantity an be used

for a urate ele tro-opti al simulations.

The standard ondu tan e methodG-Vdeveloped byNi ollianetal.[7℄ wasusedto

studytheele tri ally a tiveinterfa e statesattheSiO

2

/Siinterfa e. Furthermore,when the ondu tan e and the apa itan e is estimated, the resistan e of our devi e an be

al ulated.Thisparameterwillnallydene theRCdelayofourdevi ewhi his riti al

parameter deningthespeed of ourmodulator.

(42)

of the insulator. The interest of LRVS experiment is to be very fast and to provide

information on intrinsi aswell asextrinsi breakdownbehaviorofour devi e [58 ℄.

2.1.1 C-V and G-V measurement

The measurement of the apa itan e and the ondu tan e is arried out by

groun-ding the substrate while applying a ombination of voltages on the gate metal : a DC

omponent V

.Themeasurement wereperformedwithHP4284impedan e ana-lyzer meter (Fig. 2.1). By onne ting the two ele trodes (gate and substrate) the LCR

metermeasure the urrent (I)owing through adevi e, the voltage (V

g

) a rossthe de-vi e,andthephaseanglebetweenthemeasuredVandI.Fromthesethreemeasurements,

allimpedan eparameters anthenbe al ulated.Sin etheimpedan eis al ulated,then

the apa itan e isestimatedfrom itsimaginary part.

12345

6BC

D

67893A432

EF

Figure2.1Experimental ongurationfor measuringthe apa itan e of a MOS

stru -ture.

Fig.3.4presentsameasure C(V

g

) ofaMOS apa itor. Two operating modes anbe distinguished,thea umulationandthedepletionregime.Theboundarybetweenthetwo

(43)

frequen yistoohightoallowdiusionofminority arrier. Inthisgure,we alsopresent

asimulation of the hara teristi C(V

g

)extra ted bythemodeldes ribedinthese tion 2.1.4.

Figure 2.2 C-V hara teristi of MOS apa itor onsists of Cu/Si

3

N

4

(3nm)/SiO

2

(10nm)/p-doped Si (10

18

) at 100 kHz voltage sweep frequen y. Blue line : experimental

results,red line:simulations(see se tion 2.1.4.1).The a umulation anddepletion mode

are also presented.

We also use the same experimental set up to measure very low apa itan e, in the

rangeoffF.Duetothefa tthatthetarget apa itan eisultralow,wemadesome

alibra-tionbeforethe measurement. Firstwemeasurethe apa itan e ofareferen estru tures

onsistingoftwoopen- ir uitprobepads.Ase ondreferen estru ture onsistingof

inter- onne tsbut nottargetdevi eisalsomeasured.De-embeddingtheparasiti apa itan e

and using the open ir uit onne tions we are able to measure the apa itan e of the

target devi ewhi hisof theorderoffewfF. An exampleofsu hmeasurement isshown

intheFig.2.3.

The interfa e defe t density D

it

between sili on substrate and gate insulator is an important parameters thatdire tly inuen e the MOS devi eperforman es su h asthe

(44)

-6 -4 -2 0 2 4 6 2.5 3.0 3.5 4.0 4.5 5.0 5.5 C a p a c i t a n c e ( f F ) Voltage (V)

Figure2.3Experimentalresultsofultralow apa itan emeasuringusingtheimpedan e

analyser meter, HP 4284. The measurement was performed to the plasmostor devi e of

a tive length = 1

µ

m

leakage urrentthroughthe diele tri (i.eanin reaseofdefe t density an betranslated

asanin reaseoftheele tronstrappedstateintheoxideoranin reaseoftheweakspotin

thediele tri ,resultinginasubsequentin reaseofleakage urrent)andthegatediele tri

reliability.Inthis thesis,we usethisparameterinorderto investigate the opperindu e

ioni defe ts at the interfa e ofinsulator/Sili on.

Several methods were proposed in order to measure the interfa e defe ts. Here, we

usethestandard parallel ondu tan e method developed by Ni ollianetal. [7 ℄.To that

aim, apa itan e and ondu tan e (CâGâV) hara teristi s are performed using

the HP4284 impedan e analyzer at frequen ies ranging from 1 kHz to 1 MHz with an

applying modulatingsignal of40 mV and V

g

varying from-5 Vto +5V witha step of 0.01 V (Fig. 2.4). Subsequently, the equivalent parallel ondu tan e G

p

/

ω

is estimated by :

(45)

123

143

Figure2.4(a)AnexampleofC-V

g

andG-V

g

measurementat50kHzand100kHz.(b) Extra tionofinterfa edefe tdensityusingthe ondu tan emethoddeveloped byNi ollian

et al. [7℄.

G

p

/ω =

ωC

2 ox

G

m

C

2 m

ω

2 (C

ox

−

C

m

)

2

(2.1)

The interfa e trapdensityD

it

is thenextra ted fromtheequivalent parallel ondu -tan ethrough the relation :

D

it

≃

2.5 q

(

G

p

ω

)

max

(2.2)

whereC

ox

istheoxide apa itan e,G

m

isthemeasured ondu tan e,C

m

isthemeasured apa itan e and

ω

is theangularfrequen y (

ω

=2

π

f).Wenotethat for theextra tionof Eq. 2.2 we assume that a surfa e potential u tuation is negle ted. Fig. 2.4 shows an

example of the measured apa itan e (C

m

) and ondu tan e (G

m

) at 50 KHz and 100 KHz(2.4a) andalso theextra tedinterfa e defe ts densityatdierent energy levels(E)

with respe t to the the intrinsi Fermi level (Ei) for Cu/Si

3

N

4

(3nm)/SiO

2

(10nm)/Si MOS apa itor.

(46)

Several phenomena are responsible for the deterioration of the C-V hara teristi s

su hashighleakage urrent,highseriesresistan e(R

s

)and the reationofanundesired loss interfa e layer between the oxide/sili on or metal/sili on interfa e. These

pheno-menausually auseto an additional parasiti apa itan e andalso to anin rease ofthe

resistan e of the devi e. The latter is ofparti ular importan e for this thesis, sin e will

dene the RC time onstant of the devi e and thus its operation speed. Therefore the

phenomena des ribedhereafterareplayinga keyroleintheperforman eofour devi es.

Therst isrelatedtoleakage urrent.Whenthe insulatorthi knessbe omessmaller

than 3 nm, the leakage urrent through the insulator be omes important (around 1

A. m

−

2

[59℄) and may ae t the C-V hara teristi s espe ially in a umulation and

inversion region [60 ℄. As a result the estimation of oxide apa itan e is ina urate and

onsequently leads to unreliable extra tion of oxide thi kness, defe t densityand break

down eldofthe insulator. However, inour ase theoxide thi kness islarger than 3nm

whi hleadsto lowleakage urrent. Therefore,it anbeassumedthattunnellingisnot a

signi ant issueinour devi es.

The series resistan e is also identied one of the reason auses deterioration of C-V

[61 ℄.Theseriesresistan e(R

s

) an omefromeitherthesubstrate ortheba ksideofthe semi ondu tor/metal interfa e.For the latterit ispossible to reate a S hottky onta t

instead of ohmi and to ause in a signi ant in rease of the resistan e (

>

1k

Ω

[62℄) whi h in turn ae ts the apa itan e. The ee t of series resistan e an be limited via

the simplied three-element model [63 ℄ by measuring the ondu tan e and apa itan e

at a umulation onditions usingthesimple two-element (parallel)model.

Finally,itiswell knownthatduringfabri ation pro ess,ifthesurfa e leaningisnot

ideal, itis possible to reate an undesiredlossdiele tri layerbetween theoxide/sili on

[64 ,65 ℄or metal/sili oninterfa e. Thislayermayindu eadditionalresistan eand

para-siti apa itan e inour system. Theseee ts an alter theMOS apa itan e (espe ially

ina umulation regime) andalso an signi antly in rease theRCtime onstant ofour

(47)

123 ₁₄₃

₁₅₃

6

78

9 _A

9 B

6 B

6 ₇₈

9 B

6 _B

6 C

D

EF

9 _A

Figure2.5(a)Measured apa itan e atdierentfrequen iesillustratingthetremendous

ee toftheunwantedlossdiele tri layertakenfrom[8℄;(b)Theequivalent ir uitmodel

of thereal MOS apa itor in luding the ee tof lossydiele tri layer;( )The equivalent

ir uit model of (b) but in strong a umulation.

Thephenomenades ribedabove anbesummarizedinanele tri alequivalentmodel

(Fig. 2.5b and ). Fig.2.5b shows theele tri al model des ribing areal MOS apa itor

in ludingthe ee tofthelossdiele tri layerandalso theseriesresistan e, whereC

ox

is theoxide apa itan e, C

D

thedepletion layer apa itan e, Y

it

isthe admittan e dueto theinterfa e trap harge, C

T

and R

T

isthe apa itan e and resistan eof theunwanted losslayer, respe tively.Bybiasingthe apa itorinstronga umulation,theee tofthe

depletion apa itan e and admittan e (2.5 ) are limited and therefore we an estimate

theparasiti ee t of thelosslayer(see[8℄ for more details).

2.1.3 Linear ramp voltage stress measurement

When qualifying the reliability of a devi e, the extrinsi break down mode due to

pro essindu eddefe tsisoneofthemostimportantparametertobeinvestigated[58℄.In

thisthesisweusetheextrinsi modeofLinearRampVoltageStress(LRVS)measurement

(48)

insidetheinsulator.

Break down reliability measurement su h as LRSV onsists in applying a linearly

in reasingvoltagewithtypi alspeedof0.5V/sanddete tingthebreakdownata riti al

gate urrent of100 mA/ m

2

.Theinterest ofLRVS experiment isto be very fastandto

provide informationon intrinsi aswell asextrinsi breakdownevents. Theusualwayof

des ribing the statisti alanalysisofthebreakdownparameters istheWeibullapproa h

[66 ℄.Thisapproa his widelyusedinCMOS foundry.Fig.2.6 showsa Weibul s ale plot

of the umulative distributionfun tion (F),

ln(−ln(1 − F ))

,of thebreak downeld for a large numberof apa itors. One an distinguish two typesof statisti alpopulationas

suggested by the presen e of two linear regions. The region of high break down elds

whi h is asso iated with intrinsi quality of the insulator (e.g., bonding energy, latti e

strength) and the region with low eld to break down whi h is related to the extrinsi

behaviorsu h pro ess indu eddefe ts.

0 5 10 15 -5 -4 -3 -2 -1 0 1 2 W e i b u l s c a l e

Break down oxide field (MV/cm)

Figure2.6Typi al Weibuls aleplotof themaximumvoltage tobreakdownas obtained