vcsel arrays for high-aggregate bandwidth of up to 1.34 ... arrays for high aggregate... · vcsel...

10
VCSEL arrays for high aggregate bandwidth of up to 1.34 Tbps Martin Grabherr, Steffan Intemann, Roger King, Stefan Wabra, Roland Jäger, Michael Riedl Philips Technologie GmbH U-L-M Photonics, Lise-Meitner-Str.13, 89081 Ulm Germany Abstract Even though the lane speed of VCSEL based AOC and transceivers has reached 25 Gbps and beyond [1-7], parallel optics are getting even more important in order to meet the increasing demand for aggregate bandwidths in upcoming applications, among others, 100 Gigabit Ethernet, Infiniband EDR, or EOM (embedded optical modules). As 100 Gbps can be achieved by, e.g., 4 times 25 Gbps using standard QSFP form factor, different approaches are using large scale 2D VCSEL arrays operating at lower lane speeds. Early work on 2D VCSEL based transceivers has already been presented beginning of this century [8] and recent work also addressed the potential of this technology [9,10]. In 2013, Compass EOS has introduced a 1.34 Tbps core router solution [11,12,13] that incorporates 2D VCSEL arrays of 14x12 emitters designed and manufactured by Philips U-L-M Photonics. The VCSEL array is mounted face down onto a CMOS ASIC, directly on top of the analog area. The emission wavelength of 1000 nm allows for substrate side emission and thus for flip-chip mounting as well as the possibility of integrating 2D microlens arrays onto the stack of CMOS and VCSEL array. After briefly introducing the router with regard to the incorporated VCSEL technology we discuss the design and performance of the VCSEL array. Finally, the assembly solution for this most compact and dense transceiver solution is presented. Keywords: VCSEL, VCSEL array, EOM, flip-chip, chip-on-chip Introduction Within routers, data distribution between line cards has been identified as one main bottleneck for performance. In 2013 Compass EOS introduced the r10004 Router which incorporates an optical transceiver solution based on 12x14 channels and according fiber management for data distribution between line cards as well as racks [11]. The full duplex 1.34 Tbps transceivers provide 64 Gbps/mm² data density. Fig. 1 shows the compactness of the r10004 Router which also provides very low power consumption of only 10.2 pF/bit, incl. SERDES on smallest footprint. The transceiver is monolithically built into the CMOS ASIC by monolithic analog & digital ASIC design, and the VCSEL and PD arrays are flip-chip mounted onto the analog part of the ASIC [13]. The subassembly is then flip-chipped onto a substrate carrying the ball grid array (BGA) that allows for on board system integration. In Fig 2. the two step flip-chip assembled optical module is presented, revealing the optical interfaces of VCSEL and PD arrays in the Vertical-Cavity Surface-Emitting Lasers XVIII, edited by James K. Guenter, Chun Lei, Proc. of SPIE Vol. 9001, 900105 · © 2014 SPIE · CCC code: 0277-786X/14/$18 · doi: 10.1117/12.2039226 Proc. of SPIE Vol. 9001 900105-1

Upload: ngonguyet

Post on 31-Oct-2018

217 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: VCSEL arrays for high-aggregate bandwidth of up to 1.34 ... arrays for high aggregate... · VCSEL arrays for high aggregate bandwidth of up to 1.34 Tbps Martin Grabherr, Steffan Intemann,

VCSEL arrays for high aggregate bandwidth of up to 1.34 Tbps Martin Grabherr, Steffan Intemann, Roger King, Stefan Wabra, Roland Jäger, Michael Riedl

Philips Technologie GmbH U-L-M Photonics, Lise-Meitner-Str.13, 89081 Ulm Germany

Abstract Even though the lane speed of VCSEL based AOC and transceivers has reached 25 Gbps and beyond [1-7], parallel optics are getting even more important in order to meet the increasing demand for aggregate bandwidths in upcoming applications, among others, 100 Gigabit Ethernet, Infiniband EDR, or EOM (embedded optical modules). As 100 Gbps can be achieved by, e.g., 4 times 25 Gbps using standard QSFP form factor, different approaches are using large scale 2D VCSEL arrays operating at lower lane speeds. Early work on 2D VCSEL based transceivers has already been presented beginning of this century [8] and recent work also addressed the potential of this technology [9,10]. In 2013, Compass EOS has introduced a 1.34 Tbps core router solution [11,12,13] that incorporates 2D VCSEL arrays of 14x12 emitters designed and manufactured by Philips U-L-M Photonics. The VCSEL array is mounted face down onto a CMOS ASIC, directly on top of the analog area. The emission wavelength of 1000 nm allows for substrate side emission and thus for flip-chip mounting as well as the possibility of integrating 2D microlens arrays onto the stack of CMOS and VCSEL array. After briefly introducing the router with regard to the incorporated VCSEL technology we discuss the design and performance of the VCSEL array. Finally, the assembly solution for this most compact and dense transceiver solution is presented.

Keywords: VCSEL, VCSEL array, EOM, flip-chip, chip-on-chip

Introduction Within routers, data distribution between line cards has been identified as one main bottleneck for performance. In 2013 Compass EOS introduced the r10004 Router which incorporates an optical transceiver solution based on 12x14 channels and according fiber management for data distribution between line cards as well as racks [11]. The full duplex 1.34 Tbps transceivers provide 64 Gbps/mm² data density. Fig. 1 shows the compactness of the r10004 Router which also provides very low power consumption of only 10.2 pF/bit, incl. SERDES on smallest footprint. The transceiver is monolithically built into the CMOS ASIC by monolithic analog & digital ASIC design, and the VCSEL and PD arrays are flip-chip mounted onto the analog part of the ASIC [13]. The subassembly is then flip-chipped onto a substrate carrying the ball grid array (BGA) that allows for on board system integration. In Fig 2. the two step flip-chip assembled optical module is presented, revealing the optical interfaces of VCSEL and PD arrays in the

Vertical-Cavity Surface-Emitting Lasers XVIII, edited by James K. Guenter, Chun Lei, Proc. of SPIE Vol. 9001, 900105 · © 2014 SPIE · CCC code: 0277-786X/14/$18 · doi: 10.1117/12.2039226

Proc. of SPIE Vol. 9001 900105-1

Page 2: VCSEL arrays for high-aggregate bandwidth of up to 1.34 ... arrays for high aggregate... · VCSEL arrays for high aggregate bandwidth of up to 1.34 Tbps Martin Grabherr, Steffan Intemann,

center of arrays for

Fig. 1: Cochip-to-chVCSEL an

Stacking device denecessitywhich cansubstrate on CMOSthermal bgood highorder to wnm wavesubstrate

Fig. 3 shoGaAs as bplaced onpillars to pads are tby dielec

f the BGA. Ar fiber coupl

ompass-EOS rhip data disttrind PD arrays

Deof optical c

esign for VCy of flip-chipn be enabledremoval tec

S and the eleboundary conh speed perfwithstand thelength have is already tr

ows a cross barriers. Then 3 µm polyithe same levthermally cotric barriers.

As a subsequling.

r10004 Routeibution by usflip-chipped

esign of lachips onto CMCSELs and Pp bonding insd by either lochnologies [8evated tempenditions. Theformance, exe advanced abeen chosen

ransparent an

section of the dry etchedimide passivvel as the p-onnected by .

ent step mic

er. Optical ing 2D on ASIC.

arge scaleMOS requir

PDs comparestead or wireonger wavele8]. The extreerature by see requiremen

xcellent heat assembly pron, as they arend can act as

he VCSEL ded mesa is covvation layer. electrode paelectroplated

crolens array

Fig. 2flip-chsubsewith Btransc

e 2D bottoes advanced

ed to today’se bonding foength aboveemely compaelf-heating onts for VCSEremoval, an

ocedures. Oxe based on ms mechanica

esign. The Qvered by a fuThe cathode

ad. For improd Au-pillars

ys are placed

2: 2D VCSELhipped on CMquently flip-cBGA, represeceiver.

om emittd assembly te standard arr

or 2D arrays, the fundamact design ofof the CMOSEL arrays ar

nd sufficient xide confine

mature DBR al base as we

Quantum Weull size p-cone pad is elevoved thermato the subst

onto the VC

L and PD arraMOS ASIC anchipped on a enting a 1.34 T

er array echnology anray formats. , bottom-emi

mental absorpf stacked 2DS itself, resulre low powermechanical

ed VCSELs edesigns whi

ell as heat dis

ells are built ntact, the p-eated by elec

al managemetrate, but elec

CSEL and PD

ays (center) nd substrate Tbps optical

nd a differenDue to the

ission is requption of GaAD VCSEL arrlts in challenr consumptiorobustness i

emitting at 1ile the GaAsstributing lay

of InGaAs welectrode padtroplated Au

ent additionactrically isol

D

nt

uired As or rays nging on, in 1000 s yer.

with d is u al lated

Proc. of SPIE Vol. 9001 900105-2

Page 3: VCSEL arrays for high-aggregate bandwidth of up to 1.34 ... arrays for high aggregate... · VCSEL arrays for high aggregate bandwidth of up to 1.34 Tbps Martin Grabherr, Steffan Intemann,

Fig. 3: DQWs aprovid

Efficient an oxide results inlayer. As wettable creeping The chanthus the V

A fundamuniformitenables aof 168 VCthresholdrespectiv

Design of the and the currede improved

current conflayer, which

n high performthe VCSELcover layersof solder alo

nnel to channVCSEL array

CWmental requirty of the pera reasonable CSELs arran

d current of 0ely, at 6.0 m

1000 nm botent confinem heat flow to

finement to ah in combinamance of the

L arrays are ds built of an aong the metanel pitch of 2y dimension

W performrement for laformance payield on arra

nged in a 14x0.5 mA and a

mA laser curr

ttom-emittingment is controo the solder in

dielectr

an 8 micron ation with sme device. Thdesigned for additional Pal tracks. 250 µm is mns result in 3

mance of arge scale 2Darameters anay level. Figx12 matrix. an operatingrent.

g VCSEL. Tholled by selecnterface. Theric single lay

aperture jusmall mesa gehe substrate i

flip-chip soolyimide lay

matching the g.5 x 3.0 mm

1000 nmD VCSEL ar

nd of course g. 4 depicts thThe curves a

g voltage and

he active layctive oxidatioe substrate byer.

st above the eometries anis AR coatedldering usin

yer protect th

geometries om².

m bottom errays is an exa very high yhe room temare practicald output pow

ers are built on. Electroplback side is A

active regionnd according d with a singng eutectic Ahe chip surfa

of standard f

emitters xcellent parayield on waf

mperature LIVlly identical,

wer of 1.9 V

of InGaAs/Glated Au pilla

AR coated wi

n is achievedlow parasiti

gle dielectricAuSn depots, ace against

fiber bundles

ametric fer level thatV characteri, showing a and 1.9 mW

GaAs ars th

d by ics non-

s and

t istics

W,

Proc. of SPIE Vol. 9001 900105-3

Page 4: VCSEL arrays for high-aggregate bandwidth of up to 1.34 ... arrays for high aggregate... · VCSEL arrays for high aggregate bandwidth of up to 1.34 Tbps Martin Grabherr, Steffan Intemann,

Fig. 4:plotted

Analyzindistributiwafer (nostandard mean val

Fig. 5: Ta full wand the

: CW LIV ped in the diagr

ng the distribons of param

o upfront seldeviation ofue is calcula

Threshold cuwafer. Mean ve standard de

erformance oram and show

re

bution of parameters can belection of in-f only 0.037 ated to 0.342

urrent distribvalue results eviation is 0.0

of the 1000 nw excellent hespective volt

ameters acroe noticed. In-spec VCSELmA. For the

2 W/A, the st

bution acrosin 0.503 mA

037 mA.

m bottom-emomogeneity. tage drop is o

oss an entiren Fig. 5 the mLs) is given e slope efficitandard devi

s A

Fig.full and

mitting VCSEAt operating

only 1.9 V.

production mean value fas 0.503 mA

iency, valuesiation 0.013

. 6: Slope effiwafer. Mean

d the standar

EL array. 16g bias curren

wafer, very for all 38.304A, with an acs are shown W/A.

iciency distrin value resuld deviation i

68 channels ant of 6.0 mA

narrow 4 channels occording in Fig. 6, th

ibution acroslts in 0.342 Wis 0.013 W/A

are the

on

e

ss a W/A

.

Proc. of SPIE Vol. 9001 900105-4

Page 5: VCSEL arrays for high-aggregate bandwidth of up to 1.34 ... arrays for high aggregate... · VCSEL arrays for high aggregate bandwidth of up to 1.34 Tbps Martin Grabherr, Steffan Intemann,

Also for tthe standwafer chaachieved

Fig. 7:

In additiomeasuremaccessiblwafer. Thto allow fmeasuremremainingwafer thumeter ben

The largeFig. 8, sh

the differentard deviationannel yield hby superior

Slope efficie

on to the VCment of the Ve from the toherefore the for free standment withoutg substrate t

us can be maneath the sub

High free signal perfohowing an ey

tial resistancn is 5.0 Ohmhas been demcontrol of e

ency distribu

CSEL performVCSEL paraop side, the ewafer chuckding substrat potentially

thickness of apped identicbstrate.

equency pformance of tye diagram m

ce the distribm. The parammonstrated aepitaxial grow

ution across astandard de

mance requirameters itselfemitted powk of the wafeate during they disturbing o500 µm doecally to top-e

performathe bottom-emeasured for

ution is very

metric uniforas high as 98wth and oxid

a full wafer. Meviation is 5.

rements, anof. Whilst bot

wer can only er probing ste probing. Toptical interfes support theemitting VC

nce of 10emitting VCr a bit sequen

y narrow. Mermity is exce.9%. Such hde apertures.

Mean value r0 Ohm.

other challenth electrodesbe measured

tation must bThe latter solufaces and rese measurem

CSEL wafers

00nm boSEL at roomnce of PRBS

ean value is ellent and thehomogeneity.

results in 70.

nge is the ons, anode andd at the backbe either tranution does espective frinents mechan, but with th

ttom emim temperaturS 231-1 . The

70.0 Ohm ae according

y can only be

.0 Ohm and t

n wafer d cathode, arekside of the nsparent or nnable a

nges. The nically. The he optical pow

itters re is presentee unmounted

and on

e

the

e

needs

wer

ed in d

Proc. of SPIE Vol. 9001 900105-5

Page 6: VCSEL arrays for high-aggregate bandwidth of up to 1.34 ... arrays for high aggregate... · VCSEL arrays for high aggregate bandwidth of up to 1.34 Tbps Martin Grabherr, Steffan Intemann,

FC8500_PI_R601_total wfms 9.

failed smpls 0source

Y Align: Full Di

Mask

.075 k margin 30 %mask hits 0

margin hits 0isp total hits 0

IP.. ri. r . P' 1 rIK,. IN:i kr4i ' 6 f,;-,1

K., 'N., 4-"'

, ta T .

; ft '%'.,,,

0 i. i ri ,L,A4 1 ::. ,

'IF- O.

- ., e .J IN: Pir 11?4, ti; Pi' p4:1A SA ILA - <J' M/ WA MA ' ILA l A CA

re

currentie( ) 31.1 psie( ) 48.9 ps

minimum31.1 ps47.8 ps

maximum32.2 ps48.9 ps

total mea81

81

r . 3 %/M VO 111 8Y 111 8.

, , jqd;IW ri ïi 'i i'i i,

Ja j._ " pa 04 ra _ <

VCSEL ihigh speefall timesintroduce

Fig. 8: Eyspeed proroom temGbps 30%

The unifoarrangemidentical eye diagr

Fig. 10

is operated aed probe. Ths extracted fred VCSEL d

ye diagram mobe on a unm

mperature. At% mask marg

ormity of thement of eye d

operating corams are wid

0: Eye diagratemper

at 6.0 mA avhe resulting mrom the curv

design does p

measured witmounted chan

t a data rate gin is achieve

e high speeddiagrams in Fonditions of de open and r

ams for all 16rature, 8 Gbp

verage laser cmask marginves given in provide satis

th high nnel at of 8 ed.

d performancFig. 10. The6.0 mA averrather comp

68 channels ips with PRBS

current and n is calculateFig. 9 are 31

sfactory marg

Fig. 9fall tirespe

ce within an eye diagramrage current arable.

in a large scaS 231-1 seque

5.5 mA moded to 30%. T1 ps and 50 pgin for the ta

9: Measuremime results inectively.

array can bems have been

and 5.5 mA

ale 2D VCSEence, and 6 m

dulation currThe corresponps, respectivargeted data

ment results fon 31 ps and 5

e seen from tn measured sA modulation

EL array, meamA bias curre

rent by usingnding rise an

vely. The rate.

for rise and 50 ps,

the 14x12 sequentiallyn current. Al

asured at rooent.

g a nd

at l 168

om

Proc. of SPIE Vol. 9001 900105-6

Page 7: VCSEL arrays for high-aggregate bandwidth of up to 1.34 ... arrays for high aggregate... · VCSEL arrays for high aggregate bandwidth of up to 1.34 Tbps Martin Grabherr, Steffan Intemann,

urrent FC8500_Pl R601_.60 dB total wTms 3..09 failed smpls 06.0 ps source3.6 os Y Alun: Full Di

More meare the reeye diagrdriven byself-heatito addres

Fig. 11: Emounted VCSEL ochip temp

The resulMeasuremfor BER

SComparetypically of large ssteps requused for tcreeping soldered

A fully asthe ASCIthermal jotransceiv

aningful thaesults obtaineram of one cy the CMOS ing from the s the slightly

Eye diagram 1000 nm bot

operated at 6perature. Ma

lting eye diaments of Bitdown to 10-

System ined to the stan

die attach byscale VCSELuires a high the VCSEL/of AuSn aloonto the AS

ssembled ASI is shown inoints are ester is coverin

an eye diagraed from fullyhannel withiASIC. Due ASIC, the a

y increased t

of a flip-chipttom-emittin

6.5 mA bias aask margin is

agram again Error Rates18.

ntegrationndard assemby epoxy glueL arrays is vetemperature

/ASIC interfaong the metaIC can be se

SIC with VCn Fig. 14. In ablished for

ng only abou

ams measurey assembledin a fully assto the elevat

average laserthreshold cu

p g

and 65°C s 35%.

is wide opens (BER) and

n of 1000bly technoloe and wire bery special a

e solder for thface. Non-weal tracks towaeen in Fig. 1

CSEL and PDtotal 672 elethis optical

ut 10% of the

ed by using pd optical engsembled VCted VCSEL r current has

urrent. The m

Fig. 1total

n and providaccording ex

nm bottoogy of today’bonding is stiand demandihe first flip-cettable areasards critical 3.

D die soldereectrically funsubassembly

e total ASIC

pattern generines or trans

CSEL array, mchip temper

s been set a bmodulation cu

12: BER rateJitter of 0.5 U

des an excellextrapolation

om emitti’s QSFP or Cill state of thing. The necchip process on the VCSareas. AuSn

ed by flip-chnctional joiny. The area u

C area.

rator and higsceivers. Figmounted fliprature of 65°bit higher to urrent again

e measuremeUI at BER 1

ent mask mans show a tot

ing VCSELCXP optical he art, the flicessity of seqs. Eutectic ASEL top surfan bumps and

hip onto the nts as well asused for the

gh speed pro. 11 shows t

p-chip onto a°C caused by

6.5 mA in ois set to 5.5

ents result in0-18.

argin of 35%tal Jitter of 0

L array engines, wh

ip-chip assemquential sold

AuSn pillars aface avoid d an optical d

analog area s many moreoptical

obes the and y the order mA.

%. 0.5 UI

here mbly dering are

die

of e

Proc. of SPIE Vol. 9001 900105-7

Page 8: VCSEL arrays for high-aggregate bandwidth of up to 1.34 ... arrays for high aggregate... · VCSEL arrays for high aggregate bandwidth of up to 1.34 Tbps Martin Grabherr, Steffan Intemann,

Fig. 13: Achipping CMOS A

The opticmounted bundle isis capablerealized bmatching

AuSn bumps the VCSEL

ASIC.

cal subassemto the backs aligned ande of carryingbased on VCg CMOS tech

Fig. 15

are used forarray onto th

mbly is then sside of the syd fixed to theg 1.34 Tbps CSEL technohnology.

: 1.34 Tbps o

flip-he

soldered to aystem boarde optical subfull duplex.

ology which

optical transc

Fig. 1on thThe tthe to

a substrate ca. On the oppassembly, asSuch high dintrinsically

ceiver assem

14: Both, VCe CMOS AS

transceiver aotal ASIC ar

arrying the Bposite side ofs shown in F

density opticy enables 2D

mbled onto th

CSEL and PDIC by flip-ch

area consumeea.

BGA (see Fif the system Fig. 15. The al systems c

D arrangemen

e system boa

D arrays plachip technologes about 10%

ig. 2) and finboard, the ffinal transcean only be

nts ideally

ard.

ced gy.

% of

nally fiber eiver

Proc. of SPIE Vol. 9001 900105-8

Page 9: VCSEL arrays for high-aggregate bandwidth of up to 1.34 ... arrays for high aggregate... · VCSEL arrays for high aggregate bandwidth of up to 1.34 Tbps Martin Grabherr, Steffan Intemann,

Conclusion The r10004 Router released by Compass-EOS is making use of a primary VCSEL technology promise: the stacking of ASIC and 2D VCSEL array together with micro lens arrays for 2D fiber coupling that results in a most compact optical transceiver. The further demand for higher aggregate bandwidth will have to be addressed by increasing lane speeds even beyond 25 Gbps, but upscaling in orders of magnitude will only be possible by highly parallel solutions. Both directions need to be further enhanced simultaneously. Combining already demonstrated building blocks, like 28 Gbps single lane data rates and the presented assembly solution, the aggregate bandwidth already provides the capability to reach 4.7 Tbps.

Aknowledgements We acknowledge the cooperation and support of Compass EOS, especially through Kobi Hasharoni and Shuki Benjamin. Pictures 1,2,8,9,10,11,12,13,14,15 are courtesy of Compass EOS.

References [1] D. M. Kuchta et al., “A 55Gb/s Directly Modulated 850nm VCSEL-Based Optical Link”,

IEEE Photonics Conference 2012 (IPC 2012) Post Deadline Paper PD 1.5 (2012). [2] S.A. Blokhin et al., “850 nm VCSELs operating at bit rates up to 40 Gbit/s“, Electronics

Letters 45, pp. 501-503 (2009). [3] Toshihito Suzuki et al., “Reliability study of 1060 nm 25 Gbps VCSEL in terms of high

speed modulation”, Proc. SPIE 8276, pp. 04-01 to 04-8 (2012). [4] AVAGO Technologies, http://www.avagotech.com/pages/video_ofc2011_25g_vcsel

(2011). [5] Neinyi Li et al., “Emcore’s 1 Gb/s to 25 Gb/s VCSELs”, Proc. SPIE vol. 8276, pp. 03-1 to

03-10 (2012). [6] M. Grabherr et al., "25 Gbps and beyond: VCSEL development at Philips", Proc. of SPIE

Vol. 8639, pp. 0J-1 to 0J12 (2013). [7] Afriat Gil et al., "VCSEL based optical transceiver module operating at 25 Gb/s and using a

single CMOS IC", Proc. of SPIE Vol. 8282, paper 06, pp. 1-6 (2012). [8] M. Grabherr et al., „Comparison of approaches to 850 nm 2-D VCSEL arrays“,Proc. of

SPIE Vol. 4994, pp. 83-91 (2003). [9] Clint L. Schow et al., "A 24-Channel, 300 Gb/s, 8.2 pJ/bit, Full-Duplex Fiber-Coupled

Optical Transceiver Module Based on a Single “Holey” CMOS IC," J. Lightwave Technol. 29, 542-553 (2011).

[10] Pinxiang Duan et al., "A novel 3D stacking method for Opto-electronic dies on CMOS ICs," Opt. Express 20, B386-B392 (2012)

Proc. of SPIE Vol. 9001 900105-9

Page 10: VCSEL arrays for high-aggregate bandwidth of up to 1.34 ... arrays for high aggregate... · VCSEL arrays for high aggregate bandwidth of up to 1.34 Tbps Martin Grabherr, Steffan Intemann,

[11] Kobi Hasharoni et al., “A high end routing platform for core and edge applications based on chip to chip optical interconnect”, OTu3H.2, OFC/NFOFC (2013).

[12] S. Benjamin et al., “Assembly of 1.3 Tb/s Full Duplex Optical Module“, Proc. IEEE 63rd Electronic Components and Technology Conference (ECTC), pp. 292-296 (2013).

[13] Kobi Hasharoni et al., “A 1.3 Tb/s parallel optics VCSEL link”, Proc. SPIE vol. 8991, paper 10 (2014).

Proc. of SPIE Vol. 9001 900105-10