Evaluation of Multi-Gbps Optical Transceivers for Use in Future HEP Experiments

Evaluation of Multi-Gbps Optical Transceivers for Use in Future

HEP Experiments

Luis AmaralCERN – PH/ESE/BE – Opto

16/09/2008

[email protected] 2

Outline• Introduction

– Optical Links Upgrade for SLHC– Project Organization

• Commercial Optical Transceivers– Overview– Devices Under Test

• Performance Evaluation of Commercial Optical Transceivers– Setups and Procedures– Performance metrics– Specification Proposal for Operation at 5Gbps

• Analysis of the Data and Results– Figure of Merit– Results of the Devices Under Test

• Summary

TWEPP 2008

[email protected] 3

Optical Link Upgrade for LHC• The future SLHC upgrade is expected to increase the LHC luminosity by an

order of magnitude. This implies:– More data to be transmitted;

• Assuming more complex DAQ/TTC/SC systems.

– Higher radiation doses.• The optical links are also required to have low power dissipation and to

reduce the mass inside the detector.• The solution is to increase the bandwidth of each individual link.

– The links will be required to:• Run at multi-Gbps speeds;• Have better radiation tolerance;• Operate at low temperatures and in strong magnetic field;• Be easy to install and operate.

• The goal is to come up with a common solution that offers:– System architectures;– Basic building blocks.

TWEPP 2008

[email protected] 4

GBTXPD

LDTRx FPGA

●●

●●

●●

On-DetectorCustom Electronics and Packaging

Radiation Hard

Off-DetectorCommercial Off-The-Shelf (COTS)

Custom Protocol

GBT GBTVersatile Link

Timing and TriggerDAQ

Slow Control


Slow Control

TIA

LDD

Radiation-Hard Optical Link for Experiments

Project Organization

TWEPP 2008

[email protected] 5

Project Organization

• The GBT project covers the ASIC design, verification and packaging.

• The Versatile Link project covers system architectures and the components.

• One of the main F.E. components is a Versatile Transceiver (VTRx), which results from the customization of a commercial transceiver (TRx).

GBTSpokesperson:

P. Moreira

WP 1System

WP 2Components

WP 1.1 - P2P, SMU, J. Ye

WP 1.2 - PON WP 1.3 - Mixed

WP 2.1 - F.E. Components,

CERN, J. Troska

WP 2.2 - B.E. Components

WP 2.3 - Passive Components,

Oxford, C. Issever

Versatile LinkSpokesperson:

F. Vasey

TWEPP 2008

GBTXPD

LDTRx FPGA

●●

●●

●●

On-DetectorCustom Electronics and Packaging

Radiation Hard

Off-DetectorCommercial Off-The-Shelf (COTS)

Custom Protocol

GBT GBTVersatile Link


Slow Control


Slow Control

TIA

LDD

Radiation-Hard Optical Link for Experiments

[email protected] 6

Commercial Transceivers Overview

1211109876543210Gbps

1

0

OC3/STM1OC12/STM4

1GFCGbE

2GFCOC48/STM16

4GFCGBT (4.8Gbps)

8GFC10GBASE-R

10GBASE-W & OC192/STM64

10GFCOC192/STM64 + FEC

10GFC + FEC10GE + FEC

SFP+XFPSFP

• The goal is the VTRx customization of a commercial TRx.

• There are several families of commercial optical TRxs to target the Telecom and Datacom standards.

• The GBT project specifies a single lane running at 4.8Gbps which is not a standard.

TWEPP 2008

[email protected] 7

Commercial Transceivers Overview

1211109876543210Gbps

1

0

OC3/STM1OC12/STM4

1GFCGbE

2GFCOC48/STM16

4GFCGBT (4.8Gbps)

8GFC10GBASE-R

10GBASE-W & OC192/STM64

10GFCOC192/STM64 + FEC

10GFC + FEC10GE + FEC

• The goal is the VTRx customization of a commercial TRx.

• There are several families of commercial optical TRxs to target the Telecom and Datacom standards.

• The GBT project specifies a single lane running at 4.8Gbps which is not a standard.

SFP+ TRxs seem to be worthy of our attention.What are they made of?

LA

LDD TOSA

ROSACAGE Controller (back)

SFPs are similar but rated for lower speeds and the XFPs include a clock and data recovery circuit on both Tx and Rx paths.

TWEPP 2008

SFP+XFPSFP

[email protected] 8

Devices Under TestDevice # TRx Type Wavelength [nm] Max. Bitrate [Gbps] LD/PD type Applications

1 SFP 850 4.25 VCSEL/PIN 1/2/4GFC; 1000BASE-SX 2 SFP 1310 4.25 VCSEL/PIN 1/2/4GFC; 1000BASE-LX10 3 SFP+ 850 10.5 VCSEL/PIN 2/4/8/10GFC; 10GBASE-SR 4 SFP+ 850 10.5 VCSEL/PIN 2/4/8/10GFC; 10GBASE-SR 5 SFP+ 1310 10.5 DFB/PIN 2/4/8/10GFC, 10GBASE-LR 6 SFP+ 1310 10.5 DFB/PIN 2/4/8/10GFC, 10GBASE-LR 7 XFP 1310 10.3 DFB/PIN 10GBASE-LR/LW 8 SFP+ 1310 10 VCSEL/PIN Prototype for 10Gbps over SM fiber 9 SFP+ 1310 10 VCSEL/PIN Prototype for 10Gbps over SM fiber

10 SFP+ 1310 10 VCSEL/PIN Prototype for 10Gbps over SM fiber 11 SFP+ 1310 10 VCSEL/PIN Prototype for 10Gbps over SM fiber 12 SFP+ 1310 10 VCSEL/PIN Prototype for 10Gbps over SM fiber

Testboard

SFP+ module in

side

an assembly ca

ge

• Our evaluation was guided towards the performance at 5Gbps, i.e. slightly faster than what the GBT protocol is targeting (4.8Gbps).

• Some devices are suitable for MM links operating at 850nm and other are suitable for SM links operating at 1310nm.

• Some devices have VCSELs (850 and 1310nm) and other have DFB lasers (1310nm).

TWEPP 2008

[email protected] 9

Receiver

Transmitter SDA ScopePRBS

SourceClock

Generator PC Automation:Runs through several

bitrates and saves data

Setup A(Tx) Optical Evaluation

PRBS7 Test Pattern Jitter @5Gbps:Tj@1E-12=0.10UI, Dj=0.03UI

20%-80% Rise/Fall Times: 28/27ps

10GHz Optical Sampling Module Without Filtering0.5-12.5GHz PRBS7-PRBS31

2m Optical Fiber

Testboard and DUT

Tx Performance Evaluation – Eye Diagram

TWEPP 2008

[email protected] 10

Receiver

Transmitter SDA ScopePRBS

SourceClock

Generator PC Automation:Runs through several

bitrates and saves data

Setup A(Tx) Optical Evaluation

PRBS7 Test Pattern Jitter @5Gbps:Tj@1E-12=0.10UI, Dj=0.03UI

20%-80% Rise/Fall Times: 28/27ps

10GHz Optical Sampling Module Without Filtering0.5-12.5GHz PRBS7-PRBS31

2m Optical Fiber

Testboard and DUT

Tx Performance Evaluation – Eye Diagram

800

600

400

200

0

-200

µW

250200150100500-50ps

800

700

600

500

400

300

200

100

0

-100

-200

µW

50250

x103

800

700

600

500

400

300

200

100

0

-100

-200

µW

3002001000

x103

200

100

024020016012080400-40

ps

Unit Interval (UI) = 200ps @5Gbps

Jitter Jitter

OMAEyeOpening

Level 1

Level 0

20% V. Center Window

H. Center

Window

Open Level 1

Open Level 0

Time

Op

tica

l Po

wer

Higher # of hitsNo hits above

Higher # of hitsNo hits below

Gaussian tail fitting Gaussian tail fitting

Histogram Histogram

Histogram

Eye diagram

Specification Proposal for 5Gbps Operation# Specification Min Max Unit Note1 OMA 300 uW 2 ER 3 dB 3 Eye closure 60 %4 Rise Time 65 ps 20%-80%5 Fall Time 65 ps 20%-80%

6 Total Jitter 0.25 UI @BER=1E-12, (1)

7 Deterministic Jitter 0.12 UI (1)

(1) The Tx jitter spec. is the 4GFC Tx jitter budget, not including the jitter of the test set-up.

Tx optical output and metrics:

by processing the Tx eye raw data

TWEPP 2008


800

600

400

200

0

-200

µW

250200150100500-50

ps0.15

0.31 0.69

0

1

0.2

0.8

-0.4

1.4

0 1

0.5

0.85O

pti

ca

l P

ow

er

No

rma

lize

d t

o t

he O

MA


Tx Performance Evaluation – Mask Test• The mask defines an area which the

eye diagram must not cross.• Although based on the 4GFC Tx mask,

the jitter and slope are adjusted to the previous specifications and to the jitter of our test setup.

• The maximum overshoot is 40% to allow laser driver pre-emphasis and because no filtering is being used.

• The green arrows defines the mask margin from 0% to 100%.

Tx Relative Mask(based on the 4GFC spec.)

Specification Proposal for 5Gbps Operation# Specification Min Max Unit Note

8 Tx Mask Margin 0% % (2)

• The mask defines the limits for the overshoot and ringing but cannot be used to test jitter or rise/fall time compliance.

(2) The margins of the full and center mask must be measured as they will be used to compare different devices.

by processing the Tx eye raw data

TWEPP 2008


2m Optical Fiber

Clock Generator

PRBS Source

Optical Attenuator and Power

Meter

Testboard and DUT

2m Optical Fiber

SDA Scope

PRBS7 Test Pattern Jitter @5Gbps (both 850 and 1310nm):Tj@1E-12=0.11UI, Dj=0.04UI

Setup B(Rx) Electrical

Evaluation

850nm/1310nmOptical Tx

20GHz Electrical Sampling Module Without Filtering

0.5-12.5GHz PRBS7-PRBS31Transmitter

Receiver

20%-80% Rise/Fall Times:850nm: 38/46ps, 1310nm: 35/42ps

PC Automation:Runs through several

attenuations (and bitrates) and saves data

Rx Performance Evaluation – Eye and Mask

TWEPP 2008


2m Optical Fiber

Clock Generator

PRBS Source


Meter

Testboard and DUT

2m Optical Fiber

SDA Scope

PRBS7 Test Pattern Jitter @5Gbps (both 850 and 1310nm):Tj@1E-12=0.11UI, Dj=0.04UI

Setup B(Rx) Electrical

Evaluation


20GHz Electrical Sampling Module Without Filtering

0.5-12.5GHz PRBS7-PRBS31Transmitter

Receiver

20%-80% Rise/Fall Times:850nm: 38/46ps, 1310nm: 35/42ps


attenuations (and bitrates) and saves data

-0.3

-0.2

-0.1

0.0

0.1

0.2

0.3

V

250200150100500-50ps

0.16 0.840 1

-0.425

-0.150

0.425

0.150


Dif

fere

nti

al

Ele

ctr

ica

l A

mp

litu

de

[m

V]

Rx Performance Evaluation – Eye and Mask

Specification Proposal for 5Gbps Operation# Specification Min Max Unit Note

9 Total Jitter 0.26 UI @BER=1E-12, (3), (4)

10 Deterministic Jitter 0.11 UI (3), (4)

11 Rx Mask Pass Pass/Fail test,(4)

(3) The Rx jitter spec. is the 4GFC Rx jitter budget , not including the jitter of the test set-up.(4) Measured with an input OMA of 90uW.

Rx Absolute Mask(based on the SFP+ SFI spec. )

Defines the limits for the electrical swing.

TWEPP 2008


2m Optical Fiber

2m Optical Fiber


0.5-6.5GbpsPRBS7-PRBS31 and others

FPGA-Based BER Tester

Clock Generator

Receiver

Transmitter

Setup C(Rx) Sensitivity

Evaluation Testboard and DUT


Meter


OMAs and saves the BER

Rx Performance Evaluation – BER

TWEPP 2008


2m Optical Fiber

2m Optical Fiber


0.5-6.5GbpsPRBS7-PRBS31 and others


Clock Generator

Receiver

Transmitter

Setup C(Rx) Sensitivity

Evaluation Testboard and DUT


Meter


OMAs and saves the BER

Rx Performance Evaluation – BER

Specification Proposal for 5Gbps Operation# Specification Min Max Unit Note12 Sensitivity 45 uW (5), (6)

10-13

10-12

10-11

10-10

10-9

10-8

10-7

10-6

10-5

10-4

10-3

10-2

10-1

BE

R

-25 -24 -23 -22 -21 -20 -19 -18 -17 -16 -15 -14

OMA [dBm]

1310nm 850nm

Sensitivity@1E-12

Sensitivity@1E-12

10G Rx10G Rx

Rx BER curves – Two examples

(5) Specified for a BER of 1E-12.(6) The sensitivity is specified as an OMA. The setup reads average power and the OMA is calculated using the Tx ER.

• By measuring the Bit Error Rate (BER) at different OMAs the BER curve is created.

• The Rx sensitivity is defined as the minimum OMA which allows a 1E-12 BER.

TWEPP 2008


Evaluation of the TRx Power Dissipation

Receiver

TransmitterPRBS

SourceClock

GeneratorPC

Automation

2m Optical Fiber


Meter

Setup D(TRx) Power Dissipation 2m Optical

FiberPower Supply

Clean Testboard (no additional electronics)

and DUT

Specification Proposal for 5Gbps Operation# Specification Min Max Unit Note13 Power Dissipation 600 mW (7)

(7) End-of-life power dissipation and across all operating temperatures and all Rx optical input levels.

• To measure the power dissipation we can simply read current being supplied to the host board.

• The setup is to measure the power at different bitrates and Rx optical input levels.

• In reality we measured the non end-of-life power dissipation at room temperature.

Our experience with commercial TRxs tells us that this spec. might be very harsh for non VCSEL-based transmitters.

TWEPP 2008


Analysis of the Data

• The previous test setups generate a very large data set: raw eye diagrams, direct scope measurements, BER curves and measurements of the raw data.

• As we are interested in the TRx performance at 5Gbps we discard all other bitrates and compare the measurements with the specification.

Tx Perform

ance

at 5Gbps

Rx Perform

ance

at 5Gbps

Power

Dissipation

Flag devices non-compliant with

the spec.

• We flag the devices that do not meet the spec. and we have developed a Figure of Merit (FoM) that combines all the performance data into three numbers: for the Tx part (Tx_FoM), for the Rx part (Rx_FoM) and for the power dissipation (Pwr_FoM).

TWEPP 2008


200

150

100

50

0

Tx

_Fo

M (

Arb

itra

ry U

nit

s)

121110987654321

Figure of Merit1.0

0.8

0.6

0.4

0.2

0.0

-0.2

mW

250200150100500-50ps

There is a strong correlation between the performance of the device and the FoM number.

1.0

0.8

0.6

0.4

0.2

0.0

-0.2

mW

250200150100500-50ps

1.0

0.8

0.6

0.4

0.2

0.0

-0.2

mW

250200150100500-50ps

WorstTx

BetterTx

BestTx

TWEPP 2008

SFP+

10G 13

10VCS

EL #5

SFP+

10G 13

10VCS

EL #4

SFP+

10G 13

10VCS

EL #3

SFP+

10G 13

10VCS

EL #2

SFP+

10G 13

10VCS

EL #1

XFP 1

0G 13

10DFB

SFP+

10G 13

10DFB

#2

SFP+

10G 13

10DFB

#1

SFP+

10G 85

0VCS

EL #2

SFP+

10G 85

0VCS

EL #1

SFP 4

G 1310

VCSEL

SFP 4

G 850V

CSEL

139 171 183

RED=FAILGREEN=PASS

Tx


Results200

150

100

50

0

Tx_

Fo

M (

Arb

itra

ry U

nit

s)

121110987654321

300

250

200

150

100

50

0

Rx_

Fo

M (

Arb

itra

ry U

nit

s)

121110987654321

200

150

100

50

0Pw

r_F

oM

(A

rbit

rary

Un

its)

121110987654321

TWEPP 2008

Pow

erTx

Rx

• SFP (devices 1 and 2): Tx problems with rise/fall times or jitter. Rx jitter problems. Both and Tx and Rx masks fail.

• SFP+ 850mn VCSEL (devices 3 and 4): Good power dissipation and very good 5Gbps Tx performance. The sensitivity of the PINs at 850nm was found to be around -16dBm. The Rx jitter is good with 90uW of OMA (spec.).

• SFP+ 1310mn DFB (devices 5 and 6): Good 5Gbps Tx performance but high power dissipation. The sensitivity of the PINs at 1310nm was found to be around -19dBm.

• XFP 1310nm DFB (device 7): Great jitter performance but very high power dissipation.

• SFP+ 1310mn VCSEL (devices 8 to 12): Great power dissipation and good 5Gbps Tx performance in most of the modules. Considerable overshoot and ringing.

• The sensitivity of all 850nm PINs was found to about 3dB worse than the sensitivity of 1310nm PINs.

• The only modules that were able to meet all 13 points of our spec. (Tx, Rx and power) were 850 and 1310nn VCSEL-based SFP+ modules.

PASS PASSFAILFAIL FAIL


Summary• The future Versatile Transceiver will be

– built from radiation-qualified optoelectronic components;– customized from a commercial transceiver.

• Using commercial devices we have developed test methods for transceiver testing to – select a transceiver family for VTRx customization;– test the performance of the prototype VTRx modules.

• The results from our evaluation of several commercial transceiver modules show that– the SFP+ is the most suitable candidate for VTRx customization;– to achieve low power dissipation we need to target VCSEL-based lasers;– there are 1310nm VCSEL diodes capable of being operated at 5Gbps

with sufficient performance.TWEPP 2008


Extra Slides

TWEPP 2008


250

200

150

100

50

0

# H

its

-25 -20 -15 -10 -5 0 5 10 15 20 25

ps

LoopbackUsing Optical

Reference TxUsing a

-21.0

-20.0

-19.0

-18.0

-17.0

-16.0

dB

m

54321

TRx Device #

Using Optical Loopback

Using a Reference Tx

Rx Performance Evaluation – Alternative

Receiver

Transmitter

SDA Scope

PRBS Source

Clock Generator PC

Automation

2m Optical Fiber


Meter

Receiver

Transmitter


Clock Generator

PC Automation

2m Optical FiberTestboard and DUT


Meter

Setup B’(TRx) Electrical Evaluation

in Optical Loopback

Setup C’(Rx) Sensitivity Evaluation

in Optical Loopback

2m Optical Fiber

2m Optical Fiber

Rx eye, jitter components and bathtub,

mask test

Rx sensitivity

20GHz Electrical Sampling Module

Testboard and DUT

Jitter Histogram - Loopback vs. Reference TxExample Using of the Same Module

Sensitivity - Loopback vs. Reference TxExample Using 5 Units of the Same TRx

• The Rx performance is not decoupled from the Tx.

• The Jitter is from the entire TRx.

• The sensitivity may be different and less consistent.TWEPP 2008

[email protected] 23TWEPP 2008

FoM DefinitionPo

wer

TxRx


Results – Tx and Power

2 0 0

1 5 0

1 0 0

5 0

0

Arb

itra

ry U

nit

s

1 21 11 0987654321

2 0 0

1 5 0

1 0 0

5 0

0

Arb

itra

ry U

nit

s

1 21 11 0987654321D e v ic e #

# TRx TypePass/FailTx Power

1 SFP 4G 850VCSEL Fail Pass2 SFP 4G 1310VCSEL Fail Pass3 SFP+ 10G 850VCSEL #1 Pass Pass4 SFP+ 10G 850VCSEL #2 Pass Pass5 SFP+ 10G 1310DFB #1 Pass Fail6 SFP+ 10G 1310DFB #2 Pass Fail7 XFP 10G 1310DFB Pass Fail8 SFP+ 10G 1310VCSEL #1 Fail Pass9 SFP+ 10G 1310VCSEL #2 Fail Pass10 SFP+ 10G 1310VCSEL #3 Pass Pass11 SFP+ 10G 1310VCSEL #4 Pass Pass12 SFP+ 10G 1310VCSEL #5 Pass Pass

SFP VCSEL modules: Problems with rise/fall times and jitter. The mask fails.

XFP 1310nm DFB module: Great jitter performance but high power dissipation.

SFP+ 1310mn VCSEL modules: Great power dissipation and good 5Gbps Tx performance in most of the modules. Considerable overshoot and ringing.

SFP+ 850mn VCSEL modules: Good power dissipation and very good 5Gbps Tx performance.

SFP+ 1310mn DFB modules: Good 5Gbps Tx performance but high power dissipation.

850nm VCSEL

1310nm DFB

1310nm VCSEL

1310nm VCSEL

850nm VCSEL

PWR_

FoM

Tx_F

oM

TWEPP 2008


Results – Rx and Overall Performance

3 0 0

2 5 0

2 0 0

1 5 0

1 0 0

5 0

0

Arb

itra

ry U

nit

s

1 21 11 0987654321

2 0 0

1 5 0

1 0 0

5 0

0

Arb

itra

ry U

nit

s

1 21 11 0987654321D e v ic e #Fo

M (a

vera

ge o

f all

3)Rx

_FoM # TRx Type

Pass/FailRx All

1 SFP 4G 850VCSEL Fail Fail2 SFP 4G 1310VCSEL Fail Fail3 SFP+ 10G 850VCSEL #1 Pass Pass4 SFP+ 10G 850VCSEL #2 Pass Pass5 SFP+ 10G 1310DFB #1 Pass Fail6 SFP+ 10G 1310DFB #2 Pass Fail7 XFP 10G 1310DFB Pass Fail8 SFP+ 10G 1310VCSEL #1 Pass Fail9 SFP+ 10G 1310VCSEL #2 Pass Fail10 SFP+ 10G 1310VCSEL #3 Pass Pass11 SFP+ 10G 1310VCSEL #4 Pass Pass12 SFP+ 10G 1310VCSEL #5 Pass Pass

SFP modules: Our jitter spec. is too harsh for 4G SFP receivers. Also their electrical swing can be much higher than is allowed by our mask.

Modules with 1310nm PINs: The sensitivity at 1E-12 was found to be around -19dBm (about 3dB better than with 850nm PINs).

SFP+ with 850nm PINs: Using an input with an OMA of 90uW (spec.), the jitter seems not to be made worse by the lower sensitivity of 850nm PINs. Using this OMA the jitter seems to depend mostly on the electronics, i.e. TIA, LA and PCB design.

The only modules that were able to meet all 13 points of our spec. (Tx, Rx and power) were 850 and 1310nn VCSEL-based SFP+ modules.

Fall, Rx Jitt

er, Tx

and Rx Mask

Tx and Rx Jitter

and Mask Power

dissipationOMA and Tx

maskTx mask

TWEPP 2008

Documents

Evaluation of Multi-Gbps Optical Transceivers for Use in Future HEP Experiments