Transcript TWEPP2009_gbtia

The GBTIA, a 5 Gbit/s radiation-hard optical
receiver for the SLHC upgrades
Mohsine Menouni, CPPM - Marseille
Gui Ping, SMU - Dallas - Texas
Paulo Moreira, CERN - Genève
Outline

Introduction


Specifications of the GBTIA chip
Receiver architecture

Transimpedance design

Limiting amplifier design

Pin diode bias and leakage current effect

Measurement results

Conclusion and Perspectives
TWEPP 2009 : Paris, France / September 21-25
[email protected]
2
GBTIA Specifications

The GBTIA is a regenerative optical receiver for data transmission up to 5 Gbit/s

It provides the proper signal level for the clock recovery and deserializer stages

Main specifications:

Bit rate : 5 Gbit/s (min)

Total jitter : < 40 ps p-p

Sensitivity: 20 μA p-p (-17 dBm) for BER = 10-12

pin diode capacitance Cd ~ 400 fF

Dark current : 0 to 1 mA

Power supply : 2.5 V ± 10%

Power consumption < 250 mW

Large range of temperature : From -20 C to 80 C

Die size: 0.75 mm × 1.25 mm

Radiation tolerant (up to 200 Mrad)
TWEPP 2009 : Paris, France / September 21-25
[email protected]
3
Overview of the GBTIA design

Integrating the TIA with the LA in the same chip presents the risk to
degrade performances



A fully differential architecture

Better tolerance to the supply noise

However the input referred noise (thermal noise) is larger than for the
single ended

The power consumption is higher

The photodiode is AC coupled to the TIA
No Need of an offset control circuit at the output of the TIA

A high value of the low cut off frequency

Parasitic of the coupling integrated capacitances limits the bandwidth
Define the sensitivity of the optical receiver

Wide bandwidth

Low noise
Limiting Amplifier and output buffer (LA)




Transimpedance amplifier (TIA)


Propagation of the crosstalk noise trough power supplies or the substrate
Provide a clean signal to the output

High gain

Wide bandwidth
Offset level compensation
Additional features :


Internal voltage regulator (with enable/disable control)
leakage-current indicator

Carrier-detect and signal-strength indicators

Squelch function (with enable/disable control)
TWEPP 2009 : Paris, France / September 21-25
[email protected]
4
Transimpedance Amplifier
RF

Shunt feedback amplifier is widely used for high
speed receiver designs
CPH
IIN
VOUT
AV


To increase the bandwidth :

Decrease the feedback resistor

Increase the amplifier open loop gain

Decrease the input node capacitance
To minimize the thermal noise :

Increase the feedback resistor

Decrease the input node capacitance

Increase the amplifier transconductance
RIN 
RF
AV  1
CPAR
CT  CPH  CPAR  CIN
ZT 
BW 
vout
 AV
RF

iin
AV  1 1  jC RF
T
1  AV
1  AV
2 .RF CT
4kT 4kT  1 2 . f .RF CT 




2
RF
gm
RF  g m
2
in ,in
TWEPP 2009 : Paris, France / September 21-25
2
[email protected]




5
Bandwidth extension technique

In order to maintain a low level of noise with keeping a
large bandwidth, the shunt peaking technique is used
in the design

Shunt peaking

Introduction of an inductance in series with the load
resistance

Enhances the bandwidth

The frequency response is characterized by the
ratio m
L  m.R 2 .C
Factor m
Normalized f3dB
Response
0
1.00
No shunt peaking
0.32
1.60
Optimum group delay
0.41
1.72
Maximally flat
0.71
1.85
Maximum bandwidth
TWEPP 2009 : Paris, France / September 21-25
[email protected]
6
Implemented TIA structure


Differential structure is adopted
Inductive peaking



The target bandwidth of 3.5 GHz is achieved for the
worst case of process and temperature (simulations
including parasitic)

High transimpedance gain (RF=380 W)

Low level of input referred noise
Cascode


2V
2 nH
200 W
2 nH
200 W
Out-
Out+
Reduces the Miller effect
Current density is optimized

High current density needed to achieve high cut off
frequency for the input transistor

Input transistor size optimized for an input
capacitance of 700 fF
380 W
380 W
7.8 pF
7.8 pF
In+
In-
2 V supply required
TWEPP 2009 : Paris, France / September 21-25
[email protected]
7
Limiting Amplifier requirements

Considering the sensitivity and the gain of the TIA :

the photocurrent is converted to a minimum voltage of 12 mV pp

This voltage is amplified by the Limiting Amplifier to reach the proper
voltage necessary for the following stages

The design of the LA demonstrates a high gain to achieve the 400 mV
pp

Gain is around 40 dB in typical condition (28 dB in worst-case
scenario)

The minimum overall bandwidth is 3.5 GHz

The noise contribution of this stage is maintained negligible :

The input referred noise is maintained lower than 850 µV RMS
(12mV/14) for a BER of 10E-12

The input capacitance of the LA is sufficiently low so that it does not
reduce the TIA bandwidth

The number of stages is set to 5 (4 LA + a buffer)

Limiting Amplifier
TIA
LA1
LA2
LA3
2 mA
2 mA
4 mA
LA4
8 mA
Buffer
8 mA
More stages introduce a high power dissipation

Offset cancellation is incorporated in LA block to prevent the
mismatch in the differential amp from saturating the latter stages

In order to maintain a wide bandwidth while delivering large current to
the load, the amplifiers stages in the LA are designed to have
increasingly larger size and current

Offset cancellation

Minimize the load capacitance seen by the previous stage

Allow bandwidth extension
The gain of the first stage (LA1) is set to a high value to reduce the
noise
TWEPP 2009 : Paris, France / September 21-25
[email protected]
8
Limiting Amplifier stage

High bandwidth topology for each stage

Cherry and Hopper structure

gm stage followed by shunt-feedback stage

Second stage uses active “inductors”

by active inductive peaking, the bandwidth is
increased by 34% over a resistive loaded
topology.
TWEPP 2009 : Paris, France / September 21-25
[email protected]
9
Output Buffer Stage
Off chip 50 W transmission line
Voutp
Voutn
Vinn
50 W
Vinp
8 mA

Needs to be able to deliver 4 mA current to a 50 W load at full speed

The output stage needs to be able to fully switch 8 mA taking into consideration double termination.

The buffer has not to present too large capacitance to the preceding stage
TWEPP 2009 : Paris, France / September 21-25
[email protected]
10
Pin diode leakage current effect

The pin diode leakage current increases with the radiation dose and can
reach a value of 1 mA for a high dose level
Vdd

AC coupling is adopted for the fully differential receiver:
IDC
iAC
IDC
iAC


The AC coupling capacitance is integrated in the chip

The value is made as high as possible : 7.8 pF
In order to maintain the low cutoff frequency to a reasonable value we
need a high value for the photodiode bias resistance

Since we have to maintain a voltage across the photodiode, this resistance
is implemented with active device

Photodiode biasing


The voltage across the pin diode decreases to 0.6 V for Vdd = 2 V

The low cut-off frequency increases

The simulated cut off frequency is around 1 MHz for IDC = 1 mA

Still compatible with the GBT encoding
CC
CC
The DC level has an effect on the noise and the sensitivity

For the low level of the leakage current, the shot noise is negligible
comparing to the thermal noise

When the DC level is around 1 mA, the shot noise level becomes
comparable to the thermal noise

A sensitivity degradation is expected at the end of life of the SLHC

Simulations show a sensitivity loss of 3-4 dB
TWEPP 2009 : Paris, France / September 21-25
[email protected]
Long consecutive identical bits
11
Outline

Introduction

Receiver architecture

measurement results

Thanks to :

Chip photograph and test boards
Luis Amaral, Jan Troska and Csaba Soos

Eye diagram measurements
for their help with the test setup

Bit Error Rate estimate

Performances versus power supply

BER measurements with the GBT protocol
and error correction

Radiation effects

Influence of the optical DC level on the BER
Summary and Perspectives
TWEPP 2009 : Paris, France / September 21-25
[email protected]
12
Chip photograph and test boards

IBM CMOS8RF-LM technology, a standard eight-metal-layer

The n-MOSFET fT of this technology is in the range of
100–120 GHz

Die size: 0.75 mm × 1.25 mm

2 PCB boards were designed in order to evaluate the GBTIA
performances

Board for optical tests


Use a pin-diode as a signal source

PDCS60T-XS : high speed photodiode Pin diode from
Enablence

Top illuminated 10 Gb/s photodiode

Low capacitance: 240 fF

Responsitivity : 0.9 A/W at  = 1310 nm

The connection between the TIA and the pin diode is made
very short < 200 µm
output-
Input+
Input-
Bandgap reference
AC coupling
capacitance
0.13-μm bulk CMOS process.
Pin diode biasing

TIA
LA
output+
Board for electrical tests

PIN diode is replaced by an electrical network

Voltage source to adjust the input current

Input Capacitance is set to 500 fF

PCB parasitic capacitances were minimized

This board was used essentially for irradiation test
TWEPP 2009 : Paris, France / September 21-25
[email protected]
13
Eye diagram measurements (1/2)
PRBS
Generator
Agilent
N4903A
clock
data datab
Optical Tx
Optical
Attenuator
DC blocking connectors
ckin
Lecroy
SDA18000
ch1
outp
ch2
outn
GBTIA Test
Board

12 Gbit/s Serial Pattern Generator

Commercial 10 Gbit/s Optical transmitter

18 GHz BW serial data analyzer

Optical attenuator to adjust the optical level for the pin diode
TWEPP 2009 : Paris, France / September 21-25
[email protected]
14
Eye diagram measurements (2/2)


Measured differential eye diagrams at 5 Gbit/s
for different optical power at the input (-6 dBm
and -18 dBm)

Well opened eye diagram for -6 dBm and still
correct at 18 dBm

The test PRBS sequence length is 27-1

A constant output swing of 400 mV

For a power supply of 2 V the power consumption
is 90 mW (120 mW at 2.5 V)
For -6 dBm input :

Rise time = 30 ps

Total jitter = 0.15 UI @ BER = 10-12


Eye diagram 5 Gbit/s optical power = -6 dBm
UI = 200 ps
For -18 dBm input :

Rise time = 60 ps

Total jitter = 0.55 UI @ BER = 10-12
Eye diagram 5 Gbit/s optical power = -18 dBm
TWEPP 2009 : Paris, France / September 21-25
[email protected]
15
Bit Error Rate measurements
Bit Error Rate
Tester (BERT)
Xilinx ML421
platform
ckin
Clock
generator
ck
ckb
ckinb
din dinb
dout doutb
Optical
Tx
ckin
Optical
Attenuator
Lecroy
SDA18000
ch1
GBTIA Test
Board
ch2
DC blocking connectors

Test set up to evaluate the BER of the optical link using the GBTIA chip as a receiver

The Bit Error Rate Tester (BERT) is the Xilinx platform ML421

Virtex-4 Rocket-IO FPGA

Transceivers operating up to 6.5 Gbit/s

BER calculation is based on the comparison of the transmitted and the received data

Measuring a very low BER is time consuming

Low BER is determined using extrapolation from the measurements of BER versus the input optical power
TWEPP 2009 : Paris, France / September 21-25
[email protected]
16
Bit Error Rate Estimate
Bit Error Rate versus Input Optical Level
Data pattern PRBS7
1E-04
Vdd = 2 V and T = 25 °C

Data pattern : PRBS7

The sensitivity for a BER of 10-12 is
estimated around -19 dBm
1E-06
1E-08
BER

1E-10
1E-12
1E-14
-26
-24
-22
-20
-18
-16
-14
Optival power (dBm)
TWEPP 2009 : Paris, France / September 21-25
[email protected]
17
Performances versus power supply
BER versus Input OpticalPower
BER @1.8V
BER @2.0V
BER @2.2V
1E-02
1E-04
BER
1E-06
1E-08
1E-10
1E-12
1E-14
-30
-25
-20
-15
-10
-5
Optical Power (dBm)
TWEPP 2009 : Paris, France / September 21-25
[email protected]
18
BER measurements with the GBT encoding

The SEU on the photodiode are likely to be
the main source of errors

In the GBT chip an error correction system is
implemented



1E-02
GBT prtocol
Reed-Solomon error-correcting
encoder/decoder
For this test set up, the GBT encoder decoder
was implemented in virtex-4 FPGA used in
the BERT platform
Without error correction, the sensitivity of the
optical receiver still around -19 dBm
The sensitivity is improved by 2 dB if the
correction encoder is enabled
TWEPP 2009 : Paris, France / September 21-25
1E-04
GBT protocol with error
correction
1E-06
BER

Bit Error Rate versus Input Optical Level
Data pattern : GBT encoding
1E-08
1E-10
1E-12
1E-14
-26
-24
-22
-20
-18
-16
-14
Optival power (dBm)
[email protected]
19
Eye diagram versus the total dose
Prerad eye diagram (input=500 mV )
200 Mrad eye diagram (input=500 mV )
Prerad eye diagram (input=50 mV )
200 Mrad eye diagram (input=50 mV )

Electrical board used for irradiation test

Irradiation test done at CERN using Xray facility

Only the GBTIA chip is submitted to Xray beam

No degradation is observed after a dose rate of 200 Mrad
TWEPP 2009 : Paris, France / September 21-25
[email protected]
20
BER versus the total dose
BER versus the total dose
Prerad
1E-04
10M
100M
BER
1E-06
1E-08
1E-10
1E-12
1E-14
-24
-22
-20
-18
-16
Equivalent optical input level (dBm)
TWEPP 2009 : Paris, France / September 21-25
[email protected]
21
Influence of the optical DC level on the BER

the leakage current of the photodiode increases
to 1 mA at a high level of dose

In order to measure the influence of this
leakage current, the pin diode is illuminated by
an additional DC laser source
Bit Error Rate versus the DC Optical Level
Data pattern : GBT encoding
DC courant = 0
1E-04
DC current = 0.45 mA
DC current = 0.92 mA

In this case we checked that the integrated bias
circuit ensures a sufficient voltage across the
pin diode
We don’t observe a notable degradation of the
BER coming from the effect of the low cut-off
frequency


The value of this frequency still compatible with
the GBT encoding data
The power penalty introduced by the shot noise
of the leakage current is around 4 dB
TWEPP 2009 : Paris, France / September 21-25
1E-06
BER

1E-08
1E-10
1E-12
Power penalty
1E-14
-26
-24
-22
-20
-18
-16
-14
-12
-10
Optical power (dBm)
[email protected]
22
Conclusion and Perspectives




Main Specifications in terms of bandwidth and
sensitivity are respected

Eye diagram is well opened at 5 Gbit/s

Sensitivity of -19 dBm for a BER of 10-12
The effect of the leakage current is estimated

The sensitivity is degraded by 4 dB

The value of the low cut off frequency still compatible with
the data encoding used for the GBT
Radiation effects :

Radiation tolerance is proven up to 200 Mrad

We have to estimate the single event upset tolerance

Work has started to encapsulate the GBTIA and the
photodiode in a TO Package
A final design is scheduled to implement the additional
features :

Leakage-current and signal-strength indicators

Carrier-detect

Squelch function (with enable/disable control)
TWEPP 2009 : Paris, France / September 21-25
Bit rate
5 Gbit/s
Transimpedance gain
20 kW (typ)
Output voltage
± 0.2 V (50 W)
Sensitivity for BER =10-12
-19 dBm
Supply voltage
2.5 V ± 10%
Power consumption
120 mW
Radiation tolerance
> 200 Mrad
Penalty for high dark current
4 dB
[email protected]
23