White Paper: 7 Series FPGAs
WP419 (v1.0) March 27, 2012
Equalization for High-Speed Serial
Interfaces in Xilinx 7 Series FPGA
Transceivers
By: Harry Fu
The appetite for data is exploding, and the industry
is being forced to transition from parallel to serial
interface technology at constantly increasing speeds.
Higher data rates produce higher bandwidths and
present new demands and challenges in transceiver
design.
Serial systems today require very low bit-error rates.
As the serial signaling rate increases, channel-driven
signal distortion increases with it, while the bit
sampling time needed to process this distortion only
grows shorter. The techniques used to compensate
for signal distortion, therefore, become critical
elements in the design.
This compensation is usually called emphasis in the
transmit domain and equalization in the receive
domain. The effective implementation of emphasis
and equalization in Xilinx® 7 series FPGAs and
Zynq™-7000 Extensible Processing Platform (EPP) is
the subject of this white paper.
© Copyright 2012 Xilinx, Inc. Xilinx, the Xilinx logo, Artix, ISE, Kintex, Spartan, Virtex, Zynq, and other designated brands included herein are trademarks of Xilinx in the United
States and other countries. All other trademarks are the property of their respective owners.
WP419 (v1.0) March 27, 2012
www.xilinx.com
1
7 Series Transceiver Emphasis/Equalization Overview
7 Series Transceiver Emphasis/Equalization Overview
The Xilinx 7 series FPGA and EPP portfolios are based on the concept of a scalable,
optimized architecture, which allows all the FPGA elements to be scaled and
optimized to the market and target application. This yields the flexibility required to
port design investments to the “right fit” platform, based on current needs. The serial
transceivers are no exception. Xilinx offers a portfolio of transceivers that run from
500 Mb/s up to 28.05 Gb/s. See Figure 1.
X-Ref Target - Figure 1
Max Line RAte (Gb/s)
30
28.05
25
20
15
13.1
12.5
10
6.6
5
0
GTZ
GTX
GTH
GTP
Transceivers
Figure 1:
WP419_01_032012
Xilinx 7 Series FPGA Transceiver Portfolio
This scalable approach enables cost-effective solutions for consumer applications,
where transceivers are used at lower data rates, to the very powerful, sophisticated
transceivers required by the telecommunications equipment used to move data
around the world.
Figure 2 depicts an example 2 x 100GE line card with traffic manager design showing
three different uses of 7 series serial transceivers.
2
www.xilinx.com
WP419 (v1.0) March 27, 2012
7 Series Transceiver Emphasis/Equalization Overview
X-Ref Target - Figure 2
1
2
CFP2/4
Optical
Module
Xilinx
100GE
IP
CFP2/4
Optical
Module
Xilinx
100GE
IP
Serial
Memory
SDRAM
SDRAM
SDRAM
SDRAM
Xilinx
Traffic
Manager
IP
Xilinx
Interlaken
IP
Backplane
3
Virtex-7 FPGA XC7VH870T
WP419_02_031912
Figure 2:
2 x 100GE Traffic Manager: Example Transceiver Use Cases
The three example use cases shown in Figure 2 are:
1. GTZ transceivers (8 x 25.78 Gb/s) connect directly to CFP2 optics via an OIF VSR
channel (6 in.)
2. GTH transceivers (16 x 10.3125 Gb/s) connect to a serial memory (for storing
descriptors used by the Traffic Manager IP) (6 in.)
Note: This case uses the special “low-power mode” of the GTH transceiver to optimize
power for the short channel to the serial memory using TX emphasis and RX linear
equalization.
3. GTH transceivers (24 x 12.5 Gb/s) connect over a mezzanine card connector, midplane, or backplane using decision feedback equalization (DFE) mode (40 in.)
To enable creation of a robust high-speed serial link system with these transceivers
running at low power, equalization techniques are built into each transceiver, based on
its targeted line rate and targeted applications. See Figure 3.
X-Ref Target - Figure 3
3-Tap FIR
Serial Transceiver
TX Driver
Pre-emphasis
FPGA Logic
PISO
PLL
Hard
PCS
Logic
Serial
Channel
SIPO
RX DFE
GTX/GTH
RX
CDR
RX
Linear EQ
RX
DiffAmp
2-D Eye
Scan
Fully Adapting
Up to 20 dB
5 Fixed Taps in GTX
7 Fixed Taps plus
4 Sliding Taps in the GTH Transceiver
WP419_03_031912
Figure 3:
WP419 (v1.0) March 27, 2012
7 Series FPGA and EPP Transceiver Equalization Blocks
www.xilinx.com
3
Channel Conditions and Signal Distortion Components
Transmitter emphasis is a commonly used technique to compensate for the channel
losses between transceivers. All 7 series FPGA transceivers offer three-tap TX
emphasis.
Receiver equalization varies with different transceivers:
•
•
•
For systems running at ~6 Gb/s or below that utilize the GTP transceiver, the
Continuous Time Linear Equalization-only (CTLE) structure has proven to be
very efficient and has been used successfully over the past decade.
For systems running at ~10 Gb/s – 15 Gb/s, the channel complexity can vary
greatly, from simple chip-to-chip interfaces, to very complicated backplanes with
long traces, blind vias, and connectors. In complex interface applications running
at these line rates, signal distortions exist that make it impossible to guarantee
adequate margins with a linear equalization process like CTLE. Therefore, in
addition to CTLE, GTX/GTH transceivers offer DFE, a non-linear equalization
technique and a popular option in the industry for 10G+ serial links.
For systems running beyond 20 Gb/s, the GTZ transceiver is designed for
short-reach applications directly connecting to CFP2 and CFP4 optics
(OIF-CEI28-VSR), or for longer channels with re-timers. In this scenario, a
CTLE-only receiver is sufficient.
Channel Conditions and Signal Distortion Components
Channel insertion loss and channel discontinuity are the most widely seen conditions
that degrade signal quality. This section illustrates how each one impacts the signal.
Channel Insertion Loss
When signals travel through a transmission line, such as a board trace or backplane,
signal power is degraded and the high frequency components are attenuated more
than the low frequency components. This impact is referred to as channel insertion
loss. Channel conditions such as conductor loss, dielectric loss, and surface roughness
can all result in channel insertion loss.
Channel loss is frequency-dependent and is usually quantified in the frequency
domain at the Nyquist frequency of the data stream:
fNyquist = (data rate) / 2
Equation 1
Figure 4 shows insertion loss from two sample traces made of Nelco 4000-13. One is
16 inches long, the other 26 inches long.
4
www.xilinx.com
WP419 (v1.0) March 27, 2012
Channel Conditions and Signal Distortion Components
X-Ref Target - Figure 4
0
-5
16 Inch Nelco 4000-13
-10
dB (S)
-15
26 Inch Nelco 4000-13
-20
-25
-30
-35
-40
0
1
2
3
4
5
6
7
8
9
10
Frequency (GHz)
WP419_04_031912
Figure 4:
Differential Insertion Loss of Sample Nelco 4000-13 Traces (16-Inch, 26-Inch)
In general, the insertion loss increases as the frequency increases. Given the same
material and geometry, the insertion loss also increases as the trace length increases.
Table 1 lists the insertion loss at 3 GHz (fNyquist of 6 Gb/s data streams) and 5 GHz
(fNyquist of 10 Gb/s data streams) for the traces shown in Figure 4.
Table 1:
Insertion Losses for Sample 16-Inch and 26-Inch Nelco Traces
3 GHz
5 GHz
16 inch Nelco
~7 dB
~11.5 dB
26 inch Nelco
~12 dB
~18 dB
At a given data rate, the channel loss impact on the data stream can also be viewed in
the time domain. Figure 5 shows the non-ideal pulse responses of the channels shown
in Figure 4 at 10 Gb/s. Because of the insertion loss, one data symbol “lifts” the
neighboring data symbols. The greater the insertion loss, the greater the lift — and,
therefore, more symbols are lifted.
WP419 (v1.0) March 27, 2012
www.xilinx.com
5
Channel Conditions and Signal Distortion Components
X-Ref Target - Figure 5
0.0
16 Inch Nelco 4000-13
dB (S)
-100.0
-200.0
26 Inch Nelco 4000-13
-300.0
-400.0
-500.0
2.0
3.0
4.0
5.0
6.0
Time (ns)
Figure 5:
WP419_05_031912
Non-ideal Pulse Responses of the Nelco Traces
The impact from neighboring symbols is usually referred as “inter-symbol
interference” (ISI). Figure 6 illustrates how ISI can change the data stream in a linear
system. A received 01000 is changed by the pseudo-channel. In a linear system, a bit
stream 011110 is a summation of UI-delayed 01000s. Because the pseudo-channel
introduces ISI, the last '0' of the 011110 stream becomes a potential error bit.
X-Ref Target - Figure 6
1.4
Received 011110
1.2
Amplitude (v)
1.0
Ideal 01000
0.8
0.6
Potential Error
0.4
0.2
0
-0.2
Received 01000
UI Delayed Received 010000
-0.4
0
2
4
6
8
10
Unit Interval
Ideal Response
Real Response
UI Delayed Responses
WP419_06_031912
Figure 6:
Sample ISI Impacts on Data Stream
Channel Discontinuity
Channel discontinuity happens when the characteristic impedance of the channel
changes along the transmission line and results in signal reflection during the
6
www.xilinx.com
WP419 (v1.0) March 27, 2012
Channel Conditions and Signal Distortion Components
transmission. Figure 7 illustrates how impedance mismatches cause reflections to
occur and how the waveform splits across the interface.
Z1 and Z2 are the characteristic impedances on the two sides of the interface.
X-Ref Target - Figure 7
Z1
Vi
Z2
Vr
Vt
Vr
Vi
=
Z2 – Z1
Z2 + Z1
Vt
Vi
=
2 * Z2
Z2 + Z1
Vi: Input Signal
Vr: Reflected Signal
Vt: Transmitted Signal
WP419_07_032012
Figure 7:
Reflection Due to Impedance Mismatch
When multiple reflections occur, the transmitted signal is reshaped by the extra Vt.
Figure 8 shows an extra Vt at the Z1-Z2 interface due to the second reflection.
X-Ref Target - Figure 8
Z0
Z1
Z2
V1
Vt1
1st
2nd
Vt2
3rd
WP419_08_031912
Figure 8:
Extra Vt Due to Second Reflection
Signal reflection can result in signal amplification or signal destruction. These impacts
can be observed in the pulse response. Figure 9 shows an example of a channel
reflection appearing on the pulse response: i.e., a “bump” at ~8 ns and a “dip”
following that. The distance between the bump to the main symbol depends on the
length of the trace along which the reflected signal bounces back.
X-Ref Target - Figure 9
100.0
Volts (mV)
0.0
-100.0
-200.0
Reflection
-300.0
4.0
5.0
6.0
7.0
8.0
9.0
10.0
Time (ns)
Figure 9:
WP419 (v1.0) March 27, 2012
11.0
WP419_09_031912
Reflection on Pulse Response
www.xilinx.com
7
Techniques Used to Combat Insertion Loss and Discontinuities
Techniques Used to Combat Insertion Loss and Discontinuities
Here are a few techniques to combat insertion loss and discontinuities.
Transmitter End Techniques
TX emphasis has been used for many years as a simple solution to combat insertion
loss. Because insertion loss causes high-frequency signal components to lose more
energy than low-frequency components, TX emphasis either:
•
•
boosts the high-frequency components
suppresses the low-frequency components at the transmitter.
After channel losses degrade the high-frequency component of the TX emphasized
signal, the high-frequency and low-frequency components of the signal should appear
to be more balanced.
De-emphasis
De-emphasis suppresses the low-frequency components at the transmitter. When a
0-to-1 or 1-to-0 transition occurs, regular (nominal) voltage swing is applied to the
symbols that have a new value after the transition. The symbols that keep the same
value after the transition, however, use a reduced voltage swing. The amount of swing
voltage reduction is referred to as the “tap weight” of the de-emphasis.
Figure 10 and Figure 11 show how TX de-emphasis can reshape a signal based on
simulations with 7 series transceiver IBIS-AMI models.
Two GTX transmitters are configured to send a repeating 11110000 pattern.
Figure 10 shows the data stream without any de-emphasis. The symbols following the
transition have a peak-to-peak amplitude of ~0.28V.
Time (nsec)
Figure 10:
208.8
208.7
208.6
208.5
208.4
208.3
208.2
208.1
208.0
207.9
207.8
207.7
207.6
0.16
0.14
0.12
0.10
0.08
0.06
0.04
0.02
-0.00
-0.02
-0.04
-0.06
-0.08
-0.10
-0.12
-0.14
-0.16
207.5
Waveform
X-Ref Target - Figure 10
WP419_10_031912
11110000 Binary Pattern, No De-emphasis
Figure 11 shows the data stream with 6 dB of de-emphasis. The symbols following the
transition now have a peak-to-peak amplitude of ~0.14V. The difference is because the
de-emphasis tap weight applied to the data stream is 6 dB.
8
www.xilinx.com
WP419 (v1.0) March 27, 2012
Techniques Used to Combat Insertion Loss and Discontinuities
0.16
0.14
0.12
0.10
0.08
0.06
0.04
0.02
-0.00
-0.02
-0.04
-0.06
-0.08
-0.10
-0.12
-0.14
-0.16
Time (nsec)
Figure 11:
208.8
208.7
208.6
208.5
208.4
208.3
208.2
208.1
208.0
207.9
207.8
207.7
207.6
~6dB
207.5
Waveform
X-Ref Target - Figure 11
WP419_11_031912
11110000 Binary Pattern, 6 dB De-emphasis
The de-emphasis implementation can reduce the ISI to ensure correct signal sampling
at the receiver. Figure 12 and Figure 13 show scope shots of Kintex®-7 FPGA GTX
transceiver eye diagrams and jitter decomposition at 10 Gb/s. The signal goes through
the package, the characterization board trace, the connector, and the SMA cable. With
the de-emphasis off, the eye shown in Figure 12 is distorted, and the peak-to-peak ISI
jitter [ISIJ(p-p)] reads 8.50 ps on the scope.
X-Ref Target - Figure 12
WP419_12_031912
Figure 12: GTX Transceiver 10 Gb/s Eye Diagram without De-emphasis
WP419 (v1.0) March 27, 2012
www.xilinx.com
9
Techniques Used to Combat Insertion Loss and Discontinuities
With ~2 dB post-tap de-emphasis, the eye in Figure 13 looks cleaner, and the ISI jitter
decreases to 3.97 ps.
X-Ref Target - Figure 13
WP419_13_031912
Figure 13:
GTX Transceiver 10 Gb/s Eye Diagram with 2 dB Post-tap De-emphasis
Three-Tap TX De-emphasis
In 7 series FPGAs and EPPs, all transceivers have a three-tap FIR to implement TX
de-emphasis. The three taps include the pre-tap, main tap, and post-tap. Figure 14
shows the TX driver block diagram with three-tap de-emphasis.
X-Ref Target - Figure 14
Pre-Driver
Pre-tap
De-emphasis
Pad Driver
MGTAVTT
TXPRECURSOR[4:0]
50Ω
Pre-Driver
PISO
50Ω
TXP
Main
Pad Driver
TXN
TXDIFFCTRL[3:0]
Pre-Driver
TX Serial Clock =
Data Rate / 2
Figure 14:
Post-tap
De-emphasis
Pad Driver
TXPOSTCURSOR[4:0]
WP419_14_032712
TX Driver Block Diagram with Three-Tap De-emphasis
Figure 10 and Figure 11 show the post-tap de-emphasis impact on the signal. The
pre-tap de-emphasis is similar to post-tap. The difference is that the symbol that has
the high swing is the one before the transition rather than after the transition. Figure 15
shows, a simulation example of GTX transmitter output, how a 6 dB pre-tap
de-emphasis can reshape a signal repeating 11110000.
10
www.xilinx.com
WP419 (v1.0) March 27, 2012
Techniques Used to Combat Insertion Loss and Discontinuities
0.16
0.14
0.12
0.10
0.08
0.06
0.04
0.02
-0.00
-0.02
-0.04
-0.06
-0.08
-0.10
-0.12
-0.14
-0.16
Time (nsec)
Figure 15:
208.8
208.7
208.6
208.5
208.4
208.3
208.2
208.1
208.0
207.9
207.8
207.7
207.6
~6dB
207.5
Waveform
X-Ref Target - Figure 15
WP419_15_032712
11110000 Binary Pattern, 6 dB Pre-tap De-emphasis
Programmable TX De-emphasis
In 7 series FPGA transceivers, the tap weights are all programmable to meet different
channel conditions. For example, the GTX/GTH/GTP transceivers have 32 settings
for post-tap de-emphasis, up to 12.96 dB, and 21 settings for pre-tap de-emphasis, up
to 6.02 dB.
Note: The maximum pre-tap boost is not as high as the maximum post-tap boost. The
reason for this is that, in general, the pre-cursor degradation due to channel loss is small
compared to post-cursor degradation.
Besides the weight, the polarity of the de-emphasis taps is also programmable. This
gives more flexibility of the de-emphasis to fit different channel loss conditions.
Receiver End Techniques
At the receiver, there are also similar equalization techniques to boost the
high-frequency components of the signals. Continuous Time Linear Equalization
(CTLE) is one of the most popular linear equalization techniques at the receiver end of
the line.
RX Continuous Time Linear Equalization (CTLE)
The concept of CTLE can be explained in the frequency domain. Figure 16 shows the
conceptual frequency response of channel and CTLE and the result of the
combination. There is more channel insertion loss for high-frequency signals than
low-frequency signals. Thus, the link channel can be viewed as a low-pass filter. To
compensate for the low-pass characteristics of the channel, a high-pass filter is added
at the receiver to achieve balance between the high-frequency and low-frequency
components of the data stream.
WP419 (v1.0) March 27, 2012
www.xilinx.com
11
Techniques Used to Combat Insertion Loss and Discontinuities
X-Ref Target - Figure 16
Channel Loss
Combined Frequency
Response
CTLE
=>
+
WP419_16_032712
Figure 16:
Frequency Domain Effects of CTLE
Selecting the Correct CTLE
When selecting a CTLE to compensate for channel loss, it is important to ensure that
the peak of the CTLE frequency response is at the correct frequency and the correct
gain (boost). Incorrect CTLE selection results in under-equalization or
over-equalization, and thus, in suboptimal post-CTLE signal integrity. Figure 17
through Figure 19 show simulation examples of under-equalization,
over-equalization, and correct equalization (respectively) using the GTX receiver and
a sample 16-inch Nelco 4000-13 trace. Under-equalization does not open the eye
enough in both the horizontal and vertical directions; over-equalization can result in
high jitter. The eye with correct equalization shows low noise and low jitter.
X-Ref Target - Figure 17
0.20
0.15
0.10
Density
0.05
0.00
-.05
-0.10
-0.15
-0.20
0
20
40
60
80
100
120
140
Time (psec)
Figure 17:
12
160
180
200
WP419_17_032712
Under-Equalized Eye Diagram
www.xilinx.com
WP419 (v1.0) March 27, 2012
Techniques Used to Combat Insertion Loss and Discontinuities
X-Ref Target - Figure 18
0.4
Density
0.2
0.0
-0.2
-0.4
0
20
40
60
80
100
120
140
160
Time (psec)
Figure 18:
180
200
WP419_18_032712
Over-Equalized Eye Diagram
X-Ref Target - Figure 19
0.3
0.2
Density
0.1
-0.0
-0.1
-0.2
-0.3
0
20
40
60
80
100
120
140
Time (psec)
Figure 19:
160
180
200
WP419_19_032712
Correctly Equalized Eye Diagram
Programmable CTLE
To support varying channel conditions, the correct CTLE selection is needed. All
7 series FPGAs and EPPs offer programmable CTLE in their receivers to address this
concern. The thousands of possible frequency responses with different high-frequency
peaking can address a multitude of channel variations.
WP419 (v1.0) March 27, 2012
www.xilinx.com
13
Techniques Used to Combat Insertion Loss and Discontinuities
CTLE Auto Adaptation for Ease of Use
With thousands of settings available, proper tuning of the CTLE is critically important.
As illustrated in Figure 17 through Figure 19, under-equalization or over-equalization
can result in suboptimal signal integrity.
To resolve this issue, 7 series FPGA transceivers offer an auto-adaptation feature. The
adaptation removes the CTLE tuning burden from the user by automatically finding
the optimal setting. Based on the incoming data streams, it tests different settings and
selects one that provides the sufficient system margins.
Users can employ a one-time calibration for a given channel, or leave auto-adaptation
on for run-time calibration. Using auto-adaptation adds less than a 5% power
consumption increase to the transceiver. For applications that might experience
voltage and temperature variations during operation, the run-time calibration can be
very useful.
RX Decision Feedback Equalizer (DFE)
Linear equalization techniques such as RX CTLE have one major limitation: noise.
When noise (such as reflections or crosstalk) is present on the channel, CTLE amplifies
the high-frequency noise right along with the data.
Decision Feedback Equalizer (DFE) is a successful technique to mitigate ISI without
amplifying the noise. It works by directly removing the ISI from a previous bit,
allowing the following bits to be correctly sampled. Figure 20 shows the block
diagram of a simple DFE implementation.
X-Ref Target - Figure 20
y (t ) = u (t ) +
n
w [i ] yd (t i UI )
i =1
Equalized Output
Adder to
Feedback
Tap Weight
X
W[3]
yd(t)
y(t)
u(t)
+
RX
W[n]
Slicer for Decision
X
W[2]
X
W[1]
X
Multiplier
Z-1
Z-1
Unit Delay
Figure 20:
Z-1
WP419_20_032712
Simplified DFE Block Diagram
As shown in Figure 5, the ISI effect from this symbol applied to the following bits is
determined by the pulse response, based on channel characteristics. DFE starts with a
“decision slicer” to determine whether the current symbol is High or Low. The
resulting symbol goes through unit delays and multiplies with the tap weights. The
weighted delayed signals are added together to the input analog signals. If the tap
weights are well selected to cancel the ISI at each following symbol, the result of this
feedback loop is able to compensate for as many taps of ISI as the DFE has.
Figure 21 and Figure 22 illustrate two sample data streams equalized by the DFE.
14
www.xilinx.com
WP419 (v1.0) March 27, 2012
Techniques Used to Combat Insertion Loss and Discontinuities
X-Ref Target - Figure 21
1.4
Potential Error
Amplitude (v)
0.9
010110000
-0.4
010110000
Feedback
-0.1
Equalized
010110000
-0.6
1
2
3
4
5
6
7
8
9
10
Unit Interval
Figure 21:
WP419_21_032712
010110000 Binary Pattern with DFE Equalization Applied
X-Ref Target - Figure 22
1.4
011110000
Amplitude (v)
0.9
-0.4
011110000
Feedback
-0.1
Equalized
011110000
Potential Error
-0.6
1
2
3
4
5
6
7
8
9
10
Unit Interval
Figure 22:
WP419_22_032712
11110000 Binary Pattern with DFE Equalization Applied
The red curve in Figure 21 is the pre-equalized data stream of 010110000. Without
any equalization, the second ‘0’ could be sampled incorrectly due to the ISI. The green
curve is the feedback stream added to the original data stream with well-selected DFE
tap weights. The blue curve is the final data stream after applying the feedback stream.
Compared to the red curve, the blue curve has much more margin and is likely to be
sampled correctly.
Similarly, Figure 22 shows the DFE equalizing a data stream of 011110000. The first
‘0’ after the ‘1’s is a potential bit error. After DFE equalization, it has a large margin
and is likely to be sampled correctly.
DFE Advantages in 7 Series FPGA GTX/GTH Transceivers
One way to judge the capability of a DFE is the number of taps it supports. The DFE
compensation stops when it reaches the last tap of the feedback loop, so the more taps,
the more ISI “tail” effect can be compensated for by the DFE. It is also notable that the
reflections going through a trace with a length of more than a few inches shows up in
the pulse response after a number of taps. For example, at 10 Gb/s, reflections through
a 6-inch trace are likely to show up in the pulse response after 20 taps. However, more
DFE taps can cost more silicon area as well as more power consumption.
DFE in GTX Transceivers
The DFE capability of GTX transceivers is the best offering in the FPGA industry. GTX
transceivers support five fixed taps to cover high-frequency needs as well as short
reflections, such as those that occur within packages and connectors. Competing
devices have similar taps but have trouble delivering the full adaptation.
WP419 (v1.0) March 27, 2012
www.xilinx.com
15
Techniques Used to Combat Insertion Loss and Discontinuities
DFE in GTH Transceivers
GTH transceivers have the most DFE taps in the FPGA industry: seven fixed taps and
four sliding taps that can reach up to 63 taps. The seven fixed taps can be used for
high-frequency attenuation and short-distance reflections. The four sliding taps are
designed to compensate for reflections on channels up to 18 inches long at 10 Gb/s,
without sacrificing the power consumption of 63 fixed taps. This combination of fixed
and floating taps enables great compensatory capability while keeping the power
consumption low and costs in check.
Programmable Auto-adaptive DFE in GTX/GTH Transceivers
The tap weight selection is critical to the DFE functionality. If the tap weights are
wrong, the feedback loop can have negative impact on the data stream, which can lead
to wrong data sampling and then propagate to the following data streams. Variation of
tap weights is necessary to meet different channel characteristics. Similar to TX
emphasis, DFE tap weights in GTX/GTH transceivers are designed to be
programmable.
Choosing the right setting out of thousands of DFE tap weights combined with many
different possible CTLE settings is the major challenge for DFE designs. To make
practical use of a DFE, an adaptation algorithm is required. Both GTX and GTH
receivers support auto-adaptation when running under DFE + CTLE mode. Similar to
CTLE adaptation, users can employ a one-time DFE calibration for a given channel, or
leave auto-adaptive DFE on for constant run-time calibration.
Figure 23 illustrates DFE coverage in GTX and GTH transceivers.
X-Ref Target - Figure 23
-0.8
-0.7
-0.6
Pulse Response
-0.5
-0.4
-0.3
63 Taps
Sliding Taps
-0.2
-0.1
0
-0.1
Channel Length (UI)
-0.2
5
0
10
15
GTH Transceiver
GTX Transceiver
& Best Competition 7 Fixed + 4 Sliding Taps
Figure 23:
16
20
WP419_23_032712
DFE Taps in GTX and GTH Transceivers
www.xilinx.com
WP419 (v1.0) March 27, 2012
Techniques Used to Combat Insertion Loss and Discontinuities
Robust Performance of TX Emphasis, RX CTLE, and RX DFE
in GTX/GTH Transceivers
The DFE of GTX and GTH transceivers is designed to work together with TX
Emphasis and RX CTLE, which are well suited to mitigate some or most of the
insertion loss. The TX emphasis can boost up to 12.96 dB, and the CTLE in DFE mode
is enhanced to boost up to 20 dB between 4 GHz and 5 GHz.
DFE is the final stage to compensate for the rest of the impairments in the channel
before the incoming signal reaches Clock Data Recovery (CDR).
To avoid an “improper equalization” situation that can apply the wrong frequency
boost, amplify noise, or directly introduce bit errors, it is critical to find the right
combination of TX emphasis, RX CTLE, and RX DFE. The adaptation feature at the
receiver makes this tuning process easy. Refer to UG476, 7 Series FPGAs GTX
Transceivers User Guide for more details on how to use these equalization features.
The strategy of employing a combination of TX Emphasis, CTLE, and DFE has worked
well for backplanes, optics, plug-in cards, or even simple chip-to-chip applications.
Figure 24 shows an internal eye diagram created after the DFE, measured on a
Kintex-7 FPGA GTX receiver in a backplane system running at 10.3125 Gb/s. This eye
diagram was produced by using the on-chip Xilinx 2-D Eye-Scan feature and the
ChipScope™ toolset included in the ISE® Design Suite. The backplane system has a
total differential insertion loss of more than 28 dB at 5 GHz. The plotted eye shows the
equalization combination has well compensated the channel conditions and resulted
in plenty of margin for the backplane system.
X-Ref Target - Figure 24
120
105
90
75
BER
Amplitude Code
60
45
e-2
30
e-4
15
e-6
0
e-8
-15
e-10
-30
e-12
-45
e-14
-60
e-16
-75
-90
-105
-120
-0.5 UI
-0.375 UI
-0.25 UI
-0.125 UI
0.0 UI
Unit Interval (UI)
Figure 24:
WP419 (v1.0) March 27, 2012
0.125 UI
0.25 UI
0.375 UI
WP419_24_032712
2-D Eye Scan at the Kintex-7 FPGA Receiver through a Backplane System
www.xilinx.com
17
Summary
Summary
7 series FPGAs offer a variety of transceivers to scale and optimize across different
needs and applications. Each transceiver runs at different target line rates and has
appropriate equalization techniques for the line rates and channels expected for those
devices. The combinations of the on-chip equalization features such as TX emphasis,
RX CTLE, and RX DFE (available in the GTX and GTH transceivers) offer the best
equalization capability for even the most demanding backplane and optical
applications in the FPGA industry.
Table 2 summarizes the equalization modes in each 7 series transceiver.
Table 2:
Equalization Modes in Each 7 Series Transceiver
Transceiver
TX Emphasis
RX CTLE Only
RX CTLE + DFE
GTP
3-tap
>9 dB up to
6.6 Gb/s
GTX
3-tap
>9 dB up to
10.3125 Gb/s
Up to 20 dB CTLE and DFE (5 fixed-tap,
auto-adaptive), up to 12.5 Gb/s
GTH
3-tap
>9 dB up to
10.3125 Gb/s
Up to 20 dB CTLE and DFE (7 fixed-tap
+ 4 sliding-tap up to 63, auto-adaptive),
up to 13.1 Gb/s
GTZ
3-tap
>9dB
No
No
More information is available in the transceiver user guide for each transceiver. Also
go to: http://xilinx.com/products/technology/transceivers/index.htm.
Revision History
The following table shows the revision history for this document:
Date
Version
03/27/12
1.0
Description of Revisions
Initial Xilinx release.
Notice of Disclaimer
The information disclosed to you hereunder (the “Materials”) is provided solely for the selection and use
of Xilinx products. To the maximum extent permitted by applicable law: (1) Materials are made available
“AS IS” and with all faults, Xilinx hereby DISCLAIMS ALL WARRANTIES AND CONDITIONS,
EXPRESS, IMPLIED, OR STATUTORY, INCLUDING BUT NOT LIMITED TO WARRANTIES OF
MERCHANTABILITY, NON-INFRINGEMENT, OR FITNESS FOR ANY PARTICULAR PURPOSE; and
(2) Xilinx shall not be liable (whether in contract or tort, including negligence, or under any other theory
of liability) for any loss or damage of any kind or nature related to, arising under, or in connection with,
the Materials (including your use of the Materials), including for any direct, indirect, special, incidental,
or consequential loss or damage (including loss of data, profits, goodwill, or any type of loss or damage
suffered as a result of any action brought by a third party) even if such damage or loss was reasonably
foreseeable or Xilinx had been advised of the possibility of the same. Xilinx assumes no obligation to
correct any errors contained in the Materials or to notify you of updates to the Materials or to product
specifications. You may not reproduce, modify, distribute, or publicly display the Materials without prior
written consent. Certain products are subject to the terms and conditions of the Limited Warranties which
can be viewed at http://www.xilinx.com/warranty.htm; IP cores may be subject to warranty and support
terms contained in a license issued to you by Xilinx. Xilinx products are not designed or intended to be
fail-safe or for use in any application requiring fail-safe performance; you assume sole risk and liability
for use of Xilinx products in Critical Applications: http://www.xilinx.com/warranty.htm#critapps.
18
www.xilinx.com
WP419 (v1.0) March 27, 2012