“Xilinx” and the Xilinx logo shown above are registered trademarks of Xilinx, Inc. Any rights not expressly granted herein are reserved.
CoolRunner, RocketChips, Rocket IP, Spartan, StateBENCH, StateCAD, Virtex, XACT, XC2064, XC3090, XC4005, and XC5210 are
registered trademarks of Xilinx, Inc.
The shadow X shown above is a trademark of Xilinx, Inc.
ACE Controller, ACE Flash, A.K.A. Speed, Alliance Series, AllianceCORE, Bencher, ChipScope, Configurable Logic Cell, CORE Generator,
CoreLINX, Dual Block, EZTag, Fast CLK, Fast CONNECT, Fast FLASH, FastMap, Fast Zero Power, Foundation, Gigabit Speeds...and
Beyond!, HardWire, HDL Bencher, IRL, J Drive, JBits, LCA, LogiBLOX, Logic Cell, LogiCORE, LogicProfessor, MicroBlaze, MicroVia,
MultiLINX, NanoBlaze, PicoBlaze, PLUSASM, PowerGuide, PowerMaze, QPro, Real-PCI, RocketIO, SelectIO, SelectRAM, SelectRAM+,
Silicon Xpresso, Smartguide, Smart-IP, SmartSearch, SMARTswitch, System ACE, Testbench In A Minute, TrueMap, UIM, VectorMaze,
VersaBlock, VersaRing, Virtex-II Pro, Virtex-II EasyPath, Wave Table, WebFITTER, WebPACK, WebPOWERED, XABEL, XACTFloorplanner, XACT-Performance, XACTstep Advanced, XACTstep Foundry, XAM, XAPP, X-BLOX +, XC designated products, XChecker,
XDM, XEPLD, Xilinx Foundation Series, Xilinx XDTV, Xinfo, XSI, XtremeDSP and ZERO+ are trademarks of Xilinx, Inc.
The Programmable Logic Company is a service mark of Xilinx, Inc.
All other trademarks are the property of their respective owners.
Xilinx, Inc. does not assume any liability arising out of the application or use of any product described or shown herein; nor does it convey
any license under its patents, copyrights, or maskwork rights or any rights of others. Xilinx, Inc. reserves the right to make changes, at any
time, in order to improve reliability, function or design and to supply the best product possible. Xilinx, Inc. will not assume responsibility for
the use of any circuitry described herein other than circuitry entirely embodied in its products. Xilinx provides any design, code, or
information shown or described herein “as is.” By providing the design, code, or information as one possible implementation of a feature,
application, or standard, Xilinx makes no representation that such implementation is free from any claims of infringement. You are
responsible for obtaining any rights you may require for your implementation. Xilinx expressly disclaims any warranty whatsoever with
respect to the adequacy of any such implementation, including but not limited to any warranties or representations that the implementation
is free from claims of infringement, as well as any implied warranties of merchantability or fitness for a particular purpose. Xilinx, Inc. devices
and products are protected under U.S. Patents. Other U.S. and foreign patents pending. Xilinx, Inc. does not represent that devices shown
or products described herein are free from patent infringement or from any other third party right. Xilinx, Inc. assumes no obligation to
correct any errors contained herein or to advise any user of this text of any correction if such be made. Xilinx, Inc. will not assume any liability
for the accuracy or correctness of any engineering or software support or assistance provided to a user.
Xilinx products are not intended for use in life support appliances, devices, or systems. Use of a Xilinx product in such applications without
the written consent of the appropriate Xilinx officer is prohibited.
The contents of this manual are owned and copyrighted by Xilinx. Copyright 2001–2007 Xilinx, Inc. All Rights Reserved. Except as stated
herein, none of the material may be copied, reproduced, distributed, republished, downloaded, displayed, posted, or transmitted in any form
or by any means including, but not limited to, electronic, mechanical, photocopying, recording, or otherwise, without the prior written consent
of Xilinx. Any unauthorized use of any material contained in this manual may violate copyright laws, trademark laws, the laws of privacy and
publicity, and communications regulations and statutes.
RocketIO™ Transceiver User Guidewww.xilinx.comUG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
RocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
The following table shows the revision history for this document.
DateVersionRevision
11/20/011.0• Initial Xilinx release.
01/23/021.1• Updated for typographical and other errors found during review.
02/25/021.2• Part of Virtex-II Pro™ Developer’s Kit (March 2002 Release)
07/11/021.3• Updated “PCB Design Requirements”. Added Appendix A, “RocketIO Transceiver
Timing Model.” Changed Cell Models to Appendix B.
09/27/021.4• Added additional IMPORTANT NOTE regarding ISE revisions at the beginning of
Chapter 1
• Added material in section “CRC (Cyclic Redundancy Check).”
• Added section “Other Important Design Notes.”
• New pre-emphasis eye diagrams in section “Pre-emphasis Techniques.”
• Numerous parameter additions previously shown as “TBD” in “MGT Package Pins.”
10/16/021.5• Corrected pinouts for FF1152 package, device column 2VP20/30, LOC Constraints
rows GT_X0_Y0 and GT_X0_Y1.
• Corrected section “CRC Latency” and Ta bl e 2- 20 to express latency in terms of
TXUSRCLK and RXUSRCLK cycles.
• Corrected sequence of packet elements in Figure 2-30.
11/20/021.6• Ta bl e 1 -2 : Added support for XAUI Fibre Channel.
• Corrected max PCB drive distance to 40 inches.
• Reorganized content sequence in Chapter 2, “Digital Design Considerations.”
• Ta bl e 1- 5: Additional information in RXCOMMADET definition.
• Code corrections in VHDL Clock templates.
• “Data Path Latency” section expanded and reformatted.
• Corrections in clocking scheme drawings. Addition of drawings showing clocking
schemes without using DCM.
• Ta bl e B- 1: Corrections in Valid Data Characters.
• Ta bl e 3- 4: Data added.
• Corrections made to power regulator schematic, Figure 3-7.
• Ta bl e 2- 23 : Data added/corrected.
12/12/021.6.1• Added clarifying text regarding trace length vs. width.
03/25/032.0• Reorganized existing content
• Added new content
• Added Appendix C, “Related Online Documents”
• Added “Index”
UG024 (v3.0) February 22, 2007www.xilinx.comRocketIO™ Transceiver User Guide
Product Not Recommendedfor NewDesigns
DateVersionRevision
06/12/032.1• Ta bl e 1 -2 : Added qualifying footnote to XAUI 10GFC.
• Ta bl e 1- 5: Corrected definition of RXRECCLK.
• Section “RocketIO Transceiver Instantiations” in Chapter 1: added text briefly
explaining what the Instantiation Wizard does.
• Ta bl e 2- 14 : Changed numerics from exact values to rounded-off approximations
(nearest 5,000), and added footnote calling attention to this.
• Section “Clocking” in Chapter 2: added text recommending use of an IBUFGDS for
reference clock input to FPGA fabric.
• Section “RXRECCLK” in Chapter 2: Deleted references to SERDES_10B attribute and
to divide-by-10. (RXRECCLK is always 1/20th the data rate.).
• Section “CRC_FORMAT” in Chapter 2: Corrected minimum data length for
USER_MODE to “greater than 20”.
• Ta bl e 3- 5: Clarified the significance of the V
• Section “AC and DC Coupling” in Chapter 3: Explanatory material added regarding
V
TRX/VTTX
settings when AC or DC coupling is used.
• Ta bl e 4- 1: Corrected pinouts for FG256 and FG456.
• Ta bl e 4- 3: Corrected pinouts for FF1517 (XC2VP70).
TTX/VTRX
voltages shown in this table.
11/07/032.2• Section “Clock Signals” in Chapter 2: Added material that states:
♦ the reference clock must be provided at all times.
♦ any added jitter on the reference clock will be reflected on the RX/TX I/O.
• Figure 2-3: Added a BUFG after the IBUFGDS reference clock buffer.
• Section “RX_BUFFER_USE” in Chapter 2: Corrected erroneous “USRCLK2” to
“RXUSRCLK/RXUSRCLK2”.
• Ta bl e 2- 20 : Added footnotes qualifying the maximum receive-side latency parameters
given in the table.
• Section “FIBRE_CHAN” in Chapter 2: Added specification for minimum data length
(24 bytes not including CRC placeholder).
• Section “ETHERNET” in Chapter 2: Added note indicating that Gigabit Ethernet 802.3
frame specifications must be adhered to.
• Ta bl e 2- 23 : Corrected “External” to “Internal” loopback. Improved explanation of
Parallel Mode loopback.
• Added Figure 2-28, “Serial and Parallel Loopback Logic.”
• Section “Clock and Data Recovery” in Chapter 3: Corrected text to make clear that
RXRECCLK is always 1/20th the incoming data rate, and that CDR requires a
minimum number of transitions to achieve and maintain a lock on the received data.
• Section “Voltage Regulation” in Chapter 3: Added material defining voltage regulator
requirements when a device other than the LT1963 is used.
• Section “AC and DC Coupling” in Chapter 3: Added footnote to Ta bl e 3- 8 clarifying
V
TRX/VTTX
voltage compliance.
• Figure 3-17 and section “Epson EG-2121CA 2.5V (LVPECL Outputs)” in Chapter 3:
Added material specifying the optional use of an LVPECL buffer as an alternative to
the LVDS buffer previously specified.
• Ta bl e 4- 2: Added pinouts for FG676 package, XC2VP20 and XC2VP30.
• Ta bl e A- 5: Added BREFCLK parameters T
BREFPWH
and T
BREFPWL
• Section “Application Notes” in Appendix C: Included new Xilinx Application Notes
XAPP648, XAPP669, and XAPP670.
• Various non-technical edits and corrections.
.
RocketIO™ Transceiver User Guidewww.xilinx.comUG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
DateVersionRevision
02/24/042.3• Tabl e 2- 3, p age 41: Added FG676 row to BREFCLK Pin Numbers.
The RocketIO Transceiver User Guide provides the product designer with the detailed
technical information needed to successfully implement the RocketIO™ multi-gigabit
transceiver in Virtex-II Pro Platform FPGA designs.
RocketIO Features
The RocketIO transceiver’s flexible, programmable features allow a multi-gigabit serial
transceiver to be easily integrated into any Virtex-II Pro design:
Preface
•Variable-speed, full-duplex transceiver, allowing 600 Mb/s to 3.125 Gb/s baud
•Monolithic clock synthesis and clock recovery system, eliminating the need for
•Automatic lock-to-reference function
•Five levels of programmable serial output differential swing (800 mV to 1600 mV
•Four levels of programmable pre-emphasis
•AC and DC coupling
•Programmable 50Ω/75Ω on-chip termination, eliminating the need for external
•Serial and parallel TX-to-RX internal loopback modes for testing operability
•Programmable comma detection to allow for any protocol and detection of any 10-bit
Guide Contents
The RocketIO Transceiver User Guide contains these sections:
•Preface, “About This Guide” — This section.
•Chapter 1, “RocketIO Transceiver Overview” — An overview of the transceiver’s
•Chapter 2, “Digital Design Considerations” — Ports and attributes for the six
•Chapter 3, “Analog Design Considerations” — RocketIO serial overview; pre-
transfer rates
external components
peak-peak), allowing compatibility with other serial system voltage levels
termination resistors
character.
capabilities and how it works.
provided communications protocol primitives; VHDL/Verilog code examples for
clocking and reset schemes; transceiver instantiation; 8B/10B encoding; CRC; channel
bonding.
•Chapter 4, “Simulation and Implementation” — Simulation models; implementation
tools; debugging and diagnostics.
•Appendix A, “RocketIO Transceiver Timing Model” — Timing parameters associated
with the RocketIO transceiver core.
•Appendix B, “8B/10B Valid Characters” — Valid data and K-characters.
•Appendix C, “Related Online Documents” — Bibliography of online Application
Notes, Characterization Reports, and White Papers.
For More Information
For a complete menu of online information resources available on the Xilinx website, visit
http://www.xilinx.com/virtex2pro/
Documents.”
For a comprehensive listing of available tutorials and resources on network technologies
and communications protocols, visit http://www.iol.unh.edu/training/
Additional Resources
Preface: About This Guide
or refer to Appendix C, “Related Online
.
For additional information, go to http://support.xilinx.com. The following table lists
some of the resources you can access from this website. You can also directly access these
resources using the provided URLs.
ResourceDescription/URL
TutorialsTutorials covering Xilinx design flows, from design entry to
This document uses the following conventions. An example illustrates each typographical
and online convention.
Input and output ports of the RocketIO transceiver primitives are denoted in upper-case
letters. Attributes of the RocketIO transceiver are denoted in upper-case letters with
underscores. Trailing numbers in primitive names denote the byte width of the data path.
These values are preset and not modifiable. When assumed to be the same frequency,
RXUSRCLK and TXUSRCLK are referred to as USRCLK and can be used interchangeably.
This also holds true for RXUSRCLK2, TXUSRCLK2, and USRCLK2.
Comma Definition
A comma is a “K-character” used by the transceiver to align the serial data on a
byte/half-word boundary (depending on the protocol used), so that the serial data is
correctly decoded into parallel data.
Typographical
The following typographical conventions are used in this document:
Courier font
Courier bold
Helvetica bold
Italic font
ConventionMeaning or UseExample
Messages, prompts, and
program files that the system
displays
Literal commands that you
enter in a syntactical statement
Commands that you select
from a menu
Keyboard shortcutsCtrl+C
Variables in a syntax
statement for which you must
supply values
References to other manuals
Emphasis in text
speed grade: - 100
ngdbuilddesign_name
File → Open
ngdbuild design_name
See the Development System
Reference Guide for more
information.
If a wire is drawn so that it
overlaps the pin of a symbol,
the two nets are not connected.
An optional entry or
Square brackets [ ]
Braces { }
RocketIO™ Transceiver User Guidewww.xilinx.com19
UG024 (v3.0) February 22, 2007
parameter. However, in bus
specifications, such as
bus[7:0], they are required.
A list of items from which you
must choose one or more
ngdbuild [ option_name]
design_name
lowpwr ={on|off}
Product Not Recommendedfor NewDesigns
R
ConventionMeaning or UseExample
Preface: About This Guide
Vertical bar |
Vertical ellipsis
Horizontal ellipsis . . .
Online Document
The following conventions are used in this document:
ConventionMeaning or UseExample
Blue text
Red text
Separates items in a list of
choices
.
.
.
Repetitive material that has
been omitted
Repetitive material that has
been omitted
Cross-reference link to a
location in the current
document
Cross-reference link to a
location in another document
lowpwr ={on|off}
IOB #1: Name = QOUT’
IOB #2: Name = CLKIN’
.
.
.
allow block block_name
loc1 loc2 ... locn;
See the section “Additional
Resources” for details.
Refer to “Title Formats” in
Chapter 1 for details.
See Figure 2-5 in the Virtex-II
Handbook.
Blue, underlined text
Hyperlink to a website (URL)
Go to http://www.xilinx.com
for the latest speed files.
20www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
RocketIO Transceiver Overview
Basic Architecture and Capabilities
The RocketIO transceiver is based on Mindspeed’s SkyRail™ technology. Figure 1-1,
page 22, depicts an overall block diagram of the transceiver. Up to 20 transceiver modules
are available on a single Virtex-II Pro FPGA, depending on the part being used. Tab le 1-1
shows the RocketIO cores available by device.
Table 1-1: Number of RocketIO Cores per Device Type
DeviceRocketIO CoresDeviceRocketIO Cores
XC2VP24XC2VP400 or 12
XC2VP44XC2VP500 or 16
XC2VP78XC2VP7016 or 20
XC2VP208XC2VP1000 or 20
XC2VP308
Chapter 1
The transceiver module is designed to operate at any serial bit rate in the range of
600 Mb/s to 3.125 Gb/s per channel, including the specific bit rates used by the
communications standards listed in Tabl e 1 -2. The serial bit rate need not be configured in
the transceiver, as the operating frequency is implied by the received data, the reference
clock applied, and the SERDES_10B attribute (see Ta bl e 1 -3 ).
Table 1-2: Communications Standards Supported by RocketIO Transceiver
1. One channel is considered to be one transceiver.
2. Out-of-Band (OOB) signals are not supported with the transceiver.
3. Supported with the GT_CUSTOM primitive. Certain attributes must be modified to comply with the XAUI 10GFC specifications,
including but not limited to CLK_COR_SEQ and CHAN_BOND_SEQ.
4. Bit rate is possible with the following topology specification: maximum 6" FR4 and one Molex 74441 connector.
(2)
(3)
Channels
(Lanes)
(1)
12.5
43.1875
I/O Bit Rate
(Gb/s)
1.06
2.12
(4)
RocketIO™ Transceiver User Guidewww.xilinx.com21
UG024 (v3.0) February 22, 2007
R
PACKAGE
PINS
Product Not Recommendedfor NewDesigns
Chapter 1: RocketIO Transceiver Overview
Table 1-3: Serial Baud Rates and the SERDES_10B Attribute
Tab le 1 -4 lists the sixteen gigabit transceiver primitives provided. These primitives carry
attributes set to default values for the communications protocols listed in Ta bl e 1 -2 . Data
widths of one, two, and four bytes are selectable for each protocol.
There are two ways to modify the RocketIO transceiver:
•Static properties can be set through attributes in the HDL code. Use of attributes are
covered in detail in “Primitive Attributes,” page 29.
•Dynamic changes can be made by the ports of the primitives
Fully customizable
by user
Fibre Channel,
1-byte data path
Fibre Channel,
2-byte data path
Fibre Channel,
4-byte data path
Gigabit Ethernet,
1-byte data path
Gigabit Ethernet,
2-byte data path
Gigabit Ethernet,
4-byte data path
10-Gb Ethernet,
1-byte data path
GT_XAUI_2
GT_XAUI_4
GT_INFINIBAND_1
GT_INFINIBAND_2
GT_INFINIBAND_4
GT_AURORA_1
GT_AURORA_2
GT_AURORA_4
10-Gb Ethernet,
2-byte data path
10-Gb Ethernet,
4-byte data path
Infiniband, 1-byte
data path
Infiniband, 2-byte
data path
Infiniband, 4-byte
data path
Xilinx protocol,
1-byte data path
Xilinx protocol,
2-byte data path
Xilinx protocol,
4-byte data path
The RocketIO transceiver consists of the Physical Media Attachment (PMA) and Physical
Coding Sublayer (PCS). The PMA contains the serializer/deserializer (SERDES), TX and
RX buffers, clock generator, and clock recovery circuitry. The PCS contains the 8B/10B
encoder/decoder and the elastic buffer supporting channel bonding and clock correction.
The PCS also handles Cyclic Redundancy Check (CRC). Refer again to Figure 1-1, showing
the RocketIO transceiver top-level block diagram and FPGA interface signals.
RocketIO Transceiver Instantiations
For the different clocking schemes, several things must change, including the clock
frequency for USRCLK and USRCLK2 discussed in “Digital Clock Manager (DCM)
Examples” in Chapter 2. The data and control ports for GT_CUSTOM must also reflect this
change in data width by concatenating zeros onto inputs and wires for outputs for Verilog
designs, and by setting outputs to open and concatenating zeros on unused input bits for
VHDL designs.
HDL Code Examples
Please use the Architecture Wizard to create instantiation templates. This wizard creates
code and instantiation templates that define the attributes for a specific application.
RocketIO™ Transceiver User Guidewww.xilinx.com23
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
List of Available Ports
The RocketIO transceiver primitives contain 50 ports, with the exception of the 46-port
GT_ETHERNET and GT_FIBRE_CHAN primitives. The differential serial data ports
(RXN, RXP, TXN, and TXP) are connected directly to external pads; the remaining 46 ports
are all accessible from the FPGA logic (42 ports for GT_ETHERNET and
GT_FIBRE_CHAN).
Tab le 1 -5 contains the port descriptions of all primitives.
Chapter 1: RocketIO Transceiver Overview
Table 1-5: GT_CUSTOM
(1)
, GT_AURORA, GT_FIBRE_CHAN
(2)
, GT_ETHERNET
(2)
,
GT_INFINIBAND, and GT_XAUI Primitive Ports
PortI/O
Port
Size
Definition
BREFCLKI1This high-quality reference clock uses dedicated routing to improve
jitter for serial speeds of 2.5 Gb/s or greater. See Tab le 2 - 2, p a ge 4 0 for
usage cases.
BREFCLK2I1Alternative to BREFCLK. Can be selected by REFCLKSEL.
CHBONDDONE
(2)
O1Indicates a receiver has successfully completed channel bonding when
asserted High.
CHBONDI
(2)
I4The channel bonding control that is used only by “slaves” which is
driven by a transceiver's CHBONDO port.
CHBONDO
(2)
O4Channel bonding control that passes channel bonding and clock
correction control to other transceivers.
CONFIGENABLEI1Reconfiguration enable input (unused). Should be set to logic 0.
CONFIGINI1Data input for reconfiguring transceiver (unused). Should be set to
logic 0.
CONFIGOUTO1Data output for configuration readback (unused). Should be left
unconnected.
ENCHANSYNC
(2)
I1Comes from the core to the transceiver and enables the transceiver to
perform channel bonding
ENMCOMMAALIGNI1Selects realignment of incoming serial bitstream on minus-comma.
High realigns serial bitstream byte boundary when minus-comma is
detected.
ENPCOMMAALIGNI1Selects realignment of incoming serial bitstream on plus-comma. High
realigns serial bitstream byte boundary when plus-comma is detected.
LOOPBACKI2Selects the two loopback test modes. Bit 1 is for serial loopback and bit 0
is for internal parallel loopback.
POWERDOWNI1Shuts down both the receiver and transmitter sides of the transceiver
when asserted High. This decreases the power consumption while the
transceiver is shut down. This input is asynchronous.
24www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
List of Available Ports
R
Table 1-5: GT_CUSTOM
(1)
, GT_AURORA, GT_FIBRE_CHAN
(2)
, GT_ETHERNET
(2)
,
GT_INFINIBAND, and GT_XAUI Primitive Ports (Continued)
and multiplied for parallel/serial conversion) and clock recovery.
REFCLK frequency is accurate to ±100 ppm. When running an
asynchronous system, this accuracy must be met by both reference
clocks. This clock originates off the device, is routed through fabric
interconnect, and is selected by REFCLKSEL.
REFCLK2I1An alternative to REFCLK. Can be selected by REFCLKSEL.
REFCLKSELI1Selects the reference clock to use:
Low = selects REFCLK if REF_CLK_V_SEL = 0
selects BREFCLK if REF_CLK_V_SEL = 1
High = selects REFCLK2 if REF_CLK_V_SEL = 0
selects BREFCLK2 if REF_CLK_V_SEL = 1
See “REF_CLK_V_SEL,” page 32.
RXBUFSTATUSO2Receiver elastic buffer status. Bit 1 indicates if an overflow/underflow
error has occurred when asserted High. Bit 0 indicates that the buffer is
at least half-full when asserted High.
(3)
RXCHARISCOMMA
RXCHARISK
(3)
O1, 2, 4Similar to RXCHARISK except that the data is a comma.
O1, 2, 4If 8B/10B decoding is enabled, it indicates that the received data is a
K-character when asserted High. Included in Byte-mapping. If 8B/10B
decoding is bypassed, it remains as the first bit received (Bit “a”) of the
10-bit encoded data (see Figure 2-14, page 65).
RXCHECKINGCRCO1CRC status for the receiver. Asserts High to indicate that the receiver
has recognized the end of a data packet. Only meaningful if
RX_CRC_USE = TRUE.
RXCLKCORCNTO3Status that denotes occurrence of clock correction or channel bonding.
This status is synchronized on the incoming RXDATA. See
“RXCLKCORCNT,” page 76.
RXCOMMADETO1Signals that a comma has been detected in the data stream.
To assure signal is reliably brought out to the fabric for different data
paths, this signal may remain High for more than one
USRCLK/USRCLK2 cycle.
RXCRCERRO1Indicates if the CRC code is incorrect when asserted High. Only
meaningful if RX_CRC_USE = TRUE.
RXDATA
(3)
O8, 16, 32 Up to four bytes of decoded (8B/10B encoding) or encoded (8B/10B
bypassed) receive data.
RXDISPERR
(3)
O1, 2, 4If 8B/10B encoding is enabled it indicates whether a disparity error has
occurred on the serial line. Included in Byte-mapping scheme.
RocketIO™ Transceiver User Guidewww.xilinx.com25
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
Chapter 1: RocketIO Transceiver Overview
Table 1-5: GT_CUSTOM
(1)
, GT_AURORA, GT_FIBRE_CHAN
(2)
, GT_ETHERNET
(2)
,
GT_INFINIBAND, and GT_XAUI Primitive Ports (Continued)
PortI/O
Port
Size
Definition
RXLOSSOFSYNCO2Status related to byte-stream synchronization
(RX_LOSS_OF_SYNC_FSM)
If RX_LOSS_OF_SYNC_FSM = TRUE, RXLOSSOFSYNC indicates the
state of the FSM:
Bit 1 = Loss of sync (High)
Bit 0 = Resync state (High)
If RX_LOSS_OF_SYNC_FSM = FALSE, RXLOSSOFSYNC indicates:
Bit 1 = Received data invalid (High)
Bit 0 = Channel bonding sequence recognized (High)
(4)
RXN
RXNOTINTABLE
(3)
I1Serial differential port (FPGA external)
O1,2,4Status of encoded data when the data is not a valid character when
asserted High. Applies to the byte-mapping scheme.
RXP
(4)
I1Serial differential port (FPGA external)
RXPOLARITYI1Similar to TXPOLARITY, but for RXN and RXP. When de-asserted,
assumes regular polarity. When asserted, reverses polarity.
RXREALIGNO1Signal from the PMA denoting that the byte alignment with the serial
data stream changed due to a comma detection. Asserted High when
alignment occurs.
RXRECCLKO1Clock recovered from the data stream by dividing its speed by 20.
RXRESETI1Synchronous RX system reset that “recenters” the receive elastic buffer.
It also resets 8B/10B decoder, comma detect, channel bonding, clock
correction logic, and other internal receive registers. It does not reset the
receiver PLL.
RXRUNDISP
(3)
O1, 2, 4Signals the running disparity (0 = negative, 1 = positive) in the received
serial data. If 8B/10B encoding is bypassed, it remains as the second bit
received (Bit “b”) of the 10-bit encoded data (see Figure 2-14, page 65).
RXUSRCLKI1Clock from a DCM or a BUFG that is used for reading the RX elastic
buffer. It also clocks CHBONDI and CHBONDO in and out of the
transceiver. Typically, the same as TXUSRCLK.
RXUSRCLK2I1Clock output from a DCM that clocks the receiver data and status
between the transceiver and the FPGA core. Typically the same as
TXUSRCLK2. The relationship between RXUSRCLK and RXUSRCLK2
depends on the width of RXDATA.
TXBUFERRO1Provides status of the transmission FIFO. If asserted High, an
overflow/underflow has occurred. When this bit becomes set, it can
only be reset by asserting TXRESET.
TXBYPASS8B10B
(3)
I1, 2, 4This control signal determines whether the 8B/10B encoding is enabled
or bypassed. If the signal is asserted High, the encoding is bypassed.
This creates a 10-bit interface to the FPGA core. See the 8B/10B section
for more details.
26www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
List of Available Ports
R
Table 1-5: GT_CUSTOM
(1)
, GT_AURORA, GT_FIBRE_CHAN
(2)
, GT_ETHERNET
(2)
,
GT_INFINIBAND, and GT_XAUI Primitive Ports (Continued)
PortI/O
TXCHARDISPMODE
(3)
Port
Size
Definition
I1, 2, 4If 8B/10B encoding is enabled, this bus determines what mode of
disparity is to be sent. When 8B/10B is bypassed, this becomes the first
bit transmitted (Bit “a”) of the 10-bit encoded TXDATA bus section (see
Figure 2-13, page 65) for each byte specified by the byte-mapping.
TXCHARDISPVAL
(3)
I1,2,4If 8B/10B encoding is enabled, this bus determines what type of
disparity is to be sent. When 8B/10B is bypassed, this becomes the
second bit transmitted (Bit “b”) of the 10-bit encoded TXDATA bus
section (see Figure 2-13, page 65) for each byte specified by the bytemapping section.
TXCHARISK
(3)
I1, 2, 4If 8B/10B encoding is enabled, this control bus determines if the
transmitted data is a K-character or a Data character. A logic High
indicates a K-character.
TXDATA
(3)
I8, 16,32Transmit data that can be 1, 2, or 4 bytes wide, depending on the
primitive used. TXDATA [7:0] is always the last byte transmitted. The
position of the first byte depends on selected TX data path width.
TXFORCECRCERRI1Specifies whether to insert error in computed CRC.
When TXFORCECRCERR = TRUE, the transmitter corrupts the
correctly computed CRC value by XORing with the bits specified in
attribute TX_CRC_FORCE_VALUE. This input can be used to test
detection of CRC errors at the receiver.
TXINHIBITI1If a logic High, the TX differential pairs are forced to be a constant 1/0.
TXN = 1, TXP = 0
TXKERR
(3)
O1,2,4If 8B/10B encoding is enabled, this signal indicates (High) when the
K-character to be transmitted is not a valid K-character. Bits correspond
to the byte-mapping scheme.
TXN
TXP
(4)
(4)
O1Transmit differential port (FPGA external)
O1Transmit differential port (FPGA external)
TXPOLARITYI1Specifies whether or not to invert the final transmitter output. Able to
reverse the polarity on the TXN and TXP lines. Deasserted sets regular
polarity. Asserted reverses polarity.
TXRESETI1Synchronous TX system reset that “recenters” the transmit elastic
buffer. It also resets 8B/10B encoder and other internal transmission
registers. It does not reset the transmission PLL.
TXRUNDISP
(3)
O1, 2, 4Signals the running disparity after this byte is encoded. Low indicates
negative disparity, High indicates positive disparity.
TXUSRCLKI1Clock output from a DCM or a BUFG that is clocked with a reference
clock. This clock is used for writing the TX buffer and is frequencylocked to the reference clock.
RocketIO™ Transceiver User Guidewww.xilinx.com27
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
Chapter 1: RocketIO Transceiver Overview
Table 1-5: GT_CUSTOM
(1)
, GT_AURORA, GT_FIBRE_CHAN
(2)
, GT_ETHERNET
(2)
,
GT_INFINIBAND, and GT_XAUI Primitive Ports (Continued)
PortI/O
Port
Size
Definition
TXUSRCLK2I1Clock output from a DCM that clocks transmission data and status and
reconfiguration data between the transceiver an the FPGA core. The
ratio between TXUSRCLK and TXUSRCLK2 depends on the width of
TXDATA.
Notes:
1. The GT_CUSTOM ports are always the maximum port size.
2. GT_FIBRE_CHAN and GT_ETHERNET ports do not have the three CHBOND** or ENCHANSYNC ports.
3. The port size changes with relation to the primitive selected, and also correlates to the byte mapping.
4. External ports only accessible from package pins.
28www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
Primitive Attributes
Primitive Attributes
The primitives also contain attributes set by default to specific values controlling each
specific primitive’s protocol parameters. Included are channel-bonding settings (for
primitives supporting channel bonding), clock correction sequences, and CRC. Tab le 1 -6
shows a brief description of each attribute. Tab le 1 - 7 and Ta bl e 1 -8 have the default values
of each primitive.
Table 1-6: RocketIO Transceiver Attributes
AttributeDescription
ALIGN_COMMA_MSBTRUE/FALSE controls the alignment of detected commas within the
transceiver’s 2-byte-wide data path.
FALSE: Align commas within a 10-bit alignment range. As a result the
comma is aligned to either RXDATA[15:8} byte or RXDATA [7:0] byte in
the transceivers internal data path.
TRUE: Aligns comma with 20-bit alignment range.
As a result aligns on the RXDATA[15:8] byte.
Notes:
1. If protocols (like Gigabit Ethernet) are oriented in byte pairs with commas always in
even (first) byte formation, this can be set to TRUE. Otherwise, it should be set to
FALSE.
2. For 32-bit data path primitives, see “32-bit Alignment Design,” page 95.
3. This attribute is only modifiable in the GT_CUSTOM primitive.
R
CHAN_BOND_LIMITInteger 1-31 that defines maximum number of bytes a slave receiver can read
following a channel bonding sequence and still successfully align to that
sequence.
CHAN_BOND_MODESTRING
OFF, MASTER, SLAVE_1_HOP, SLAVE_2_HOPS
OFF: No channel bonding involving this transceiver.
MASTER: This transceiver is master for channel bonding. Its CHBONDO
port directly drives CHBONDI ports on one or more SLAVE_1_HOP
transceivers.
SLAVE_1_HOP: This transceiver is a slave for channel bonding.
SLAVE_1_HOP’s CHBONDI is directly driven by a MASTER transceiver
CHBONDO port. SLAVE_1_HOP’s CHBONDO port can directly drive
CHBONDI ports on one or more SLAVE_2_HOPS transceivers.
SLAVE_2_HOPS: This transceiver is a slave for channel bonding.
SLAVE_2_HOPS CHBONDI is directly driven by a SLAVE_1_HOP
CHBONDO port.
CHAN_BOND_OFFSETInteger 0-15 that defines offset (in bytes) from channel bonding sequence for
realignment. It specifies the first elastic buffer read address that all channelbonded transceivers have immediately after channel bonding.
CHAN_BOND_WAIT specifies the number of bytes that the master
transceiver passes to RXDATA, starting with the channel bonding sequence,
before the transceiver executes channel bonding (alignment) across all
channel-bonded transceivers.
CHAN_BOND_OFFSET specifies the first elastic buffer read address that all
channel-bonded transceivers have immediately after channel bonding
(alignment), as a positive offset from the beginning of the matched channel
bonding sequence in each transceiver.
For optimal performance of the elastic buffer, CHAN_BOND_WAIT and
CHAN_BOND_OFFSET should be set to the same value (typically 8).
CHAN_BOND_ONE_SHOTTRUE/FALSE that controls repeated execution of channel bonding.
FALSE: Master transceiver initiates channel bonding whenever possible
(whenever channel-bonding sequence is detected in the input) as long as
input ENCHANSYNC is High and RXRESET is Low.
TRUE: Master transceiver initiates channel bonding only the first time it
is possible (channel bonding sequence is detected in input) following
negated RXRESET and asserted ENCHANSYNC. After channel-bonding
alignment is done, it does not occur again until RXRESET is asserted and
negated, or until ENCHANSYNC is negated and reasserted.
Always set Slave transceivers CHAN_BOND_ONE_SHOT to FALSE.
Chapter 1: RocketIO Transceiver Overview
CHAN_BOND_SEQ_*_*11-bit vectors that define the channel bonding sequence. The usage of these
vectors also depends on CHAN_BOND_SEQ_LEN and
CHAN_BOND_SEQ_2_USE. See “Receiving Vitesse Channel Bonding
Sequence,” page 65, for format.
CHAN_BOND_SEQ_2_USEControls use of second channel bonding sequence.
FALSE: Channel bonding uses only one channel bonding sequence
defined by CHAN_BOND_SEQ_1_1...4.
TRUE: Channel bonding uses two channel bonding sequences defined
by:
CHAN_BOND_SEQ_1_1...4 and
CHAN_BOND_SEQ_2_1...4
as further constrained by CHAN_BOND_SEQ_LEN.
CHAN_BOND_SEQ_LENInteger 1-4 defines length in bytes of channel bonding sequence. This
defines the length of the sequence the transceiver matches to detect
opportunities for channel bonding.
CHAN_BOND_WAITInteger 1-15 that defines the length of wait (in bytes) after seeing channel
bonding sequence before executing channel bonding.
CLK_COR_INSERT_IDLE_FLAGTRUE/FALSE controls whether RXRUNDISP input status denotes running
disparity or inserted-idle flag.
FALSE: RXRUNDISP denotes running disparity when RXDATA is
decoded data.
TRUE: RXRUNDISP is raised for the first byte of each inserted (repeated)
clock correction (“Idle”) sequence (when RXDATA is decoded data).
CLK_COR_KEEP_IDLETRUE/FALSE controls whether or not the final byte stream must retain at
least one clock correction sequence.
FALSE: Transceiver can remove all clock correction sequences to further
recenter the elastic buffer during clock correction.
TRUE: In the final RXDATA stream, the transceiver must leave at least
one clock correction sequence per continuous stream of clock correction
sequences.
CLK_COR_REPEAT_WAITInteger 0 - 31 controls frequency of repetition of clock correction operations.
This attribute specifies the minimum number of RXUSRCLK cycles without
clock correction that must occur between successive clock corrections. If this
attribute is zero, no limit is placed on how frequently clock correction can
occur.
R
CLK_COR_SEQ_*_*11-bit vectors that define the sequence for clock correction. The attribute
used depends on the CLK_COR_SEQ_LEN and CLK_COR_SEQ_2_USE.
CLK_COR_SEQ_2_USETRUE/FALSE controls use of second clock correction sequence.
FALSE: Clock correction uses only one clock correction sequence defined
by CLK_COR_SEQ_1_1...4.
TRUE: Clock correction uses two clock correction sequences defined by:
CLK_COR_SEQ_1_1...4 and
CLK_COR_SEQ_2_1...4
as further constrained by CLK_COR_SEQ_LEN.
CLK_COR_SEQ_LENInteger that defines the length of the sequence the transceiver matches to
detect opportunities for clock correction. It also defines the size of the
correction, since the transceiver executes clock correction by repeating or
skipping entire clock correction sequences.
CLK_CORRECT_USETRUE/FALSE controls the use of clock correction logic.
FALSE: Permanently disable execution of clock correction (rate
matching). Clock RXUSRCLK must be frequency-locked with
RXRECCLK in this case.
TRUE: Enable clock correction (normal mode).
COMMA_10B_MASKThis 10-bit vector defines the mask that is ANDed with the incoming serial
bit stream before comparison against PCOMMA_10B_VALUE and
MCOMMA_10B_VALUE.
CRC_END_OF_PKT NOTE: This attribute is only valid when CRC_FORMAT = USER_MODE.
K28_0, K28_1, K28_2, K28_3, K28_4, K28_5, K28_6, K28_7, K23_7, K27_7,
K29_7, K30_7. End-of-packet (EOP) K-character for USER_MODE CRC.
Must be one of the 12 legal K-character values.
selection. Modifiable only for GT_AURORA_n, GT_XAUI_n, and
GT_CUSTOM. USER_MODE allows user definition of Start of Packet (SOP)
and End of Packet (EOP) K-characters.
CRC_START_OF_PKTNOTE: This attribute is only valid when CRC_FORMAT = USER_MODE.
K28_0, K28_1, K28_2, K28_3, K28_4, K28_5, K28_6, K28_7, K23_7, K27_7,
K29_7, K30_7. Start-of-packet (SOP) K-character for USER_MODE CRC.
Must be one of the twelve legal K-character values.
DEC_MCOMMA_DETECTTRUE/FALSE controls the raising of per-byte flag RXCHARISCOMMA on
minus-comma.
Chapter 1: RocketIO Transceiver Overview
DEC_PCOMMA_DETECTTRUE/FALSE controls the raising of per-byte flag RXCHARISCOMMA on
plus-comma.
DEC_VALID_COMMA_ONLYTRUE/FALSE controls the raising of RXCHARISCOMMA on an invalid
comma.
FALSE: Raise RXCHARISCOMMA on:
0011111xxx
(if DEC_PCOMMA_DETECT is TRUE)
and/or on:
1100000xxx
(if DEC_MCOMMA_DETECT is TRUE)
regardless of the settings of the xxx bits.
TRUE: Raise RXCHARISCOMMA only on valid characters that are in the
8B/10B translation.
MCOMMA_10B_VALUEThis 10-bit vector defines minus-comma for the purpose of raising
RXCOMMADET and realigning the serial bit stream byte boundary. This
definition does not affect 8B/10B encoding or decoding. Also see
COMMA_10B_MASK.
MCOMMA_DETECTTRUE/FALSE indicates whether to raise or not raise RXCOMMADET when
minus-comma is detected.
PCOMMA_10B_VALUEThis 10-bit vector defines plus-comma for the purpose of raising
RXCOMMADET and realigning the serial bit stream byte boundary. This
definition does not affect 8B/10B encoding or decoding. Also see
COMMA_10B_MASK.
PCOMMA_DETECTTRUE/FALSE indicates whether to raise or not raise RXCOMMADET when
plus-comma is detected.
REF_CLK_V_SEL1/0:
1: Selects BREFCLK/BREFCLK2 for 2.5 Gb/s or greater serial speeds.
0: Selects REFCLK/REFCLK2 for serial speeds under 2.5 Gb/s.
Integer (1, 2, or 4). Relates to the data width of the FPGA fabric interface.
TX_DATA_WIDTH
RX_DECODE_USEThis determines if the 8B/10B decoding is bypassed. FALSE denotes that it
is bypassed.
RX_LOS_INVALID_INCRPower of two in a range of 1 to 128 that denotes the number of valid
characters required to “cancel out” appearance of one invalid character for
loss of sync determination.
RX_LOS_THRESHOLDPower of two in a range of 4 to 512. When divided by
RX_LOS_INVALID_INCR, denotes the number of invalid characters
required to cause FSM transition to “sync lost” state.
RX_LOSS_OF_SYNC_FSM TRUE/FALSE denotes the nature of RXLOSSOFSYNC output.
TRUE: RXLOSSOFSYNC outputs the state of the FSM bits.
See “RXLOSSOFSYNC,” page 26, for details.
SERDES_10BDenotes whether the reference clock is 1/10 or 1/20 the serial bit rate.
TRUE: 1/10
range.)
(Refer to XAPP572 when considering running in this serial
FALSE: 1/20
FALSE supports a serial bitstream range of 1.0 Gb/s to 3.125 Gb/s.
TRUE supports a range of 600 Mb/s to 1.0 Gb/s.
See “Half-Rate Clocking Scheme,” page 54.
TERMINATION_IMPInteger (50 or 75). Termination impedance of either 50Ω or 75Ω. Refers to
both the RX and TX.
TX_BUFFER_USEAlways set to TRUE.
TX_CRC_FORCE_VALUE8-bit vector. Value to corrupt TX CRC computation when input
TXFORCECRCERR is High. This value is XORed with the correctly
computed CRC value, corrupting the CRC if TX_CRC_FORCE_VALUE is
nonzero. This can be used to test CRC error detection in the receiver
downstream.
TX_DIFF_CTRLAn integer value (400, 500, 600, 700, or 800) representing 400 mV, 500 mV,
600 mV, 700 mV, or 800 mV of voltage difference between the differential
lines. Twice this value is the peak-peak voltage.
TX_PREEMPHASISAn integer value (0-3) that sets the output driver pre-emphasis to improve
output waveform shaping for various load conditions. Larger value denotes
stronger pre-emphasis. See pre-emphasis values in Table 3-2, page 104.
RocketIO™ Transceiver User Guidewww.xilinx.com33
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
Chapter 1: RocketIO Transceiver Overview
Modifiable Primitives
As shown in Tab le 1 -7 and Ta bl e 1- 8, only certain attributes are modifiable for any
primitive. These attributes help to define the protocol used by the primitive. Only the
GT_CUSTOM primitive allows the user to modify all of the attributes to a protocol not
supported by another transceiver primitive. This allows for complete flexibility. The other
primitives allow modification of the analog attributes of the serial data lines and several
channel-bonding values.
Table 1-8: Default Attribute Values: GT_FIBRE_CHAN, GT_INFINIBAND,
and GT_XAUI (Continued)
R
Attribute
Default
GT_FIBRE_CHAN
CLK_COR_KEEP_IDLEFALSE
CLK_COR_REPEAT_WAIT2
(1)
(1)
Default
GT_INFINIBAND
FALSE
(1)
(1)
1
Default
GT_XAUI
FALSE
(1)
(1)
1
CLK_COR_SEQ_1_1001101111000010001110000100011100
CLK_COR_SEQ_1_2000100101010000000000000000000000
CLK_COR_SEQ_1_3000101101010000000000000000000000
CLK_COR_SEQ_1_4000101101010000000000000000000000
CLK_COR_SEQ_2_1000000000000000000000000000000000
CLK_COR_SEQ_2_2000000000000000000000000000000000
CLK_COR_SEQ_2_3000000000000000000000000000000000
CLK_COR_SEQ_2_4000000000000000000000000000000000
CLK_COR_SEQ_2_USEFALSEFALSEFALSE
CLK_COR_SEQ_LEN411
CLK_CORRECT_USETRUETRUETRUE
COMMA_10B_MASK111111100011111110001111111000
CRC_END_OF_PKT
Note (3)Note (3)
CRC_FORMATFIBRE_CHANINFINIBANDUSER_MODE
CRC_START_OF_PKT
Note (3)Note (3)
K29_7
K27_7
(1)
(1)
(1)
DEC_MCOMMA_DETECTTRUETRUETRUE
DEC_PCOMMA_DETECTTRUETRUETRUE
DEC_VALID_COMMA_ONLYTRUETRUETRUE
Lane ID(INFINBAND ONLY)NA00000000000
(1)
NA
MCOMMA_10B_VALUE110000000011000000001100000000
MCOMMA_DETECTTRUETRUETRUE
PCOMMA_10B_VALUE001111100000111110000011111000
PCOMMA_DETECTTRUETRUETRUE
REF_CLK_V_SEL000
RX_BUFFER_USETRUETRUETRUE
RX_CRC_USEFALSE
(1)
FALSE
(1)
FALSE
(1)
RX_DATA_WIDTHN
RocketIO™ Transceiver User Guidewww.xilinx.com37
UG024 (v3.0) February 22, 2007
(2)
N
(2)
N
(2)
Product Not Recommendedfor NewDesigns
R
Table 1-8: Default Attribute Values: GT_FIBRE_CHAN, GT_INFINIBAND,
and GT_XAUI (Continued)
Chapter 1: RocketIO Transceiver Overview
Attribute
Default
GT_FIBRE_CHAN
Default
GT_INFINIBAND
Default
GT_XAUI
RX_DECODE_USETRUETRUETRUE
RX_LOS_INVALID_INCR1
RX_LOS_THRESHOLD4
(1)
(1)
RX_LOSS_OF_SYNC_FSMTRUE
SERDES_10BFALSE
TERMINATION_IMP50
(1)
(1)
(1)
(1)
1
(1)
4
TRUE
FALSE
(1)
50
(1)
(1)
(1)
1
(1)
4
TRUE
FALSE
50
TX_BUFFER_USETRUETRUETRUE
TX_CRC_FORCE_VALUE11010110
TX_CRC_USEFALSE
TX_DATA_WIDTHN
TX_DIFF_CTRL500
TX_PREEMPHASIS0
Notes:
1. Modifiable attribute for specific primitives.
2. Depends on primitive used: either 1, 2, or 4.
3. CRC_EOP and CRC_SOP are not applicable for this primitive.
(1)
(2)
(1)
(1)
(1)
11010110
FALSE
500
N
(1)
(1)
(2)
(1)
(1)
0
11010110
FALSE
N
500
(1)
0
(1)
(1)
(1)
(1)
(1)
(2)
(1)
Byte Mapping
Most of the 4-bit wide status and control buses correlate to a specific byte of TXDATA or
RXDATA. This scheme is shown in Tab le 1 -9 . This creates a way to tie all the signals
together regardless of the data path width needed for the GT_CUSTOM. All other
primitives with specific data width paths and all byte-mapped ports are affected by this
situation. For example, a 1-byte wide data path has only 1-bit control and status bits
(TXKERR[0]) correlating to the data bits TXDATA[7:0]. Footnote 3 in Tab le 1 -5 shows the
ports that use byte mapping.
Table 1-9: Control/Status Bus Association to Data Bus Byte Paths
Control/Status BitData Bits
[0][7:0]
[1][15:8]
[2][23:16]
[3][31:24]
38www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
Digital Design Considerations
Clocking
Clock Signals
There are eight clock inputs into each RocketIO transceiver instantiation (Tabl e 2 -1 ).
REFCLK and BREFCLK are reference clocks generated from an external source and
presented to the FPGA as differential inputs. The reference clocks connect to the REFCLK
or BREFCLK ports of the RocketIO multi-gigabit transceiver (MGT). While only one of
these reference clocks is needed to drive the MGT, BREFCLK or BREFCLK2 must be used
for serial speeds of 2.5 Gb/s or greater. (See “BREFCLK,” page 41.)
Chapter 2
To clock the serial data, the PLL architecture for the transceiver uses the reference clock as
the interpolation source. Removing the reference clock stops the RX and TX PLLs from
working. Therefore, a reference clock must be provided at all times. This is especially
important at the end of configuration when the PMA portion of the MGT requires a
reference clock in order to properly initialize. If a reference clock is not available at this
point, the user should toggle the POWERDOWN pin when the reference clock becomes
available to ensure the PMA is properly initialized.
The reference clock also clocks a Digital Clock Manager (DCM) or a BUFG to generate all of
the other clocks for the MGT. Never run a reference clock through a DCM, since unwanted
jitter will be introduced. Any additional jitter on the reference clock will be transferred to
the transceiver’s RX and TX serial I/O.
It is recommended that all reference clock sources into the FPGA be LVDS or LVPECL
IBUFGDS. The DCI or DT attributes of LVDS are optional. Refer to the Virtex-II Pro Platform
FPGA User Guide (Chapter 3, “Design Considerations”) for a complete listing and
discussion of IBUFGDS and other available I/O primitives. Also see section “Reference
Clock” in Chapter 3 of this Guide.
Typically, TXUSRCLK = RXUSRCLK and TXUSRCLK2 = RXUSRCLK2. The transceiver
uses one or two clocks generated by the DCM. As an example, USRCLK and USRCLK2
clocks run at the same speed if the 2-byte data path is used. The USRCLK must always be
frequency-locked to the reference clock of the RocketIO transceiver when SERDES_10B =
FALSE (full-rate operation).
Note:
rate operation) with a duty cycle between 45% and 55%, and should have a frequency stability of
The reference clock must be at least 50 MHz (for full-rate operation only; 60 MHz for half-
RocketIO™ Transceiver User Guidewww.xilinx.com39
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
±100 ppm or better, with jitter as low as possible. Module 3 of the Virtex-II Pro data sheet gives
further details.
Table 2-1: Clock Ports
ClockI/OsDescription
BREFCLKInputReference clock used to read the TX FIFO and multiplied by
20 for parallel-to-serial conversion (20X)
BREFCLK2InputAlternative to BREFCLK
RXRECCLKOutputRecovered clock (from serial data stream) divided by 20.
Clocks data into the elastic buffer.
REFCLKInputReference clock used to read the TX FIFO and multiplied by
20 for parallel-to-serial conversion (20X)
REFCLK2InputAlternative to REFCLK.
REFCLKSELInputSelects which reference clock is used. 0 selects REFCLK;
1 selects REFCLK2.
RXUSRCLKInputClock from FPGA used for reading the RX Elastic Buffer.
Clock signals CHBONDI and CHBONDO into and out of the
transceiver. This clock is typically the same as TXUSRCLK.
TXUSRCLK
(1)
InputClock from FPGA used for writing the TX Buffer. This clock
must be frequency locked to REFCLK for proper operation.
Chapter 2: Digital Design Considerations
RXUSRCLK2InputClock from FPGA used to clock RX data and status between
the transceiver and FPGA fabric. The relationship between
RXUSRCLK2 and RXUSRCLK depends on the width of the
receiver data path. RXUSRCLK2 is typically the same as
TXUSRCLK2.
(1)
TXUSRCLK2
InputClock from FPGA used to clock TX data and status between
the transceiver and FPGA fabric. The relationship between
TXUSRCLK2 and TXUSRCLK depends on the width of the
transmission data path.
Notes:
1. TXUSRCLK and TXUSRCLK2 must be driven by clock sources, even if only the receiver of the MGT is
being used.
Table 2-2: Reference Clock Usage
Data RateRouting
600 Mb/s –
2.499 Gb/s
REFCLK√√
BREFCLK√√
2.500 Gb/s –
3.125 Gb/s
Can Route
Across Chip?
(2)
Note (1)Note (1)
Can Route
Through BUFG?
(2)
√
Notes:
1. Because of dedicated routing to reduce jitter, BREFCLK cannot be routed through the fabric.
2. While this option is available in the silicon, this topography adds extra jitter to the reference clock
which can affect the overall performance of the transceiver.
40www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Clocking
Product Not Recommendedfor NewDesigns
R
BREFCLK
At speeds of 2.5 Gb/s or greater, REFCLK configuration introduces more than the
maximum allowable jitter to the RocketIO transceiver. For these higher speeds, BREFCLK
configuration is required. The BREFCLK configuration uses dedicated routing resources
that reduce jitter.
BREFCLK must enter the FPGA through dedicated clock I/O. BREFCLK can connect to the
BREFCLK inputs of the transceiver and the CLKIN input of the DCM for creation of
USRCLKs. If all the transceivers on a Virtex-II Pro FPGA are to be used, two BREFCLKs
must be created, one for the top of the chip and one for the bottom. These dedicated clocks
use the same clock inputs for all packages:
BREFCLK
PGCLK4S
BREFCLK
PGCLK6P
NGCLK5PNGCLK7S
Top
PGCLK2S
BREFCLK2
Bottom
PGCLK0P
BREFCLK2
NGCLK3PNGCLK1S
An attribute (REF_CLK_V_SEL) and a port (REFCLKSEL) determine which reference clock
is used for the MGT PMA block. Figure 2-1 shows how REFCLK and BREFCLK are
selected through use of REFCLKSEL and REF_CLK_V_SEL.
refclk
REF_CLK_V_SEL
0
1.5V
refclk2
1
0
REFCLKSELrefclk_out
to PCS and PMA
brefclk
1
0
2.5V
brefclk2
1
ug024_35_091802
Figure 2-1: REFCLK/BREFCLK Selection Logic
Tab l e 2 - 3 shows the BREFCLK pin numbers for all packages. Note that these pads must be
used for BREFCLK operations.
Table 2-3: BREFCLK Pin Numbers
TopBottom
Package
BREFCLK
Pin Number
BREFCLK2
Pin Number
BREFCLK
Pin Number
BREFCLK2
Pin Number
FG256A8/B8B9/A9R8/T8T9/R9
FG456C11/D11D12/C12W11/Y11Y12/W12
FG676B13/C13C14/B14AD13/AE13AE14/AD14
RocketIO™ Transceiver User Guidewww.xilinx.com41
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
Table 2-3: BREFCLK Pin Numbers
Clock Ratio
USRCLK2 clocks the data buffers. The ability to send/receive parallel data to/from the
transceiver at three different widths requires the user to change the frequency of
USRCLK2. This creates a frequency ratio between USRCLK and USRCLK2. The falling
edges of the clocks must align. Ta bl e 2 -4 shows the ratios for each of the three data widths.
Chapter 2: Digital Design Considerations
TopBottom
Package
BREFCLK
Pin Number
BREFCLK2
Pin Number
BREFCLK
Pin Number
BREFCLK2
Pin Number
FF672B14/C14C13/B13AD14/AE14AE13/AD13
FF896F16/G16G15/F15AH16/AJ16AJ15/AH15
FF1152H18/J18J17/H17AK18/AL18AL17/AK17
FF1148N/AN/AN/AN/A
FF1517E20/D20J20/K20AR20/AT20AL20/AK20
FF1704G22/F22F21/G21AU22/AT22AT21/AU21
FF1696N/AN/AN/AN/A
Table 2-4: Data Width Clock Ratios
Data WidthFrequency Ratio of USRCLK\USRCLK2
1 byte1:2
2 byte1:1
4 byte2:1
Notes:
1. Each edge of the slower clock must align with the falling edge of the faster clock.
Digital Clock Manager (DCM) Examples
With at least three different clocking schemes possible on the transceiver, a DCM is the best
way to create these schemes.
Tab le 2 -5 shows typical DCM connections for several transceiver clocks. REFCLK is the
input reference clock for the DCM. The other clocks are generated by the DCM. The DCM
establishes a desired phase relationship between TXUSRCLK, TXUSRCLK2, etc. in the
FPGA core and REFCLK at the pad.
NOTE: The reference clock may be any of the four MGT clocks, including the BREFCLKs.
(1)
(1)
42www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
Clocking
Table 2-5: DCM Outputs for Different DATA_WIDTHs
R
SERDES_10B
TX_DATA_WIDTH
RX_DATA_WIDTH
REFCLK
TXUSRCLK
RXUSRCLK
TXUSRCLK2
RXUSRCLK2
FALSE1CLKINCLK0CLK2X180
FALSE2CLKINCLK0CLK0
FALSE4CLKINCLK180
(1)
TRUE1CLKINCLKDV (divide by 2)CLK180
CLKDV (divide by 2)
(1)
TRUE2CLKINCLKDV (divide by 2)CLKDV (divide by 2)
TRUE4CLKINCLKFX180 (divide by 2)CLKDV (divide by 4)
Notes:
1. Since CLK0 is needed for feedback, it can be used instead of CLK180 to clock USRCLK or USRCLK2 of the transceiver with the use
of the transceiver’s local inverter, saving a global buffer (BUFG).
Example 1a: Two-Byte Clock with DCM
The following HDL codes are examples of a simple clock scheme using 2-byte data with
both USRCLK and USRCLK2 at the same frequency. USRCLK_M is the input for both
USRCLK and USRCLK2.
--------------------------------------------------------------------library IEEE;
use IEEE.std_logic_1164.all;
--
-- pragma translate_off
library UNISIM;
use UNISIM.VCOMPONENTS.ALL;
-- pragma translate_on
-entity TWO_BYTE_CLK is
port (
REFCLKIN : in std_logic;
RocketIO™ Transceiver User Guidewww.xilinx.com43
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
RST : in std_logic;
USRCLK_M: buffer std_logic;
REFCLK: buffer std_logic;
LOCK : out std_logic
);
end TWO_BYTE_CLK;
-architecture TWO_BYTE_CLK_arch of TWO_BYTE_CLK is
--
-- Components Declarations:
component BUFG
port (
I: in std_logic;
O : out std_logic
);
end component;
-component IBUFG
port (
I : in std_logic;
O : out std_logic
);
end component;
-component DCM
port (
CLKIN: in std_logic;
CLKFB : in std_logic;
DSSEN : in std_logic;
PSINCDEC : in std_logic;
PSEN : in std_logic;
PSCLK : in std_logic;
RST : in std_logic;
CLK0 : out std_logic;
CLK90 : out std_logic;
CLK180 : out std_logic;
CLK270 : out std_logic;
CLK2X : out std_logic;
CLK2X180 : out std_logic;
CLKDV : out std_logic;
CLKFX : out std_logic;
CLKFX180 : out std_logic;
LOCKED : out std_logic;
PSDONE : out std_logic;
STATUS : out std_logic_vector ( 7 downto 0 )
);
end component;
--
-- Signal Declarations:
-signal GND : std_logic;
signal CLK0_W: std_logic;
If TXDATA and RXDATA are not clocked off the FPGA using the respective USRCLK2s,
then the DCM may be removed from the two-byte clocking scheme, as shown in
Figure 2-3:
MGT for 2-Byte Data Path (no DCM)
GT_std_2
REFCLKSEL
0
REFCLK_P
IBUFGDS BUFG
REFCLK_N
Note: Implementation tools automatically
instantiate the BUFG. There is no need to
explicitly instantiate in HDL code.
REFCLK
TXUSRCLK2
RXUSRCLK2
TXUSRCLK
RXUSRCLK
ug024_02b_062404
Figure 2-3: Two-Byte Clock without DCM
Example 2: Four-Byte Clock
If a 4-byte or 1-byte data path is chosen, the ratio between USRCLK and USRCLK2
changes. The time it take for the SERDES to serialize the parallel data requires the change
in ratios.
The DCM example (Figure 2-4) is detailed for a 4-byte data path. If 3.125 Gb/s is required,
REFCLK is 156 MHz and USRCLK2_M runs at only 78 MHz, including the clocking for
46www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Clocking
Product Not Recommendedfor NewDesigns
R
any interface logic. Both USRCLK and USRCLK2 are aligned on the falling edge, since
USRCLK_M is 180° out of phase when using local inverters with the transceiver.
Note:
in the FOUR_BYTE_CLK templates.
Clocks for 4-Byte Data Path
VHDL Template
-- Module:FOUR_BYTE_CLK
-- Description:VHDL submodule
-- DCM for 4-byte GT
--
-- Device:Virtex-II Pro Family
--------------------------------------------------------------------library IEEE;
use IEEE.std_logic_1164.all;
--
-- pragma translate_off
library UNISIM;
use UNISIM.VCOMPONENTS.ALL;
-- pragma translate_on
-entity FOUR_BYTE_CLK is
port (
end FOUR_BYTE_CLK;
-architecture FOUR_BYTE_CLK_arch of FOUR_BYTE_CLK is
--
-- Components Declarations:
component BUFG
port (
end component;
-component IBUFG
port (
These local MGT clock input inverters, shown and noted in Figure 2-4, are not included
REFCLKIN: in std_logic;
RST: in std_logic;
USRCLK_M: out std_logic;
USRCLK2_M: out std_logic;
REFCLK: out std_logic;
LOCK : out std_logic
);
I : in std_logic;
O : out std_logic
);
RocketIO™ Transceiver User Guidewww.xilinx.com47
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
I : in std_logic;
O : out std_logic
);
end component;
-component DCM
port (
CLKIN : in std_logic;
CLKFB : in std_logic;
DSSEN : in std_logic;
PSINCDEC : in std_logic;
PSEN : in std_logic;
PSCLK : in std_logic;
RST: in std_logic;
CLK0 : out std_logic;
CLK90 : out std_logic;
CLK180 : out std_logic;
CLK270 : out std_logic;
CLK2X : out std_logic;
CLK2X180 : out std_logic;
CLKDV : out std_logic;
CLKFX : out std_logic;
CLKFX180 : out std_logic;
LOCKED : out std_logic;
PSDONE : out std_logic;
STATUS : out std_logic_vector ( 7 downto 0 )
);
end component;
--
-- Signal Declarations:
-signal GND : std_logic;
signal CLK0_W: std_logic;
signal CLKDV_W: std_logic;
signal USRCLK2_M_W: std_logic;
This is the 1-byte data path width clocking scheme example. USRCLK2_M is twice as fast
as USRCLK_M. It is also phase-shifted 180° for falling edge alignment.
--------------------------------------------------------------------library IEEE;
use IEEE.std_logic_1164.all;
50www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Clocking
Product Not Recommendedfor NewDesigns
R
--
-- pragma translate_off
library UNISIM;
use UNISIM.VCOMPONENTS.ALL;
-- pragma translate_on
-entity ONE_BYTE_CLK is
port (
REFCLKIN : in std_logic;
RST : in std_logic;
USRCLK_M: out std_logic;
USRCLK2_M: out std_logic;
REFCLK: out std_logic;
LOCK : out std_logic
);
end ONE_BYTE_CLK;
-architecture ONE_BYTE_CLK_arch of ONE_BYTE_CLK is
--
-- Components Declarations:
component BUFG
port (
I: in std_logic;
O: out std_logic
);
end component;
-component IBUFG
port (
I: in std_logic;
O: out std_logic
);
end component;
-component DCM
port (
CLKIN: in std_logic;
CLKFB : in std_logic;
DSSEN : in std_logic;
PSINCDEC : in std_logic;
PSEN : in std_logic;
PSCLK : in std_logic;
RST : in std_logic;
CLK0 : out std_logic;
CLK90 : out std_logic;
CLK180 : out std_logic;
CLK270 : out std_logic;
CLK2X : out std_logic;
CLK2X180 : out std_logic;
CLKDV : out std_logic;
CLKFX : out std_logic;
CLKFX180 : out std_logic;
LOCKED : out std_logic;
PSDONE: out std_logic;
STATUS : out std_logic_vector ( 7 downto 0 )
);
end component;
--
-- Signal Declarations:
RocketIO™ Transceiver User Guidewww.xilinx.com51
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
-signal GND : std_logic;
signal CLK0_W: std_logic;
signal CLK2X180_W : std_logic;
signal USRCLK2_M_W : std_logic;
signal USRCLK_M_W : std_logic;
Some applications require serial speeds between 600 Mb/s and 1 Gb/s. (Refer to XAPP572
when considering running in this serial range.) The transceiver attribute SERDES_10B,
which sets the REFCLK multiplier to 10 instead of 20, enables the half-rate speed range
when set to TRUE. With this configuration, the clocking scheme also changes. The figures
below illustrate the three clocking scheme waveforms when SERDES_10B = TRUE.
Chapter 2: Digital Design Considerations
Clocks for 1-Byte Data Path
(SERDES_10B = TRUE)
REFCLK
TXUSRCLK
RXUSRCLK
TXUSRCLK2
RXUSRCLK2
Figure 2-6: One-Byte Data Path Clocks, SERDES_10B = TRUE
Figure 2-8: Four-Byte Data Path Clocks, SERDES_10B = TRUE
54www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Clocking
Product Not Recommendedfor NewDesigns
Multiplexed Clocking Scheme with DCM
Following configuration of the FPGA, some applications might need to change the
frequency of its REFCLK depending on the protocol used. Figure 2-9 shows how the
design can use two different reference clocks connected to two different DCMs. The clocks
are then multiplexed before input into the RocketIO transceiver.
User logic can be designed to determine during auto negotiation if the reference clock used
for the transceiver is incorrect. If so, the transceiver must then be reset and another
reference clock selected.
Use of 2 DCMs is required
to maintain correct
IBUFG/DCM/BUFGMUX topology
for clock skew compensation
UG024_05a_112202
Figure 2-9: Multiplexed REFCLK with DCM
Multiplexed Clocking Scheme without DCM
As with “Example 1b: Two-Byte Clock without DCM”, the DCMs shown in Figure 2-9 may
be removed if TXDATA and RXDATA are not clocked off the FPGA. (See Figure 2-10.)
However, the transceiver must still be reset when clocks are switched.
RXRECCLK is a recovered clock derived by dividing by 20 the received data stream bit rate
(whether full-rate or half-rate). If clock correction is bypassed, it is not possible to
compensate for differences in the clock embedded in the received data and the REFCLKcreated USRCLKs. In this case, RXRECCLK is used to generate the RXUSRCLKs, as shown
in Figure 2-11.
RXRECCLK changes monotonically when it changes from being locked to the reference
clock to being locked to data and vice versa. The recovered bit clock jumps by a maximum
of 1/16th of a bit period every eight RXRECCLK cycles (20 ps for a data rate of 3.125 Gb/s
with a 320-ps bit period) in the interpolator. RXRECCLK is derived from this bit clock
through a divide-by-20 process. When the data input is kept static, however, the recovered
clock does not frequency-lock to the reference clock exactly, but can deviate from it by up
to 400 ppm.
Chapter 2: Digital Design Considerations
Figure 2-11: Using RXRECCLK to Generate RXUSRCLK and RXUSRCLK2
Note:
and routing to global clock resources is uncertain and may cause unreliable performance.
Bypassing the RX elastic buffer is not recommended, as the skew created by the DCM
Clock Dependency
All signals used by the FPGA fabric to interact between user logic and the transceiver
depend on an edge of USRCLK2. These signals all have setup and hold times with respect
to this clock. For specific timing values, see Module 3 of the Virtex-II Pro data sheet. The
timing relationships are further discussed and illustrated in Appendix A, “RocketIO
Transceiver Timing Model.”
REFCLK_P
REFCLK_N
IBUFGDS
CLKIN
CLKFB
RST
DCM
CLK0
BUFG
BUFG
0
REFCLKSEL
REFCLK
TXUSRCLK
TXUSRCLK2
RXUSRCLK
RXUSRCLK2
RXRECCLK
UG024_38_112202
56www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
Reset/Power Down
Data Path Latency
With the many configurations of the MGT, the both transmit and receive data path
latencies vary. Below are several tables that provide approximate latencies for common
configurations.
Table 2-6: Latency through Various Transmitter Components/Processes
Component/ProcessLatency
R
1 Byte Data Path:
TX Fabric/GT Interface
included7 TXUSRCLK cycles
TX CRC
bypassed1 TXUSRCLK cycle
included1 TXUSRCLK cycle
8B/10B Encoder
bypassed1 TXUSRCLK cycle
TX FIFO4 TXUSRCLK cycles (
TX SERDES
Table 2-7: Latency through Various Receiver Components/Processes
Component/ProcessLatency
RX SERDES1.5 recovered clock (RXRECCLK) cycles
Comma Detect/Realignment
included1 recovered clock cycle
8B/10B Decoder
bypassed1 recovered clock cycle
2.5 TXUSRCLK2 cycles
1.25 TXUSRCLK cycles
SERDES_10B = FALSE:
1.5 TXUSRCLK cycles
2.5 or 3.5 recovered clock cycles
(some bits bypass one register, depending on comma alignment)
2 Byte Data Path:
1 TXUSRCLK2 cycle
1 TXUSRCLK cycle
±0.5)
SERDES_10B = TRUE:
0.5 TXUSRCLK cycles (approx.)
4 Byte Data Path:
1.25 TXUSRCLK2 cycles
2.5 TXUSRCLK cycles
RX FIFO18 RXUSRCLK cycles (
1 Byte Data Path:
RX GT/Fabric Interface
2.5 RXUSRCLK2 cycles
1.25 RXUSRCLK cycles
±0.5)
2 Byte Data Path:
1 RXUSRCLK2 cycle
1 RXUSRCLK cycle
4 Byte Data Path:
1.25 RXUSRCLK2 cycles
2.5 RXUSRCLK cycles
Reset/Power Down
The receiver and transmitter have their own synchronous reset inputs. The transmitter
reset recenters the transmission FIFO, and resets all transmitter registers and the 8B/10B
encoder. The receiver reset recenters the receiver elastic buffer, and resets all receiver
registers and the 8B/10B decoder. Neither reset signal has any effect on the PLLs.
After the DCM-locked signal is asserted, the resets can be asserted. The resets must be
asserted for two USRCLK2 cycles to ensure correct initialization of the FIFOs. Although
both the transmit and receive resets can be attached to the same signal, separate signals are
preferred. This allows the elastic buffer to be cleared in case of an over/underflow without
affecting the ongoing TX transmission. The following example is an implementation that
resets all three data-width transceivers.
RocketIO™ Transceiver User Guidewww.xilinx.com57
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
Additional reset and power control descriptions are given in Tab le 2 -8 and Tab le 2- 9.
Table 2-8: Reset and Power Control Descriptions
PortsDescription
RXRESETSynchronous receive system reset recenters the receiver elastic buffer, and resets
the 8B/10B decoder, comma detect, channel bonding, clock correction logic,
and other receiver registers. The PLL is unaffected.
TXRESETSynchronous transmit system reset recenters the transmission FIFO, and resets
the 8B/10B encoder and other transmission registers. The PLL is unaffected.
POWERDOWNShuts down the transceiver (both RX and TX sides).
In POWERDOWN mode, transmit output pins TXP/TXN are not driven, but
biased by the state of transmit termination supply VTTX. If VTTX is not
powered, TXP/TXN float to a high-impedance state. Receive input pins
RXP/RXN respond similarly to the state of receive termination supply VTRX.
Table 2-9: Power Control Descriptions
POWERDOWNTransceiver Status
0Transceiver in operation
1Transceiver temporarily powered down
Chapter 2: Digital Design Considerations
Notes:
1. Unused transceivers are automatically configured as powered-down by the implementation tools.
VHDL Template
-- Module: gt_reset
-- Description: VHDL submodule
-- reset for GT
--
-- Device: Virtex-II Pro Family
--------------------------------------------------------------------LIBRARY IEEE;
USE IEEE.std_logic_1164.all;
use IEEE.STD_LOGIC_ARITH.ALL;
use IEEE.Numeric_STD.all;
use IEEE.STD_LOGIC_UNSIGNED.ALL;
--
-- pragma translate_off
library UNISIM;
use UNISIM.VCOMPONENTS.ALL;
-- pragma translate_on
-entity gt_reset is
port (
USRCLK2_M : in std_logic;
LOCK : in std_logic;
REFCLK: out std_logic;
DCM_LOCKED: in std_logic;
RST: out std_logic);
end gt_reset;
-architecture RTL of gt_reset is
- signal startup_count : std_logic_vector (7 downto 0);
begin
58www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Reset/Power Down
Product Not Recommendedfor NewDesigns
R
process (USRCLK2_M, DCM_LOCKED)
begin
if (USRCLK2_M' event and USRCLK2_M = '1') then
if(DCM_LOCKED = '0') then
startup_count <= "00000000";
elsif (DCM_LOCKED = '1') then
startup_count <= startup_count + "00000001";
end if;
end if;
if (USRCLK2_M' event and USRCLK2_M = '1') then
if(DCM_LOCKED = '0') then
RST <= '1';
elsif (startup_count = "00000010") then
RST <= '0';
end if;
end if;
end process;
end RTL;
Verilog Template
// Module:gt_reset
// Description: Verilog Submodule
//reset for4-byte GT
//
// Device: Virtex-II Pro Family
always @ ( posedge USRCLK2_M or negedge DCM_LOCKED )
if ( !DCM_LOCKED )
RST <= 1'b1;
else
RST <= ( startup_counter != 8'h02 );
endmodule
RocketIO™ Transceiver User Guidewww.xilinx.com59
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
8B/10B Encoding/Decoding
Overview
The RocketIO transceiver has the ability to encode eight bits into a 10-bit serial stream
using standard 8B/10B encoding. This guarantees a DC-balanced, edge-rich serial stream,
facilitating DC- or AC-coupling and clock recovery. Table 2-10, page 62, shows the
significance of 8B/10B ports that change purpose, depending on whether 8B/10B is
bypassed or enabled.
8B/10B Encoder
A bypassable 8B/10B encoder is included in the transmitter. The encoder uses the same
256 data characters and 12 control characters (shown in Appendix B, “8B/10B Valid
Characters”) that are used for Gigabit Ethernet, XAUI, Fibre Channel, and InfiniBand.
The encoder accepts 8 bits of data along with a K-character signal for a total of 9 bits per
character applied. If the K-character signal is High, the data is encoded into one of the
twelve possible K-characters available in the 8B/10B code. (See Table B-2, page 143.) If the
K-character input is Low, the 8 bits are encoded as standard data. If the K-character input
is High and a user applies other than one of the twelve possible combinations, TXKERR
indicates the error.
Chapter 2: Digital Design Considerations
8B/10B Decoder
An optional 8B/10B decoder is included in the receiver. A programmable option allows the
decoder to be bypassed. When it is bypassed, the 10-bit character order is as shown in
Figure 2-14, page 65. The decoder uses the same table that is used for Gigabit Ethernet,
Fibre Channel, and InfiniBand.
The decoder separately detects both “disparity errors” and “out-of-band” errors. A
disparity error occurs when a 10-bit character is received that exists within the 8B/10B table
(Table B-1, page 135), but has an incorrect disparity. An out-of-band error occurs when a 10-
bit character is received that does not exist within the 8B/10B table. It is possible to obtain
an out-of-band error without having a disparity error. The proper disparity is always
computed for both legal and illegal characters. The current running disparity is available at
the RXRUNDISP signal.
The 8B/10B decoder performs a unique operation if out-of-band data is detected. Should
this occur, the decoder signals the error, passes the illegal 10 bits through, and places them
on the outputs. This can be used for debugging purposes if desired.
The decoder also signals reception of one of the twelve valid K-characters (Ta bl e B-2 ,
page 143) by way of the RXCHARISK port.
In addition, a programmable comma detect is included. The comma detect signal
RXCOMMADET registers a comma on the receipt of any plus-comma, minus-comma, or
both. Since the comma is defined as a 7-bit character, this includes several out-of-band
characters. RXCHARISCOMMA allows the decoder to detect only the three defined
commas (K28.1, K28.5, and K28.7) as plus-comma, minus-comma, or both. In total, there
are six possible options, three for valid commas and three for “any comma.”
Note that all bytes (1, 2, or 4) at the RX FPGA interface each have their own individual
8B/10B indicators (K-character, disparity error, out-of-band error, current running
disparity, and comma detect).
60www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
8B/10B Encoding/Decoding
Ports and Attributes
TXBYPASS8B10B,
RX_DECODE_USE
One port and one attribute enable 8B/10B encoding/decoding in the transceiver.
TXBYPASS8B10B is a byte-mapped port that is 1, 2, or 4 bits wide, depending on the data
width of the transceiver primitive being used. These bits correlate to each byte of the data
path. To enable 8B/10B encoding in the transmitter, these bits must be set Low. In this
mode, the transmit data input to the TXDATA port is non-encoded data of either 8, 16, or
32 bits wide. However, if other encoding schemes are preferred, the encoder capabilities
can be bypassed by setting all bits High. In this mode, the data input to TXDATA is either
10, 20, or 40 bits wide. The extra bits are fed through the TXCHARDISPMODE and
TXCHARDISPVAL buses (shown in Tab le 2 -1 0).
The decoder is controlled by the attribute RX_DECODE_USE. When this attribute is set to
TRUE, the decoder is enabled and should coincide with TXBYPASS8B10B being set Low. In
this mode, the received data output from the RXDATA port is decoded data, either 8, 16, or
32 bits wide. However, when the attribute is set to FALSE, the decoder is disabled. In this
mode, the received data is 10, 20, or 40 bits wide, and the extra bits are provided by
RXCHARISK and RXRUNDISP (shown in Tab le 2 -10 ).
R
If this pair is not matched, the data is not received correctly. Figure 2-12 shows the
encoding/decoding blocks of the transceiver and how the data passes through these
blocks. Ta bl e 2 -1 0 shows the significance of 8B/10B ports that change purpose depending
on whether 8B/10B is bypassed or enabled.
Tr ansceiver Module
32/16/8 bits
TXDATA
50 – 156.3 MHz
32/16/8bits
RXDATA
Physical Coding SublayerPhysical Media Attachment
C
R
C
REFCLK
Channel Bonding
Clock Correction
CRC
Elastic
Buffer
8B/10B
Encode
and
8B/10B
Decode
Comma Detect
F
I
F
O
Loop-back (parallel)
Mindspeed IP
Serializer
TX Clock Generator
Tr ansmitter
20X Multiplier
Receiver
RX Clock Recovery
Deserializer
Tr ansmit
Buffer
Receive
Buffer
TX+
TX−
Loop-back
RX+
RX−
UG024_09_020507
Figure 2-12: 8B/10B Data Flow
RocketIO™ Transceiver User Guidewww.xilinx.com61
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
Table 2-10: 8B/10B Bypassed Signal Significance
Chapter 2: Digital Design Considerations
Function
TXBYPASS8B10B
TXCHARDISPMODE,
TXCHARDISPVAL
RXCHARISKReceived byte is a K-characterPart of 10-bit encoded byte
RXRUNDISP
8B/10B encoding is enabled (not bypassed). 1, 2, or 4 bits, mapped to number of
0
bytes of data path width.
8B/10B encoding bypassed (disabled). 1, 2, or 4 bits, mapped to number of bytes
1
of data path width.
Function, 8B/10B EnabledFunction, 8B/10B Bypassed
00Maintain running disparity normally
Invert the normally generated running
01
disparity before encoding this byte.
Set negative running disparity before
10
encoding this byte.
Set positive running disparity before
11
encoding this byte.
Indicates running disparity is
0
NEGATIVE
Indicates running disparity is
1
POSITIVE
Part of 10-bit encoded byte
(see Figure 2-13):
TXCHARDISPMODE[0]
(or: [1] / [2] / [3])
TXCHARDISPVAL[0]
(or: [1] / [2] / [3])
TXDATA[7:0]
(or: [15:8] / [23:16] / [31:24])
(see Figure 2-14):
RXCHARISK[0]
(or: [1] / [2] / [3])
RXRUNDISP[0]
(or: [1] / [2] / [3])
RXDATA[7:0]
(or: [15:8] / [23:16] / [31:24])
RXDISPERRDisparity error occurred on current
byte
TXCHARISKTransmitted byte is a K-characterUnused
RXCHARISCOMMAReceived byte is a commaUnused
Unused
TXCHARDISPVAL,
TXCHARDISPMODE
TXCHARDISPVAL and TXCHARDISPMODE are dual-purpose ports for the transmitter
depending upon whether 8B/10B encoding is enabled. Tab le 2 -1 0 shows this dual
functionality. When encoding is enabled, these ports function as byte-mapped control
ports controlling the running disparity of the transmitted serial data.
In the encoding configuration, the disparity of the serial transmission can be controlled
with the TXCHARDISPVAL and TXCHARDISPMODE ports. When TXCHARDISPMODE
is set High, the running disparity is set before encoding the specific byte.
TXCHARDISPVAL determines if the disparity is negative (set Low) or positive (set High).
Tab le 2 -11 illustrates this.
62www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
8B/10B Encoding/Decoding
Table 2-11: Running Disparity Control
R
{TXCHARDISPMODE,
TXCHARDISPVAL}
00Maintain running disparity normally
01
10Set negative running disparity before encoding this byte
11Set positive running disparity before encoding this byte
When TXCHARDISPMODE is set Low, the running disparity is maintained if
TXCHARDISPVAL is also set Low, but the disparity is inverted before encoding the byte
when TXCAHRDISPVAL is set High.
Most applications will use the mode where both TXCHARDISPMODE and
TXCHARDISPVAL are set Low. Some applications may use other settings if special
running disparity configurations are required, such as in the “Vitesse Disparity Example”
below.
In the bypassed configuration, TXCHARDISPMODE [0] becomes bit 9 of the 10 bits of
encoded data. TXCHARDISPMODE [1:3] are bits 19, 29, and 39 in the 20- and 40-bit wide
buses. TXCHARDISPVAL becomes bits 8, 18, 28, and 38 of the transmit data. See
Figure 2-13.
Invert normally generated running disparity before
encoding this byte
Function
TXCHARISK
TXCHARISK is a byte-mapped control port that is used only when the 8B/10B encoder is
implemented. This port controls whether the byte of TXDATA is to be encoded as a control
(K) character (when asserted High) or as a data character (when de-asserted). When
8B/10B encoding is bypassed, this port is undefined.
TXRUNDISP
TXRUNDISP is a status port that is byte-mapped to TXDATA. This port indicates the
running disparity after the byte of TXDATA is encoded. When High, the disparity is
positive. When Low, the disparity is negative.
TXKERR
TXKERR is a status port that is byte-mapped to TXDATA. This port is defined only if
8B/10B encoding is enabled. If a bit is asserted High, it means that TXDATA and
TXCHARISK have combined to create an invalid control (K) character. The transmission,
reception, and decode of this invalid character will create unexpected RXDATA results in
the RocketIO receiver, or in other transceivers.
RXCHARISK,
RXRUNDISP
RXCHARISK and RXRUNDISP are dual-purpose ports for the receiver depending
whether 8B/10B decoding is enabled. Tab le 2 -10 shows this dual functionality. When
decoding is enabled, the ports function as byte-mapped status ports for the received data.
RocketIO™ Transceiver User Guidewww.xilinx.com63
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
In the 8B/10B decoding configuration, RXCHARISK asserted High indicates the received
byte of data is a control (K) character. Otherwise, the received byte of data is a data
character. See Appendix B, “8B/10B Valid Characters”.
The RXRUNDISP port indicates the disparity of the received byte is either negative or
positive. RXRUNDISP asserted High indicates positive disparity. This is used in cases like
the “Vitesse Disparity Example” below. When CLK_COR_INSERT_IDLE_FLAG = TRUE,
RXRUNDISP is asserted to flag the presence of an inserted clock correction sequence.
In the bypassed configuration, RXCHARISK and RXRUNDISP are additional data bits for
the 10-, 20-, or 40-bit buses, similar to the configuration on the transmit side. RXCHARISK
[0:3] relates to bits 9, 19, 29, and 39, while RXRUNDISP pertains to bits 8, 18, 28, and 38 of
the data bus. See Figure 2-14.
RXDISPERR
RXDISPERR is a status port for the receiver that is byte-mapped to RXDATA. When a bit in
RXDISPERR is asserted High, it means that a disparity error has occurred in the received
data. This usually indicates data corruption (bit errors) or transmission of an invalid
control character. It can also occur in cases where normal disparity is not required, such as
in the “Vitesse Disparity Example”.
Chapter 2: Digital Design Considerations
RXNOTINTABLE
RXNOTINTABLE is a status port for the receiver that is byte-mapped to RXDATA. When it
is asserted High, it means that the received data is not in the 8B/10B tables. This port is
only used when the 8B/10B decoder is enabled.
Vitesse Disparity Example
To support other protocols, the transceiver can affect the disparity mode of the serial data
transmitted. For example, Vitesse channel-to-channel alignment protocol sends out:
K28.5+ K28.5+ K28.5- K28.5- or K28.5- K28.5- K28.5+ K28.5+
instead of:
K28.5+ K28.5- K28.5+ K28.5- or K28.5- K28.5+ K28.5- K28.5+
The logic must assert TXCHARDISPVAL to cause the serial data to send out two negative
running disparity characters.
Note:
version of the channel bonding sequence. This is the same as the clock correction sequence
shown in Table 2-15, page 74.
Transmitting Vitesse Channel Bonding Sequence
If bypassing 8B/10B encoding/decoding, the remaining 10 bits will be the 10-bit-encoded
The RocketIO core receives this data, but for cases where TXCHARDISPVAL is set High
during data transmission, the disp_err bit in CHAN_BOND_SEQ must also be set High.
Receiving Vitesse Channel Bonding Sequence
On the RX side, the definition of the channel bonding sequence uses the disp_err bit to
specify the flipped disparity.
When 8B/10B encoding is bypassed, the TXCHARDISPVAL and TXCHARDISPMODE
bits become bits “b” and “a”, respectively, of the 10-bit encoded data that the transceiver
must transmit to the receiving terminal. Figure 2-13 illustrates the TX data map during
8B/10B bypass.
TXCHARDISPMODE[0]
TXCHARDISPVAL[0]
TXDATA[7] . . .. . . TXDATA[0]
ghjfiedcab
7896543201
First transmittedLast transmitted
UG024_10a_051602
Figure 2-13: 10-Bit TX Data Map with 8B/10B Bypassed
During receive when 8B/10B decoding is enabled, the running disparity of the serial
transmission can be read by the transceiver from the RXRUNDISP port, while the
RXCHARISK port indicates presence of a K-character. When 8B/10B decoding is
bypassed, these bits remain as Bits “b” and “a”, respectively, of the 10-bit encoded data that
the transceiver passes on to the user logic. Figure 2-14 illustrates the RX data map during
8B/10B bypass.
RXCHARISK[0]
RXRUNDISP[0]
RXDATA[7] . . .. . . RXDATA[0]
ghjfiedcab
7896543201
First receivedLast received
UG024_10b_051602
Figure 2-14: 10-Bit RX Data Map with 8B/10B Bypassed
RocketIO™ Transceiver User Guidewww.xilinx.com65
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
8B/10B Serial Output Format
The 8B/10B encoding translates a 8-bit parallel data byte to be transmitted into a 10-bit
serial data stream. This conversion and data alignment are shown in Figure 2-15. The serial
port transmits the least significant bit of the 10-bit data “a” first and proceeds to “j”. This
allows data to be read and matched to the form shown in Appendix B, “8B/10B Valid
Characters.”
Chapter 2: Digital Design Considerations
ABCDEFHG
Parallel
01234576
8B/10B
ghjfiedcab
Serial
7896543201
First transmittedLast transmitted
UG024_10_021102
Figure 2-15: 8B/10B Parallel to Serial Conversion
The serial data bit sequence is dependent on the width of the parallel data. The most
significant byte is always sent first, regardless of the whether 1-byte, 2-byte, or 4-byte paths
are used. The least significant byte is always last. Figure 2-16 shows a case when the serial
data corresponds to each byte of the parallel data. TXDATA [31:24] is serialized and sent
out first, followed by TXDATA [23:16], TXDATA [15:8], and finally TXDATA [7:0]. The
2-byte path transmits TXDATA [15:8] and then TXDATA [7:0].
H3 − A
3
TXDATA 31:24TXDATA 23:16TXDATA 15:8TXDATA 7:0
H2 − A
2
H1 − A
1
H0 − A
0
8B/10B
LSB
a3 − j
3
st
1
Sent
Encoded
3
LSB
a2 − j
2
2nd Sent
Encoded
LSB
a1 − j
1
3rd Sent
Encoded
1
2
LSB
a0 − j
0
4th Sent
Encoded
U024_11_020802
0
Figure 2-16: 4-Byte Serial Structure
HDL Code Examples: Transceiver Bypassing of 8B/10B Encoding
8B/10B encoding can be bypassed by the transceiver. The TXBYPASS8B10B is set to 1111;
the RXDECODE attribute is set to FALSE to create the extra two bits needed for a 10-bit
data bus; and TXCHARDISPMODE, TXCHARDISPVAL, RXCHARISK, and RXRUNDISP
are added to the 8-bit data bus.
Please use the Architecture Wizard to create instantiation templates. This wizard creates
code and instantiation templates that define the attributes for a specific application.
66www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
SERDES Alignment
SERDES Alignment
Overview
Serializer
The multi-gigabit transceiver multiplies the reference frequency provided on the reference
clock input (REFCLK) by 20, or by 10 if half-rate operation is selected. Data is converted
from parallel to serial format and transmitted on the TXP and TXN differential outputs.
The electrical polarity of TXP and TXN can be interchanged through the TXPOLARITY
port. This option can either be programmed or controlled by an input at the FPGA core TX
interface. This facilitates recovery from situations where printed circuit board traces have
been reversed.
Deserializer
The RocketIO transceiver core accepts serial differential data on its RXP and RXN inputs.
The clock/data recovery circuit extracts clock phase and frequency from the incoming data
stream and re-times incoming data to this clock. The recovered clock is presented on
output RXRECCLK at 1/20 of the received serial data rate.
R
The receiver is capable of handling either transition-rich 8B/10B streams or scrambled
streams, and can withstand a string of up to 75 non-transitioning bits without an error.
Word alignment is dependent on the state of comma detect bits. If comma detect is
enabled, the transceiver recognizes up to two 10-bit preprogrammed characters. Upon
detection of the character or characters, RXCOMMADET is driven High and the data is
synchronously aligned. If a comma is detected and the data is aligned, no further
alignment alteration takes place. If a comma is received and realignment is necessary, the
data is realigned and RXREALIGN is asserted. The realignment indicator is a distinct
output. The transceiver continuously monitors the data for the presence of the 10-bit
character(s). Upon each occurrence of the 10-bit character, the data is checked for word
alignment. If comma detect is disabled, the data is not aligned to any particular pattern.
The programmable option allows a user to align data on plus-comma, minus-comma, both,
or a unique user-defined and programmed sequence.
The electrical polarity of RXP and RXN can be interchanged through the RXPOLARITY
port. This can be useful in the event that printed circuit board traces have been reversed.
Ports and Attributes
Comma definition can be accomplished using the attributes discussed below. This method
of definition makes the MGT extremely flexible in implementing different protocols.
ALIGN_COMMA_MSB
This attribute determines where the commas will reside in the parallel received data. The
comma indicates to the deserializer how to parallelize the data. However, with the
multiple data path widths available, the PCS portion must determine where to place the
comma in the parallel data bytes.
When ALIGN_COMMA_MSB is FALSE, the PCS may place the comma in any of the
RXDATA bytes. In the 1-byte mode, of course, there is only one location in which the
comma can be placed. In the 2-byte and 4-byte paths, some uncertainty exists as to which
byte will contain the comma, as shown in Tabl e 2 -12 .
RocketIO™ Transceiver User Guidewww.xilinx.com67
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
When ALIGN_COMMA_MSB is TRUE, the PCS places the comma into the most
significant byte (MSB) of RXDATA in the 2-byte mode. Because the PCS is optimized for
the 2-byte mode, some uncertainty exists in the 4-byte mode as to which byte will contain
the comma, as shown in Tab le 2 -1 2. See “Receive Data Path 32-bit Alignment” for more
details on this case.
Table 2-12: Possible Locations of Comma Character
Chapter 2: Digital Design Considerations
Data Path Width:
ALIGN_COMMA_MSB:
1 byte2 bytes4 bytes
[7:0][15:8][7:0][31:24] [23:16] [15:8][7:0]
TRUE√√√√
FALSE√√√√√√√
ENPCOMMAALIGN,
ENMCOMMAALIGN
These two alignment ports control how the PMA aligns incoming serial data. It can align
on a minus-comma (negative disparity), a plus-comma (positive disparity), both, or
neither if comma alignment is not desired. These signals are latched inside the transceiver
with RXRECCLK.
Care must be taken not to de-assert these signals at the improper time. Comma detection
may be vulnerable to spurious realignment if RXRECCLK occurs at the wrong time. To
avoid this problem, ENPCOMMAALIGN and ENMCOMMAALIGN should be passed
through a flip-flop that is clocked with RXRECCLK. These flip-flops should be located
near the MGT, and RXRECCLK should use local interconnect (not global clock resources)
to reduce skew. For both top and bottom edges, the best slices to use are in the CLB
immediately to the left of the transceiver, next to the bottom of the transceiver. For the top
side of the chip, this is the fourth CLB row; for the bottom side, the bottom CLB row. For
example, for the XC2VP7, here are the best slices to use for two of the transceivers:
•For GT_X0Y1 (top edge), the best slices are SLICE_X15Y72 and SLICE_X15Y73.
•For GT_X0Y0 (bottom edge), the best slices are SLICE_X14Y0 and SLICE_X14Y1.
This must be done for each MGT. Figure 2-17 shows this recommendation.
PCOMMA_CONTROL
MCOMMA_CONTROL
DDQ
GT_std_
ENPCOMMAALIGN
RXRECCLK
Q
ENMCOMMAALIGN
*
UG024_39_013103
Figure 2-17: Synchronizing Comma Align Signals to RXRECCLK
68www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
SERDES Alignment
R
Figure 2-18 and Figure 2-19 show floorplanner layouts for the two examples given above.
ug024_43_031303
Figure 2-18: Top MGT Comma Control Flip-Flop Ideal Locations
ug024_44_031303
Figure 2-19: Bottom MGT Comma Control Flip-Flop Ideal Locations
RocketIO™ Transceiver User Guidewww.xilinx.com69
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
PCOMMA_DETECT,
MCOMMA_DETECT
These two control attributes define when RXCOMMADET signals that a comma has been
received. When only PCOMMA_DETECT is TRUE, RXCOMMADET signals when a pluscomma is received, but not a minus-comma. When only MCOMMA_DET is TRUE,
RXCOMMADET signals when a minus-comma is received, but not a plus-comma. If both
attributes are TRUE, RXCOMMADET will signal when either comma character is received.
The RocketIO transceiver allows the user to define a comma character using these three
attributes. The COMMA_10B_MASK bits are used in conjunction with
PCOMMA_10B_VALUE (to define a plus-comma) or MCOMMA_10B_VALUE (to define a
minus-comma) to define some number of recognized comma characters. High bits in the
mask condition the corresponding bits in PCOMMA_10B_VALUE or
MCOMMA_10B_VALUE to matter, while Low bits in the mask function as a “don’t care”
conditioner.
Chapter 2: Digital Design Considerations
For example, with COMMA_10B_MASK set to 1111111000 (meaning the three least
significant bits don’t matter) and PCOMMA_10B_VALUE is 0011111000, the comma
detection unit will recognize the following characters as plus-commas:
Using the same value in PCOMMA_10B_VALUE but setting COMMA_10B_MASK to
1111111111 (meaning all the bits in PCOMMA_10B_VALUE matter), the comma
detection unit will recognize only the 0011111000 (K28.7) sequence, which matches the
value of PCOMMA_10B_VALUE exactly.
These signals only pertain to the 8B/10B decoder, not the comma alignment circuitry. The
DEC_PCOMMA_DETECT and DEC_MCOMMA_DETECT control the 8B/10B decoder to
signal the RXCHARISCOMMA port if a plus-comma or minus-comma is received. This is
described in the table below.
DEC_VALID_COMMA_ONLY, for most applications, should be set to TRUE. If valid data
is being transmitted and hence received, then an invalid comma would arise only in the
case of a bit error, in which case RXCHARISCOMMA would not be asserted in the
presence of bit errors. If set to FALSE, then RXCHARISCOMMA will be asserted for
invalid K-characters.
RXREALIGN
This status signal indicates whenever, the serial data is realigned from a comma character
in the data stream. This signal will not necessarily go High after the transceiver is reset. If
70www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Clock Recovery
Product Not Recommendedfor NewDesigns
R
ENPCOMMAALIGN and ENMCOMMAALIGN are both set to zero then this signal
should not go High. See Tab le 2 -1 3.
RXCHARISCOMMA
This signal is similar to RXCHARISK, except that it signals that a specific byte of RXDATA
is a comma character. However, this definition only holds true for when 8B/10B
encoding/decoding is enabled. This port is controlled by the DEC_* attributes and is
shown in Tab le 2 -1 3. If the 8B/10B decoder is bypassed, this port is undefined.
RXCOMMADET
This signal indicates if a comma character has been detected in the serial data. The
definition of this port is defined by the PCOMMA_DETECT and MCOMMA_DETECT
attributes. This signal is clocked off RXRECCLK, and to reliably have the signal pulse for
all the data width configurations, this pulse may change with respect to the USRCLKs.
Table 2-13: Effects of Comma-Related Ports and Attributes
Clock Recovery
Overview
Clock Synthesizer
Synchronous serial data reception is facilitated by a clock/data recovery circuit. This
circuit uses a fully monolithic Phase-Locked Loop (PLL), which does not require any
external components. The clock/data recovery circuit extracts both phase and frequency
from the incoming data stream. The recovered clock is presented on output RXRECCLK at
1/20 of the serial received data rate.
The gigabit transceiver multiplies the reference frequency provided on the reference clock
input (REFCLK) by 20.
No fixed phase relationship is assumed between REFCLK, RXRECCLK, and/or any other
clock that is not tied to either of these clocks. When the 4-byte or 1-byte receiver data path
is used, RXUSRCLK and RXUSRCLK2 have different frequencies (1:2), and each edge of
the slower clock is aligned to a falling edge of the faster clock. The same relationships
apply to TXUSRCLK and TXUSRCLK2. See Table 2-5, page 43, for details.
RocketIO™ Transceiver User Guidewww.xilinx.com71
UG024 (v3.0) February 22, 2007
l
)
)
Product Not Recommendedfor NewDesigns
R
Clock and Data Recovery
The clock/data recovery (CDR) circuits lock to the reference clock automatically if the data
is not present. For proper operation, TXUSRCLK must have the exact same frequency as
REFCLK. REFCLK, RXUSRCLK, and the incoming stream (RXRECCLK) must not exceed
±100 ppm of frequency variation.
It is critical to keep power supply noise low in order to minimize common and differential
noise modes into the clock/data recovery circuitry. See “PCB Design Requirements,”
page 109, for more details.
Clock Correction
Clock RXRECCLK (the recovered clock) reflects the data rate of the incoming data. Clock
RXUSRCLK defines the rate at which the FPGA core consumes the data. Ideally, these rates
are identical. However, since the clocks typically have different sources, one of the clocks is
faster than the other. The receiver buffer accommodates this difference between the clock
rates. See Figure 2-20.
Chapter 2: Digital Design Considerations
Read
RXUSRCLK
Read
Removable sequence
"Nominal" condition: buffer half-ful
Read
Buffer less than half -full (emptying
Repeatable sequence
Buffer more than half-full (filling up
Write
Write
RXRECCLK
Write
DS083-2_15_100901
Figure 2-20: Clock Correction in Receiver
Nominally, the buffer is always half-full. This is shown in the top buffer, where the shaded
area represents buffered data not yet read. Received data is inserted via the write pointer
under control of RXRECCLK. The FPGA core reads data via the read pointer under control
of RXUSRCLK. The half-full/half-empty condition of the buffer gives a cushion for the
differing clock rates. This operation continues indefinitely, regardless of whether or not
“meaningful” data is being received. When there is no meaningful data to be received, the
incoming data consists of IDLE characters or other padding.
If RXUSRCLK is faster than RXRECCLK, the buffer becomes more empty over time. The
clock correction logic corrects for this by decrementing the read pointer to reread a
repeatable byte sequence. This is shown in the middle buffer, Figure 2-20, where the solid
read pointer decrements to the value represented by the dashed pointer. By decrementing
the read pointer instead of incrementing it in the usual fashion, the buffer is partially
refilled. The transceiver inserts a single repeatable byte sequence when necessary to refill a
buffer. If the byte sequence length is greater than one, and if attribute
CLK_COR_REPEAT_WAIT is 0, then the transceiver can repeat the same sequence
multiple times until the buffer is refilled to the half-full condition.
72www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Clock Recovery
Product Not Recommendedfor NewDesigns
R
Similarly, if RXUSRCLK is slower than RXRECCLK, the buffer fills up over time. The clock
correction logic corrects for this by incrementing the read pointer to skip over a removable
byte sequence that need not appear in the final FPGA core byte stream. This is shown in the
bottom buffer, Figure 2-20, where the solid read pointer increments to the value
represented by the dashed pointer. This accelerates the emptying of the buffer, preventing
its overflow. The transceiver design skips a single byte sequence, when necessary, to
partially empty a buffer. If attribute CLK_COR_REPEAT_WAIT is 0, the transceiver can
also skip four consecutive removable byte sequences in one step, to further empty the
buffer when necessary.
These operations require the clock correction logic to recognize a byte sequence that can be
freely repeated or omitted in the incoming data stream. This sequence is generally an IDLE
sequence, or other sequence comprised of special values that occur in the gaps separating
packets of meaningful data. These gaps are required to occur sufficiently often to facilitate
the timely execution of clock correction.
The clock correction logic has the ability to remove up to four IDLE sequences during a
clock correction. How many IDLEs are removed depends on several factors, including
how many IDLEs are received and whether CLK_COR_KEEP_IDLE is TRUE or FALSE.
For example, if three IDLEs are received and CLK_COR_KEEP_IDLE is set to TRUE, at
least one IDLE sequence must remain after clock correction has been completed. This
limits the clock correction logic to remove only two of the three IDLE sequences. If
CLK_COR_KEEP_IDLE is FALSE, then all three IDLEs can be removed.
Tab le 2 -1 4 illustrates the relationship between the number of IDLE sequences removed, the
inherent stability of REFCLK, and the number of bytes allowed between clock correction
sequences.
Table 2-14: Data Bytes Allowed Between Clock Corrections as a Function of
REFCLK Stability and IDLE Sequences Removed
REFCLK
Stability
100 ppm5,00010,00015,00020,000
50 ppm10,00020,00030,00040,000
20 ppm25,00050,00075,000100,000
Notes:
1. All numbers are approximate.
2. IDLE = the defined clock correction sequence.
Ports and Attributes
CLK_CORRECT_USE
This attribute controls whether the PCS will repeat/skip the clock correction sequences
(CCS) from the elastic buffer to compensate for differences between the clock recovered
from serial data and the reference clocks. When this attribute is set to TRUE, the clock
correction is enabled. If set to FALSE, clock correction is disabled. When clock correction is
disabled, RXRECCLK must drive the receive logic in the fabric. Otherwise, the elastic
buffer may over/underflow.
Bytes Allowed Between Clock Correction Sequences
Remove 1 IDLE
Sequence:
(2)
Remove 2 IDLE
Sequences:
Remove 3 IDLE
Sequences:
(1)
Remove 4 IDLE
Sequences:
RocketIO™ Transceiver User Guidewww.xilinx.com73
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
Clock correction may be used with other encoding protocols, but they must have a 10-bit
alignment scheme. This is required so the comma detection logic can properly align the
data in the elastic buffer, allowing the clock correction logic to properly read out data to the
FPGA fabric.
RX_BUFFER_USE
The RX_BUFFER_USE attribute controls if the elastic buffer is bypassed or not. Most
applications use this buffer for clock correction and channel bonding. (See “Channel
Bonding (Channel Alignment),” page 79.) It is recommended that this attribute always be
set to TRUE, since this buffer allows a way to cross the clock domains of RXRECCLK and
the fabric RXUSRCLK/RXUSRCLK2.
CLK_COR_SEQ_*_*
To accommodate many different protocols, the MGT features programmability that allows
it to detect a 1-, 2-, or 4-byte clock correction sequence (CCS), such as may be used in
Gigabit Ethernet (2-byte) or Fibre Channel (4-byte). The attributes CLK_COR_SEQ_*_*
and CLK_COR_SEQ_LEN (below) define the CCS that the PCS recognizes. Both SEQ_1
and SEQ_2 can be used at the same time if multiple CCSs are required. As shown in
Tab le 2 -1 5, the example CCS has two possible modes, one for when 8B/10B encoding is
used, the other for when 8B/10B encoding is bypassed. The most significant bit of the CCS
determines whether it is applicable to an 8-bit (encoded) or a 10-bit (unencoded) sequence.
Chapter 2: Digital Design Considerations
These sequences require that the encoding scheme allows the comma detection and
alignment circuitry to properly align data in the elastic buffer. (See
“CLK_CORRECT_USE”, above). The bit definitions are the same as shown earlier in the
Vitesse channel-bonding example. (See “Receiving Vitesse Channel Bonding Sequence.”)
Table 2-15: Clock Correction Sequence / Data Correlation for 16-Bit Data Port
Attribute Settings
CLK_COR_SEQ8-Bit Data Mode
CLK_COR_SEQ_1_1
CLK_COR_SEQ_1_20001001010111010100010D21.4095
CLK_COR_SEQ_1_30001011010111010101010D21.50Β5
CLK_COR_SEQ_1_40001011010111010101010D21.50Β5
0011011110010011111010K28.51
10-Bit Data Mode
(8B/10B Bypass)
CharacterCHARISK
TXDATA
(hex)
BC
74www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Clock Recovery
Product Not Recommendedfor NewDesigns
R
CLK_COR_SEQ_LEN
To define the CCS length, this attribute takes the integer value 1, 2, 3, or 4. Tab le 2 -1 6 shows
which sequences are used for the four possible settings of CLK_COR_SEQ_LEN.
Table 2-16: Applicable Clock Correction Sequences
CLK_COR_SEQ_LEN
11_12_1
21_1, 1_22_1, 2_2
31_1, 1_2, 1_32_1, 2_2, 2_3
41_1, 1_2, 1_3. 1_42_1, 2_2, 2_3, 2_4
Notes:
1. Applicable only if CLK_COR_SEQ_2_USE is set to TRUE.
These attributes help control how clock correction is implemented.
CLK_COR_INSERT_IDLE_FLAG is a TRUE/FALSE attribute that defines the output of
the RXRUNDISP port. When set to TRUE, RXRUNDISP is raised for the first byte of each
inserted (repeated) clock correction sequence (8B/10B decoding enabled). When set to
FALSE (default), RXRUNDISP denotes the running disparity of RXDATA (8B/10B
decoding enabled).
CLK_COR_KEEP_IDLE is a TRUE/FALSE attribute that controls whether or not the final
byte stream must retain at least one clock correction sequence. When set to FALSE
(default), the clock correction logic is allowed to remove all clock correction sequences if
needed to recenter the elastic buffer. When set to TRUE, it forces the clock correction logic
to retain at least one clock correction sequence per continuous stream of clock correction
sequences.
(1)
Example: Elastic buffer is 75% full and clock correction is needed. (IDLE is the defined clock
correction sequence.)
Data stream written into elastic buffer:
D0IDLEIDLEIDLEIDLED1D2
Data stream read out of elastic buffer (CLK_COR_KEEP_IDLE = FALSE)
D0D1D2
Data stream read out of elastic buffer (CLK_COR_KEEP_IDLE = TRUE)
D0
CLK_COR_REPEAT_WAIT is an integer attribute (0-31) that controls frequency of
repetition of clock correction operations. This attribute specifies the minimum number of
RXUSRCLK cycles without clock correction that must occur between successive clock
corrections. For example, if this attribute is 3, then at least three RXUSRCLK cycles without
RocketIO™ Transceiver User Guidewww.xilinx.com75
UG024 (v3.0) February 22, 2007
IDLED1D2
Product Not Recommendedfor NewDesigns
R
clock correction must occur before another clock correction sequence can occur. If this
attribute is 0, no limit is placed on how frequently clock correction can occur.
Example: Elastic buffer is 25% full, clock correction is needed, and one sequence is repeated
per clock correction. (IDLE is the defined clock correction sequence.)
Data stream written into elastic buffer:
Chapter 2: Digital Design Considerations
D0
Data stream read out of elastic buffer (CLK_COR_REPEAT_WAIT = 0):
D0
Data stream read out of elastic buffer (CLK_COR_REPEAT_WAIT = 1):
D0
The percent that the buffer is full, together with the value of CLK_COR_REPEAT_WAIT,
determines how many times the clock correction sequence is repeated during each clock
correction.
IDLEIDLEIDLED1D2
IDLEIDLEIDLEIDLEIDLEIDLED1D2
IDLEIDLEIDLEIDLEIDLED1D2
Synchronization Logic
Overview
For some applications, it is beneficial to know if incoming data is valid or not, and if the
MGT is synchronized on the data. For applications using the 8B/10B encoding scheme, the
RX_LOSS_OF_SYNC FSM does this. It can be programmed to lose sync after a specified
number of invalid data characters are received.
Ports and Attributes
RXCLKCORCNT
Clock correction count (RXCLKCORCNT) is a three-bit signal. It signals if clock correction
has occurred, and whether the elastic buffer realigned the data by skipping or repeating
data in the buffer. It also signals if channel bonding has occurred. Tab le 2 -1 7 defines the
eight binary states of RXCLKCORCNT.
Table 2-17: RXCLKCORCNT Definition
RXCLKCORCNT[2:0]Significance
000
001
010
011
No channel bonding or clock correction occurred for current
RXDATA
Elastic buffer skipped one clock correction sequence for
current RXDATA
Elastic buffer skipped two clock correction sequence for
current RXDATA
Elastic buffer skipped three clock correction sequence for
current RXDATA
76www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
Synchronization Logic
R
Table 2-17: RXCLKCORCNT Definition (Continued)
RXCLKCORCNT[2:0]Significance
100
101Elastic buffer executed channel bonding for current RXDATA
110
111
Elastic buffer skipped four clock correction sequence for
current RXDATA
Elastic buffer repeated two clock correction sequences for
current RXDATA
Elastic buffer repeated one clock correction sequences for
current RXDATA
RX_LOS_INVALID_INCR,
RX_LOS_THRESHOLD
These two signals determine how fast an invalid character advances the RXLOSSOFSYNC
FSM counter before loss of sync is considered to have occurred. RX_LOS_INVALID_INCR
determines how quickly the occurrence of invalid characters is “forgotten” in the presence
of subsequent valid characters. For example, RX_LOS_INVALID_INCR = 4 means that
four consecutive valid characters after an invalid character will reset the counter.
RX_LOS_THRESHOLD determines when the counter has reached the point where the link
is considered to be “out of sync.”
RX_LOSS_OF_SYNC_FSM
The transceiver’s FSM is driven by RXRECCLK and uses status from the data stream prior
to the elastic buffer. This is intended to give early warning of possible problems well before
corrupt data appears on RXDATA. RX_LOSS_OF_SYNC_FSM, a TRUE/FALSE attribute,
indicates what the output of the RXLOSSOFSYNC port (see below) means.
Note:
The values are undefined if the PLL is not locked.
RX_LOSS_OF_SYNC_FSM only indicates valid data. It is assumed that the PLL is locked.
RXLOSSOFSYNC
If RX_LOSS_OF_SYNC_FSM = FALSE, then RXLOSSOFSYNC[1] High indicates that the
transceiver has received an invalid character, and RXLOSSOFSYNC[0] High indicates that
a channel-bonding sequence has been recognized.
If RX_LOSS_OF_SYNC_FSM = TRUE, then the two bits of RXLOSSOFSYNC reflect the
state of the RXLOSSOFSYNC FSM. The state machine diagram in Figure 2-21 and the three
subsections following describe the three states of the RXLOSSOFSYNC FSM
RocketIO™ Transceiver User Guidewww.xilinx.com77
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
.
Chapter 2: Digital Design Considerations
SYNC_ACQUIRED
count < RX_LOS_THRESHOLD
00
count = RX_LOS_THRESHOLD
valid data + 4 RXRECCLK cycles
channel alignment
or
comma realignment
valid data + < 4 RXRECCLK cycles
invalid data
no comma received
0110
RESYNCLOSS_OF_SYNC
Figure 2-21: RXLOSSOFSYNC FSM States
SYNC_ACQUIRED (RXLOSSOFSYNC = 00)
In this state, a counter is decremented by 1 (but not past 0) for a valid received symbol and
incremented by RX_LOS_INVALID_INCR for an invalid symbol. If the count reaches or
exceeds RX_LOS_THRESHOLD, the FSM moves to state LOSS_OF_SYNC. Otherwise, if a
channel bonding (alignment) sequence has just been written into the elastic buffer, or if a
comma realignment has just occurred, the FSM moves to state RESYNC. Otherwise, the
FSM remains in state SYNC_ACQUIRED.
comma received
UG024_40_031803
RESYNC (RXLOSSOFSYNC = 01)
The FSM waits in this state for four RXRECCLK cycles and then goes to state
SYNC_ACQUIRED, unless an invalid symbol is received, in which case the FSM goes to
state LOSS_OF_SYNC.
LOSS_OF_SYNC (RXLOSSOFSYNC = 10)
The FSM remains in this state until a comma is received, at which time it goes to state
RESYNC.
78www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
Channel Bonding (Channel Alignment)
Channel Bonding (Channel Alignment)
Overview
Some gigabit I/O standards such as XAUI specify the use of multiple transceivers in
parallel for even higher data rates. Words of data are split into bytes, with each byte sent
over a separate channel (transceiver). See Figure 2-22.
In Transmitters:
Full word SSSS sent over four channels, one byte per channel
R
PQRS T
PQRS T
PQRS T
PQRS T
Read
RXUSRCLK
PQRS T
PQRS T
PQRS T
PQRS T
Before channel bondingAfter channel bonding
Channel (lane) 0
Channel (lane) 1
Channel (lane) 2
Channel (lane) 3
In Receivers:
Read
RXUSRCLK
PQRS T
PQRS T
PQRS T
PQRS T
DS083-2_16_010202
Figure 2-22: Channel Bonding (Alignment)
The top half of the figure shows the transmission of words split across four transceivers
(channels or lanes). PPPP, QQQQ, RRRR, SSSS, and TTTT represent words sent over the
four channels.
The bottom-left portion of the figure shows the initial situation in the FPGA’s receivers at
the other end of the four channels. Due to variations in transmission delay—especially if
the channels are routed through repeaters—the FPGA core might not correctly assemble
the bytes into complete words. The bottom-left illustration shows the incorrect assembly of
data words PQPP, QRQQ, RSRR, etc.
To support correction of this misalignment, the data stream includes special byte
sequences that define corresponding points in the several channels. In the bottom half of
Figure 2-22, the shaded “P” bytes represent these special characters. Each receiver
recognizes the “P” channel bonding character, and remembers its location in the buffer. At
some point, one transceiver designated as the Master instructs all the transceivers to align
to the channel bonding character “P” (or to some location relative to the channel bonding
character). After this operation, the words transmitted to the FPGA core are properly
aligned: RRRR, SSSS, TTTT, etc., as shown in the bottom-right portion of Figure 2-22. To
ensure that the channels remain properly aligned following the channel bonding
operation, the Master transceiver must also control the clock correction operations
described in the previous section for all channel-bonded transceivers.
RocketIO™ Transceiver User Guidewww.xilinx.com79
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
Channel Bonding (Alignment) Operation
Channel bonding is the technique of tying several serial channels together to create one
aggregate channel. Several channels are fed on the transmit side by one parallel bus and
reproduced on the receive side as the identical parallel bus. The maximum number of
serial differential pairs that can be bonded is 24. For implementation guidelines, see
“Implementation Tools,” page 121.
Channel bonding allows those primitives that support it to send data over multiple
“channels.” Among these primitives are GT_CUSTOM, GT_INFINIBAND, GT_XAUI, and
GT_AURORA. To “bond” channels together, there is always one “Master.” The other
channels can either be a SLAVE_1_HOP or SLAVE_2_HOPS. SLAVE_1_HOP is a Slave to a
Master that can also be daisy chained to a SLAVE_2_HOPS. A SLAVE_2_HOPS can only be
a Slave to a SLAVE_1_HOP and its CHBONDO does not connect to another transceiver. To
designate a transceiver as a Master or a Slave, the attribute CHAN_BOND_MODE must be
set to one of three designations: Master, SLAVE_1_HOP, or SLAVE_2_HOPS. To shut off
channel bonding, set the transceiver attribute to “off.” The possible values that can be used
are shown in Ta bl e 2 -1 8.
Table 2-18: Bonded Channel Connections
Chapter 2: Digital Design Considerations
ModeCHBONDICHBONDO
OFFNANA
MASTERNASlave 1 CHBONDI
SLAVE_1_HOPMaster CHBONDOSlave 2 CHBONDI
SLAVE_2_HOPSSlave 1 CHBONDONA
Note:
than or equal to 4 bytes between clock correction and channel bonding sequences. If a user
creates his/’her own protocol that uses clock correction and channel bonding, the user must
ensure that there is at least a 4 byte gap between the sequences.
All standards that use both clock correction and channel bonding require a gap greater
The channel bonding sequence is similar in format to the clock correction sequence. This
sequence is set to the appropriate sequence for the primitives supporting channel bonding.
The GT_CUSTOM is the only primitive allowing modification to the sequence. These
sequences are comprised of one or two sequences of length up to 4 bytes each, as set by
CHAN_BOND_SEQ_LEN and CHAN_BOND_SEQ_2_USE. Other control signals include
the attributes:
An MGT can be designated as one of three types when used in a channel-bonding scheme.
The type is designated by CHAN_BOND_MODE, the three values of which are MASTER,
SLAVE_1_HOP, and SLAVE_2_HOPS. (A fourth mode, OFF, is used when channel
bonding is not being performed.) The Master always controls, for itself and for Slaves of
either type, when channel bonding and clock correction will occur.
R
MasterSlave
Masters are always connected directly to a SLAVE_1_HOP, and indirectly to a
SLAVE_2_HOPS via daisy-chain through a SLAVE_1_HOP. This topology improves the
timing characteristics of the CHBONDO and CHBONDI buses.
ENCHANSYNC
ENCHANSYNC controls when channel bonding is enabled. Tab le 2- 19 shows the
recommended settings for Master and Slaves. To counter the possibility of a bit error
causing a false channel bonding sequence to occur, this port is usually de-asserted once a
group of channels have been successfully aligned.
CHAN_BOND_ONE_SHOT
As with ENCHANSYNC, many applications will require that the channels be aligned only
once. CHAN_BOND_ONE_SHOT = TRUE allows the Master to initiate a channel bonding
only once. This remains true even if more channel bonding sequences are received. (The
channels may be aligned again if RXRESET is asserted and then deasserted, and
ENCHANSYNC is deasserted and then reasserted.)
CHAN_BOND_ONE_SHOT may be set to FALSE when very few channel bonding
sequences appear in the data stream. (For Slave instantiations, this attribute should always
be set to FALSE. See Ta bl e 2 -1 9 .) When the channel bonding sequence appears frequently
in the data stream, however, it is recommended that this attribute be set to TRUE in order
to prevent the RX buffer from over- or underflowing.
The channel bonding sequence (CBS) is similar in format to the clock correction sequence. The
CBS is set to the appropriate sequence for the primitives supporting channel bonding.
GT_CUSTOM is the only primitive allowing modification to the sequence. These
sequences are comprised of one or two sequences of length up to 4 bytes each, as set by
CHAN_BOND_SEQ_LEN and CHAN_BOND_SEQ_2_USE.
RocketIO™ Transceiver User Guidewww.xilinx.com81
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
These CBSs should be unique from other delimiters in the data stream, including Clock
Correction Sequence, IDLE, Start of Frame, and End of Frame. As with clock correction,
there are multiple sequences that can be defined (GT_CUSTOM only). The primary CBS is
defined by CHAN_BOND_SEQ_1_*, where * = a number from 1 to 4.
If a second CBS is required, CHAN_BOND_SEQ_2_USE must be set to TRUE, and
CHAN_BOND_SEQ_2_* used to define the second CBS; otherwise,
CHAN_BOND_SEQ_2_USE should be left at its default value, FALSE. See “Receiving
Vitesse Channel Bonding Sequence,” page 65, for the bit breakdown of the sequence
definition.
Finally, CHAN_BOND_SEQ_LEN defines the CBS length as 1 to 4 bytes. When set to
anything other than 4, only those sequences are defined. For example, if
CHAN_BOND_SEQ_LEN is set to 2, only CHAN_BOND_SEQ_1_1 and
CHAN_BOND_SEQ_1_2 need to be defined.
CHAN_BOND_WAIT,
CHAN_BOND_OFFSET,
CHAN_BOND_LIMIT
These three attributes define how the Master performs channel alignment of the RX buffer.
The typical values of these attributes are:
Chapter 2: Digital Design Considerations
CHAN_BOND_WAIT = 8
CHAN_BOND_WAIT roughly defines the maximum number of bytes by which the Slave
can lag the Master. Due to internal pipelining, the equation should be
(CHAN_BOND_WAIT - 3.5) bytes = # of bytes Slave may lag Master. For example, if
CHAN_BOND_WAIT = 8, the Slave may lag the Master by 4.5 bytes. While this type of lag
is equivalent to approximately 14 ns at 3.125 Gb/s, it is recommended that channel links be
matched as closely as possible.
The equation that produces this maximum lag time result is
lag time [ns] = (1 / serial speed [Gb/s] )
or, for schemes that do not use 8B/10B encoding,
(1 / serial speed [Gb/s] )
In the example above, 1/3.125 Gb/s
The recommended setting of 8 is set for protocols such as Infiniband and XAUI, which can
repeat the CBS every 16 and 17 bytes respectively. However, CHAN_BOND_WAIT can
grow accordingly if CBSs are spaced farther apart.
• number of 10-bit lag characters • 10 bits/character
• 4.5 bytes • 10 bits/byte = 14.4 ns.
• number of lag bytes • 10 bits/byte
CHAN_BOND_OFFSET = CHAN_BOND_WAIT
CHAN_BOND_OFFSET measures the number of bytes past the beginning of the channel
bonding sequence. However, this value must always equal CHAN_BOND_WAIT.
CHAN_BOND_LIMIT = 2X CHAN_BOND_WAIT
CHAN_BOND_LIMIT defines the expiration time after which the Slave will invalidate the
most recently seen CBS location in the RX buffer. For proper alignment, this value must
always be set to two times CHAN_BOND_WAIT.
82www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
Channel Bonding (Channel Alignment)
CHBONDDONE
This port indicates when a channel alignment has occurred in the MGT. When it is
asserted, RXDATA is valid after RXCLKCORCNT goes to a 101.
R
Note:
was successful or not. To determine if channel bonding was successful, check both this signal
and RXCLKCORCNT.
CHBONDI,
CHBONDO
These two 4-bit ports are used by the Master MGT to control its clock correction and
channel bonding, as well as those of any Slaves bonded to it. CHBONDO of the Master is
connected to CHBONDI of a SLAVE_1_HOP. The signal is then daisy-chained from
SLAVE_1_HOP CHBONDO to a SLAVE_2_HOPS CHBONDI. See Figure 4-1 and
Figure 4-2, page 122, and Tab le 2 - 18, p age 80, for examples. The three least significant bits
correlate to the value of the RXCLKCORCNT port. These four bits allow the Master to
control when the Slaves perform clock correction. This keeps channels from going out of
sync if, for instance, one Slave repeated a CCS while another skipped.
RXCLKCORCNT,
RXLOSSOFSYNC
These signals are mainly used for clock correction. However, they can convey some
information relevant to channel bonding as well. Refer to “RXCLKCORCNT” and
“RXLOSSOFSYNC,” page 77.
Troubleshooting
The Slave's RXCLKCORCNT will go to 101 regardless of whether the channel bonding
Factors that influence channel bonding include:
•Skew between Master and Slave CBS arrival time, both Master-lags-Slave and Slave-lags-
Master cases. The larger the separation, the larger CHAN_BOND_WAIT needs to be.
•Arrival time between consecutive CBSs. The smaller the separation is between
consecutive CBSs, the smaller CHAN_BOND_WAIT needs to be set to ensure that the
Master aligns to the intended sequence instead of the one after or the one before.
There are several possibilities that could cause unsuccessful channel bonding:
•Slave’s CBS lagging the master by too much. Essentially, the Slave does not see a CBS
when CHBONDO is asserted.
•Master CBS lags the slave by too much. In this case, the slave’s CBS sequence has
exceeded CHAN_BOND_LIMIT and has expired.
•CBS sequences appear more frequently than CHAN_BOND_LIMIT allows, causing the
Slave to align to a CBS before or after the expected one.
RocketIO™ Transceiver User Guidewww.xilinx.com83
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
CRC (Cyclic Redundancy Check)
Overview
Cyclic Redundancy Check (CRC) is a procedure to detect errors in the received data. The
RocketIO transceiver CRC logic supports the 32-bit invariant CRC calculation used by
Infiniband, Fibre Channel, and Gigabit Ethernet.
CRC Operation
On the transmitter side, the CRC logic recognizes where the CRC bytes should be inserted
and replaces four placeholder bytes at the tail of a data packet with the computed CRC. For
Gigabit Ethernet and Fibre Channel, transmitter CRC can adjust certain trailing bytes to
generate the required running disparity at the end of the packet. This is discussed further
in the “FIBRE_CHAN” and “ETHERNET” sections under “CRC_FORMAT,” page 85.
On the receiver side, the CRC logic verifies the received CRC value, supporting the same
standards as above.
Upon reset, the CRC logic starts with an initial value of all 1s.
Chapter 2: Digital Design Considerations
SOP
CRC Generation
RocketIO transceivers support a 32-bit invariant CRC (fixed 32-bit polynomial shown
below) for Gigabit Ethernet, Fibre Channel, Infiniband, and user-defined modes.
32x26x23x22x16x12x11x10x8x7x5x4x2x1
x
The CRC recognizes the SOP (Start of Packet), EOP (End of Packet), and other packet
features to identify the beginning and end of data. These SOP and EOP are defined by
CRC_FORMAT for ETHERNET, INFINIBAND, and FIBRE_CHAN, and in these cases the
user does not need to set CRC_START_OF_PKT and CRC_END_OF_PKT. Where
CRC_FORMAT is USER_MODE (user-defined), CRC_START_OF_PKT and
CRC_END_OF_PKT are used to define SOP and EOP.
……
4 Bytes
Figure 2-23: CRC Packet Format
…
1++++++++++++++
IdleEOPDataCRC
UG024_07_021102
The transmitter computes 4-byte CRC on the packet data between the SOP and EOP
(excluding the CRC placeholder bytes). The transmitter inserts the computed CRC just
before the EOP. The transmitter modifies trailing Idles or EOP if necessary to generate
correct running disparity for Gigabit Ethernet and Fibre Channel. The receiver recomputes
CRC and verifies it against the inserted CRC. Figure 2-23 shows the packet format for CRC
generation. The empty boxes are only used in certain protocols (Ethernet). The user logic
must create a four-byte placeholder for the CRC by placing it in TXDATA. Otherwise, data
is overwritten.
84www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
CRC (Cyclic Redundancy Check)
CRC Latency
Enabling CRC increases the transmission latency from TXDATA to TXP and TXN. The
enabling of CRC does not affect the latency from RXP and RXN to RXDATA. The typical
and maximum latencies, expressed in TXUSRCLK/RXUSRCLK cycles, are shown in
Tab le 2 -2 0. For timing diagrams expressing these relationships, please see Module 3 of the
Virtex-II Pro Data Sheet
Table 2-20: Effects of CRC on Transceiver Latency
R
.
(1)
CRC Disabled8112542
CRC Enabled14172542
Notes:
1. See Ta bl e 2 -6 and Ta bl e 2 - 7 for all MGT block latency parameters.
2. This maximum may occur when certain conditions are present, and clock correction and channel
bonding are enabled. If these functions are both disabled, the maximum will be near the typical
values.
3. To further reduce receive-side latency, refer to Appendix C, “Related Online Documents.”
Ports and Attributes
TX_CRC_USE,
RX_CRC_USE
These two attributes control whether the MGT CRC circuitry is enabled or bypassed. When
set to TRUE, CRC is enabled. When set to FALSE, CRC is bypassed and must be
implemented in the FPGA fabric.
CRC_FORMAT
TXDATA to TXP and TXN
in TXUSRCLK Cycles
RXP and RXN to RXDATA
in RXUSRCLK Cycles
TypicalMaximumTypicalMaximum
(2)
(2)
(3)
There are four possible CRC modes: USER_MODE, FIBRE_CHAN, ETHERNET, and
INFINIBAND. This attribute is modifiable only for the GT_XAUI and GT_CUSTOM
primitives. Each mode has a Start of Packet (SOP) and End of Packet (EOP) setting to
determine where to start and end the CRC monitoring. USER_MODE allows the user to
define the SOP and EOP by setting the CRC_START_OF_PKT and CRC_END_OF_PKT to
one of the valid K-characters (Tabl e B -2 , p ag e 1 43 ). The CRC is controlled by RX_CRC_USE
and TX_CRC_USE. Whenever these attributes are set to TRUE, CRC is used.
The four modes are defined in the subsections following.
USER_MODE
USER_MODE is the simplest CRC methodology. The CRC checks for the SOP and EOP,
calculates CRC on the data, and leaves the four remainders directly before the EOP. The
CRC form for the user-defined mode is shown in Figure 2-24, along with the timing for
when RXCHECKINGCRC and RXCRCERR are asserted High with respect to the incoming
data.
To check the CRC error detection logic in a testing mode such as serial loopback, a CRC
error can be forced by setting TXFORCECRCERR to High, which incorporates an error into
the transmitted data. When that data is received, it appears “corrupted,” and the receiver
RocketIO™ Transceiver User Guidewww.xilinx.com85
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
signals an error by asserting RXCRCERR High at the same time RXCHECKINGCRC goes
High. User logic determines the procedure that is invoked when a CRC error occurs.
Chapter 2: Digital Design Considerations
Note:
to operate correctly, at least four gap bytes are required between EOP of one packet and SOP of
the next packet. The gap may contain clock correction sequences, provided that at least 4 bytes
of gap remain after all clock corrections.
Data length must be greater than 20 bytes for USER_MODE CRC generation. For CRC
FIBRE_CHAN
The FIBRE_CHAN CRC is similar to USER_MODE CRC (Figure 2-24), with one exception:
In FIBRE_CHAN, SOP and EOP are predefined protocol delimiters. Unlike USER_MODE,
FIBRE_CHAN does not need to define the attributes CRC_START_OF_PKT and
CRC_END_OF_PKT. Both USER_MODE and FIBRE_CHAN, however, disregard SOP and
EOP in CRC computation.
SOPDATAR
RXCHECKINGCRC
RXCRCERR
Figure 2-24: USER_MODE / FIBRE_CHAN Mode
Designs should generate only the EOP frame delimiter for a beginning running disparity
(RD) that is negative. (These are the frame delimiters that begin with /K28.5/D21.4/ or
/K28.5/D10.4/.) Never generate the EOP frame delimiter for a beginning RD that is
positive. (These are the frame delimiters that begin with /K28.5/D21.5/ or
/K28.5/D10.5/.) When the RocketIO CRC determines that the running disparity must be
inverted to satisfy Fibre Channel requirements, it will convert the second byte of the EOP
frame delimiter (D21.4 or D10.4) to the value required to invert the running disparity
(D21.5 or D10.5).
R
0
R
1
2
R
EOP
3
UG024_12_022803
Note that CRC generation for EOP requires that the transmitted K28.5 be left-justified in
the MGT’s internal two-byte data path. Observing the following restrictions assures
correct alignment of the packet delimiters:
•4-byte data path: K28.5 must appear in TXDATA[31:24] or TXDATA[15:8].
•2-byte data path: K28.5 must appear in TXDATA[15:8].
•1-byte data path: K28.5 must be strobed into the MGT on rising TXUSRCLK2 only
when TXUSRCLK is High.
Note:
Minimum data length for this mode is 24 bytes, not including the CRC placeholder.
Note: When CRC_FORMAT=FIBRE_CHAN, TX_CRC_USE must be set to TRUE. Otherwise,
occasional errors will occur in the transmitted data stream. RX_CRC_USE can be either TRUE
or FALSE in this usage.
ETHERNET
The Ethernet CRC is more complex (Figure 2-25) . T he S OP, EO P, a n d P r e a m bl e are
neglected by the CRC. The extension bytes are special “K” characters in special cases. The
extension bytes are untouched by the CRC as are the Trail bits, which are added to
maintain packet length.
86www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
CRC (Cyclic Redundancy Check)
R
SOPSOFDATAR0R1R2R
Preamble
n Bytes
Pad BitsEOPTrail Bits
2 to 3 Bytes
Figure 2-25: Ethernet Mode
Designs should generate only the /K28.5/D16.2/ IDLE sequence for transmission, never
/K28.5/D5.6/. When the RocketIO CRC determines that the running disparity must be
inverted to satisfy Gigabit Ethernet requirements, it will convert the first /K28.5/D16.2/
IDLE following a packet to /K28.5/D5.6/, performing the necessary conversion.
Note:
to the 64-byte minimum packet length. For packets that are already 64 bytes or longer, pad bits
are not used.
As noted in Figure 2-25, pad bits are used to assure that the header, data, and CRC total
Note that CRC generation for IDLE requires that the transmitted K28.5 be left-justified in
the MGT’s internal two-byte data path. Observing the following restrictions assures
correct alignment of the packet delimiters:
•4-byte data path: K28.5 must appear in TXDATA[31:24] or TXDATA[15:8].
•2-byte data path: K28.5 must appear in TXDATA[15:8].
•1-byte data path: K28.5 must be strobed into the MGT on rising TXUSRCLK2 only
when TXUSRCLK is High.
Note:
Minimum data length for this mode is defined by the protocol requirements.
Note: For correct operation of the Gigabit Ethernet CRC function, transmitted and received
frames must comply with the 802.3 specification regarding Gigabit Ethernet. This includes the
preamble maximum length.
3
UG024_13_101602
INFINIBAND
The Infiniband CRC is the most complex mode, and is not supported in the CRC generator.
Infiniband CRC contains two computation types: an invariant 32-bit CRC, the same as in
Ethernet protocol; and a variant 16-bit CRC, which is not supported in the hard core.
Infiniband CRC must be implemented entirely in the FPGA fabric.
There are also two Infiniband Architecture (IBA) packets, a local and a global. Both of these
IBA packets are shown in Figure 2-26.
Local IBA
LRHSOP
Global IBA
BTH
GRHLRHSOP
Packet
Payload
BTH
R
R
R
R
2
3
R
R
1
2
Packet
Payload
0
1
R
0
Figure 2-26: Infiniband Mode
Variant CRCEOP
R
Variant CRCEOP
3
UG024_14_020802
RocketIO™ Transceiver User Guidewww.xilinx.com87
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
The CRC is calculated with certain bits masked in LRH and GRH, depending on whether
the packet is local or global. The size of these headers is shown in Ta bl e 2 -2 1.
Table 2-21: Global and Local Headers
PacketDescriptionSize
LRHLocal Routing Header8 Bytes
GRHGlobal Routing Header40 Bytes
BTHIBA Transport Header12 Bytes
The CRC checks the LNH (Link Next Header) of the LRH. LRH is shown in Figure 2-27,
along with the bits the CRC uses to evaluate the next packet.
B
B17 − B10B
0
2
Chapter 2: Digital Design Considerations
B
B
3
B
4
B
5
B
6
7
B11, B1
1 1 IBA Global Packet
0
1 0 IBA Local Packet
0 1 Raw Packet (CRC does not insert remainder)
0 0 Raw Packet (CRC does not insert remainder)
UG024_15_020802
Figure 2-27: Local Route Header
Note:
Minimum data length for this mode is defined by the protocol requirements.
Because of the complexity of the CRC algorithms and implementations, especially with
Infiniband, a more in-depth discussion is beyond the scope of this manual.
CRC_START_OF_PACKET,
CRC_END_OF_PACKET
When implementing USER_MODE CRC, Start of Packet (SOP) and End of Packet (EOP)
must be defined for the CRC logic. These delimiters must be one of the defined
K-characters (see Table B-2, p age 1 43). These must be different than a clock correction
sequence (CCS) or IDLE sequence; otherwise, the CRC will mistake the CCS or IDLE for
SOP/EOP.
Note:
These attribute are not applicable to the other CRC formats.
RXCHECKINGCRC,
RXCRCERR
These two signals are status ports for the CRC circuitry.
RXCHECKINGCRC is asserted within several USRCLKs of the EOF being received from
RXDATA. This signals that the CRC circuitry has identified the SOF and the EOF.
If a CRC error occurred, RXCRCERR will be asserted at the same time that
RXCHECKINGCRC goes High.
88www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
Fabric Interface (Buffers)
TXFORCECRCERR,
TX_CRC_FORCE_VALUE
To test the CRC logic in either the MGT or the FPGA fabric, TXFORCECRCERR and
TX_CRC_FORCE_VALUE may be used to invoke a CRC error. When TXFORCECRCERR
is asserted High for at least one USRCLK2 cycle during data transmission (between SOP
and EOP), the CRC circuitry is forced to XOR TXDATA with TX_CRC_FORCE_VALUE,
creating a bit error. This should cause the receiver to register that a CRC error has occurred.
RocketIO CRC Support Limitations
There are limitations to the CRC support provided by the RocketIO transceiver core:
•RocketIO CRC support is implementable for single-channel use only. Computation
•The RocketIO transceiver does not compute the 16-bit variant CRC used for
•All CRC formats have minimum allowable packet sizes. These limits are larger than
R
and byte-striping of CRC across multiple bonded channels is not supported. For that
usage, the CRC logic can be implemented in the FPGA fabric.
Infiniband, and thus does not fulfill the Infiniband CRC requirement. Infiniband CRC
can be computed in the FPGA fabric.
those set by the user mode, and are defined by the specific protocol.
Fabric Interface (Buffers)
Overview: Transmitter and Elastic (Receiver) Buffers
Both the transmitter and the receiver include buffers (FIFOs) in the data path. This section
gives the reasons for including the buffers and outlines their operation.
Transmitter Buffer (FIFO)
The transmitter buffer’s write pointer (TXUSRCLK) is frequency-locked to its read pointer
(REFCLK). Therefore, clock correction and channel bonding are not required. The purpose
of the transmitter's buffer is to accommodate a phase difference between TXUSRCLK and
REFCLK. Proper operation of the circuit is only possible if the FPGA clock (TXUSRCLK) is
frequency-locked to the reference clock (REFCLK). Phase variations of up to one clock
cycle are allowable. A simple FIFO suffices for this purpose. A FIFO depth of four permits
reliable operation with simple detection of overflow or underflow, which might occur if
the clocks are not frequency-locked. Overflow or underflow conditions are detected and
signaled at the interface.
Receiver Buffer
The receiver buffer is required for two reasons:
•To accommodate the slight difference in frequency between the recovered clock
RXRECCLK and the internal FPGA core clock RXUSRCLK (clock correction)
•To allow realignment of the input stream to ensure proper alignment of data being
read through multiple transceivers (channel bonding)
The receiver uses an elastic buffer, where “elastic” refers to the ability to modify the read
pointer for clock correction and channel bonding.
RocketIO™ Transceiver User Guidewww.xilinx.com89
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
Ports and Attributes
TXBUFERR
When High, this port indicates that a transmit buffer underflow or overflow has occurred.
Once set High, TXRESET must be asserted to clear this bit.
TX_BUFFER_USE
This attribute allows the user to bypass the transmit buffer. A value of FALSE bypasses the
buffer, while a TRUE keeps the buffer in the data path. This attribute should always be set
to TRUE.
RXBUFSTATUS
This 2-bit port indicates the status of the receiver elastic buffer. RXBUFSTATUS[1] High
indicates if an overflow/underflow error has occurred. (Once set High, the assertion of
RXRESET or RXREALIGN clears this bit.) RXBUFSTATUS[0] High indicates that the elastic
buffer is at least half-full.
Chapter 2: Digital Design Considerations
RX_BUFFER_USE
When set to FALSE, this attribute causes the receive buffer to be bypassed. It should
normally be set to TRUE, since channel bonding and clock correction use the receive buffer
for realignment. When the buffer is bypassed, the user logic must be clocked with
RXRECCLK.
Miscellaneous Signals
Ports and Attributes
Several ports and attributes of the MGT have very unique functionality. The following do
not have large roles in the other functionality discussed so far:
RX_DATA_WIDTH,
TX_DATA_WIDTH
These two attributes define the data width in bytes of RXDATA and TXDATA respectively.
The possible values of each attribute are 1, 2, and 4, which correspond to 8-, 16-, and 32-bit
data buses when 8B/10B encoding/decoding is used. (See “8B/10B Encoding/Decoding,”
page 60.) The bus widths are 10, 20, and 40 bits when 8B/10B encoding/decoding is
bypassed.
SERDES_10B
This attribute allows the MGT to expand its serial speed range. The normal operational
speed range of 1.0 Gb/s to 3.125 Gb/s (20 times the reference clock rate) is obtained when
this attribute is set to FALSE. When set to TRUE, the MGT serial data will run at 10 times
the reference clock rate, producing a speed range of 600 Mb/s to 1 Gb/s.
90www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
Miscellaneous Signals
TERMINATION_IMP
R
Table 2-22: Serial Speed Ranges as a Function of SERDES_10B
SERDES_10B Reference Clock RangeSerial Speed Range
TRUE60 – 100 MHz600 Mb/s – 1.0 Gb/s
FALSE 50 – 156.25 MHz1.0 Gb/s – 3.125 Gb/s
Receive Termination
On-chip termination is provided at the receiver, eliminating the need for external
termination. The receiver includes programmable on-chip termination circuitry for 50Ω
(default) or 75Ω impedance.
Transmit Termination
On-chip termination is provided at the transmitter, eliminating the need for external
termination. Programmable options exist for 50Ω (default) and 75Ω termination.
TXPOLARITY,
RXPOLARITY,
TXINHIBIT
A differential pair has a positive-designated and a negative-designated component. If for
some reason the polarity of these components is switched between two transceivers, the
data will not be passed properly. If this occurs, TXPOLARITY will invert the definition of
the TXN and TXP pins. On the receiver side of the MGT, the RXPOLARITY port can invert
the definition of RXN and RXP.
For some protocols, the MGT must turn off the TXN/TXP pins. The TXINHIBT port shuts
off the transmit pins and forces them to a constant value (TXN = 0, TXP = 1). Asserting
TXINHIBIT also disables internal serial loopback.
TX_DIFF_CTRL,
PRE_EMPHASIS
These two attributes control analog functionality of the MGT.
The TX_DIFF_CTRL attribute is used to compensate for signal attenuation in the link
between transceivers. It has five possible values of 400, 500, 600, 700, and 800 mV. These
values represent the peak-to-peak amplitude of one component of the differential pair; the
full differential peak-to-peak amplitude is two times these values.
The PRE_EMPHASIS attribute has four values—10%, 20%, 25%, and 33%—which are
designated by 0, 1, 2, and 3 respectively. Pre-emphasis is discussed in greater detail in
Chapter 3, “Analog Design Considerations.”
LOOPBACK
To facilitate testing without the requirement to apply patterns or measure data at gigahertz
rates, two programmable loopback features are available.
One option, serial loopback, places the gigabit transceiver into a state where transmit data
is directly fed back to the receiver. An important point to note is that the feedback path is
at the output pads of the transmitter. This tests the entirety of the transmitter and receiver.
RocketIO™ Transceiver User Guidewww.xilinx.com91
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
The second loopback path is a parallel path that checks only the digital circuitry. When the
parallel option is enabled, the serial loopback path is disabled. However, the transmitter
outputs remain active and data is transmitted over the serial link. If TXINHIBIT is asserted,
TXN is forced High and TXP is forced Low until TXINHIBIT is de-asserted.
LOOPBACK allows the user to send the data that is being transmitted directly to the
receiver of the transceiver. Tab le 2 -2 3 shows the three loopback modes.
Table 2-23: LOOPBACK Modes
Chapter 2: Digital Design Considerations
Input
Val ue
ModeDescription
00Normal Mode
Internal Parallel
01
Mode
Internal Serial
10
Mode
Normal Mode is selected during normal operation. The
transmitted data is sent out the differential transmit ports
(TXN, TXP) and are sent to another transceiver without
being sent to its own receiver logic. During normal
operation, LOOPBACK should be set to 00.
Internal Parallel Mode allows testing the transmit and
receive interface logic PCS without having to go into the
PMA section of the transceiver, or to another transceiver.
See Figure 2-28.
Internal Serial Mode is used to check that the entire
transceiver is working properly, including testing of 8B/10B
encoding/decoding. This emulates what another
transceiver would receive as data from this specific
transceiver design.
Since the TXP/TXN pins are still being driven during this
loopback mode, PCB traces on these pins should be
terminated to remove reflections; otherwise, loopback bit
errors could result. Termination can be accomplished by any
of a variety of methods. Two examples:
• Connect SMA terminators on the TXP/TXN SMA
connectors (if applicable), or simply use 50Ω resistors on
the transmitter backplane pins.
• Connect the unterminated TXP/TXN to the RXP/RXN of
another instantiated transceiver, allowing its receiver
inputs to terminate the transmitter outputs.
92www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
Other Important Design Notes
R
TXDATA
RXDATA
TX PCSTX SERIALIZER
RX PCSMUX
Figure 2-28: Serial and Parallel Loopback Logic
Other Important Design Notes
Receive Data Path 32-bit Alignment
The RocketIO transceiver uses the attribute ALIGN_COMMA_MSB to align protocol
delimiters with the use of comma characters (special K-characters K28.5, K28.1, and K28.7
for most protocols). Setting ALIGN_COMMA_MSB to TRUE/FALSE determines where
the comma characters appear on the RXDATA bus. When ALIGN_COMMA_MSB is set to
FALSE, the comma can appear in any byte lane of RXDATA in the 2- and 4-byte primitives.
When ALIGN_COMMA_MSB is set to TRUE, the comma appears in RXDATA[15:8] for the
2-byte primitives, and in either RXDATA[15:8] or RXDATA[31:24] for the 4-byte primitives.
(See “ALIGN_COMMA_MSB,” page 67.)
PARALLEL
LOOPBACK = 01
RX DESERIALIZER
TXP/TXN
SERIAL
LOOPBACK = 10
MUX
RXP/RXN
UG024_25_110503
In the case of a 4-byte primitive, the transceiver sets comma alignment with respect to its
2-byte internal data path, but it does not constrain the comma to appear only in
RXDATA[31:24]. Logic must be designed in the FPGA fabric to handle comma alignment
for the 32-bit primitives when implementing certain protocols. (Note that FPGA logic is not
required for 1-byte and 2-byte configurations.)
One such protocol is Fibre Channel. Delimiters such as IDLES, SOF, and EOF are four bytes
long, and are assumed by the protocol logic to be aligned on a 32-bit boundary. The Fibre
Channel IDLE delimiter is four bytes long and is composed of characters K28.5, D21.4,
D21.5, and D21.5. The comma, K28.5, is transmitted in TXDATA[31:24], which the protocol
logic expects to be received in RXDATA[31:24].
Using Table B-1, page 135, and Tabl e B-2 , pag e 143 , the IDLE delimiter can be translated
into a hexadecimal value 0xBC95B5B5 that represents the 32-bit RXDATA word. On the
32-bit RXDATA interface, the received word is either 32-bit aligned or misaligned, as
shown in Tab le 2 -2 4. In the table, “pp” indicates a byte from a previous word of data.
RocketIO™ Transceiver User Guidewww.xilinx.com93
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
Table 2-24: 32-bit RXDATA, Aligned versus Misaligned
Chapter 2: Digital Design Considerations
RXDATA
[31:24]
RXDATA
[23:16]
RXDATA
[15:8]
RXDATA
[7:0]
32-bit alignedBC95B5B5
CHARISCOMMA1000
32-bit misalignedppppBC95
CHARISCOMMA0010
When RXDATA is 32-bit aligned, the logic should pass RXDATA though to the protocol
logic without modification. A properly aligned data flow is shown in Figure 2-29.
TXDATA
RXDATA
ALIGNED_DATA
BC95B5B5
BC95B5B5
pppppppp
FDB53737
FDB53737
BC95B5B5
45674893
45674893
FDB53737
nnnnnnnnnnnnnnnn
nnnnnnnnnnnnnnnn
45674893
nnnnnnnn
ug024_33_091602
Figure 2-29: RXDATA Aligned Correctly
When RXDATA is 32-bit misaligned, the word requiring alignment is split between
consecutive RXDATA words in the data stream, as shown in Figure 2-30. (RXDATA_REG
in the figure refers to the design example code in “32-bit Alignment Design,” page 95.)
TXDATA
RXDATA
RXDATA_REG[15:0]
ALIGNED_DATA
BC95B5B5
pppp
pppp
pppppppp pppppppp
BC95
FDB53737
B5B5FDB5
BC95
45674893
37374567
FDB54567
BC95B5B5
nnnnnnnnnnnnnnnn
4893
nnnn
FDB53737
nnnnnnnn
nnnn
45674893
ug024_34_091602
Figure 2-30: Realignment of RXDATA
This conditional shift/delay operation on RXDATA also must be performed on the status
outputs RXNOTINTABLE, RXDISPERR, RXCHARISK, RXCHARISCOMMA, and
RXRUNDISP in order to keep them properly synchronized with RXDATA.
It is not possible to adjust RXCLKCORCNT appropriately for shifted/delayed RXDATA,
because RXCLKCORCNT is summary data, and the summary for the shifted case cannot
be recalculated.
94www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
Other Important Design Notes
32-bit Alignment Design
The following example code illustrates one way to create the logic to properly align 32-bit
wide data with a comma in bits [31:24] For brevity, most status bits are not included in this
example design; however, these should be shifted in the same manner as RXDATA and
RXCHARISK.
Note that when using a 40-bit data path (8B/10B bypassed), a similar realignment scheme
may be used, but it cannot rely on RXCHARISCOMMA for comma detection.
Verilo g
/*********************************************************************
*
* XILINX IS PROVIDING THIS DESIGN, CODE, OR INFORMATION “AS IS”
* AS A COURTESY TO YOU, SOLELY FOR USE IN DEVELOPING PROGRAMS AND
* SOLUTIONS FOR XILINX DEVICES. BY PROVIDING THIS DESIGN, CODE,
* OR INFORMATION AS ONE POSSIBLE IMPLEMENTATION OF THIS FEATURE,
* APPLICATION OR STANDARD, XILINX IS MAKING NO REPRESENTATION
* THAT THIS IMPLEMENTATION IS FREE FROM ANY CLAIMS OF INFRINGEMENT,
* AND YOU ARE RESPONSIBLE FOR OBTAINING ANY RIGHTS YOU MAY REQUIRE
* FOR YOUR IMPLEMENTATION. XILINX EXPRESSLY DISCLAIMS ANY
* WARRANTY WHATSOEVER WITH RESPECT TO THE ADEQUACY OF THE
* IMPLEMENTATION, INCLUDING BUT NOT LIMITED TO ANY WARRANTIES OR
* REPRESENTATIONS THAT THIS IMPLEMENTATION IS FREE FROM CLAIMS OF
* INFRINGEMENT, IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
* FOR A PARTICULAR PURPOSE.
*
* (c) Copyright 2002 Xilinx Inc.
* All rights reserved.
*
*********************************************************************/
// Virtex-II Pro RocketIO comma alignment module
//
// This module reads RXDATA[31:0] from a RocketIO transceiver
// and copies it to
// its output, realigning it if necessary so that commas
// are aligned to the MSB position
// [31:24]. The module assumes ALIGN_COMMA_MSB is TRUE,
// so that the comma
// is already aligned to [31:24] or [15:8].
//
// Outputs
//
// aligned_data[31:0] -- Properly aligned 32-bit ALIGNED_DATA
// sync -- Indicator that aligned_data is properly aligned
// aligned_rxisk[3:0] - properly aligned 4 bit RXCHARISK
// Inputs - These are all RocketIO inputs or outputs
// as indicated:
//
// usrclk2 -- RXUSRCLK2
// rxreset -- RXRESET
// rxisk[3:0] RXCHARISK[3:0]
// rxdata[31:0] RXDATA[31:0] -- (commas aligned to
// [31:24] or [15:8])
// rxrealign -- RXREALIGN
// rxcommadet -- RXCOMMADET
// This process maintains wait_to_sync and count,
// which are used only to
// maintain output sync; this provides some idea
// of when the output is properly
// aligned, with the comma in aligned_data[31:24]. The
// counter is set to a high value
// whenever the elastic buffer is reinitialized;
// that is, upon asserted RXRESET or
// RXREALIGN. Count-down is enabled whenever a
// comma is known to have
// come through the comma detection circuit,
// that is, upon an asserted RXREALIGN
// or RXCOMMADET.
always @ ( posedge usrclk2 )
begin
if ( rxreset )
begin
wait_to_sync <= 4'b1111;
count <= 1'b0;
end
else if ( rxrealign )
begin
wait_to_sync <= 4'b1111;
count <= 1'b1;
96www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
Other Important Design Notes
end
else
begin
if ( count && ( wait_to_sync != 4'b0000 ) )
wait_to_sync <= wait_to_sync - 4'b0001;
if ( rxcommadet )
count <= 1'b1;
end
end
// This process maintains output sync, which indicates
// when outgoing aligned_data
// should be properly aligned, with the comma in aligned_data[31:24].
// Output aligned_data is
// considered to be in sync when a comma is seen on
// rxdata (as indicated
// by rxchariscomma3 or 1) after the counter wait_to_sync
// has reached 0, indicating
// that commas seen by the comma detection circuit
// have had time to propagate to
// aligned_data after initialization of the elastic buffer.
R
always @ ( posedge usrclk2 )
begin
if ( rxreset | rxrealign )
sync <= 1'b0;
else if ( ( wait_to_sync == 4'b0000 ) &
( rxchariscomma3 | rxchariscomma1 ) )
sync <= 1'b1;
end
// This process generates aligned_data with commas aligned in [31:24],
// assuming that incoming commas are aligned to [31:24] or [15:8].
// Here, you could add code to use ENPCOMMAALIGN and
// ENMCOMMAALIGN to enable a move back into the byte_sync=0 state.
always @ ( posedge usrclk2 or posedge rxreset )
begin
if ( rxreset )
begin
rxdata_reg <= 16'h0000;
aligned_data <= 32'h0000_0000;
rxisk_reg <= 2'b00;
aligned_rxisk <= 4'b0000;
byte_sync <= 1'b0;
end
else
begin
rxdata_reg[15:0] <= rxdata[15:0];
rxisk_reg[1:0] <= rxisk[1:0];
if ( rxchariscomma3 )
begin
aligned_data[31:0] <= rxdata[31:0];
aligned_rxisk[3:0] <= rxisk[3:0];
byte_sync <= 1'b0;
end
else
if ( rxchariscomma1 | byte_sync )
begin
RocketIO™ Transceiver User Guidewww.xilinx.com97
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
aligned_data[31:0] <= { rxdata_reg[15:0], rxdata[31:16]
};
aligned_rxisk[3:0] <= { rxisk_reg[1:0], rxisk[3:2] };
byte_sync <= 1'b1;
end
else
begin
aligned_data[31:0] <= rxdata[31:0];
aligned_rxisk <= rxisk;
end
end
end
-- Inputs - These are all RocketIO inputs or outputs
-- as indicated:
--
98www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
Other Important Design Notes
-- usrclk2 -- RXUSRCLK2
-- rxreset -- RXRESET
-- rxdata[31:0] RXDATA[31:0] -- (commas aligned to
-- [31:24] or [15:8])
-- rxisk[3:0] - RXCHARISK[3:0]
-- rxrealign -- RXREALIGN
-- rxcommadet -- RXCOMMADET
-- rxchariscomma3 -- RXCHARISCOMMA[3]
-- rxchariscomma1 -- RXCHARISCOMMA[1]
-LIBRARY IEEE;
USE IEEE.std_logic_1164.all;
use IEEE.STD_LOGIC_ARITH.ALL;
use IEEE.Numeric_STD.all;
use IEEE.STD_LOGIC_UNSIGNED.ALL;
ENTITY align_comma_32 IS
PORT (
aligned_data : OUT std_logic_vector(31 DOWNTO 0);
aligned_rxisk : OUT std_logic_vector(3 DOWNTO 0);
sync : OUT std_logic;
usrclk2 : IN std_logic;
rxreset : IN std_logic;
rxdata : IN std_logic_vector(31 DOWNTO 0);
rxisk : IN std_logic_vector(3 DOWNTO 0);
rxrealign : IN std_logic;
rxcommadet : IN std_logic;
rxchariscomma3 : IN std_logic;
rxchariscomma1 : IN std_logic);
END ENTITY align_comma_32;
R
ARCHITECTURE translated OF align_comma_32 IS
SIGNAL rxdata_reg : std_logic_vector(15 DOWNTO 0);
SIGNAL rxisk_reg : std_logic_vector(1 DOWNTO 0);
SIGNAL byte_sync : std_logic;
SIGNAL wait_to_sync : std_logic_vector(3 DOWNTO 0);
SIGNAL count : std_logic;
SIGNAL rxdata_hold : std_logic_vector(31 DOWNTO 0);
SIGNAL rxisk_hold : std_logic_vector(3 DOWNTO 0);
SIGNAL sync_hold : std_logic;
BEGIN
aligned_data <= rxdata_hold;
aligned_rxisk <= rxisk_hold;
sync <= sync_hold;
-- This process maintains wait_to_sync and count,
-- which are used only to
-- maintain output sync; this provides some idea
-- of when the output is properly
-- aligned, with the comma in aligned_data[31:24].
-- The counter is set to a high value
-- whenever the elastic buffer is reinitialized;
-- that is, upon asserted RXRESET or
-- RXREALIGN. Count-down is enabled whenever a
-- comma is known to have
-- come through the comma detection circuit, that
RocketIO™ Transceiver User Guidewww.xilinx.com99
UG024 (v3.0) February 22, 2007
Product Not Recommendedfor NewDesigns
R
-- is, upon an asserted RXREALIGN
-- or RXCOMMADET.
PROCESS (usrclk2)
BEGIN
IF (usrclk2'EVENT AND usrclk2 = '1') THEN
IF (rxreset = '1') THEN
wait_to_sync <= “1111”;
count <= '0';
ELSE
IF (rxrealign = '1') THEN
wait_to_sync <= “1111”;
count <= '1';
ELSE
IF (count = '1') THEN
IF (wait_to_sync /= “0000”) THEN
wait_to_sync <= wait_to_sync - “0001”;
END IF;
END IF;
IF (rxcommadet = '1') THEN
count <= '1';
END IF;
END IF;
END IF;
END IF;
END PROCESS;
Chapter 2: Digital Design Considerations
-- This process maintains output sync, which
-- indicates when outgoing aligned_data
-- should be properly aligned, with the comma
-- in aligned_data[31:24]. Output aligned_data is
-- considered to be in sync when a comma is seen
-- on rxdata (as indicated
-- by rxchariscomma3 or 1) after the counter
-- wait_to_sync has reached 0, indicating
-- that commas seen by the comma detection circuit
-- have had time to propagate to
-- aligned_data after initialization of the elastic buffer.
PROCESS (usrclk2)
BEGIN
IF (usrclk2'EVENT AND usrclk2 = '1') THEN
IF ((rxreset OR rxrealign) = '1') THEN
sync_hold <= '0';
ELSE
IF (wait_to_sync = “0000”)THEN
IF ((rxchariscomma3 OR rxchariscomma1) = '1') THEN
sync_hold <= '1';
END IF;
END IF;
END IF;
END IF;
END PROCESS;
-- This process generates aligned_data with commas
-- aligned in [31:24],
-- assuming that incoming commas are aligned
-- to [31:24] or [15:8].
-- Here, you could add code to use ENPCOMMAALIGN and
-- ENMCOMMAALIGN to enable a move back into the
100www.xilinx.comRocketIO™ Transceiver User Guide
UG024 (v3.0) February 22, 2007
Loading...
+ hidden pages
You need points to download manuals.
1 point = 1 manual.
You can buy points or you can get point for every manual you upload.