Sprabi 2 D

Application Report
SPRABI2D – November 2010 – Revised March 2019
Hardware design guide for KeyStone™ I devices
ABSTRACT
This application report describes hardware system design considerations for the KeyStone I family of
processors. This document is intended to be used as an aid during the development of application
hardware. Other aids including, but not limited to, device-specific data manuals and explicit collateral
should also be used. Where discrepancies occur between the data manual and this application report,
always defer to the latest device-specific data manual.
Contents
1 Introduction ................................................................................................................... 3
2 Power Supplies .............................................................................................................. 5
3 Clocking ..................................................................................................................... 24
4 JTAG ......................................................................................................................... 38
5 Device Configurations and Initialization ................................................................................. 42
6 Peripherals Section ........................................................................................................ 45
7 I/O Buffers and Termination .............................................................................................. 64
8 Power Saving Modes ...................................................................................................... 67
9 Mechanical .................................................................................................................. 70
10 Routing Guidelines ......................................................................................................... 72
11 Simulation and Modeling .................................................................................................. 78
12 References .................................................................................................................. 79
Appendix A ....................................................................................................................... 81
Appendix B ....................................................................................................................... 92
List of Figures
1 KeyStone I Power Supply Planes (Rails) ................................................................................. 6
2 LM10011 CVDD Supply Block Diagram .................................................................................. 9
3 VCNTL Level Translation ................................................................................................. 10
4 DDR3 VREFSSTL Voltage Divider ...................................................................................... 12
5 AVDDAx Power Supply Filter ............................................................................................. 14
6 VDDRn Power Supply Filter .............................................................................................. 14
7 VDDTn Power Supply Filter............................................................................................... 15
8 ESR Plot for Delta Current to Tolerance #1 ............................................................................ 17
11 LJCB LVDS Clock Source ................................................................................................ 28
12 SerDes Reference Clock Jitter Masks ................................................................................... 30
13 Phase Jitter Template ..................................................................................................... 31
14 Unused Clock Input Connection.......................................................................................... 33
15 Clock Fan Out - for Single Device ....................................................................................... 35
16 Clock Fan Out - Multiple DSPs ........................................................................................... 36
17 Clock Termination Location ............................................................................................... 37
18 Emulator With Trace Solution #1 ......................................................................................... 40
19 Emulator With Trace Solution #2 ......................................................................................... 40
SPRABI2D – November 2010 – Revised March 2019 Hardware design guide for KeyStone™ I devices 1
Submit Documentation Feedback
Copyright © 2010–2019, Texas Instruments Incorporated
www.ti.com
20 Emulation Voltage Translation With Buffers ............................................................................ 41

21 Emulator Voltage Translation With Switches ........................................................................... 41
22 HyperLink Bus DSP-to-DSP Connection ................................................................................ 48
23 I2C Voltage Level Translation ............................................................................................ 50
24 SGMII MAC to MAC Connection ......................................................................................... 52
25 Unused LVDS Inputs ...................................................................................................... 58
26 UART Connections With Autoflow Support ............................................................................. 60
27 SPI Connections............................................................................................................ 61
28 LJCB Clock Inputs ......................................................................................................... 65
29 Basic Series Termination.................................................................................................. 66
30 Non-Solder-Mask Defined (NSMD) PCB Land ......................................................................... 70
31 Filtered Voltage Plane ..................................................................................................... 73
32 Stripline and Microstrip Topologies ...................................................................................... 74
33 Trace Width to Current Specifications ................................................................................... 78
34 Phase Noise Plot - CECE62002/5 @ 40 MHz ......................................................................... 81
35 Phase Noise Plot - CECE62002/5 @ 66.667 MHz .................................................................... 82
40 Reflections Timing Analysis............................................................................................... 89
41 Transmission Line Geometries ........................................................................................... 93
42 Transmission Line Geometries ........................................................................................... 94
43 Microstrip Transmission Line Construction ............................................................................. 94
44 Stripline Transmission Line Construction ............................................................................... 96
45 Mitred Bend ................................................................................................................. 97
List of Tables
1 SmartReflex VID Value Mapping ......................................................................................... 7
2 SmartReflex VID Value Mapping for 1400 MHz Devices ............................................................... 8
3 Power Rail Filter Recommendations .................................................................................... 13
4 Bulk, Intermediate, and Bypass Capacitor Recommendations (Absolute Minimums) ........................... 20
5 Total Decoupling Capacitance............................................................................................ 22
6 KeyStone I System PLL Clock Inputs ................................................................................... 24
7 KeyStone I SerDes PLL Reference Clock Inputs ..................................................................... 25
8 Clocking Requirements - Slew Rate, Duty Cycle, Accuracy.......................................................... 26
9 Clocking Requirements - Clock Source and Termination ............................................................. 27
10 Phase Jitter Template Endpoints ........................................................................................ 31
11 Phase Noise Mask Integrated Values ................................................................................... 32
12 Mask and Phase Noise Adjustments .................................................................................... 32
13 Mask and Phase Noise Adjustments .................................................................................... 32
14 Managing Unused Clock Inputs ......................................................................................... 34
15 Clock Sequencing ......................................................................................................... 38
16 GPIO Pin Strapping Configurations ...................................................................................... 46
17 HyperLink Rate Scale Values ............................................................................................ 49
18 HyperLink PLL Multiplier Settings ........................................................................................ 49
19 SGMII Rate Scale Values ................................................................................................. 51
20 SGMII PLL Multiplier Settings ............................................................................................ 51
21 SRIO Rate Scale Values ................................................................................................. 53
2 Hardware design guide for KeyStone™ I devices SPRABI2D – November 2010 – Revised March 2019
www.ti.com Introduction
22 SRIO PLL Multiplier Settings ............................................................................................. 54
23 PCIe (GPIO) Configuration Table ........................................................................................ 55
24 PCIe Rate Scale Values .................................................................................................. 56
25 PCIE PLL Multiplier Settings ............................................................................................. 56
26 AIF SerDes Clocking Options ............................................................................................ 57
27 AIF2 Timer Module Configuration Options .............................................................................. 57
28 Slew Rate Control .......................................................................................................... 59
29 Unused UART Connections .............................................................................................. 60
30 Unused SPI Connections ................................................................................................. 62
31 Standby Peripheral Status ............................................................................................... 67
32 Hibernation Mode Peripheral Status .................................................................................... 68
33 Maximum Device Compression - Leaded Solder Balls ............................................................... 72
34 Maximum Device Compression - Lead-Free Solder Balls ............................................................ 72
35 Copper Weight and Thickness ........................................................................................... 77
36 Reflections in ps ........................................................................................................... 87
37 Possible First 50 Reflection Combinations ............................................................................. 90
38 Microstrip Signal Delay Through Various Materials ................................................................... 95
39 Stripline Signal Delay Through Various Materials ..................................................................... 97
Trademarks
KeyStone, SmartReflex are trademarks of Texas Instruments.
All other trademarks are the property of their respective owners.
1 Introduction
This document provides a starting point for the engineer designing with one of the KeyStone I devices. It
shows a road map for the design effort and highlights areas of significant importance that must be
addressed. This document does not contain all the information that is needed to complete the design. In
many cases, it refers to the device-specific data manuals or to various user's guides for specific
information.
This application report is generic to the entire family of KeyStone I devices, and as such, may include
information for subsystems that are not present on all devices. Designers should begin by reviewing the
device-specific data manual for their intended device to determine which sections of this document are
relevant.
This document is organized in a sequential manner. It moves from decisions that must be made in the
initial planning stages of the design, through the selection of support components, to the mechanical,
electrical, and thermal requirements. For the greatest success, each of the issues discussed should be
resolved before moving to the next section.
1.1 Before Getting Started

The KeyStone I family of processors provides a wide variety of capabilities, not all of which will be used in
every design. Consequently, the requirements for different designs using the same device can vary widely
depending on how that part is used. Many designers simply copy the evaluation module (EVM) for the
device without determining their requirements. While the EVM is a good example, it is not a reference
design and may not be the optimum design for every customer. You will need to understand your
requirements before determining the details of the design. In addition, your design may require additional
circuitry to operate correctly in the target environment. Take some time to review the data manual for your
KeyStone I device and determine the following:
• Which peripherals will be used to move data in and out of the processor
• What is the speed and organization of the DDR3 memory interface that will be used
• How much processing will each of the cores in your KeyStone I device be performing
• How you will boot the KeyStone I device
• What are the expected environmental conditions for your KeyStone I device
Introduction www.ti.com
1.2 Design Documentation

Periodically throughout this document, it is recommended that a design document be generated based on
your requirements. Generating and storing this information provides you with the foundation for your
documentation package; this design document will be needed if you are seeking support from TI.
Examples of many of these can be found in the schematic package provided with the EVM boards for your
KeyStone I device.
1.3 Acronyms Used in This Document
Acronym Description
AIF Antenna Interface
AMI IBIS Algorithmic Modeling Interface
BGA Ball Grid Array
CML Current Mode Logic, I/O type
Data Manual Also referred to as the Data Sheet
DDR3 Double Data Rate 3 (SDRAM Memory)
DSP Digital Signal Processor
EMIF External Memory Interface
EVM Evaluation Module
FC-BGA Flip-Chip BGA
GPIO General-Purpose I/O
I2C Inter-IC Control Bus
IBIS Input Output Buffer Information Specification, or ANSI/EIA-656-A
IO Input/Output
JEDEC Joint Electronics Device Engineering Council
LJCB Low Jitter Clock Buffer: Differential clock input buffer type, compatible with LVDS and LVPECL
LVDS Low Voltage Differential Swing, I/O type
McBSP Multi-Channel Buffered Serial Port
MDIO Management Data Input/Output
NSMD Non-Solder Mask Defined BGA Land
OBSAI Open Base Station Architecture Initiative
PCIe Peripheral Component Interconnect Express
PHY Physical Layer of the Interface
PTV Process/Temperature / Voltage
RIO Rapid IO, also referred to as SRIO
Seating Plane The maximum compression depth of the BGA for a given package design
SerDes Serializer/De-Serializer
SGMII Serial Gigabit Media Independent Interface
SPI Synchronous Serial Input/Output (port)
SRIO Serial RapidIO
TBD To Be Determined. Implies something is currently under investigation and will be clarified in a later
version of the specification.
UART Universal Asynchronous Receiver/Transmitter
UI Unit Interval
XAUI 10 Gigabit (X) Attachment Unit Interface standard
www.ti.com Power Supplies
2 Power Supplies
The first requirement for a successful design is to determine the power needs for your KeyStone device.
All KeyStone I devices operate with four main voltage levels requiring four power supply circuits. Some
devices will require an additional voltage level, requiring a fifth power supply. To determine all the voltages
needed, check the device-specific data manual.
2.1 Determining Your Power Requirements

The maximum and minimum current requirements for each of these voltage rails are not found in the
device-specific data manual for the part. These requirements are highly application-dependent and must
be calculated for your specific product. There is a power consumption model in the product folder for each
KeyStone I device. You must use the model for the specific KeyStone I device that you have
selected to get accurate results.
There is a link in the model spreadsheet (http://www.ti.com/lit/zip/sprm545) to Power Consumption
Summary for KeyStone C66x Devices that explains how to populate the necessary parameters. Review
this application report carefully and determine the values needed for your application. Once you have
entered these values, you will have the maximum power requirement for each of the four voltage rails.
These values are divided into activity and baseline components. The baseline power portion is associated
with leakage, clock tree, and phase-locked loop (PLL) power. This value will not change based on the
processor utilization. The activity power reflects the additional power required due to the processing load
defined in the spreadsheet. The baseline plus the activity defines your maximum power requirement.
Power supplies should be designed to transition from the baseline level to the baseline + activity level
within one CPU clock.
Note that power consumption can vary greatly based on process and temperature. The values provided by
the spreadsheet are the maximum average power consumption for all production devices. Measurements
of power consumption on a single device may be considerably lower than the values provided by the
model, but may not be representative of a wide sample of devices. To ensure that your power supply
design is adequate, use the values provided by the model.
The power consumption model populated with the values for your application should be saved
with your design documentation.
Power Supplies www.ti.com
2.2 Power Rails

The power consumption model provides the requirements for four separate voltage rails as shown in
Figure 1.
Primary Voltage Source

KeyStone I
Device
Bulk Caps
CVDD
Adjustable Core
Supply Bulk Caps Bypass Caps
Bypass Caps
CVDD1
Bulk Caps Bypass Caps
Filter VDDT1
Fixed Core Bypass Caps
Supply (1.0-V)
Bypass Caps
Filter VDDTn
Bypass Caps
DVDD18
Ferrite AVDDA1
Fixed 1.8-V Bypass Caps
Supply
Bypass Caps
Ferrite AVDDAn
Bypass Caps
DVDD15
Filter VDDR1
Fixed 1.5-V Bypass Caps
Supply
Bypass Caps
Filter VDDRn
Bypass Caps
DVDD15
VREFSSTL
Bypass Caps
DDR3
Bypass Caps Termination VREF for
Supply Bypass Caps DDR3 SDRAMs
VTT for
Bypass Caps DDR3 SDRAMs
Figure 1. KeyStone I Power Supply Planes (Rails)
These four rails (CVDD, CVDD1, 1.8 V, and 1.5 V) provide the basic power requirement for all KeyStone I
devices. In addition, each requires a number of filtered connections to the main rails to provide power for
noise-sensitive portions of the device. These include the VDDRn, VDDTn, and AVDDAn pins. The power
requirements for each of the filtered voltages are included in the values calculated by the power
consumption models.
Figure 1 includes a DDR3 termination power supply. DDR3 address and command signals are terminated
using a source and sink-tracking LDO designed to provide VTT and a low-noise reference. Figure 1 also
includes this power supply as a source for the VREFSSTL reference voltage.
Each of these voltage rails and filtered supplies are described in the following sections.
2.2.1 CVDD
CVDD is the adjustable supply used by the core logic for the KeyStone I device. KeyStone I devices use
adaptive voltage scaling (AVS) to compensate for variations in performance from die to die, and from
wafer to wafer. The actual voltage for CVDD can vary across the range specified in the device-specific
data manual and will be different for each device. Each KeyStone I device needs an independent power
supply to generate the CVDD voltage needed by that device. That CVDD power supply must use a
SmartReflex™-compliant circuit. SmartReflex is described in Section 2.2.1.1. This power supply circuit
must initialize to a level of 1.1 V and then adjust to the voltage requested by the SmartReflex circuit in the
KeyStone I device.
2.2.1.1 SmartReflex
To reduce device power consumption, SmartReflex allows the core voltage to be optimized (scaled) based
on the process corners of each device. KeyStone I devices use Class 0 SmartReflex. This class of
operation uses a code for a single ideal voltage level determined during manufacturing tests. At the end of
these tests, the code for the lowest acceptable voltage (while still meeting all performance requirements)
is established and permanently programmed into each die. This 6-bit code, referred to as the VID value,
represents the optimal fixed voltage level for that device.
Table 1 shows the voltage level associated with each of the possible 6-bit VID values.
Table 1. SmartReflex VID Value Mapping

VID No VID (5:0) CVDD VID No VID (5:0) CVDD VID No VID (5:0) CVDD VID No VID (5:0) CVDD
0 00 0000b 0.700 16 01 0000b 0.802 32 10 0000b 0.905 48 11 0000b 1.007
1 00 0001b 0.706 17 01 0001b 0.809 33 10 0001b 0.911 49 11 0001b 1.014
2 00 0010b 0.713 18 01 0010b 0.815 34 10 0010b 0.918 50 11 0010b 1.020
3 00 0011b 0.719 19 01 0011b 0.822 35 10 0011b 0.924 51 11 0011b 1.026
4 00 0100b 0.726 20 01 0100b 0.828 36 10 0100b 0.930 52 11 0100b 1.033
5 00 0101b 0.732 21 01 0101b 0.834 37 10 0101b 0.937 53 11 0101b 1.039
6 00 0110b 0.738 22 01 0110b 0.841 38 10 0110b 0.943 54 11 0110b 1.046
7 00 0111b 0.745 23 01 0111b 0.847 39 10 0111b 0.950 55 11 0111b 1.052
8 00 1000b 0.751 24 01 1000b 0.854 40 10 1000b 0.956 56 11 1000b 1.058
9 00 1001b 0.758 25 01 1001b 0.860 41 10 1001b 0.962 57 11 1001b 1.065
10 00 1010b 0.764 26 01 1010b 0.866 42 10 1010b 0.969 58 11 1010b 1.071
11 00 1011b 0.770 27 01 1011b 0.873 43 10 1011b 0.975 59 11 1011b 1.078
12 00 1100b 0.777 28 01 1100b 0.879 44 10 1100b 0.982 60 11 1100b 1.084
13 00 1101b 0.783 29 01 1101b 0.886 45 10 1101b 0.988 61 11 1101b 1.090
14 00 1110b 0.790 30 01 1110b 0.892 46 10 1110b 0.994 62 11 1110b 1.097
15 00 1111b 0.796 31 01 1111b 0.898 47 10 1111b 1.001 63 11 1111b 1.103
NOTE: Not all ranges or voltage levels are supported by every KeyStone I device. The intended
range of operation is defined in the device-specific data manual. Operation outside that
range may impact device reliability or performance.
An advanced speed class was added for some devices that require an extra 50mV on the CVDD power
supply. To see whether the speed grade of the device implemented needs this increased voltage, see the
device-specific data manual.
Table 2. SmartReflex VID Value Mapping for 1400 MHz Devices

VID No VID (5:0) CVDD VID No VID (5:0) CVDD VID No VID (5:0) CVDD VID No VID (5:0) CVDD
0 00 0000b 0.750 16 01 0000b 0.852 32 10 0000b 0.955 48 11 0000b 1.057
1 00 0001b 0.756 17 01 0001b 0.859 33 10 0001b 0.961 49 11 0001b 1.064
2 00 0010b 0.763 18 01 0010b 0.865 34 10 0010b 0.968 50 11 0010b 1.070
3 00 0011b 0.769 19 01 0011b 0.872 35 10 0011b 0.974 51 11 0011b 1.076
4 00 0100b 0.776 20 01 0100b 0.878 36 10 0100b 0.980 52 11 0100b 1.083
5 00 0101b 0.782 21 01 0101b 0.884 37 10 0101b 0.987 53 11 0101b 1.089
6 00 0110b 0.788 22 01 0110b 0.891 38 10 0110b 0.993 54 11 0110b 1.096
7 00 0111b 0.795 23 01 0111b 0.897 39 10 0111b 1.000 55 11 0111b 1.102
8 00 1000b 0.801 24 01 1000b 0.904 40 10 1000b 1.006 56 11 1000b 1.108
9 00 1001b 0.808 25 01 1001b 0.910 41 10 1001b 1.012 57 11 1001b 1.115
10 00 1010b 0.814 26 01 1010b 0.916 42 10 1010b 1.019 58 11 1010b 1.121
11 00 1011b 0.820 27 01 1011b 0.923 43 10 1011b 1.025 59 11 1011b 1.128
12 00 1100b 0.827 28 01 1100b 0.929 44 10 1100b 1.032 60 11 1100b 1.134
13 00 1101b 0.833 29 01 1101b 0.936 45 10 1101b 1.038 61 11 1101b 1.140
14 00 1110b 0.840 30 01 1110b 0.942 46 10 1110b 1.044 62 11 1110b 1.147
15 00 1111b 0.846 31 01 1111b 0.948 47 10 1111b 1.051 63 11 1111b 1.153
NOTE: These increased voltages must not be used on a device not rated for this as it will cause the
device to operate outside its rated current and thermal ranges that may result in reduced
reliability.
2.2.1.2 SmartReflex VCNTL Interface

The VCNTL[3:0] interface is used to transmit the 6-bit VID value to the SmartReflex power supply circuit.
The VCNTL[3:0] pins are open-drain IOs that require 4.7-kΩ pull-up resistors to the DVDD18 rail of the
KeyStone I device. The VCNTL[3:0] pins are used for a 4-pin, 6-bit, dual-phase interface to provide the 6-
bit VID value to the power supply circuit. In 4-pin, 6-bit dual-phase mode, VCNTL[3] acts as a command
signal while VCNTL[2:0] act as a 3-bit data bus. VCNTL[3] will transition low and then high to indicate the
presence of the lower three bits of the VID value, followed by the upper three bits.
The timing of this interface can be found in the device-specific data manual. It is important to note that the
VID values are not latched on the falling and rising edge of the VCNTL[3] signal. The VID values will be
available only after a specified amount of time following the transition, as defined in the device-specific
data manual. As stated previously, this transition will occur only a single time after a power-up reset has
occurred. The connection of the VCNTL interface to the power supply circuit is discussed in later sections.
2.2.1.3 Supported Power Supply Solutions

Texas Instruments provides a number of power supply solutions for CVDD, which fall into one of two
classes. The first uses the LM10011 to interface the VCNTL pins with a number of analog power supply
controllers. The second is a solution based on the UCD92xx family of digital power supply controllers.
Texas Instruments provides reference designs for both possible power supply solutions. These can be
accessed through the power management page on TI.com:
• http://www.ti.com/processorpower
Choose C6000 from the processor type pull down menu and then choose the KeyStone I part number
from the processor family pull down menu. This brings you to a page containing the documentation for the
possible power supply solutions that are supported for that part. The page includes power supply
requirements. These are generally the requirements calculated for the EVM platform and do not reflect the
maximum requirements for the part. Your requirements should be calculated using the Power
Consumption Model for your KeyStone I device.
The power management solutions are managed by a different group in TI. Questions on the power supply
solutions should be directed to the appropriate power group. Questions about the LM10011-based solution
can be posted to the non-isolated DC/DC forum and questions about the UCD92xx-based solutions can
be posted to the Digital Power Forum. If you need additional information, contact your local Texas
Instruments FAE.
2.2.1.4 LM10011 and Analog Controller Solutions

The LM10011 captures the VID value presented on the VCNTL interface and uses it to control the voltage
level of a DC/DC converter. The VID value is used to define the output of a current DAC that is connected
to the feedback pin of a DC/DC switching regulator. This adjusts the output voltage of the regulator to the
level required by the KeyStone I device. The LM10011 can be used to control any DC/DC regulator with a
compatible feedback circuit. The current requirements calculated using the power consumption model
should be used to determine the best device for your application.
A block diagram of how the LM10011 VID interface should be connected to a switching power supply like
the TPS56121 is shown in Figure 2.
CVDD SW VIN Vin
DC/DC
FB
COMP
DVDD18
IDAC_OUT VIN
CVDD
Keystone I LM10011
Device
VCNTL0 VIDA SET

VCNTL1 VIDB
VCNTL2 VIDC RSET
VCNTL3 VIDS
Figure 2. LM10011 CVDD Supply Block Diagram
VIDA, VIDB, VIDC, and VIDS are connected to VCNTL0, VCNTL1, VCNTL2, and VCNTL3, respectively.
The LM10011 uses a supply voltage in a range from 3 V to 5.5 V. However, the VID inputs are compatible
with the 1.8-V logic levels used by the KeyStone I device and no voltage translator is needed. Remember
that the VCNTL outputs are open-drain and require a 4.7-kΩ pull-up resistor to the KeyStone I device
DVDD18 rail.
2.2.1.5 UCD92xx Digital Controller Solutions

The UCD9222 and the UCD9244 are synchronous buck digital PWM controllers designed for DC/DC
power applications. The UCD9222 provides control for two separate voltage rails and the UCD9244 can
control four rails. The controller can be programmed to support either an AVS voltage rail controlled by the
KeyStone I device or a fixed voltage rail. Because they include multiple VID interfaces that are compatible
with the timing for the KeyStone I VCNTL interface, the UCD92xx can be used to control the AVS voltage
for multiple KeyStone I devices if more then one is used.
The UCD92xx controller is designed to be paired with UCD7xxx family synchronous buck power drivers.
One UCD7xxx is needed for each voltage rail controlled by the UCD92xx device, therefore, the UCD9222
requires two UCD7xxx devices and the UCD9244 requires four UCD7xxx devices. Each device in the
UCD7xxx family supports different maximum current levels. The power supply solutions link will help you
select the best component for your application.
For additional details, see the device-specific data sheets, application reports, and user's guides. TI
continues to evolve its power supply support for the devices. For additional details and updates, check the
product folder, updates to this document, and your local FAE.
The UCD92xx components are highly programmable, allowing you to optimize the operation of your
supply using the Fusion Digital Power Designer software. The software may be downloaded from:
http://focus.ti.com/docs/toolsw/folders/print/fusion_digital_power_designer.html
This Windows-based graphical user interface (GUI) allows the designer to configure the system operating
parameters and store the configuration into non-volatile memory. Examples of the XML configuration files
for the device-specific KeyStone I device EVMs can be found in the technical documents section for the
EVM: http://www.ti.com/tool/tmdsevm6678. Note that these files are specific to the EVM design and
require modifications depending upon the components selected. Communication between the Fusion
software and the UCD92xx is achieved using a three-wire PMBUS interface. A connector for this interface
should be included in your design to allow programming and monitoring of the operation of the UCD92xx.
The Fusion software uses a USB-to-GPIO evaluation pod to facilitate this communication.
Information for this pod can be found at:
http://www.ti.com/tool/usb-to-gpio
2.2.1.5.1 UCD92xx Communications

The KeyStone I device communicates the VID value to the UCD92xx device using the VCNTL interface.
The VCNTL operates at 1.8 V and the UCD92xx VID interface operates at 3.3 V, requiring a voltage
translator. The voltage translator is shown in with VIDA, VIDB, VIDC, and VIDS on the 3.3-V side
connected to VCNTL0, VCNTL1, VCNTL2, and VCNTL3 on the 1.8-V side, respectively. The open-drain
VCNTL interface requires 4.7-kΩ pull-up resistors be connected to DVDD18.
DVDD18 3.3V
DVDD18 3.3V
74AVC4T245
PWR
VCCB VCCA
DVDD18
Power Good
Keystone I UCD92XX
OE DIR
Device
VCNTL0 1B1 1A1 VIDA

VCNTL1 1B2 1A2 VIDB
VCNTL2 2B1 2A1 VIDC
VCNTL3 2B2 2A2 VIDS
Figure 3. VCNTL Level Translation
To avoid any glitch on the VID interface, 4.7-kΩ pull-up resistors connected to 3.3 V should be placed on
the 3.3-V side of the converter and the converter output enable should be held high until DVDD18 has
reached a valid level. If the output enable is not held off until DVDD18 has stabilized, the converter will
drive the VID interface low, followed by a transition to high as DVDD18 ramps to 1.8 V. This causes an
invalid voltage code to be latched into the UCD92xx and may prevent the KeyStone I device from booting
correctly.
The UCD92xx components can be used to control multiple rails. A rail used for the CVDD supply should
be configured to monitor the VID interface. This is done using the Fusion Digital Power Designer software.
In the configuration section, there is a VID Config tab. In this tab, select the VID mode labeled 6-Bit VID
Code via VID Pins. Set the VID Vout low to 0.7 V and the VID Vout high to 1.103 V. Note that this defines
the full range of VID values shown in Table 1 and not the CVDD voltage range of the KeyStone device as
defined in the device-specific data manual. These values must be used to ensure that the correct voltage
is presented for the VID code received. Set the VID Code Init value to 63, which will provide an initial
voltage of 1.103 V. This voltage is generated until the KeyStone I device presents the required VID value
on the VCNTL interface.
2.2.1.6 CVDD for Designs With Multiple KeyStone I Devices

As stated previously, each KeyStone I device must have a separate power supply voltage for CVDD even
if there are multiple KeyStone I devices on the same board. If a single device is used, then the LM10011-
based solution or the UCD9222 can be used. The UCD9222 is a dual controller but the other digital
controller in the device can be used to generate one of the other voltages needed by the KeyStone I
device. On the KeyStone I EVM platforms, the second controller is used to generate the CVDD1 1-V fixed
voltage. If your design includes multiple KeyStone I devices, you have the option of using any of the
recommended power supply designs. A single UCD9222 has two independent VID interfaces and can be
used to generate the CVDD voltage for two separate KeyStone I devices. The UCD9244 includes four VID
interfaces and can be used for up to four KeyStone I devices.
2.2.2 CVDD1
CVDD1 is the fixed 1-V supply for the internal memory arrays and as a termination voltage for the SerDes
interfaces. This supply must meet the requirements for stability specified in the device-specific data
manual. This supply cannot be connected to the AVS supply used for CVDD and must be generated by a
separate power circuit. If multiple KeyStone I devices are used in your design, a single CVDD1 supply
may be used for all of the devices as long as it is scaled to provide the current needed.
2.2.2.1 VDDTn Filters

The VDDTn pins provide the analog termination voltage for the SerDes interfaces in the device. The
VDDTn pins are grouped into two or more VDDTn groups. The number of groups varies from device to
device. Each group of pins provides the termination voltage for one or more SerDes interfaces. If any of
the SerDes interfaces in a group are implemented, then the VDDTn group should be connected to the
fixed 1.0-V supply through a capacitive filter. The number of bypass capacitors between the filter and the
VDDTn pins will vary depending on the number of pins on the device. If none of the SerDes interfaces in a
group are used, then the VDDTn pins can be connected directly to the CVDD1 supply.
2.2.3 DVDD18
DVDD18 is the 1.8-V supply for the LVCMOS buffers and for the PLLs. This supply must meet the
requirements for stability shown in the device-specific data manual. If multiple KeyStone I devices are
used in your design, a single 1.8-V supply may be used for all of the devices as long as it is scaled to
provide the current needed.
2.2.3.1 AVDDAn
The AVDDA pins provide the supply voltage for the PLL modules in the KeyStone I device. The number of
AVDDA pins will vary from device to device depending on the number of PLLs present. Each AVDDA pin
should be connected to DVDD18 through a filter circuit. A separate filter should be used for each pin. For
details on the filters, see section Section 2.3. Note that the power estimate generated by the power
consumption model includes the power for the AVDDA pins in the total for DVDD18. A current limit value
of 50 mA can be used as for each AVDDA pin when selecting filter components. Recommendations for
the filter can be found later in this document.
2.2.4 DVDD15
DVDD15 is the 1.5-V supply for the DDR3 IO buffers and for the SerDes regulation in the KeyStone I
device. This supply must meet the requirements for stability shown in the device-specific data manual. If
multiple KeyStone I devices are used in your design, a single 1.5-V supply may be used for all of the
devices as long as it is scaled to provide the current needed.
2.2.4.1 VDDRn
The VDDR pins provide the supply voltage for the SerDes modules in the device. The number of VDDR
pins will vary from device to device depending on the number of SerDes interfaces present. Each VDDR
pin should be connected to DVDD15 through a filter circuit. A separate filter should be used for each pin.
For details on the filter, see Section 2.3. Note that the power estimate generated by the power
consumption model includes the power for the VDDR pins in the total for DVDD15. A current limit value of
50 mA can be used for each VDDR pin when selecting filter components. Recommendations for the filter
can be found later in this document.
2.2.5 Other Supplies

The four supplies listed above provide the majority of the power required for KeyStone I devices, however,
your design may require additional supplies. Information for these supplies is provided here to advise the
designer of their necessity.
2.2.5.1 VTT Termination Supply

The DDR3 interface requires a VTT termination at the end of the fly-by chain for the DDR3 address and
control signals. The 0.75-V termination voltage is generated using a special push/pull termination regulator
specifically designed to meet the VTT requirements for DDR3. The Texas Instruments TPS51200 is one
device that meets these requirements. Although the VTT supply is not connected to the KeyStone I
devices, it is necessary for the DDR3 interface connected to the device.
2.2.5.2 VREF Supply

The KeyStone I device requires a 0.75-V reference voltage connected to the VREFSSTL pin for the DDR3
SDRAM interface. Many VTT termination supplies provide a separate supply output pin specifically for a
reference voltage. This is true for the TPS51200, which provides the reference voltage on the REFOUT
pin. If the reference voltage is provided by the VTT termination supply, it should be connected to the
VREFSSTL pin. If it is not supplied, a simple voltage divider as shown in Figure 4 can be used.
DVDD15
1 kW
0.1 mF
1%
VREFSSTL
1 kW
0.1 mF
1%
Figure 4. DDR3 VREFSSTL Voltage Divider
All resistive components must be 1% or better tolerance. The VREFSSTL voltage created by the voltage
divider should be routed to the KeyStone I device and the memory components using a trace width of 20
mil or greater. Additional bypass capacitors should be placed next to the connections at the KeyStone I
device and each of the memory devices.
If a resistor divider is used, the VREFSSTL source voltage MUST be generated from the DVDD15. VTT is
susceptible to large amounts of switching noise, so never connect the VREFSSTL to the VTT voltage rail.
2.2.5.3 VPP
Some KeyStone I devices include a VPP voltage rail. For non-secure devices, the VPP pin should be left
unconnected.
2.3 Power Supply Filters

The following section defines the recommended power plane filters for the KeyStone I devices as well as
the implementation details and requirements.
Filters are recommended for the AVDDAn, VDDRn, and VDDTn voltage rails of the device. An overview of
the recommended power supply generation architecture is shown in Figure 5 (only AVDDAn supply rails
are shown).
2.3.1 Filters Versus Ferrite Beads

In previous devices, EMI filters (commonly referred to as T-filters) have been used to condition power rails
that may be susceptible to induced noise. These filters acted as low-pass, band-pass, or high-pass filters
(depending on selection), limiting parasitic coupled noise on each respective power plane and supply.
These EMI filters solve many problems typically overlooked by designers, which may result in device-
related problems on production boards.
Ferrite beads also were evaluated based on performance, physical form factor, and cost. Ferrite beads did
offer a similar benefit at a lower cost point and, therefore, are required by TI for the PLL supplies. For
recommended ferrite beads, see Table 3. Key to selecting the EMI filter and ferrite beads are cutoff
frequency and current rating.
Table 3. Power Rail Filter Recommendations

Power Rail Voltage Filter (p/n) Mfg Current Insertion Loss (dB)/ Impedance Capacitance/DCR
(2)
AVDDAn 1.8 BLM18HE601SN1D Murata 200 mA 470 Ω or greater @ 100 MHz 250 mΩ or less
VDDRn (1) 1.5 BLM18HE601SN1D Murata 200 mA 470 Ω or greater @ 100 MHz (2)
250 mΩ or less
VDDTn (1) 1.0 NFM18CC223R1C3 Murata 1000 mA 65 dB @ ~140 MHz 22,000 pF
(1) Filter not required if peripheral not used.

(2) Requires the addition of a 100-nF and 10-nF ceramic capacitor between the ferrite and respective device pins. Routing must be
kept short and clean.
Table 3 shows the characteristics necessary for selecting the proper noise suppression filter. The following
specifications must be adhered to for proper functionality. For additional details necessary to design an
acceptable power supply system, see the power supply bulk and decoupling capacitor section.
2.3.2 AVDDAn Filters

The AVDDAn pins provide the power for the PLL modules in the KeyStone I device. The number of
AVDDA pins on the device equal the number of PLL modules and each is a power source for an individual
PLL. If a PLL is used, the associated AVDDA pin should be connected to the fixed 1.8-V supply through a
ferrite bead with a 10-nF and a 100-nF decoupling capacitor placed between the ferrite and the AVDDA
pin as shown in Figure 5. If the PLL is not used, the associated AVDDA pin can be connected directly to
the fixed 1.8-V supply without the ferrite or the associated decoupling capacitors.

KeyStone I
Device
100 mF 10 mF 4.7 mF 100 nF 10 nF 1 nF 560 pF

DVDD18
100 nF 100 nF 10 nF
Fixed 1.8-V
Ferrite AVDDA1
Supply
100 nF 100 nF 10 nF
Ferrite AVDDAn
Figure 5. AVDDAx Power Supply Filter
2.3.3 VDDRn Filters

The VDDRn pins provide the power for the PLLs within the SerDes controllers in the KeyStone I device.
The number of VDDRn pins on the device equal the number of SerDes interfaces and each is a power
source for an individual SerDes PLL. If a SerDes interface is used, the associated VDDRn pin should be
connected to the fixed 1.5-V supply through a ferrite bead with a 10-nF and a 100-nF decoupling capacitor
placed between the ferrite bead and the KeyStone I VDDRn pin as shown in Figure 6. If the PLL is not
used by the KeyStone I device, the associated VDDRn pin can be connected directly to the fixed 1.5-V
supply without the ferrite bead or the associated decoupling capacitors.

KeyStone I
Device
100 mF 10 mF 4.7 mF 100 nF 10 nF 1 nF 560 pF

DVDD15
Fixed 1.5-V 100 nF 100 nF 10 nF

Ferrite VDDR1
Supply
100 nF 100 nF 10 nF
Ferrite VDDRn
Figure 6. VDDRn Power Supply Filter
2.3.4 VDDTn Filters

The VDDTn pins provide the analog termination voltage for the SerDes interfaces in the device. The
VDDTn pins are grouped into two or more VDDTn groups. The number of groups varies from device to
device. Each group of pins provides the termination voltage for one or more SerDes interfaces. If any of
the SerDes interfaces in a group are implemented, then the VDDTn group should be connected to the
fixed 1.0-V supply through a capacitive filter. The number of bypass capacitors between the filter and the
VDDTn pins will vary depending on the number of pins on the device. If none of the SerDes interfaces in a
group are used, then the VDDTn pins can be connected directly to the CVDD1 supply.

KeyStone I
Device
100 mF 10 mF 4.7 mF 100 nF 10 nF 1 nF 560 pF

CVDD1
Fixed Core 100 nF 100 nF 100 nF

Filter VDDT1
Supply (1.0-V)
VDDT1
VDDT1
100 nF 100 nF 100 nF

Filter VDDTn
VDDTn
VDDTn
Figure 7. VDDTn Power Supply Filter
2.4 Power-Supply Decoupling and Bulk Capacitors

To properly decouple the supply planes from system noise, decoupling and bulk capacitors are required.
Component parasitics play an important role on how well noise is decoupled. For this reason, bulk
capacitors should be of the low ESR/ESL type and all decoupling capacitors (unless otherwise specified)
should be ceramic and 0402 size (largest recommended). As always, power rating and component and
PCB parasitics must be taken into account when developing a successful power supply system. Proper
board design and layout allow for correct placement of all capacitors. For a minimum set of capacitor
recommendations and values, see the reference table in Section 2.4.4.
Bulk capacitors are used to minimize the effects of low-frequency current transients (see Section 2.4.1)
and decoupling or bypass capacitors are used to minimize higher frequency noise (see Section 2.4.3).
Proper printed circuit board design is required to assure functionality and performance.
One key element to consider during the circuit board (target) design is added lead inductance, or the pad-
to-plane length. Where possible, attachment for decoupling and bypass capacitors to the respective power
plane should be made using multiple vias in each pad that connects the pad to the respective plane. The
inductance of the via connect can eliminate the effectiveness of the capacitor, so proper via connections
are important. Trace length from the pad to the via should be no more than 10 mils (0.01") or 0.254 mm,
and the width of the trace should be the same width as the pad.
As with selection of any component, verification of capacitor availability over the product's production
lifetime should be considered. In addition, the effects of the intended operating environment (temperature,
humidity, and so forth) should also be considered when selecting the appropriate decoupling and bulk
capacitors.
NOTE: All values and recommendations are based on a single KeyStone I device, TI high
performance SWIFT power supplies, and the New UCD9222/44 digital controller. The use of
alternate, non-specified on-board power supply modules, alternate power supplies, and
alternate decoupling/bulk capacitor values and configurations require additional evaluation.
2.4.1 Selecting Bulk Capacitance

This section defines the key considerations necessary to select the appropriate bulk capacitors for each
rail:
• Effective impedance for the power plane to stay within the voltage tolerance
• Amount of capacitance needed to provide power during the entire period when the voltage regulator
cannot respond (sometimes referred to as the transient period)
The effective impedance of the core power plane is determined by:
• (Allowable Voltage Deviation due to Current Transients) / (Max Current)
Calculating for the variable core supply (as an example only), the allowable deviation (transient tolerance)
is 25 mV (based on 2.5% AC [ripple] and 2.5% DC tolerance [voltage]) of the nominal 1-V rail. Using an
allowable 25-mV tolerance and a maximum transient current of 10 amps, the maximum allowable
impedance can be calculated as follows:
• 25 mV/10 Amps = 2.5 mΩ
The effective ESR for all bulk capacitors (in the above example) should not exceed this impedance value.
The combination of good-quality and multiple bulk capacitors in parallel help to achieve the overall ESR
required. Therefore, to limit the maximum transient voltage to a peak deviation of 25 mV, the power supply
output impedance, which is a function of the power supply bandwidth and the low impedance output
capacitance, should not exceed 2.5 mΩ.
The following three plots show the recommended maximum impedance to current transient for three
different tolerances (15 mV, 25 mV, and 35 mV). The first plot covers an allowable current between
500 mA and 3500 mA. The second plot covers 3700 mA to 6900 mA, and the final plot covers 7000 mA to
13000 mA.
The expected maximum current for the KeyStone I device (core) depends heavily on the use case and
design of the application hardware. The recommended components are designed to support up to 17 A (at
86% efficiency) and up to 19 A when using the recommended components and implementing them
correctly at 86% efficiency. It does not include current transients that occur during power on (yet these
must be considered). Given the recommended topology, only a reduced amount of bulk capacitors is
possible, provided they are chosen for there ESR/ESL and voltage rating properly. The majority of all core
bulk capacitors must be located in close proximity to the UCD7242 dual FETs.
Capacitance values should not be less than those specified in the applicable table in Section 2.4.4. Some
intermediate size ceramic bulk capacitors (for example, 22 µF and 47 µF as listed under Section 2.4.2) are
recommended to cover the response time between the bypass capacitors and the larger bulk capacitors.
Max Impedance to Current Load
0.0700 0.0700
0.0650
0.0600
0.0550
0.0500 0.0500 0.0500
0.0450
ESR (ohms)
0.0400
0.0389
0.0350 0.0357
0.0318
0.0300 0.0300
0.0278 0.0269 `
0.0250
0.0227 0.0233
0.0214 0.0206
0.0200 0.0192 0.0184
0.0167 0.0167 0.0167
0.0150 0.0147 0.0152 0.0140
0.0136 0.0132 0.01300.0121
0.0119 0.0113 0.0106
0.0100 0.0115 0.0109 0.0100 0.0100
0.0093 0.0086
0.0100 0.0088 0.0081
0.0079 0.0071 0.0065 0.0076 0.0071
0.0050 0.0060 0.0056
0.0052 0.0048 0.0045
0.0043
0.0000
0.5000
0.7000
0.9000
1.1000
1.3000
1.5000
1.7000
1.9000
2.1000
2.3000
2.5000
2.7000
2.9000
3.1000
3.3000
3.5000
Current (A)
15mV 25mV 35mV
Figure 8. ESR Plot for Delta Current to Tolerance #1
0.0100
0.0095 0.0095
0.0090 0.0090
0.0085 0.0085
0.0081
0.0080
0.0078
0.0075 0.0074
0.0071
0.0070
0.0068 0.0069
ESR (ohms)
0.0065 0.0066
0.0064 0.0064
0.0061 0.0061
0.0060 0.0059
0.0058 0.0057
0.0055 0.0056 0.0056
0.0053 0.0054
0.0051 0.0052
0.0050 0.0051
0.0049
0.0047
0.0045 0.0045
0.0044
0.0042
0.0040 0.0041 0.0041
0.0038 0.0040 0.0038
0.0037 0.0037
0.0036
0.0035 0.0035
0.0033
0.0032
0.0030 0.0031 0.0029
0.0028 0.0027
0.0026 0.0025
0.0025 0.0025 0.0024 0.0022
0.0023
0.0020 0.0022
3.7000
3.9000
4.1000
4.3000
4.5000
4.7000
4.9000
5.1000
5.3000
5.5000
5.7000
5.9000
6.1000
6.3000
6.5000
6.7000
6.9000
Load Current (A)
15mV 25mV 35mV
0.0050 0.0050
0.0047
0.0045
0.0044
0.0041
0.0040
0.0039
0.0037
0.0036
0.0035 0.0035
0.0033 0.0033
ESR (ohms)
0.0031 0.0032
0.0030 0.0030
0.0029 0.0029
0.0028 0.0028
0.0026 0.0027
0.0025 0.0025
0.0024
0.0021 0.0023
0.0022
0.0021
0.0020 0.0019 0.0020
0.0020 0.0019
0.0017
0.0018 0.0015
0.0015 0.0016 0.0014
0.0014 0.0013
0.0013 0.0012 0.0012
0.0010
7.0000
7.5000
8.0000
8.5000
9.0000
9.5000
10.0000
10.5000
12.0000
12.5000
13.0000
11.0000
11.5000
Current Load (A)
15mV 25mV 35mV
2.4.1.1 Bulk Capacitor Details and Placement

Placement of the bulk capacitors is highly dependent on the power supply solution selected. It is important
to review the application documents associated with the power supply selected to determine the best
placement of the bulk capacitors for your design. Generally, the bulk capacitors are placed in close
proximity to the power supply. For the purpose of this document and related devices, a bulk capacitor is
defined as any capacitor ≥ 22 µF, unless otherwise noted.
2.4.2 Selecting Intermediate Value Capacitors

All intermediate capacitors must be positioned as close to the device as possible, decoupling capacitors
taking precedence where placement between the two form factors and values come into question.
Intermediate capacitor values are considered to be those between 22 µF and 47 µF, and they should be
low ESR and ceramic, where possible.
2.4.3 Selecting Decoupling Capacitors

All decoupling or bypass capacitors must be positioned close to the device. In practice, each decoupling
capacitor should be placed a maximum distance of 10 mm from the respective pins. Where possible, each
decoupling capacitor should be tied directly between the respective device pin and ground without the use
of traces. Any parasitic inductance limits the effectiveness of the decoupling capacitors; therefore, the
physically smaller the capacitors are (0402 or 0201 are recommended), the better the power supply will
function.
Proper capacitance values are also important. All small decoupling or bypass capacitors (560 pF, 0.001
μF, and 0.01 μF) must be placed closest to the respective power pins on the target device. Medium
bypass capacitors (0.1 uF or as large as can be obtained in a small package such as an 0402) should be
the next closest. TI recommends placing decoupling capacitors immediately next to the BGA vias, using
the interior BGA space and at least the corners of the exterior.
The inductance of the via connect can eliminate the effectiveness of the capacitor so proper via
connections are important. Trace length from the pad to the via should be no more than 10 mils and the
width of the trace should be the same width as the pad. If necessary, placing decoupling capacitors on the
back side of the board is acceptable provided the placement and attachment is designed correctly.
2.4.3.1 Decoupling Capacitor Details and Placement

All decoupling capacitors should be placed in close proximity to the device power pins. For the purpose of
this document and related devices, a decoupling or bypass capacitor shall be defined as any capacitor < 1
µF, unless otherwise noted.
Table 4 shows minimum recommended decoupling capacitor values. Each decoupling or bypass capacitor
should be directly coupled to the device via. Where direct coupling is impractical, use a trace 10 mil
(0.010"/0.254 mm) or shorter, and having the same width as the capacitor pad, is strongly recommended.
The table values are minimums; many variables have an effect on the performance of the power supply
system. Re-evaluate the individual power supply design for all variables before implementing the final
design.
2.4.4 Example Capacitance

Table 4 establishes the bulk and decoupling capacitance requirements for the various device voltage rails.
In many cases these rails originate from a common source, which accounts for a reduction in bulk
capacitance for many rails. Table 4 does not cover the additional capacitance (decoupling or bulk)
required by any other components or the filters outlined in previous sections. All capacitance identified
assumes the use of the recommended power supplies (especially for the variable core), and switching
frequencies as indicated. For more information, see the table notes following Table 4. The recommended
values are in addition to those required for proper operation of the recommended power supplies.
The values identified in Table 4 are per device and assume a typical loading per device IO. In the event
multiple devices are used and a voltage rail is common or shared (where applicable), the capacitor
requirements must be re-evaluated (on a case-by-case basis) and take into account key concepts
including current (load), minimum and maximum current requirements, voltage step, power supply
switching frequency, and component ESR. Other constraints also apply.
Table 4. Bulk, Intermediate, and Bypass Capacitor Recommendations (Absolute Minimums)

Total
Voltage Supply Capacitors Qty ESR (mΩ) (1) Capacitance (7) Description Notes
(2) (7) (3) (8) (14)
CVDD 560 pF Ceramic 20 1662.9112 µF Variable Core Supply , , ,
(7) (8) (14)
100 nF Ceramic 8 , ,
(7) (8) (14)
10 nF Ceramic 10 , ,
22 µF Low ESR 1 9 mΩ (1) (16)
47 µF Low ESR 4 3 mΩ (1) (7) (8) (14)

, ,
(1) (16)
47 µF Low ESR 2 3 mΩ
220 µF Low ESR 2 7 mΩ (1) (7) (8) (14)
, ,
(1) (7) (8) (14)
470 µF Low ESR 2 12 mΩ , ,
Table 4. Bulk, Intermediate, and Bypass Capacitor Recommendations (Absolute Minimums) (continued)
Total
Voltage Supply Capacitors Qty ESR (mΩ) (1) Capacitance (7) Description Notes
(4) (7) (8) (14)
DVDD18 560 pF Ceramic 8 125.03748 µF 1.8-V IO Supply , , ,
(4) (7) (8) (14)
100 nF Ceramic 3 , , ,
(4) (7) (8) (14)
(4) (7) (8) (14)
1 µF Ceramic 0
(1) (4) (7) (8) (14)
4.7 µF Ceramic 1 10 mΩ , , ,
10 µF Ceramic 2 9 mΩ (1) (4) (7) (8) (14)
, , ,
47 µF Low ESR 0
100 µF Low ESR 1 9 mΩ (1) (4) (7) (8) (14)
, , ,
(4) (7) (8) (14)
220 µF Low ESR 0 , , ,
(11)
10 µF Ceramic 0
(4)(5) (7) (8) (14)
CVDD1 560 pF Ceramic 10 277.4556 µF 1.0-V Fixed Supply , , ,
(5) (7) (8) (14)
(5) (7) (8) (14)
1 µF Ceramic 0
(1) (16)
10 µF Ceramic 3 9 mΩs
(1) (5) (7) (8) (14)
47 µF Low ESR 1 3 mΩs , , ,
(1) (5) (7) (8) (14)
100 µF Low ESR 2 9 mΩs , , ,
(6) (7) (8) (12) (14)
DVDD15 560 pF Ceramic 8 5.16448 µF 1.5-V DDR Supply , , , ,
(6) (7) (8) (12) (14)
100 nF Ceramic 4 , , , ,
(6) (7) (8) (12) (14)
10 nF Ceramic 6 , , , ,
(1) (6) (7) (8) (12) (14)
4.7 µF Ceramic 1 10 mΩs , , , ,
10 µF Low ESR 0
22 µF Low ESR 0
47 µF Low ESR 0
100 µF Low ESR 0
220 µF Low ESR 0
VREFSSTL 100 nF Ceramic 2 0.2000 µF 1.5-V DDR VREF
Supply
TOTAL 2281.40528 µF
(1) Refers to each individual capacitor ESR Value (and is a maximum value per capacitor).
(2) Capacitor selection for CVDD variable core (as well as all other supplies) is highly dependant upon design and must be
evaluated on a case by case scenario. The bulk capacitors shown are with respect to a 470-nH inductor for 10 A with a
maximum delta current of 85%, other configurations may apply.
(3) Estimates provided assume 10 W @ 1 V for the variable (scalable) core.
(4) Estimates provided assume 0.75 W @ 1.8 V for the 1.8-V IOs.
(5) Estimates provided assume 5 W @ 1 V for the fixed core.
(6) Estimates provided assume 1.0 W @ 1.5 V for the DDR3 IOs (SDRAM not included in estimations).
(7) Always denotes a minimum capacitance.
(8) Fsw is 1e6 Hz.
(9) Fsw is 550e3 Hz
(10) Not required if common input to rail to supply (bulk is applied to AVDDA1) and distance is short between supply and device
pin.
(11) Not required if adequate bulk capacitance is supplied for CVDD1 common input.
(12) Additional capacitance required when adding SDRAM (must be common supply per device).
(13) To be placed on the input side of the ferrite or filter.
(14) Total value is calculated on a 1.5-pF device pin capacitance, variations in pin capacitance require re-calculation.
(15) Not required if adequate bulk capacitance is supplied for DVDD15 common input.
(16) Required for all devices within the same family having higher current requirements beyond that identified (future devices).
(17) Not required if supplied for VDDT1 and is a common input source.
NOTE: The capacitor values provided are a MINIMUM recommended amount.
The final capacitor value should not be less than the value specified in Section 2.4.4. Final selection for
the core rails are determined using the provided capacitor selection tool and power estimation application
guide, which take into account the board impedance, variation in the CVDD supply voltage, use case, and
the ESR of the bulk and decoupling capacitors selected.
2.4.4.1 Calculating the Value of Bypass Capacitors

The following is a brief example of how the decoupling (bypass) capacitance was obtained. Final capacitor
values should not be less than the values specified in Table 4. The final selection must take into account
key variables including PCB board impedance, supply voltage, PCB topologies, component ESR, layout
and design, as well as the end-use application, which should include crosstalk and overall amount of AC
ripple allowed.
1. Assume all gates are switching at the same time (simultaneously) in the system, identify the maximum
expected step change in power supply current ΔI
2. Estimate the maximum amount of noise that the system can tolerate.
3. Divide the noise voltage by the current change results in the maximum common path impedance or:
V
Zmax = n
DI (1)
4. Compute the inductance (LPSW) of the power supply wiring. Use this and the maximum common path
Z
impedance (Zmax) to determine the frequency, fPSW = fcorner = f3db = (2PLmax ) below which the power supply
PSW
wiring is fine (power supply wiring noise <Vn.
5. Calculate the capacitance (Cbypass) of the bypass capacitor:
1
Cbypass = * fPSW * Zmax
2P (2)
NOTE: If the operating frequency foper < fPSW, then no bypass capacitors are necessary. If foper < fPSW,
then the bypass or decoupling capacitors are needed.
Table 5 provides an example of the capacitance needed for each of the power supply inputs of the
TMS320C6670. The decoupling needed for your application must be calculated based on the KeyStone I
device that you have selected and the application that you are running. Examples for nominal usage for
each KeyStone I device may be found in the schematic for the EVM.
Table 5. Total Decoupling Capacitance

Rail Voltage Pins nF Notes
CVDD 1 81 773.490 SmartReflex rail
AVDDA1 to AVDDAn 1.8 N 110.00 each
DVDD18 1.8 18 14.3239
VDDT1 to VDDTn 1 X
CVDD1 1 43 76.3944 Fixed 1-V rail
VDDR1 to VDDRn 1.5 N 110.00 each
DVDD15 1.5 31 309.397
VrefSSTL 0.75 1 17.9049
Assumptions (Core):
• 81 core power pins
• 2.5% ripple allowed
• Supply voltage is 1.0 V
• Each pin has a capacitive load of 1.5 pF

• Power supply inductance is 100 nH (approximately from PCB stack up)
• Power supply switching frequency is 1000 kHz
1. Calculate the current:
12 10
nC VCC 81 1.5 1.0 1.066
'I o 'I o 'I o 'I 0.1215
't 19 19 (3)
2. Maximum impedance:
25 mW
Zmax o Zmax 0.205761316872 :
0.1215 (4)
3
3. Insert the power supply switching frequency: fPSW = 1000
4. Calculate the bypass value(s) needed:
1
Cbypass min o
2 3 fPSW Zmax
1
Cbypass min o
2 pi le6 0.2057613168 7242
1 7
Cbypass min o Cbypass min 7.7349302 or 0.77349302 PF
1292836.483
(5)
In theory, decoupling capacitors should be placed on all pins, or at a minimum, on pins that share a
common via to the power rail.
2.5 Power Supply Sequencing

A detailed representation of the power supply sequencing for KeyStone I devices can be found in the
device-specific data manual. All power supplies associated with the KeyStone I devices should allow for
controlled sequencing using on-board logic or a power supply sequence controller such as the UCD9090.
Two sequencing orders are supported with the KeyStone I devices.
The first is a core voltage before IO voltage sequence with the power supplies enabled in the following
order:
• CVDD → CVDD1 → DVDD18 → DVDD15
The second sequencing is an 1.8-V IO before core voltage sequence with the power supplies enabled in
the following order:
• DVDD18 → CVDD → CVDD1 → DVDD15
A power supply should reach a valid voltage level and declare the supply output in a power good state
before enabling the next supply in the sequence.
2.6 Power Supply Block Diagram

Now that the power supply components have been selected, create a power supply block diagram similar
to the one found in the EVM schematic for the KeyStone I devices. This block diagram should include all
of the power supplies in the design, the output voltage associated with each, the current limit, and the
signal used to enable the supply. A second diagram showing the sequencing of the supplies should also
be generated. These documents will be requested by Texas Instruments if support is needed. They should
be added to the design folder and it is good practice to include them in the schematics for reference.
Clocking www.ti.com
3 Clocking
The next requirement for the KeyStone I design is proper clocking. KeyStone I devices include a number
of internal PLLs that are used to generate the clocks within the part. These include PLLs for generating
system clocks and PLLs for the high speed SerDes interfaces. For successful operation of the design, all
clocks must meet the specified stability and jitter requirements from the device-specific data manual. This
section describes the clocks found in the KeyStone I devices and the requirements for these clocks.
Detailed clock input requirements are found in the device-specific data manual. Not all of the clocks
described in this section will appear on every KeyStone I device. To determine which clocks are present in
your KeyStone I device, see the device-specific data manual.
3.1 System PLL Clock Inputs

The system PLL clock inputs for KeyStone I devices are shown in Table 6.
Table 6. KeyStone I System PLL Clock Inputs

Clock Frequency Range (MHz) Drivers Supported Description
CORECLKp 40-312.5 LVDS, LVPECL Used to clock the Main PLL
CORECLKn (AC Coupled)
ALTCORECLKp 40-312.5 LVDS, LVPECL Used to clock the Main PLL if CORECLKSEL = 1
ALTCORECLKn (AC Coupled)
DDRCLKp 40-312.5 LVDS, LVPECL Used to clock the DDR PLL
DDRCLKn (AC Coupled)
PASSCLKp 40-312.5 LVDS, LVPECL Used to clock the packet accelerator subsystem PLL if
PASSCLKn (AC Coupled) PASSCLKSEL = 1
All system PLL clock inputs are differential and must be driven by one of the specified differential driver
types. All the differential clock inputs specified in Table 6 are implemented with Texas Instruments low
jitter clock buffers (LJCBs). These input buffers include a 100-Ω parallel termination (P to N) and common-
mode biasing. Because the common-mode biasing is included, the clock source must be AC coupled.
Low-voltage differential swing LVDS and LVPECL clock sources are compatible with the LJCBs. A
separate, properly terminated clock driver must be provided for each clock input. The proper termination
for the clock driver selected must be included. Not all clock drivers are terminated in the same manner.
For the proper termination of the clock driver you have selected, see the device-specific data manual. For
additional information on AC termination schemes, see the AC-coupling between differential LVPECL,
LVDS, and CML. Note that the LJCB clock input is assumed to be a CML input in this document.
These clock inputs are used by system PLLs to generate the internal clocks needed to operate the device.
These PLLs are compatible with a wide range of clock input frequencies, which can be multiplied to the
valid operating frequencies of the device. Detailed information on programming the PLLs can be found in
the Phase-locked loop (PLL) for KeyStone devices user's guide.
The main PLL is driven in one of two manners, depending on the device.
• It is driven by CORECLKP/N if a dedicated main PLL clock input is present on the device.
• It is driven by either the ALTCORECLKP/N or the SYSCLKP/N depending on the state of the
CORECLKSEL configuration input.
Some KeyStone I devices allow one of two clock inputs to be used as a source for the main PLL. If
ALTCORECLKP/N is selected, it acts as a dedicated clock input for the main PLL. SYSCLKP/N is also
used as a clock source for the AIF interface and is considered a SerDes PLL clock source. If SYSCLKP/N
is selected it will be the clock source for both the main PLL and the AIF interface.
www.ti.com Clocking
3.2 SerDes PLL Reference Clock Inputs

KeyStone I devices contain several high-performance serial interfaces commonly known as SerDes
interfaces. The Serializer/Deserializer (SerDes) PLL reference clock inputs for KeyStone I devices are
shown in Table 7.
Table 7. KeyStone I SerDes PLL Reference Clock Inputs

Clock Valid Input Frequencies (MHZ) Description
SYSCLKp 122.88, 153.60, 307.20 Used to clock the AIF PLL. Used to clock the main PLL if
SYSCLKn CORECLKSEL= 0.
SRIOSGMIICLKp 156.25, 250, 312.5 Used to clock the SGMII PLL and the SRIO PLL.
SRIOSGMIICLKn
PCIECLKp 100, 156.25, 250, 312.5 Used to clock the PCIexpress PLL.
PCIECLKn
MCMCLKp 156.25, 250, 312.5 Used to clock the HyperLink PLL.
MCMCLKn
Not all clock inputs are present on all KeyStone devices. For a complete list of clock inputs, see the
device-specific data manual. All clock inputs are differential and must be driven by one of the specified
differential driver types.
All of the SerDes reference differential clock inputs are implemented with Texas Instruments low jitter
clock buffers (LJCBs). These input buffers include a 100-Ω parallel termination (P to N) and common-
mode biasing. Because the common mode biasing is included, the clock source must be AC coupled.
Low-voltage differential swing LVDS and LVPECL clock sources are compatible with the LJCBs.
Termination on the driver side of the AC coupling capacitors may still be needed. For the driver selected
for the proper termination see the device-specific data manual.
Clock drivers may source only one clock input, so a separate clock driver must be provided for each clock
input. In addition, the proper termination for the clock driver selected must be included. For additional
information on AC termination schemes, see the AC-Coupling Between Differential LVPECL, LVDS, and
CML Application Report . Note that the LJCB clock input is assumed to be a CML input in this document.
In addition, the LJCB will support a PCI express-compliant HCSL clock input. This clock is terminated in
the same manner as an LVDS clock driver. Spread spectrum clocks are commonly used with PCI express
and may be present on a PCI express bus connector. The PCIECLKP/N input is compatible with PCI
express-compliant spread-spectrum clocking. Note that a common clock for a PCI express root complex
and endpoint are not required unless a spread-spectrum clock is used. If a spread-spectrum clock is used
by one device at the far end of a PCI express link, the same clock must be used as a reference by the
KeyStone I device.
These clock inputs are used by PLLs in the reference SerDes interface. These interfaces require high-
quality clocks with low phase noise and are specified only to operate with specific reference clock input
frequencies. Detailed information on programming the PLL associated with a reference clock input can be
found in the user's guide for that SerDes interface.
Clocking www.ti.com
3.3 Input Clock Waveform Requirements

Table 8 shows the specific input clocking requirements for the KeyStone I device including maximum rise
and fall times, duty cycle and accuracy. Note that all LJCB and LVDS differential input clock buffers
contain a 100-Ω internal termination resistor (no external resistor is required).
Table 8. Clocking Requirements - Slew Rate, Duty Cycle, Accuracy

Clock Input Trise/Tfall Duty Cycle % Accuracy
(1)
SRIOSGMIICLKp 50 – 350 ps 45 / 55 ± 100 PPM
SRIOSGMIICLKn
(if SRIO is used)
(1)
SRIOSGMIICLKp 50 – 350 ps 45 / 55 ± 100 PPM
SRIOSGMIICLKn
(SRIO not used)
(1)
PCIECLKp 50 – 350 ps 45 / 55 ± 100 PPM
PCIECLKn
RP1CLKp 350 ps max 45 / 55 ± 100 PPM
RP1CLKn
(1)
MCMCLKp 50 – 350 ps 45 / 55 ± 100 PPM
MCMCLKn
(1)
ALTCORECLKp 50 – 350 ps 45 / 55 ± 100 PPM
ALTCORECLKn
(1)
PASSCLKp 50 – 350 ps 45 / 55 ± 100 PPM
PASSCLKn
(1)
SYSCLKp 50 – 350 ps 45 / 55 ± 100 PPM
SYSCLKn
(1)
DDRCLKp 50 – 350 ps 45 / 55 ± 100 PPM
DDRCLKn
(1)
CORECLKp 50 – 350 ps 45 / 55 ± 100 PPM
CORECLKn
(1) Swing is rated for a 250 mV (pk-pk) at zero crossing where 10% – 90% of TRISE and TFALL must occur within the prescribed 50-
350 pS time frame. The minimum slew specified is relative to the minimum input signal value. An input signal must transition
through:
MInimum:
Vtransition = 250 mV *(90% 10%)
Vtransition = 250 mV * 80%
Vtransition = 200 mVin →in 350 pS
Maximum
Vtransition = 250 mV *(90% 10%)
Vtransition = 250 mV * 20%
Vtransition = 200 mVin →in 50 pS
The two major concerns pertaining to differential reference clocks are low jitter and having the proper
termination. LVDS, CML, or LVPECL clock sources can be used but they require different termination
strategies. The input buffer sets its own common mode voltage, so AC coupling is necessary. It also
includes a 100-Ω differential termination resistor, eliminating the need for an external 100-Ω termination
when using an LVDS driver. For additional information on AC termination schemes, see AC-Coupling
Between Differential LVPECL, LVDS, and CML. For information on DC coupling, see DC-Coupling
Between Differential LVPECL, LVDS, HSTL, and CM.
www.ti.com Clocking
3.4 Input Clock Termination Requirements

Table 9 shows the input clock type, termination, and termination location for the device. For additional
termination details, see Section 7.
Table 9. Clocking Requirements - Clock Source and Termination

Clock Input Logic Termination Termination Location
(1)
SRIOSGMIICLK LJCB or LVDS AC Couple 0.1 μF Destination Side
SRIOSGMIICLKn
(if SRIO is used)
(1)
SRIOSGMIICLKp LJCB or LVDS AC Couple 0.1 μF Destination Side
SRIOSGMIICLKn
(SRIO not used)
(1)
PCIeCLKp LJCB or LVDS AC Couple 0.1 μF Destination Side
PCIeCLKn
RP1CLKp LVDS AC Couple 0.1 μF Destination Side
RP1CLKn
(1)
MCMCLKp LJCB or LVDS AC Couple 0.1 μF Destination Side
MCMCLKn
(1)
ALTCORECLKp LJCB or LVDS AC Couple 0.1 μF Destination Side
ALTCORECLKn
(1)
PASSCLKp LJCB or LVDS AC Couple 0.1 μF Destination Side
PASSCLKn
(1)
SYSCLKp LJCB or LVDS AC Couple 0.1 μF Destination Side
SYSCLKn
(1)
DDRCLKp LJCB or LVDS AC Couple 0.1 μF Destination Side
DDRCLKn
(1)
CORECLKp LJCB or LVDS AC Couple 0.1 μF Destination Side
CORECLKn
(1) Can be DPECL or LVPECL if properly terminated and biased.
Additional SerDes clock requirements that must be met and followed in order to assure a functional
system are identified in Section 3.8.6. Texas Instruments has designed and evaluated specific clock
sources (see Section 3.8) that meet or exceed the performance requirements necessary for a functional
system.
Calculation for proper AC termination is based on the following formula:
1
C³ * P * R * FreqMHz
2 (6)
Where,
• R = impedance of the net (50 Ω single ended and 100 Ω differential) – each net is 50 Ω
• C = AC capacitor value derived from formula (results are in nF)
• Freq = input frequency in MHz
• Pi = rounded to 3.141592654
NOTE: The pole associated with the high-pass response should be placed at a minimum of two
decades below the input clock frequency.
Two examples are provided, the first for an input clock of 50 MHz and the second for an input clock
frequency of 312.50 MHz.
Although the results denoted in Equation 7 and Equation 8 provide a minimum value, it is recommended
that a minimum of 10 nF or a 0.1 μF (100 nF) ceramic capacitor be used as the AC termination value.
Clocking www.ti.com
Equation 7: 50 MHz
1
C³ * P * RW * FreqMHz / 100
2
1
C³
(2 * 3.141592654 * (50 ) * (50e6 / 100 ))
1
C³
157079632.7
C ³ 6.636619 nF (7)
Equation 8: 312.50 MHz
1
C³ * (2 P * RW * FreqMHz / 100 )
2
1
C³
(2 * 3.141592654 * 50 * (312.50e6 / 100 ))
1
C³
981747704.375
C ³ 1.01859e - 9F
or 1.01859 nF (8)
Figure 11 shows the typical LVDS-based solution, including the appropriate AC termination.
Low Jitter
LVDS Source
+ +
Clock
Source DSP
- -
Figure 11. LJCB LVDS Clock Source
AC termination values will vary depending on many conditions, but should be in the range of 0.01 µF to
1.0 µF to help ensure minimal amplitude and phase degradation of the incoming clock. Capacitor
placement is critical and highly dependent upon design topology. Determination on whether the AC
coupling (DC blocking) capacitors are to be staggered or in parallel and whether they are to be placed
near the receiver end or elsewhere in the nets should be determined through simulation and modeling or a
signal-analysis tool typically incorporated in most common layout packages.
Ultra high speed interfaces such as SerDes require special consideration. The parasitic capacitance of the
capacitor bodies (and mounting pads) to one another and to the reference plane beneath may require a
change in traditional layout techniques. Placement of the AC capacitors should also be evaluated for
impact on signal integrity due to reflections.
For additional clocking solutions, see the Clocking Design Guide for KeyStone Devices.
3.5 SerDes Reference Clock Jitter Requirements

The SerDes architecture allows for reliable data transmission without a need for common synchronized
clocks. However, the architecture does require high-quality clock sources that have very little phase noise.
Phase noise is also commonly referred to as clock jitter.
Note that PCIe can be operated with a common clock, which allows for the use of spread-spectrum clock
sources that have higher levels of bounded phase noise. HyperLink is also defined as an interface
requiring a common clock because it is expected to be a very short-reach interface between DSPs located
on the same board.
www.ti.com Clocking
3.5.1 BER and Jitter

Excessive jitter can cause data errors. For this reason, especially in high-performance data transmission
links, a BER (bit error rate) limit is required to allow reliable communication. Since error-correcting codes
are normally used, reliable data transmission can occur as long as links operate within their BER limit.
Jitter is the sum of both random and deterministic sources. All jitter is undesirable and capable of causing
bit errors. Random jitter can be defined as the sum of all noise sources, including from components, used
to construct and/or implement the data path between source and target. Random jitter is considered
unbounded and measurements of it will increase as larger and larger time periods are tested. Some
sources of random or unbounded jitter include:
• Shot noise – This is electron and hole noise in a semiconductor. It increases due to bias current and
measurement bandwidth.
• Thermal noise – This is associated with electron flow in conductors. It has variations due to
temperature, bandwidth, and noise resistance.
• Flicker or pink noise – This noise has an amplitude proportional to 1/ƒ.
Deterministic jitter, unlike random Jitter, is the total jitter induced by the characteristics of the data path
between source and target. Deterministic jitter, also called bounded jitter, obtains its maximum phase
deviation within a specified time interval. Sources of deterministic jitter include (most prominently):
• Electromagnetic Interference (EMI) coupled in from power supplies, AC power lines, and RF-signal
sources.
• Crosstalk, both inductive and capacitive, where signals from one transmission line or component
change the waveform on a nearby signal. Inductive crosstalk results when the magnetic field generated
from the current in one signal line induces a current in another signal line. Similarly, capacitive
crosstalk results from electrical charge changes on one signal caused by voltage changes on a nearby
trace or component separated by an insulator (dielectric), similar to the operation of a capacitor.
• Inter-Symbol Interference (ISI) is where the amplitude and/or phase of one data bit is affected by the
levels of preceding data bits within a single data stream. This is caused by the bandwidth limiting
effects of transmission lines.
• Simultaneous Switch Output (SSO) jitter results when multiple output drivers change state
simultaneously. These induce current spikes on power and/or ground sources causing a voltage
threshold level shift. This will increase rise and/or fall times of various amounts which manifest as jitter.
• Reflections along the signal path or transmission line, due to conductor stubs or improper termination,
also create jitter. Reflections result from energy flowing back through the conductor that sum with the
original signal causing an amplitude variation. If these reflections are unbalanced on each conductor
within a differential pair, this will result in time variation of the crossover (sampling) points.
Clocking www.ti.com
3.5.2 Random Jitter

Phase noise amplitude is not the only concern. The frequency content of this phase noise is also
significant. Therefore, masks are provided as a means of indicating an acceptable phase noise tolerance.
Each SerDes interface has a slightly different mask. SerDes data rate, input clock rate, and required bit-
error rate affect the phase noise tolerance.
Figure 12 shows the basic phase jitter tolerance masks for the KeyStone I SerDes interfaces. The mask
shown is for the highest performance operating mode defined for each SerDes interface. Board designers
must provide a reference clock that has clock jitter below the appropriate mask across the entire
frequency range.
- 10.00
- 20.00
SRIO @ 5GBd mask

- 30.00
AIF @ 6.144GBd mask
- 40.00 X HyperLink @ 12.5GBd mask
PCIe @ 2.5GBd common clk mask

50.00
PCIe @ 2.5GBd mask
60.00
Phase Noise (dBc/Hz)
PCIe @ 5GBd mask
70.00
80.00
90.00
X
100.00
X
110.00 X
- 120.00 X
X
- 130.00
X
- 140.00 X
X
- 150.00
1.00E+04 1.00E+05 1.00E+06 1.01E+07
Frequency (Hz)
Figure 12. SerDes Reference Clock Jitter Masks
www.ti.com Clocking
This information is also shown as a straight-line approximation with the end points and the knee defined
for ease of understanding. This template is shown in Figure 13 below and the endpoints in Table 10. The
mask level at specific frequencies can now be estimated based on the straight-line approximation. This is
useful when comparing clock sources from various vendors who specify jitter masks in different ways.
P1
Phase Noise (dBc/Hz)
P2
F1 F2 F3
Frequency (Hz)
Figure 13. Phase Jitter Template
Table 10. Phase Jitter Template Endpoints

Reference
Clock
Frequency P1 P2
Interface & Rate (GBd) (MHz) BER F1 (kHz) (dBc/Hz) F2 (MHz) (dBc/Hz) F33 (MHz)
-15
SRIO @ 5 250 1 10 -87.1 1.56 -130.8 20
AIF @ 6.144 256 1-12 10 -86.1 1.91 -131.6 20
HyperLink @ 12.5 250 1-17 10 -92.1 3.43 -142.9 20
-12
PCIe @ 2.5 250 1 10 -11.4 2.96 -117.7 20
(common)
PCIe @ 2.5 250 1-12 10 -71.5 2.18 -117.7 20
-12
PCIe @ 5 250 1 10 -71.7 3.76 -123.1 20
Clocking www.ti.com
A single value is also commonly provided as the limit for phase noise. This is an integration of the phase
noise across a frequency range. Table 11 shows the integrated value for each of the phase noise masks
across the 10-kHz to 20-MHz frequency range and across the 1-MHz to 20-MHz frequency range. These
limits are also useful for evaluating high-performance clocks sources from some vendors.
Table 11. Phase Noise Mask Integrated Values

Interface and Rate Reference Clock Rj (ps) 10 kHz to 20 Rj (ps) 1 MHz to 20
(GBd) Frequency (MHz) Bit Error Rate (BER) MHz MHz
SRIO @ 5 250 1-15 4.12 1.17
-12
AIF @ 6.144 256 1 4.56 1.09
HyperLink @ 12.5 250 1-17 2.26 0.33
PCIe @ 2.5 250 (common) 1-12 14030 8.99
-12
PCIe @ 2.5 250 1 24.59 5.62
PCIe @ 5 250 1-12 23.65 3.37
Note that not all clock sources with jitter values below the single values shown in Table 11 will meet the
templates shown above. These figures were generated assuming that the reference clock spectrum is
shaped in the same manner as the template. Further verification may be needed.
All of the masks shown in Table 11 assume the input clock rate is about 250 MHz. The input clock rate
affects the phase noise tolerance. This is because the reference clock phase noise is multiplied within the
SerDes PLL. Higher levels of phase noise can be tolerated if the input reference clock rate is higher. The
entire mask shifts proportional to the frequency difference. This mask offset is approximated by the
equation 20*log(F2/F1). The integrated phase noise value also shifts by the same ratio. It is approximated
by the equation F2/F1. Because all of the masks shown above are for 250 MHz reference clocks, the
mask and phase noise adjustments from 250 MHz are shown in Table 12.
Table 12. Mask and Phase Noise Adjustments

Reference Clock Frequency (MHz) Phase Noise Mask Offset (dB) Rj Multiplier
100 (1) -8.0 0.40
122.88 -6.2 0.49
153.60 -4.2 0.61
156.25 -4.1 0.63
250 0.0 1.00
312.5 1.9 1.25
(1)
Only supported for PCIe
The phase noise masks provided are at the maximum supported data rates for SRIO, AIF, and HyperLink.
Additional phase noise margin is gained when the data rate is reduced. This is because the margin for
jitter increases as the data eye gets longer when the data rate decreases. Once again, the mask offset is
approximated by the equation 20*log(F2/F1) and the integrated phase noise value is approximated by the
equation F2/F1.
Table 13 shows the mask and phase noise adjustments for the available SRIO operating rates. Similar
numbers can be computed for AIF and HyperLink. Note that this does not apply for PCIe. The phase noise
masks are defined within the PCIe specification and the masks converge at low frequency regardless of
data rate as shown in Figure 13.
Table 13. Mask and Phase Noise Adjustments

SerDes Data Rate (GBd) Phase Noise Mask Offset (dB) Rj Multiplier
5 0.00 1.00
3.125 4.08 1.60
2.5 6.02 2.00
1.25 12.04 4.00
www.ti.com Clocking
3.5.3 Deterministic Jitter

Reference clock sources may also contain deterministic jitter in the form of spurs that must also be
accounted for. The effects of spurs vary depending upon whether they are correlated or uncorrelated. An
exact determination of the effects of the spurs would be very complicated. (There are published papers
concerning the complexities of combining the noise power of spurs.) A simplified formula is shown below:
2 SpursN (dBc )
Jitterspurs = å1N 10 £ 0.1 ps
2 g P Freference 10 (9)
The limit of 0.1 ps is a level chosen to provide a reasonable margin without complex analysis. This
equation has sufficient margin to be used for all interface types and baud rates.
3.6 Unused Clock Inputs

Any unused LJCB or LVDS differential clock inputs should be connected to the appropriate rails to
establish a valid logic level.
The recommended connections are shown in Figure 14. The added 1-kΩ resistor is designed to reduce
power consumption. The positive terminal should be connected to the respective power rail.
Power
Rail
DSP
100 W
N
1 kW
Figure 14. Unused Clock Input Connection
Clocking www.ti.com
Table 14 shows the specific rails to which each unused clock input should be connected (in accordance
with Figure 14).
Table 14. Managing Unused Clock Inputs

(1)
Clock Input Power Supply Rail Input Buffer Type
SYSCLKP CVDD LJCB
SYSCLKN
PASSCLKP CVDD LJCB
PASSCLKN
ALTCORECLKP CVDD LJCB
ALTCORECLKN
SRIOSGMIICLKP CVDD LJCB
SRIOSGMIICLKN
DDRCLKP CVDD LJCB
DDRCLKN
PCIECLKP CVDD LJCB
PCIECLKN
MCMCLKP CVDD LJCB
MCMCLKN
RP1CLKP DVDD18 LVDS
RP1CLKN
RP1FBP DVDD18 LVDS
RP1FBN
(1) The power rails must be identical to those directly supporting the intended device.
3.7 Single-Ended Clock Sources

Unlike some previous Texas Instruments DSPs, KeyStone I devices do not allow single-ended clock input
sources. Differential clock sources provide better noise immunity and signal integrity, and are required by
KeyStone I devices.
NOTE: It is possible to use a single-ended clock source, however, the complexity and difficulty in
properly biasing the circuit as well as the potential effects on signal quality, slew rates,
overshoots and undershoots make the single-ended solution undesirable.
3.8 Clocking and Clock Trees (Fanout Clock Sources)

For systems with multiple high-performance processor devices, it is not recommended that a single clock
source be used (single output only). Excessive loading, reflections, and noise will adversely impact
performance. Instead, it is recommended that differential clocking be used, which includes a differential
clock tree.
Texas Instruments has developed a specific line of clock sources to meet the challenging requirements of
today’s high performance devices. In most applications the use of these specific clock sources eliminates
the need for external buffers, level translators, external jitter cleaners, or multiple oscillators.
A few of the recommended parts include:
• CDCM6208 - Single PLL, 8 Differential Output Clock Tree with 4 Fractional Dividers
• CDCE62005 - Single PLL, 5 Differential Output Clock Tree with Jitter Cleaner
• CDCE62002 - Single PLL, 2 Differential Output Clock Tree with Jitter Cleaner
NOTE: These parts are specifically designed for the KeyStone I devices. To minimize cost and
maximize performance, all parts accept a differential, single-ended, or crystal input clock
source.
For clock tree input requirements and functionality, see the device-specific data manuals.
www.ti.com Clocking
3.8.1 Clock Tree Fanout Solution

The following section provides examples for routing between a CDCM6208 to one and multiple KeyStone I
devices. A similar topology should be used regardless of the end-use application hardware. Figure 15
shows an example of a CDCM6208 used to provide all the clocks needed for a KeyStone I device.
DSP
OSC OUT DDR

CLK IN
PLL Div OSC OUT SYSCLK
Xtal
OSC OUT SRIO
OSC IN
SRIO SW
PCIe
SCL AltCoreClk
F-Div OSC OUT Alt Clk

SDA
Cntrl F-Div OSC OUT cPLD
RST
F-Div OSC OUT USB
PDN
F-Div OSC OUT RPI
CDCM6208
Figure 15. Clock Fan Out - for Single Device
The CDCM6208 incorporates fractional dividers. When assigning fractional divider outputs from the
CDCM6208, it will be necessary to verify that the input jitter and performance meet or exceed the
requirements for the respective device input (select the correct clock outputs for the proper clock inputs).
Clocking www.ti.com
Figure 16 shows the recommended clock source as applied to multiple DSPs using alternate TI clock
sources (CDCL6010 and CDCE62005). Terminations are not shown, but must be included where
required. Additional application hardware topologies may dictate different configurations based on trace
lengths and routing rules. It is always recommended to model the clocking topologies to confirm that the
design has been optimized.
66.67 MHz 2
2
DDRCLKP/N
25.00 MHz 66.67 MHz
CDCE62005 – A.1 66.67 MHz 2
DDRCLKP/N
XTAL DDRCLKP/N
66.67 MHz 2
DDRCLKP/N
122.88 MHz 2
122.88 MHz 2
ALTCORECLKP/N
122.88 MHz 2
ALTCORECLKP/N
122.88 MHz 2
ALTCORECLKP/N
30.72 MHz
ALTCORECLKP/N
OSC CDCL6010-A 122.88 MHz 2
2
PA_SS
122.88 MHz PA_SS
122.88 MHz 2
PA_SS
122.88 MHz 2
PA_SS
156.25 MHz 2
PCIeCLKP/N
156.25 MHz 2
156.25 MHz 2
PCIeCLKP/N
PCIeCLKP/N
156.25 MHz 2
30.72 MHz PCIeCLKP/N
OSC CDCL6010-A 312.50 MHz 2
2
MCMCLKP/N
312.50 MHz MCMCLKP/N
312.50 MHz 2
312.50 MHz 2
MCMCLKP/N
MCMCLKP/N
312.50 MHz 2
2
SRIOSGMIICLKP/N
25.00 MHz 312.50 MHz
XTAL CDCE62005 - D 312.50 MHz 2
SRIOSGMIICLKP/N
SRIOSGMIICLKP/N
312.50 MHz 2
SRIOSGMIICLKP/N
153.60 MHz 2 SYSCLKP/N
30.72 MHz CDCE62005 - B 153.60 MHz 2
XTAL SYSCLKP/N
30.72 MHz 2
RP1CLKP/N
30.72 MHz 2
RP1CLKP/N
30.72 MHz CDCE62005 - D 30.72 MHz 2
RP1CLKP/N
XTAL 30.72 MHz 2 RP1CLKP/N
DSP 0 DSP 1 DSP 2 DSP 3
Figure 16. Clock Fan Out - Multiple DSPs
3.8.1.1 Clock Tree Layout Requirements

Key considerations when routing between clock sources and the device include the problems associated
with reflections, crosstalk, noise, signal integrity, signal levels, biasing, and ground references. To
establish the optimal clocking source solution, the requirements in the sections below should be followed.
As with all high-performance applications, it is strongly recommended that you model (simulate) the
clocking interface to verify the design constraints in the end-use application. Application board stack up,
component selection, and the like all have an impact on the characteristics identified below.
For additional details pertaining to routing and interconnection, see the Clocking Design Guide for
KeyStone Devices Application Report.
3.8.2 Clock Tree Trace Width

Clocking trace widths are largely dependent on frequency and parasitic coupling requirements for adjacent
nets on the application board. As a general rule of thumb, differential clock signals must be routed in
parallel and the spacing between the differential pairs should be a minimum 2 times (2×) the trace width.
Single-ended nets should have a spacing of 1.5× the distance of the widest parallel trace width.
3.8.3 Clock Tree Spacing

Given the variation in designs, the minimum spacing between any adjacent signals should be a minimum
of 1× the width. A spacing of 1× is typically suitable for control signals (or static signals) and not data or
clock nets. Single-ended or differential nets should be spaced from other nets a minimum of 1.5× and 2×,
respectively. A detailed cross-talk and coupling analysis (signal integrity) should be performed before any
design is released to production. An additional rule of thumb would be to maintain trace spacing ≥ three
times the dielectric height (reduces ground bounce and crosstalk).
www.ti.com Clocking
3.8.4 Clock Transmission Lines - Microstrip versus Stripline

For additional detailed information regarding microstrip and stripline features and selection criteria, see
Section 10.6.
3.8.5 Clock Component Selection and Placement

A proper clock source is critical to ensure the KeyStone I device maintains its intended functionality. TI
strongly recommends the use of the clock sources developed to complement the device (see Section 3.8).
These clock sources have been optimized for performance, jitter, and form-factor. The recommended
clock source (CDCM6208) has been designed to accept either a low frequency crystal, LVPECL, LVDS,
HCSL, or LVCMOS input and is capable of sourcing either a LVPECL, LVDS, HCSL, or LVCMOS clock to
the KeyStone I device and peripheral logic.
All clock sources should be placed as close to the targeted devices as possible. Excessive trace length
induces many problems that are costly to correct in the end product and usually require a respin of the
application hardware.
Alternative clock sources have been identified and include the CDCE62002 and CDCE62005. These
alternative clock sources will provide functionally identical resources when implemented correctly.
3.8.6 Clock Termination Type and Location

All input clocks are presumed to be of the differential type and require proper terminations or coupling.
Each differential pair must be AC-Coupled using a 0.1-µF ceramic capacitor (0402 size or smaller
recommended). In some cases, depending on topology, a 1-µF capacitor can be used. All AC-coupling
capacitors should be located at the destination end as close to the device as possible.
Figure 17 shows the intended connection between the clock source and DSP.
Low Jitter
Clock Source
+ +
Clock
Source DSP
- -
Figure 17. Clock Termination Location
Clocking www.ti.com
3.9 Clock Sequencing

Table 15 shows the requirements for input clocks to the KeyStone I device. Time frames indicated are
minimum. Maximum time frames are defined in the device-specific data manual and should not be
violated. For specific timing requirements, see the device-specific data manual.
NOTE: All clock drivers must be in a high-impedance state until CVDD (at a minimum) is at a valid
level. After CVDD is at a valid level, all clock inputs must either be active or in a static state.
Table 15. Clock Sequencing

Clock Condition Sequencing
DDRCLK None Must be present 16 µs before POR transitions high.
SYSCLK CORECLKSEL = 0 SYSCLK used to clock the core PLL. It must be present 16 µs before POR transitions high.
CORECLKSEL = 1 SYSCLK only used for AIF. Clock most be present before the reset to the AIF is removed.
CORECLK None CORECLK used to clock the core PLL. It must be present 16 µs before POR transitions
high.
ALTCORECLK CORECLKSEL = 0 ALTCORECLK is not used and should be terminated as shown in Table 14 and Figure 14.
CORECLKSEL = 1 ALTCORECLK is used to clock the core PLL. It must be present 16 µs before POR
transitions high.
PASSCLK PASSCLKSEL = 0 PASSCLK is not used and should be terminated as shown in Table 14 and Figure 14.
PASSCLKSEL = 1 PASSCLK is used as a source for the PA_SS PLL. It must be present before the PA_SS
PLL is removed from reset and programmed.
SRIOSGMIICLK An SGMII port will be SRIOSGMIICLK must be present 16 µs before POR transitions high.
used
SGMII will not be used. SRIOSGMIICLK must be present 16 µs before POR transitions high.
SRIO will be used as a
boot device
SGMII will not be used. SRIOSGMIICLK is used as a source to the SRIO SerDes PLL. It must be present before
SRIO will be used after the SRIO is removed from reset and programmed.
boot
SGMII will not be used. SRIOSGMIICLK is not used and should be terminated as shown in Table 14 and Figure 14.
SRIO will not be used.
PCIECLK PCIE will be used as a PCIECLK must be present 16 µs before POR transitions high.
boot device.
PCIE will be used after PCIECLK is used as a source to the PCIE SerDes PLL. It must be present before the PCIE
boot. is removed from reset and programmed.
PCIE will not be used. PCIECLK is not used and should be terminated as shown in Table 14 and Figure 14.
4 JTAG
The JTAG interface on KeyStone I devices is used to communicate with test and emulation systems.
Although JTAG is not required for operation, it is strongly recommended that a JTAG connection be
included in all designs.
4.1 JTAG / Emulation

Relevant documentation for the JTAG/Emulation:
• Emulation and Trace Headers Technical Reference Manual (but note the differences defined below)
• Boundary Scan Test Specification (IEEE-1149.1)
• AC Coupled Net Test Specification (IEEE-1149.6)
• Clocking Design Guide for KeyStone Devices Application Report
www.ti.com JTAG
4.1.1 Configuration of JTAG/Emulation

The IEEE Standard 1149.1-1990, IEEE Standard Test Access Port and Boundary-Scan Architecture
(JTAG) interface can be used for boundary scan and emulation. The boundary scan implementation is
compliant with both IEEE-1149.1 and 1149.6 (for SerDes ports). Boundary scan can be used regardless of
the device configuration.
As an emulation interface, the JTAG port can be used in various modes:
• Standard emulation: requires only five standard JTAG signals
• HS-RTDX emulation: requires five standard JTAG signals plus EMU0 and/or EMU1. EMU0 and/or
EMU1 are bidirectional in this mode.
• Trace port: The trace port allows real-time dumping of certain internal data. The trace port uses the
EMU[18:00] pins to output the trace data, however, the number of pins used is configurable.
Emulation can be used regardless of the device configuration.
For supported JTAG clocking rates (TCLK), see the device-specific data manual. The EMU[18:00] signals
can operate up to 166 Mbps, depending on the quality of the board-level design and implementation.
Any unused emulation port signals can be left floating.
NOTE: Not all KeyStone I devices contain the same number of emulation or trace pins. For the
number of emulation pins provided in that particular device, see the device-specific data
manual.
4.1.2 System Implementation of JTAG / Emulation

For most system-level implementation details, see the Emulation and Trace Headers Technical Reference
Manual. However, there are a few differences for KeyStone I device implementation compared to the
information in this document:
• Although the document implies 3.3-V signaling, 1.8-V signaling is supported as long as the TVD source
is 1.8 V.
For a single device connection in which the trace feature is not used, the standard non-buffered
connections can be used along with the standard 14-pin connector. If the trace feature is used (which
requires the 60-pin emulator connector), the five standard JTAG signals should be buffered and TCLK and
RTCLK should be buffered separately. It is recommended to have the option for an AC parallel termination
on TCLK because it is critical that the TCLK have a clean transition. EMU0 and EMU1 should not be
buffered because these are used as bidirectional signals when used for HS-RTDX.
For a system with multiple DSPs that do not use the trace analysis features, the JTAG signals should be
buffered as described above but the standard 14 pin connector can be used.
There are two recommended solutions if trace analysis is desired in a system with multiple DSPs:
• Emulator with trace, solution #1: trace header for each DSP (see Figure 18).
– Pros
• Most simple
• Most clean solution, electrically
– Cons
• Expensive (multiple headers)
• Takes up board real estate
• No global breakpoints, synchronous run/halt
JTAG www.ti.com
DSP #1 DSP #2 DSP #n
EMU[2:n]
EMU[1:0]
EMU[2:n]
EMU[1:0]
EMU[2:n]
EMU[1:0]
TDI, TDO, TMS, TCK, JTAG
TDI, TDO, TMS, TCK, JTAG
JTAG
TDI, TDO, TMS, TCK,
TCKRTn, TRSTn
TCKRTn, TRSTn
TCKRTn, TRSTn
EMU[2:n]
EMU[1:0]
EMU[2:n]
EMU[1:0]
EMU[2:n]
EMU[1:0]
60 Pin Emulation 60 Pin Emulation 60 Pin Emulation
Header Header Header
Figure 18. Emulator With Trace Solution #1
• Emulator with trace, solution #2: single trace header (see Figure 19).
– Pros
• Fairly clean solution, electrically
• Supports global breakpoints, synchronous run/halt
– Cons
• Supports trace on only one device
• Less bandwidth for trace (EMU0 used for global breakpoints)
• Loss of AET action points on EMU1 (only significant if EMU1 has been used as a trigger
input/output between devices. EMU0 can be used instead if needed).
DSP #1 DSP #2 DSP #n

EMU[2:n]
EMU[1:0]
EMU[2:n]
EMU[1:0]
EMU[2:n]
EMU[1:0]
JTAG
JTAG
JTAG
EMU[2:n]
EMU[1:0]
EMU[2:n]
EMU[1:0]
TDI, TDO, TMS, TCK,

TDI, TDO, TMS, TCK,
TDI, TDO, TMS, TCK,
TCKRTn, T RSTn
TCKRTn, T RSTn
TCKRTn, TRSTn
EMU[2:n]
EMU[1:0]
60 Pin Emulation
Header
Figure 19. Emulator With Trace Solution #2
Although no buffer is shown, for multiple DSP connections, it is recommended to buffer the five standard
JTAG signals (TDI, TDO, TMS, TCK, and TRST).
If trace is used, it is not recommended to add both a 60-pin header and a 14-pin header because of signal
integrity concerns. 60-pin to 14-pin adapters are available to allow connection to emulators that support
only the 14-pin connector (although 14-pin emulators do not support trace features).
Some emulators may not support 1.8-V I/O levels. Check emulators intended to operate with the
KeyStone I device to verify that they support the proper I/O levels. If 1.8-V levels are not supported, a
voltage translator circuit is needed or a voltage converter board may be available. If the KeyStone I device
is in a JTAG chain with devices that have a different voltage level than 1.8 V (for example, 3.3 V), voltage
translation is needed.
www.ti.com JTAG
Figure 20 shows a dual voltage JTAG solution using buffers to perform the voltage translation. The ALVC
family for 3.3 V and the AUC family for 1.8 V are both used because they have similar propagation delays.
3.3V
3.3V
TDI
Optional Emulation Header
/ Emulator
TCKRTn JTAG
3.3V Signals 1.8V Signals
3.3V 1.8V
TMS, TCK, TRSTn TDI, TMS, TCK, TRSTn

ALVC125 AUC125
3.3V Device DSP #n

3.3V
TDO TDI TDO

AVC2T45
1.8V
Figure 20. Emulation Voltage Translation With Buffers
Figure 21 shows an example of using FET switch devices for voltage translation. If EMU0 and EMU1 are
connected, use the following approach because these are bidirectional signals.
3.3V
3.3V
TDI
Optional Emulation Header
/ Emulator
TCKRTn JTAG
3.3V Signals ALVC125 1.8V Signals
3.3V 1.8V
TMS, TCK, TRSTn TDI, TMS, TCK, TRSTn
CBTLV
3.3V Device or TVC DSP #n
Device
TDO TDI TDO
Figure 21. Emulator Voltage Translation With Switches
If the trace signals are supported, they may need to have voltage translation as well. Because of the
speed of the trace signals, it is recommended that these signals be connected directly to the emulation
header.
4.1.3 Unused Emulation Pin Requirement

The TI device contains both a JTAG and emulation pin interface. JTAG is a serial scan mechanism
allowing for communication between a debugger and the device. The emulation pins provide for a high-
performance output of select data captured from the device and transferred to an emulator and debugger.
Device Configurations and Initialization www.ti.com
If the JTAG and emulation interface is not used, all pins except TRST can be left floating. TRST must be
pulled low to ground through a 4.7-kΩ resister.
In the event that the JTAG interface is used and the emulation (or a subset of the emulation pins) interface
is not used, the unused emulation pins can be left floating.
For additional connectivity requirements, see the device-specific data manual and also emulation
documentation.
5 Device Configurations and Initialization

Once the voltage rails and the required clocks are present, the KeyStone I device may be released from
reset and initialized.
Boot modes and certain device configuration selections are latched at the rising edge of RESETFULL via
specific pins. The configuration and boot mode inputs are multiplexed with general-purpose I/O (GPIO)
pins or other pins. Once the level on these pins is latched into the configuration registers, these pins are
released to be used for their primary function. In addition to these inputs, some peripheral usage
(enabled/disabled) may also be determined by peripheral configuration registers after DSP reset. Most of
the peripherals on the device are enabled after reset, however, some are configured as disabled by
default. The basic information on configuration options, boot modes options, and the use of the power
configuration registers can be found in the device-specific KeyStone I data manual.
5.1 Device Reset

The KeyStone I device can be reset in several ways. The methods are defined and described in greater
detail in the device-specific data manual. The device incorporates four external reset pins (POR, RESET,
LRESET, and RESETFULL) and one reset status pin (RESETSTAT). Additional reset modes are available
through internal registers, emulation, and a watchdog timer. Each of the reset pins (POR, RESET,
LRESET, and RESETFULL) must have either a high or low logic level applied at all times. They should
not be left floating.
NOTE: For device reset requirements, see the device-specific data manual.
5.1.1 POR
POR is an active-low signal that must be asserted during device power-up to reset all internal
configuration registers to a known good state. In addition, POR must be asserted (low) on a power-up
while the clocks and power planes become stable. Most output signals will be disabled (Hi-Z) while POR is
low.
During a POR condition, RESET should be de-asserted before POR is de-asserted on power-up,
otherwise the device comes up in the warm reset condition.
Proper POR use is required and must be configured correctly. For additional POR details, see the device-
specific data manual.
The internal POR circuit is unable to detect inappropriate power supply levels, including power supply
brown-outs. It is necessary that all power supplies be at valid levels (stable) prior to the release of POR.
The use of power-good logic in the controlling POR circuitry reduces the risk of device degradation due to
power failures. Power-good logic can re-assert POR low when a power supply failure or brown-out occurs
to prevent over-current conditions.
5.1.2 RESETFULL
RESETFULL is an active-low signal that will reset all internal configuration registers to a known good
state. RESETFULL performs all the same functions as POR and is also used to latch the BOOTMODE
and configuration pin data. RESETFULL must be asserted (low) on a power-up while the clocks and
power planes become stable, and RESETFULL must remain low until the specified time after POR has
been released. For the timing requirements of POR and RESETFULL, see the device-specific data
manual. Once POR has been released, the RESETFULL signal may be toggled at any time to place the
KeyStone I device into a default state.
www.ti.com Device Configurations and Initialization
Both POR and RESET should be high before RESETFULL is deasserted, otherwise the device may not
boot correctly.
The proper use of RESETFULL is required and must follow all timing requirements. For additional
RESETFULL details, see the device-specific data manual. RESETFULL and POR must be controlled
separately. The KeyStone I part will not operate correctly if they are tied together.
If POR is asserted due to an incorrect power supply voltage, then RESETFULL must also be asserted
until the prescribed time after POR to assure that the proper configuration is latched into the device.
5.1.3 RESET
RESET is a software-configurable reset function to the device that is available either externally via a pin,
or internally through an MMR (memory-mapped register). When asserted (active low), RESET functions to
reset everything on the device except for the test, emulation, and logic blocks that have reset isolation
enabled.
When RESET is active low, all 3-state outputs are placed in a high-impedance (Hi-Z) state. All other
outputs are driven to their inactive level. The PLL multiplier and PLL controller settings return to default
and must be reprogrammed if required. If necessary, the reset isolation register within the PLL controller
can be modified to block this behavior. This setting is mandatory for reset isolation to work. For additional
details on reset isolation, see the applicable PLL Controller Specification.
NOTE: RESET does not latch-boot mode and configuration pins. The previous values latched or
programmed into the Device Status Register remain unchanged.
5.1.4 LRESET
The LRESET signal can be used to reset an individual CorePac or all CorePacs simultaneously. This
signal, along with the NMI signal, can be used to affect the state of an individual CorePac without
changing the state of the rest of the part. It is always used in conjunction with the CORESEL inputs and
the LRESETNMIEN input for proper operation. LRESET does not reset the peripheral devices, change the
memory contents, or change the clock alignment. LRESET does not cause RESETSTAT to be driven low.
LRESET must be asserted properly to function correctly. All setup, hold, and pulse-width timing for the
proper implementation of LRESET can be found in the KeyStone I device-specific data manual. For proper
operation, the following steps should be implemented:
1. Select the CorePac to be reset by driving the code for that CorePac onto the CORESEL input pins.
The code values can be found in the device-specific data manual.
2. Drive the LRESET input low. This can be done simultaneously with the CORESEL pins.
3. Once both the CORESEL inputs and the LRESET input have been valid for the specified setup time,
drive the LRESETNMIEN input low.
4. Keep the LRESETNMIEN input low for the specified minimum pulse width time.
5. Drive LRESETNMIEN high.
6. Keep the values of CORESEL and LRESET valid for the specified hold time.
5.1.4.1 Unused LRESET Pin

If the LRESET is not used, TI recommends that it be pulled high to DVDD18 (1.8 V) with an external 4.7-
kΩ resistor to ensure that the CorePacs are not unintentionally reset. (This is a recommendation, not a
requirement. It adds additional noise margin to this critical input.) The LRESETNMIEN and CORESEL
inputs are also used in conjunction with the NMI to send an interrupt to one of the CorePacs. If neither
LRESET or NMI are used, then it is recommended that the LRESET, NMI, and LRESETNMIEN inputs be
pulled high to DVDD18 (1.8-V) with 4.7-kΩ resistors. The CORESEL pins can be left unconnected, relying
on their internal resistors to hold them in a steady state.
Device Configurations and Initialization www.ti.com
5.1.5 Reset Implementation Considerations

The sequencing of the POR, RESETFULL, and RESET signals requires some logic to ensure that the
signals are of proper length and are properly spaced. POR should be driven by a power management
circuit that releases the signal after all power supplies have been enabled and are stable. If any of the
power supplies fail during normal operation, the power management circuit should reassert the POR
signal. RESETFULL can be used to reset the part through control logic in your design. RESETFULL can
be asserted while POR is high but the control logic should monitor the POR signal and reassert
RESETFULL if POR is driven low. RESET must be high before POR and RESETFULL are deasserted.
5.1.6 RESETSTAT
The RESETSTAT signal indicates the internal reset state. The RESETSTAT pin is driven low by almost all
reset initiators and this pin remains low until the device completes initialization. The only resets that do not
cause RESETSTAT to be driven low are initiators of local CorePac reset such as the LRESET. More
information is available in the device-specific data manual. Access to the RESETSTAT signal is important
for debugging reset problems with the device. This signal should be routed to external logic or to a test
point.
5.1.7 EMULATION RESET

The KeyStone I family of devices supports the emulation reset function as part of its standard offering.
Emulation reset supports both a hard and soft reset request. The source for this reset is on-chip emulation
logic. This reset behavior is the same as hard and soft reset based on the reset requester type. This reset
is non-blockable.
5.2 Device Configuration

Several device configuration strapping options are multiplexed with the GPIO pins. For pin assignments,
see the device-specific data manual. There are dedicated configuration pins, such as CORECLKSEL and
DDRSLRATE1:0. The state of these pins is not latched and must be held at the desired state at all times.
For details on the configuration options, see the device-specific data manual.
Device configuration pins should be pulled to ground or to 1.8 V (DVDD18) to establish a configuration
level.
If the GPIO signals are not used, the internal pull-up (to DVDD18) and pull-down (to ground) resistors can
be used to set the input level and external pull-up/down resistors are needed only if the opposite setting is
desired. If the GPIO pins are connected to other components or a test point, the internal pull-up/pull-down
resistor should not be relied upon. External pull-up and pull-down resistors are recommended for all
desired settings. If the pin is pulled in the same direction as the internal pull, then a 4.7-kΩ resistor should
be used. If the pin is pulled in the opposite direction as the internal pull, then a 1-kΩ resistor should be
used.
If multiple configuration inputs are tied together from one or more KeyStone I devices (allowed only for
input-only pins), the external resistor should have a maximum value of 1000/N where N is the number of
input pins. If the boot mode or configuration pin is used after the boot mode is captured, then individual
resistors should be used.
The phase-locked loop (PLL) multipliers can be set only by device register writes. For details on
configuration of the PLL, see the KeyStone Architecture Phase-Locked Loop (PLL) User's Guide
(SPRUGV2) and the device-specific data manual.
5.3 Peripheral Configuration

In addition to the device reset configuration, which is covered in the previous section, there are several
configuration pins to set. These include: bit ordering (endian), PCI mode, and PCI enable. For correct pin
configuration, see the device-specific data manual and Section 6.1.
www.ti.com Device Configurations and Initialization
Any additional configurations not listed in this or previous sections are done through register accesses.
Peripherals that default to disabled can be enabled using the peripheral configuration registers. If the boot
mode selection specifies a particular interface for boot (SRIO, Ethernet, I2C), it is automatically enabled
and configured. For additional details on peripheral configurations, see the Device Configuration section of
the device-specific data manual.
In some cases, the peripheral operating frequency is dependent on the device core clock frequency or
boot mode inputs. This should be taken into account when configuring the peripheral.
5.4 Boot Modes

The KeyStone I device contains multiple interfaces that support boot loading. Examples include: Ethernet
Media Access Controller (EMAC), Serial Rapid IO (SRIO), Peripheral Component Interconnect – express
(PCIe), Emulation, Inter-Integrated Circuit (I2C), Serial Peripheral Interface (SPI), and HyperLink.
For details regarding boot modes, see the KeyStone Architecture Bootloader User's Guide and the device-
specific data manual. Regardless of the boot mode selected, a connected emulator can always reset the
device and acquire control.
6 Peripherals Section
This section covers each of the KeyStone I device peripherals and modules (some may not be included,
see the device-specific data manual). This section is intended to be used in addition to the information
provided in the device-specific data manual, application reports, and the peripheral and module guides.
The four types of documents should be used as follows:
• Data Manual: AC Timings, register offsets
• Module Guide: Functional Description, Programming Guide
• Applications Reports: System level issues
• This Section: Configuration, system level issues not covered in a device-specific application report
Each peripheral section includes recommendations on how to handle pins on interfaces that are disabled
or for unused pins on interfaces that are enabled. Generally, if internal pull-up or pull-down resistors are
included, the pins can be left unconnected. Any pin that is output-only can always be left unconnected.
Normally, if internal pull-up and pull-down resistors are not included, pins can still be floated with no
functional issues for the device. However, this normally causes additional leakage currents that can be
eliminated if external pull-up or pull-down resistors are used. Inputs that are not floating have a leakage
current of approximately 100 µA per pin. Leakage current is the same for a high- or low-input (either pull-
up or pull-down resistors can be used). When the pins are floating, the leakage can be several milliamps
per pin. Connections directly to power or ground can be used only if the pins can be assured to never be
configured as outputs and the boundary scan is not run on those pins.
6.1 GPIO/Device Interrupts

Documentation for GPIO/Interrupts:
• KeyStone Architecture General Purpose Input/Output (GPIO) User's Guide
• KeyStone IBIS Model File (specific to device part number)
• Using IBIS Models for Timing Analysis
6.1.1 Configuration of GPIO/Interrupts

All GPIOs pins are multiplexed with other functions, with several being used for device configuration
strapping options. The strapping options are latched by RESETFULL going high. After RESETFULL is
high, the pins are available as GPIO pins.
NOTE: All pins must be driven to a logical state during boot and through RESETFULL after being
released.
The GPIO pins are read during a RESETFULL boot function and, if required, are available for GPIO
functionality after a successful boot.
Peripherals Section www.ti.com
GPIOs are enabled at power-up and default to inputs.

All GPIOs can be used as interrupts and EDMA events to any of the cores.
6.1.2 System Implementation of GPIO/Interrupts

It is recommended that GPIOs used as outputs have a series resistance (22 Ω or 33 Ω being typical
values). The value (or need) for the series resistor can be determined by simulating with the IBIS models.
If you want to have a GPIO input default to a particular state (low or high), use an external resistor. A pull-
up resistor value of 1 kΩ to 1.8 V (DVDD18) is recommended to make sure that it overrides the internal
pull-down resistor present on some GPIOs. If this GPIO is also used as a strapping option, the default
state needs to also be the desired boot strapping option. The RESETSTAT signal can be used to Hi-Z
logic that drives a GPIO boot strapping state during the POR transition.
6.1.3 GPIO Strapping Requirements

Table 16 shows the strapping requirements for one of the KeyStone I family devices. For the pin numbers,
see the device-specific data manual. Included is the pin-to-function correlation needed to properly
configure the device.
Table 16. GPIO Pin Strapping Configurations

Primary Function
Pin Name Boot Mode Pull-up Pull-dn Secondary Function Comment
AJ20 GPIO00 LENDIAN Little Endian Big Endian GPIO (after RESETFULL)
AG18 GPIO01 BOOTMODE00 Boot Device Boot Device GPIO (after RESETFULL) For additional details, see
Section 2.5
AD19 GPIO02 BOOTMODE01 Boot Device Boot Device GPIO (after RESETFULL) For additional details, see
Section 2.5
AE19 GPIO03 BOOTMODE02 Boot Device Boot Device GPIO (after RESETFULL) For additional details, see
Section 2.5
AF18 GPIO04 BOOTMODE03 Device Cfg Device Cfg SR ID Device-specific configurations
AE18 GPIO05 BOOTMODE04 Device Cfg Device Cfg SR ID Device-specific configurations
AG20 GPIO06 BOOTMODE05 Device Cfg Device Cfg GPIO (after RESETFULL) Device-specific configurations
AH19 GPIO07 BOOTMODE06 Device Cfg Device Cfg GPIO (after RESETFULL) Device-specific configurations
AJ19 GPIO08 BOOTMODE07 Device Cfg Device Cfg GPIO (after RESETFULL) Device-specific configurations
AE21 GPIO09 BOOTMODE08 Device Cfg Device Cfg GPIO (after RESETFULL) Device-specific configurations
AG19 GPIO10 BOOTMODE09 Device Cfg Device Cfg GPIO (after RESETFULL) Device-specific configurations
AD20 GPIO11 BOOTMODE10 PLL Multiplier/I2C PLL Multiplier/I2C GPIO (after RESETFULL) Device-specific configurations
AE20 GPIO12 BOOTMODE11 2 2 GPIO (after RESETFULL) Device-specific configurations
PLL Multiplier/I C PLL Multiplier/I C
AF21 GPIO13 BOOTMODE12 PLL Multiplier/I2C PLL Multiplier/I2C GPIO (after RESETFULL) Device-specific configurations
AH20 GPIO14 PCIESSMODE Endpt/RootComplex Endpt/RootComplex GPIO (after RESETFULL) Device-specific configurations
0
AD21 GPIO15 PCIESSMODE Endpt/RootComplex Endpt/RootComplex GPIO (after RESETFULL) Device-specific configurations
1
NOTE: For additional details, see the Boot and Configuration Specification in addition to the device-
6.1.4 Unused GPIO Pin Requirement

All GPIO pins serve multiple functions, so each of the GPIO pins are required to be pulled to a logical
state during boot (as defined by the boot method and as listed in the Bootloader Specification). After boot
is complete (POR is deasserted), each GPIO pin is released and made available to the user as a general
purpose IO. Each GPIO pin (after boot is complete) defaults to its identified state, unless otherwise
configured. For default pull-up and pull-down conditions, see the device-specific data manual. All
application hardware must incorporate an appropriate design to permit flexibility between device
configuration, boot and functional GPIO requirements.
www.ti.com Peripherals Section
If an external resistor is used to obtain a specific post boot state for any GPIO pin, use a 1-kΩ to 4.7-kΩ
resistor to 1.8 V (DVDD18).
NOTE: If any of the GPIO pins are attached to a trace, test point, another device, or something other
than a typical no-connect scenario, a pull-up or pull-down resistor is mandatory. The default
internal pull up resistor for GPIO00 and the default internal pull down resistors for GPIO01
through GPIO15 are suitable only if nothing is connected to the pin.
6.2 HyperLink Bus

Documentation for HyperLink:
• KeyStone Architecture HyperLink User's Guide
• TMS320C66x DSP CPU and Instruction Set Reference Guide
• KeyStone IBIS Model File (see the device-specific part number)
• SerDes Implementation Guidelines for KeyStone I Devices
6.2.1 System Implementation of HyperLink Interface

The HyperLink interface is a high-performance TI-developed interface intended for high speed data
communication. For additional configuration details, see the KeyStone Architecture HyperLink User's
Guide (SPRUGW8).
For details, guidelines, and routing requirements, see the SerDes Implementation Guidelines for KeyStone
I Devices (SPRABC1). This interface is intended to be a device-to-device interconnection. It is not
intended for a board-to-board interconnection. Currently, HyperLink is supported only by TI and select
third party FPGA IPs.
6.2.2 HyperLink-to-HyperLink Connection

The HyperLink interface bus is intended to be a short-reach high-performance bus supporting a total of
four interconnecting lanes between two DSPs operating at up to 10.0 GHz per lane (other configurations
may exist). HyperLink data is passed using up to four SerDes lanes. In addition, the HyperLink includes a
number of sideband signals used for flow control and power management control. The sideband signals
use LVCMOS buffers and operate at speeds up to 166 MHz. Both the SerDes lanes and the sideband
signals must be connected for proper operation.
0.1uF
MCMTXP 0 MCMRXP 0
0.1uF
MCMTXN 0 MCMRXN 0
0.1uF
DSP MCMTXP 1 MCMRXP 1 DSP
0.1uF
MCMTXN 1 MCMRXN 1
0.1uF
MCMTXP 2 MCMRXP 2
0.1uF
MCMTXN 2 MCMRXN 2
0.1uF
MCMTXP 3 MCMRXP 3
0.1uF
MCMTXN 3 MCMRXN 3
0.1uF
MCMRXP 0 MCMTXP 0
0.1uF
MCMRXN 0 MCMTXN 0
0.1uF
MCMRXP 1 MCMTXP 1
0.1uF
MCMRXN 1 MCMTXN 1
0.1uF
MCMRXP 2 MCMTXP 2
0.1uF
MCMRXN 2 MCMTXN 2
0.1uF
MCMRXP 3 MCMTXP 3
0.1uF
MCMRXN 3 MCMTXN 3
MCMTXFLCLK MCMRXFLCLK
MCMTXFLDAT MCMRXFLDAT
MCMTXPMCLK MCMRXPMCLK
MCMTXPMDAT MCMRXPMDAT
MCMRXFLCLK MCMTXFLCLK
MCMRXFLDAT MCMTXFLDAT
MCMRXPMCLK MCMTXPMCLK
MCMRXPMDAT MCMTXPMDAT
Figure 22. HyperLink Bus DSP-to-DSP Connection
6.2.3 Configuration of HyperLink

If the HyperLink interface is used, a MCMCLKP/N clock must be provided and the SerDes PLL must be
configured to generate the desired line rate. The line rate is determined using the MCMCLK frequency, the
PLL multiplier, and the rate scale setting. The PLL multiplier value is found in the PLL configuration
register and the rate scale setting is found in the transmit and receive configuration registers. The PLL
output frequency is equal to the frequency of the MCMCLKP/N multiplied by the PLL multiplier.
• PLL Output (MHz) = PLL multiplier × MCMCLK (MHz)
The VCO frequency range of the HyperLink PLL is 1562.5 MHz to 3125 MHz, so the PLL output frequency
must fall within this range.
The highest HyperLink line rates supported by KeyStone I devices is 10000 Mbps. The rate scale is used
to translate the output of the SRIO PLL to the line rate using the following formula:
• Line rate Mbps = PLL output/rate scale
The HyperLink SerDes interface supports the four rate scales and the associated rate scale value are
shown in Table 17.
Table 17. HyperLink Rate Scale Values

Rate Scale Rate Scale Value Line Rate (Mbps)
Full 0.25 PLL Output / 0.25
Half 0.5 PLL Output / 0.5
Quarter 1 PLL Output / 1
8th 2 PLL Output / 2
The possible PLL multiplier settings for the three recommended MCMCLKP/N clock frequencies are
shown in Table 18. The table includes only the settings for the highest rate. The HyperLink interface can
be operated at slower rates. The equations above can be used to generate the settings for lower rates.
Table 18. HyperLink PLL Multiplier Settings

Reference Clock MHz PLL Multiplier PLL Output (MHz) Rate Scale Line Rate (Mpbs)
156.25 16 2500 0.25 10000
250.00 10 2500 0.25 10000
312.50 8 2500 0.25 10000
6.2.4 Unused HyperLink Pin Requirement

All HyperLink clock and data pins (MCMRXFLCLK, MCMRXFLDAT, MCMTXFLCLK, MCMTXFLDAT,
MCMRXPMCLK, MCMRXPMDAT, MCMTXPMCLK, MCMTXPMDAT) can be left floating when the entire
HyperLink peripheral is not used. Each unused HyperLink clock and data pin will be pulled low when not
in use through internal pull-down resistors.
When only a single lane is used instead of four, all clock and data pins must be connected. Unused lanes
(both transmit and receive) can be left floating.
If the HyperLink interface is not being used, the peripheral should be disabled in the MMR.
If the HyperLink interface is not required, the HyperLink regulator power pin must still be connected to the
correct supply rail with the appropriate decoupling capacitance applied. In most devices, this is the
VDDR1 supply pin. To identify the supply pin, see the device-specific data manual. The EMI/noise filter is
not required when this interface is not used.
If unused, the primary HyperLink clock input must be configured as indicated in Section 3.6. Pull-up must
be to the CVDD core supply.
6.3 Inter-Integrated Circuit (I2C)

Documentation for I2C:
• KeyStone Architecture Inter-IC Control Bus (I2C) User's Guide
• IBIS Model File for KeyStone Devices (device-specific part number)
• Using IBIS Models for Timing Analysis Application Report
• Philips I2C Specification, Version 2.1
6.3.1 Configuration of I2C

The I2C peripheral powers up enabled. The input clock for the I2C module is SYSCLK, which is further
divided down using an internal PLL. There is a prescaler in the I2C module that must be set up to reduce
this frequency. For additional information and details, see the Bootloader for KeyStone Architecture User’s
Guide and Inter-Integrated Circuit (I2C) for KeyStone Devices User's Guide.
6.3.2 System Implementation of I2C

External pull-up resistors to 1.8 V are needed on the device I2C signals (SCL, SDA). The recommended
pull-up resistor value is 4.7 kΩ.
Multiple I2C devices can be connected to the interface, but the speed may need to be reduced (400 kHz is
the maximum) if many devices are connected.
The I2C pins are not 2.5 V or 3.3 V tolerant. For connection to 2.5-V or 3.3-V I2C peripherals, the
PCA9306 can be used. For more information, see the PCA9306 Dual Bidirectional I2C Bus and SMBus
Voltage-Level Translator Data Sheet . Section 6.3.3 describes the interconnection in greater detail.
6.3.3 I2C Interface

KeyStone I I2C pins are open-drain IOs, and require pull-up resistors to the DVDD18 rail (same rail as the
device 1.8-V supply).
Figure 23 shows the intended configuration inclusive of the PCA9306, DSP, and, as an example, an
EEPROM. This configuration shows the connection topology for a single DSP. In the case in which
multiple DSPs are used, each I2C interface must include the same configuration.
1.8V Proper Voltage Level
DSP H I2C Interface

Logic / Glue Logic
SD PCA9306 SD
SC SC
Figure 23. I2C Voltage Level Translation
6.3.4 Unused I2C Pin Requirement

If the I2C signals are not used, it is recommended that the SDA and SCL pins be pulled up to 1.8 V
(DVDD18). These pins can be left floating, but this will result in a slight increase in power consumption
due to increased leakage.
6.4 Ethernet
Documentation for EMAC:
• KeyStone Architecture Gigabit Ethernet (GbE) Switch Subsystem User's Guide
• SGMII Specification (ENG-46158), Version 1.8, dated April 27, 2005 [7]
6.4.1 Configuration of EMAC, SGMII and MDIO

The EMAC interface is compliant with the SGMII Specification (ENG-46158), Version 1.8 that specifies
LVDS signals. Only the data channels are implemented, so the connected device must support clock
recovery and not require a separate clock signal. When the EMAC is enabled, the MDIO interface is
enabled.
The MDIO interface (MDCLK, MDIO) uses 1.8-V LVCMOS buffers. The EMAC must be enabled via
software before it can be accessed unless the Boot over Ethernet-boot mode is selected.
If the EMAC is used, a SRIOSGMIICLKP/N clock must be provided and the SerDes PLL must be
configured to generate a 1.25-Gbps line rate. The line rate is determined using the SRIOSGMIICLK
frequency, the PLL multiplier, and the rate scale setting. The PLL multiplier value is found in the PLL
configuration register and the rate scale setting is found in the transmit and receive configuration registers.
The PLL output frequency is equal to the frequency of the SRIOSGMIICLKP/N multiplied by the PLL
multiplier.
• PLL Output (MHz) = PLL Multiplier × SRIOSGMIICLK (MHz)
The VCO frequency range of the SGMII PLL is 1500 MHz to 3125 MHz, so the PLL output frequency must
fall within this range.
The line rate for SGMII is 1250 Mbps. The rate scale is used to translate the output of the SGMII PLL to
the line rate using the following formula:
• 1250 Mbps = PLL Output / rate scale
The SGMII SerDes interface supports the four rate scales and the associated rate scale value are shown
in Table 19.
Table 19. SGMII Rate Scale Values

Half 1 PLL Output / 1
32nd 16 PLL Output / 16
The possible PLL multiplier settings for the three recommended SRIOSGMIICLKP/N clock frequencies are
shown in Table 20. Although the SGMII SerDes share a reference clock with the SRIO SerDes, they have
separate PLLs that can be configured with different multipliers.
Table 20. SGMII PLL Multiplier Settings

156.25 16 2500 2 1250
250.00 10 2500 2 1250
312.50 8 2500 2 1250
For more detailed information, see the device-specific data manual, user's guide, and application reports.
If EMAC is not used, the SerDes signals and MDIO signals can be left unconnected. If both EMAC and
SRIO are not used, the RIOSGMIICLKP/N pins should be terminated as shown in Section 3.6. For
additional detail, see Section 6.4.4.
6.4.2 System Implementation of SGMII

SGMII uses LVDS signaling. The KeyStone I device uses a CML-based SerDes interface that requires AC
coupling to interface to LVDS levels. Texas Instruments recommends the use of 0.1-µF AC coupling
capacitors for this purpose. The SerDes receiver includes a 100-Ω internal termination, so an external
100-Ω termination is not needed. Examples of SerDes-to-LVDS connections appear in the following
sections.
If the connected SGMII device does not provide common-mode biasing, external components need to be
added to bias the LVDS side of the AC-coupling capacitors to the nominal LVDS offset voltage, normally
1.2 V. Additional details regarding biasing can be found in Clocking Design Guide for KeyStone Devices.
For information regarding supported topologies and layout guidelines, see the SerDes Implementation
Guidelines for KeyStone I Devices .
The SGMII interface supports hot-swap, where the AC-coupled inputs of the device can be driven without
a supply voltage applied.
SRIO/SGMII SerDes power planes and power filtering requirements are covered in Section 7.3.
6.4.3 SGMII MAC to MAC Connection

The SGMII interface can be connected from the KeyStone I device to a PHY or from the KeyStone I
device to a MAC, including KeyStone-to-KeyStone device direct connects. An example of the hardware
connections is shown in Figure 24. For auto-negotiation purposes, the KeyStone I device can be
configured as a master or a slave, or it can be set up for fixed configuration. If the KeyStone I device is
connected to another MAC, the electrical compatibility must be evaluated to determine if additional
terminations are needed.
NOTE: Not all devices contain the implementation shown. Your specific KeyStone I device should
be properly configured for the correct number of SGMII interfaces shown in the device-
0.1µF
SGMIITXP0 SGMIIRXP0
DSP SGMIITXN0
0.1µF
SGMIIRXN0
DSP
0.1µF
SGMIIRXP0 SGMIITXP0
0.1µF
SGMIIRXN0 SGMIITXN0
0.1µF
SGMIITXP1* SGMIIRXP1*
0.1µF
SGMIITXN1* SGMIIRXN1*
0.1µF
SGMIIRXP 1* SGMIITXP1*
0.1µF
SGMIIRXN1* SGMIITXN1*
Note: * Optional on some DSPs
Figure 24. SGMII MAC to MAC Connection
6.4.4 Unused SGMII Pin Requirement

All unused SGMII pins must be left floating. As an additional note, should the SGMII interface not be
required, the peripherals should be disabled. In the event only one of the two SGMII interfaces is required,
the unused SGMII interface must be disabled through the MMR of the device.
The SGMII shares a common input clock with the SRIO interface. If either of these interfaces is used, the
respective power must still be properly connected in accordance with this specification. In addition, if the
SGMII interface is used, the appropriate amount of decoupling/bulk capacitance must be included.
If both of the SGMII interfaces are not used, the SGMII regulator power pin (VDDR3_SGMII) must still be
connected to the correct supply rail with the appropriate decoupling capacitance applied. The EMI/Noise
filter is not required when this interface is not used.
If both SRIO and SGMII are unused, the primary SRIOSGMIICLK clock input must be configured as
indicated in Section 3.6. Pull-up must be to the CVDD core supply.
NOTE: Both interfaces must be unused in order to pull-up the unused clock signal.
6.5 Serial RapidIO (SRIO)

Relevant documentation for SRIO:
• KeyStone Architecture Serial RapidIO (SRIO) User’s Guide
• RapidIO Interconnect Part VI: Physical Layer 1×/4× LP-Serial Specification, Version 2.0.1 [10]
6.5.1 Configuration of SRIO

Detailed information on the configuration of the SRIO logic block can be found in the functional operation
section of the KeyStone Architecture Serial RapidIO (SRIO) User's Guide .
If the SRIO interface is used, a SRIOSGMIICLKP/N clock must be provided and the SerDes PLL must be
configured to generate the desired line rate. The line rate is determined using the SRIOSGMIICLK
frequency, the PLL multiplier, and the rate scale setting. The PLL multiplier value is found in the PLL
configuration register and the rate scale setting is found in the transmit and receive configuration registers.
The PLL output frequency is equal to the frequency of the SRIOSGMIICLKP/N multiplied by the PLL
multiplier:
• PLL Output (MHz) = PLL Multiplier × SRIOSGMIICLK (MHz)
The VCO frequency range of the SRIO PLL is 1562.5 MHz to 3125 MHz, so the PLL output frequency
must fall within this range.
The SRIO line rates supported by KeyStone I devices are 1250 Mbps, 2500 Mbps, 3125 Mbps, and
5000 Mbps. The rate scale is used to translate the output of the SRIO PLL to the line rate using the
following formula:
• Line rate Mbps = PLL Output/rate scale
The SRIO SerDes interface supports the four rate scales and the associated rate scale value shown in
Table 21.
Table 21. SRIO Rate Scale Values

Half 0.5 PLL Output / 0.5
8th 2 PLL Output / 2
The possible PLL multiplier settings for the three recommended SRIOSGMIICLKP/N clock frequencies are
shown in Table 22. Although the SRIO SerDes share a reference clock with the SGMII SerDes, they have
separate PLLs that can be configured with different multipliers.
Table 22. SRIO PLL Multiplier Settings

156.25 16 2500 2 1250
250.00 10 2500 2 1250
312.50 8 2500 2 1250
156.25 16 2500 1 2500
250.00 10 2500 1 2500
312.50 8 2500 1 2500
156.25 10 1562.5 0.5 3125
156.25 20 3125 1 3125
250.00 12.5 3125 1 3125
312.50 5 1562.5 0.5 3125
312.50 10 3125 1 3125
156.25 16 2500 0.5 5000
250.00 10 2500 0.5 5000
312.50 8 2500 0.5 5000
It is possible to configure the KeyStone I device to boot load application code over the SRIO interface.
Boot over SRIO is a feature that is selected using boot strapping options. For details on boot strapping
options, see the device-specific KeyStone I data manual.
If the SRIO peripheral is not used, the SRIO link pins can be left floating and the SerDes links should be
left in the disabled state.
The SRIO SerDes ports support hot-swap, in which the AC-coupled inputs of the device can be driven
without a supply voltage applied.
If the SRIO peripheral is enabled but only one link is used, the pins of the unused link can be left floating.
6.5.2 System Implementation of SRIO

The Serial RapidIO implementation is compliant with the RapidIO Interconnect Part VI: Physical Layer
1×/4× LP-Serial Specification, Version 2.0.1.
Guidelines for KeyStone I Devices .
Suggestions on SRIO reference clocking solutions can be found in Section 3.7.
SRIO/SGMII SerDes power planes and power filtering requirements are covered in Section 2.
6.5.3 Unused SRIO Pin Requirement

All unused SRIO lanes must be left floating and the unused lanes properly disabled through the MMR. If
the entire SRIO interface is not required, the SRIO peripheral should be disabled.
The SRIO shares a common input clock with the SGMII interface. If either of these interfaces are used,
the respective power is still required to be connected in accordance with this specification and the
appropriate decoupling/bulk capacitance included.
If the SRIO interface is not required, the SRIO regulator power pin (VDDR4_SRIO) must still be connected
to the correct supply rail with the appropriate decoupling capacitance applied. The EMI/Noise filter is not
required when this interface is not used.
If both SRIO or SGMII are unused, the primary SRIOSGMIICLK clock input must be configured as
indicated in Section 3.6. Pull-up must be to the CVDD core supply.
NOTE: Both interfaces must be unused in order to pull-up the unused clock signal.
6.6 Peripheral Component Interconnect Express (PCIe)

Relevant documentation for PCIe:
• KeyStone Architecture Peripheral Component Interconnect Express (PCIe) User's Guide
• PCI Express Base Specification, revision 1.0
• PCI Express Base Specification, Version 2.0.1
• OCP Specification, revision 2.2
6.6.1 PCIe Features Supported

PCI Express is a point-to-point serial signaling protocol incorporating two links (consisting of one TX and
one RX differential pair, each). Supported is a single ×2 configuration or a single ×1 configuration (2
separate ×1 links are not supported). TI’s PCIe interface supports (depending upon the protocol version)
either 2.5 Gbps or 5.0 Gbps data transfer in each direction (less 8b/10b encoding overhead).
The device PCIe interface supports a dual-operating mode (either Root complex (RC) or end point (EP)).
6.6.2 Configuration of PCIe

The PCIe mode is controlled by pinstrapping the two MSB GPIO bits (GPIO15:14). Table 23 shows the
mode configuration. Specific details and proper configuration of the PCIe (if used) are identified in detail in
the KeyStone Architecture Peripheral Component Interconnect Express (PCIe) User's Guide.
Table 23. PCIe (GPIO) Configuration Table

GPIO15 GPIO14
PCIESSMODE1 PCIESSMODE0 Selection
0 0 PCIe in End-Point Mode
0 1 PCIe in Legacy End-Point Mode
1 0 PCIe in Root-Complex Mode
1 1 NOT VALID – Do Not Use
If the PCIe interface is used, a PCIECLKP/N clock must be provided and the SerDes PLL must be
configured to generate the desired line rate. The line rate is determined using the PCIECLK frequency, the
PLL multiplier, and the rate scale setting. The PLL multiplier value is found in the PLL configuration
register and the rate scale setting is found in the transmit and receive configuration registers. The PLL
output frequency is equal to the frequency of the PCIECLKP/N multiplied by the PLL multiplier:
• PLL Output (MHz) = PLL Multiplier × PCIECLK (MHz)
The VCO frequency range of the SGMII PLL is 1500 MHz to 3125 MHz, so the PLL output frequency must
fall within this range.
The supported line rates for PCIe are 2500 Mbps and 5000 Mbps. The rate scale is used to translate the
output of the PCIE PLL to the line rate using the following formula:
• Line rate Mbps = PLL Output/rate scale
The PCIE SerDes interface supports the four rate scales and the associated rate scale value shown in
Table 24.
Table 24. PCIe Rate Scale Values

Half 1 PLL Output / 1
32nd 16 PLL Output / 16
The possible PLL multiplier settings for the three recommended PCIECLKP/N clocks frequencies are
shown in Table 25.
Table 25. PCIE PLL Multiplier Settings

156.25 16 2500 1 2500
250.00 10 2500 1 2500
312.50 8 2500 1 2500
156.25 16 2500 0.5 5000
250.00 10 2500 0.5 5000
312.50 8 2500 0.5 5000
For more detailed information, see the device-specific data manual, user's guide, and application reports.
6.6.3 Unused PCIe Pin Requirement

All unused PCIe lanes must be left floating and the unused lanes properly configured in the MMR. If the
entire PCIe interface is not required, the PCIe peripheral should be disabled in the MMR.
If the PCIe interface is not required, the PCIe regulator power pin (VDDR2_PCIe) must still be connected
to the correct supply rail with the appropriate decoupling capacitance applied. The EMI/Noise filter is not
required when this interface is not used.
If unused, the PCIe clock must be configured as indicated in Section 3.6. Pull-up must be to the CVDD
core supply.
6.7 Antenna Interface (AIF)

Relevant documentation for AIF:
• KeyStone I Architecture Antenna Interface 2 (AIF2) User's Guide
• OBSAI Reference Point 4 Specification, Version 1.1 [1]
• CPRI Specification, Version 4.1 [5]
• Electrical Specification (IEEE-802.3ae-2002), dated 2002 [9]
6.7.1 Configuration of AIF

The AIF defaults to disabled and with the internal memories for the module in a sleep state. The
memories, followed by the module, must be enabled by software before use.
There are two protocol modes supported on the AIF interface: OBSAI and CPRI. The mode is selected by
software after power-up. All links are the same mode.
The AIF module requires the AIF reference clock (SYSCLKP/N) to drive the SerDes PLLs and requires
frame sync timing signals provided by the frame sync module. The frame sync clock provided to the FSM
has the following requirements:
• RP1 mode:
– RP1CLKP/N must be 30.72 MHz (8× UMTS chip rate)
– RP1FBP/N must provide a UMTS frame boundary signal
• Non-RP1 mode:
– RADSYNC must be between 1 ms and 10 ms
– PHYSYNC must provide UMTS frame boundary pulse
For proper operation of the AIF, the SYSCLKP/N (which is the antenna interface SerDes reference clock)
and the frame sync clock (either RP1CLKP/N or ALTFSYNCCLK) must be generated from the same clock
source and must be assured not to drift relative to each other.
The AIF reference clock and the SerDes PLL multiplier are used to select the link rates. Both CPRI and
OBSAI have three supported line rates that run at 2×, 4×, and 8× the base line rate. The SerDes line rates
can be operated at half rate, quarter rate, or eighth rate of the PLL output. For that reason, it is suggested
that the AIF SerDes PLL be run at the 8× line rate. Each link pair can be configured as half rate (8×),
quarter rate (4×), or eighth rate (2×).
Table 26 shows the possible clocking variations for the AIF SerDes.
Table 26. AIF SerDes Clocking Options

Protocol Mode Reference Clock (MHz) PLL Multiplier 8× (Gbps) 4× (Gbps) 2× (Gbps)
122.88 20 4.9152 2.4576 1.2288
CPRI 153.60 16 4.9152 2.4576 1.2288
307.20 8 4.9152 2.4576 1.2288
122.88 25 6.250 3.072 1.536
OBSAI 153.60 20 6.250 3.072 1.536
307.20 10 6.250 3.072 1.536
To ensure the required data throughput is present, a minimum core frequency of 1 GHz should be used if
the antenna interface is configured for OBSAI 8× and 4× links. If an OBSAI 2× link is used, or if CPRI is
used, the minimum core frequency is 800 MHz.
The AIF SerDes ports support hot-swap, where the AC-coupled inputs of the device can be driven without
a supply voltage applied.
6.7.2 System Implementation of AIF

In OBSAI RP3 mode, the interface is electrically compatible with the OBSAI RP3 Specification, Version
4.1.
In CPRI mode, the interface is electrically compatible with the XAUI Electrical Specification (IEEE-
802.3ae-2002).
Guidelines for KeyStone I Devices.
Suggestions on AIF reference clocking solutions can be found in Section 3.2. AIF SerDes power planes
and power filtering requirements are covered in Section 2.3.
Table 27 highlights the various uses of the sync timer inputs for the AIF module.
Table 27. AIF2 Timer Module Configuration Options

RAD Timer Sync PHYTimer Sync RP1Timer Sync RP1 Timer Clock Intended Use
RP1FBP/N RP1CLKP/N RP1 or differential sync & clock
RADSYNC PHYSYNC Non-RP1 single-ended sync
6.7.3 Unused AIF Pin Requirement

All unused AIF transmit and receive lanes must be left floating and the unused lanes properly disabled in
the MMR. If the entire AIF interface is not required, the AIF peripheral should be disabled in the MMR and
all link pins can be left floating. The reference clock should be terminated as shown in Section 3.6.
If the AIF interface is not required, the AIF regulator power pins (VDDR5_AIF1 and VDDR6_AIF2) must
still be connected to the correct supply rail with the appropriate decoupling capacitance applied. The
EMI/Noise filters are not required when this interface is not used.
If the LVDS inputs (RP1CLKP/N and RP1FBP/N) are not used, external connections should be provided
to generate a valid logic level. The recommended connections for unused LVDS inputs are shown in
Figure 25. The 1-kΩ resistor is used on the N pin to reduce power consumption, and the P pin should be
connected to the DVDD18 supply rail.
Power
Rail
DSP
100 W
N
1 kW
Figure 25. Unused LVDS Inputs
6.7.4 System Implementation of RP1

For an RP1-compliant interface, the frame sync clock and frame sync burst signal defined by the RP1
specification should be connected to RP1CLKP/N and RP1FBP/N, respectively.
LVDS 1:N buffers (such as the SN65LVDS108 or CDCL1810) can be used to connect these signals to
multiple devices on a single board, however it is strongly recommended that the new high performance
CDCM6208 low-jitter, low-power clock sources be used. The 100-Ω LVDS termination resistor is included
in the KeyStone I device LVDS receiver, so an external 100-Ω resistor is not needed. These are standard
LVDS inputs, so, unlike the LJCB inputs, these inputs should not be AC coupled. They should be driven
directly by an LVDS-compliant driver.
The EXTFRAMEEVENT output is available to generate a frame sync output to other devices based on the
frame clock and frame sync inputs. Its switching frequency and offset are programmable in the FSM. This
output edge is not aligned with the frame clock input so take care if this output needs to be latched based
on this clock.
For information on clocking distribution options for the single-ended frame sync clocks, see Section 10.4.
6.7.5 Unused RP1 Pin Requirement

If the RP1 interface is not used it is recommended that the RP1 clk input pins be connected in accordance
with Figure 25. The EXTFRAMEEVENT output pin can be left floating.
6.8 DDR3
Relevant documentation for DDR3:
• KeyStone Architecture DDR3 Memory Controller User's Guide
• DDR3 Design Requirements for KeyStone Devices
• JEDEC JESD79-3C [19]
6.8.1 Configuration of DDR3

The DDR3 peripheral is enabled at power-up.
The DDR3 output clock is derived from the DDR3 PLL that uses DDRCLKP/N as a reference clock. The
DDR3 output clock is 1/2 of the PLL output clock. For example, a 66.667-MHz reference clock multiplied
times 20 results in a 1,333-MHz PLL output and a DDR3 output clock of 667 MHz for DDR3-1333 support.
6.8.2 System Implementation of DDR3

For information regarding supported topologies and layout guidelines, see the DDR3 Design Guide for
KeyStone Devices
Suggestions for DDR3 reference clocking solutions can be found in Section 3.2.
6.8.3 Slew Rate Control

The KeyStone I device incorporates two pins necessary to control the DDR3 slew rate. There are four
possible slew rate combinations. The slew rate pins (DDRSLRATE1:0) must be pulled low or high (to 1.8
V (DVDD18) at all times (they are not latched). Pulling both DDRSLRATE input pins low selects the
fastest slew rate. If the DDRSLRATE pins are both pulled high, the resulting slew rate is the slowest. For
normal full speed operation, the DDRSLRATE pins should be pulled low. Table 28 shows the possible
combinations when controlling the slew rate for the DDR3 interface.
Table 28. Slew Rate Control

Setting Speed DDRSLRATE1 DDRSLRATE0
00 Fastest Pull-down Pull-down
01 Fast Pull-down Pull-up
10 Slow Pull-up Pull-down
11 Slowest Pull-up Pull-up
NOTE: For specific details, see the DDR3 implementation guide. Slew rates denoted in Table 28 are
relative and are highly dependant upon topologies, component selection, and overall design
implementation.
6.8.4 Unused DDR3 Pin Requirement

If the DDR3 peripheral is disabled, all interface signals can be left floating and the input buffers are
powered down.
All unused DDR3 address pins must be left floating. All unused error correction pins must also be left
floating. All remaining control lines must also be left floating when unused. Unused byte lanes must be
properly disabled in the appropriate MMR.
If the DDR3 interface is not required, the DDR3 PLL supply (AVDDA2) must still be connected and the
ferrite bead installed to minimize noise.
Even if the DDR3 interface is unused, a reference clock is still required to ensure that the DDR peripheral
is initialized to a proper state.
6.9 UART
Relevant documentation for the UART:
• KeyStone Architecture Universal Asynchronous Receiver/Transmitter (UART) User's Guide
• TMS320C66x DSP CPU and Instruction Set Reference Guide
• TMS320C66x CorePac User's Guide
The UART peripheral performs serial-to-parallel conversion to data received from a peripheral device
connecting to the DSP and parallel-to-serial data conversion to data connecting between the DSP and
peripheral device.
Figure 26 shows the concept of peripheral-to-DSP connection with autoflow control support.
DSP DSP
UARTRXD UARTTXD
UARTCTS UARTRTS
UARTTXD UARTRXD
UARTRTS UARTCTS
Figure 26. UART Connections With Autoflow Support
6.9.1 Configuration of the UART

The UART peripheral is based on the industry standard TL16C550 asynchronous communications
element, which is a functional upgrade of the TL16C450. Key configuration details are described in the
KeyStone Architecture Universal Asynchronous Receiver/Transmitter (UART) User's Guide.
6.9.2 System Implementation of the UART

The UART interface is intended to operate at 1.8 V. Connections between the device UART interface and
an external peripheral must be at a 1.8-V level to assure functionality and avoid damage to either the
external peripheral or the KeyStone device.
6.9.3 Unused UART Pin Requirements

The UART interface to the device contains four pins (CTS, RTS, TXD, and RXD). These pins (defined with
respect to the device) are LVCMOS, which when not used should be connected as shown in Table 29.
Table 29. Unused UART Connections

Signal Direction If Unused Internal Resistor
UARTRTS Output Leave Floating Internal Pull-down
UARTCTS Input Leave Floating Internal Pull-down*
UARTTXD Output Leave Floating Internal Pull-down
UARTRXD Input Leave Floating Internal Pull-down*
NOTE: The above recommendations are based on the peripheral being unused and no traces
attached. If traces are attached to any of the device input pins, added pull-up (to DVDD18) or
pull-down (to VSS) resistors (4.7 kΩ) are needed.
For additional connectivity requirements, see the device-specific data manual and the KeyStone
Architecture Universal Asynchronous Receiver/Transmitter (UART) User's Guide.
6.10 Serial Port Interface (SPI)

Relevant documentation for the SPI Interface:
• KeyStone Architecture Serial Peripheral Interface (SPI) User's Guide
• TMS320C66x CPU and Instruction Set Reference Guide
• TMS320C66x CorePac User's Guide
The SPI peripheral is a high-speed synchronous serial input/output port that allows a serial bit stream of
programmed length (1 to 32 bits) to be shifted into and out of the device at a programmed bit-transfer rate.
The device SPI interface can support multiple SPI slave devices (varies by device type and select/enable
pins). The device SPI interface can operate only as a master device.
The SPI is normally used for communication between the device and common external peripherals.
Typical applications include an interface to external I/O or peripheral expansion via devices such as shift
registers, display drivers, SPI EEPROMs, and analog-to-digital converters.
The DSP must be connected only to SPI-compliant slave devices. Figure 27 shows the concept of a DSP
to two-peripheral connection.
DSP1 Peripheral 0 Peripheral 1
SPICLK SCLK SCLK
SPISOMI (in) MISO (out) MISO (out)
SPISIM (out) MOSI (in) MOSI (in)
SPISCS0 SS (CS)
SPISCS1 SS (CS)
Figure 27. SPI Connections
6.10.1 Configuration of the SPI

The SPI peripheral is required to be configured correctly prior to use for a proper bit stream transfer.
6.10.2 System Implementation of the SPI

The SPI interface is intended to operate at 1.8 V. Connection between the device SPI interface and
peripheral must be at a 1.8-V level to assure functionality and avoid damage to either the external
peripheral or KeyStone device.
Two key controls are necessary to make the SPI to function:
• Chip select control (SPISCS1:0)
• Clock polarity and phase
For specific and detailed configuration, see the KeyStone Architecture Serial Peripheral Interface (SPI)
User's Guide.
6.10.3 Unused SPI Pin Requirements

The SPI interface to the device contains five pins: SPISCS0, SPICS1, SPICLK, SPIDIN, and SPIDOUT.
These device-specific pins are 1.8-V LVCMOS, which, when not used, should be connected as shown in
Table 30.
Table 30. Unused SPI Connections

Signal Direction If Unused Internal Resistor
SPICS0 Output Leave floating Internal pull-up
SPICS1 Output Leave floating Internal pull-up
SPICLK Output Leave floating Internal pull-down
SPIDOUT Output Leave floating Internal pull-down
SPIDIN Input Leave floating Internal pull-down (1)
(1) These recommendations are based on the peripheral being unused and no traces attached. If traces are attached to any of the
peripheral input pins, added pull-up (to DVDD18) or pull-down (to VSS) resistors (4.7 kΩ) are needed.
For additional connectivity requirements, see the device-specific data manual and the KeyStone
Architecture Serial Peripheral Interface (SPI) User's Guide.
6.11 External Memory Interface 16 (EMIF16)

Documentation for EMIF16:
• KeyStone Architecture External Memory Interface (EMIF16) User's Guide
• KeyStone Architecture DSP Bootloader User's Guide
• BSDL Model for your device
• IBIS Model File for your device
• Using IBIS Modes for Timing Analysis
6.11.1 Configuration of EMIF16

The EMIF16 peripheral can be disabled using boot strapping options, as defined in the device-specific
KeyStone I Data Manual and the KeyStone Architecture DSP Bootloader User's Guide. If the EMIF16 is
enabled via boot strapping, it still must be enabled via software after a reset.
The clock to be used for EMIF16 can either be supplied externally to the AECLKIN pin or can be
generated internally from the device core clock. The internally-generated clock is called SYSCLK6. After a
reset, SYSCLK6 is device core clock / 6. The SYSCLK6 divider can be changed via software register
accesses to the PLLDIV6 register. For details, see the KeyStone Architecture Phase-Locked Loop (PLL)
User's Guide.
One of the boot modes provides for booting using the EMIF16. In this mode, after reset, the device
immediately begins executing from the base address of CS2 in 16-bit asynchronous mode. This would
support booting from simple asynchronous memories such as NOR FLASH.
6.11.2 System Implementation of EMIF16

Generally, series resistors should be used on the EMIF16 signals to reduce overshoot and undershoot.
They are strongly recommended for any output signal used as a synchronous latch. Acceptable values are
10 Ω, 22 Ω, or 33 Ω. To determine the optimum value, simulations using the IBIS models should be
performed to check for signal integrity and AC timings. Significant signal degradation can occur when
multiple devices are connected on the EMIF16. Simulations are the best mechanism for determining the
best physical topologies and the highest frequency obtainable with that topology.
6.11.3 Unused EMIF16 Pin Requirements

If the EMIF16 peripheral is not used, the EMIF16 inputs can be left unconnected. Internal pull-up and pull-
down resistors are included on this interface so leakage will be minimal. If only a portion of the peripheral
is used, data pins that are not used can be left floating.
NOTE: The above recommendations are based on the peripheral being unused and no traces
attached. If traces are attached to any of the peripheral input pins, added pull-up (to
DVDD18) or pull-down (to VSS) resistors (4.7 kΩ) are needed.
6.12 Telecom Serial Interface Port (TSIP)

Documentation for TSIP:
• KeyStone Architecture Telecom Serial Interface Port (TSIP) User's Guide
• BSDL Model for your device
• IBIS Model File for your device
• Using IBIS Models for Timing Analysis
6.12.1 TSIP Configuration

TSIP0 and TSIP1 are independent serial interface ports. Each one has two clock inputs, two frame sync
inputs, eight serial data inputs, and eight serial data outputs. All interface timing is derived from the clock
and frame sync inputs. Each TSIP module can be individually configured to operate at either 8.192 Mbps,
16.384 Mbps, or 32.768 Mbps. All eight lanes are used at 8.192 Mbps, only the first four lanes are used at
16.384 Mbps, and only the first two lanes are used at 32.768 Mbps.
TSIP outputs are configurable to simplify multiplexing of TDM streams. Unused output timeslots can be
configured to drive high, drive low, or be high impedance. Low-speed links can use simple wire-OR
multiplexing when unused timeslots are high impedance. Higher speed links can be combined with OR
logic when unused timeslots are driven low or AND logic can combine output streams when unused
timeslots are driven high.
6.12.2 TSIP System Implementation

Series resistors should be used on clock sources because the TSIP clock inputs are edge-sensitive.
These resistors need to be placed near the signal source. This prevents glitches on the clock signal
transitions that could potentially disrupt TSIP timing. Most implementations do not need series
terminations on the other TSIP control and data lines as long as the traces are kept short and routed
cleanly.
The maximum TSIP performance is achievable only when using point-to-point connections. Termination
resistors are rarely needed in this topology. In a bused architecture in which several devices are
connected to the same TSIP interface, the traces should be routed from device to device in a single, multi-
drop route without branches. Then a single termination can be implemented at the end of the bus to
reduce over-shoot and under-shoot, which minimizes EMI and crosstalk. This need is even more likely
whenever a TSIP bus passes through board-to-board connectors. To determine the optimum value,
simulations using the IBIS models should be performed to validate signal integrity and AC timings.
Multiple devices can be connected to a common TSIP bus using TDM mode. The additional loads require
a reduction in the operating frequency. Also, the specific routing topology becomes much more significant
as additional devices are included. The way to determine the best topology and maximum operating
frequency is by performing IBIS simulations.
6.12.3 Unused TSIP Pin Requirements

If any of the TSIP inputs are not used, they can be left floating because all TSIP pins have internal pull-
down resistors.
I/O Buffers and Termination www.ti.com
7 I/O Buffers and Termination

This section discusses buffer impedances, I/O timings, terminators, and signaling standards.
7.1 Process, Temperature, Voltage (PTV) Compensated Buffers

The impedance of I/O buffers is affected by process, temperature, and voltage. For the DDR3 interface,
these impedance changes can impact performance and make it difficult to meet the JEDEC specifications.
For this reason, the KeyStone I device uses PTV-compensated I/O buffers for the DDR3 interface. The
PTV compensation works by adjusting internal impedances to nominal values based on an external
referenced resistance. This is implemented by connecting a resistor between the PTV pin and VSS. The
PTV resistor must be a 1% 45.3-Ω SMT resistor and connected between the pin and VSS (ground). The
resistor should be placed adjacent to the KeyStone device, as close to the pin as possible. For details,
see the device-specific data manual and the DDR3 Design Requirements for KeyStone Devices.
7.2 I/O Timings

The I/O timings in the device-specific KeyStone I data manual are provided for, and based on, the tester
test load. These timings need to be adjusted based on the actual board topologies. It is highly
recommended that timing for all high-speed interfaces (including the high-performance SerDes interfaces)
on the KeyStone I design be checked using IBIS simulations (at a minimum). Simulating the high
performance interfaces (SerDes) will require IBIS 5.0 AMI-compliant models. IBIS models and parasitics
incorporated in the IBIS models are based on IBIS loads and not the tester load.
7.3 External Terminators

Series termination is not always needed but is useful for some interfaces to avoid over-shoot/under-shoot
problems. Check the recommendations in the peripherals sections and/or perform IBIS or Matlab®
simulations on each interface.
7.4 Signaling Standards

This section covers signaling standards for various interfaces.
7.4.1 1.8-V LVCMOS

All LVCMOS I/O (input and output or bidirectional) buffers are JEDEC-compliant 1.8-V LVCMOS I/Os as
defined in JESD 8-5. Several types of LVCMOS buffers are used in KeyStone I devices, depending mainly
on whether an internal pull-up or pull-down resistor is implemented. There are also some differences in
the drive strength and impedance for some I/Os. For details on different LVCMOS buffer types, see the
device-specific KeyStone data manual.
These internal pull-up and pull-down resistors can be can be considered a 100-µA current source (with an
actual range of 45 µA to 170 µA). This equates to a nominal pull-up/pull-down resistor of 18 kΩ (with a
possible range of 10 kΩ to 42 kΩ) The 1.8-V LVCMOS interfaces are NOT 2.5-V or 3.3-V tolerant, so
connections to 2.5-V or 3.3-V CMOS logic require voltage translation.
For input buffers operating at less than 10 MHz, TI's LVC logic family can be operated at 1.8 V and is 3.3-
V tolerant. For faster signaling, TI's AUC family is optimized to operate at 1.8 V and is also 3.3-V tolerant.
A good option for voltage translation for 1.8 V outputs that need to drive 2.5 V or higher inputs would be
the CBT3245A. Some useful TI application reports on voltage translation options are:
• Flexible Voltage-Level Translation With CBT Family Devices
• Selecting the Right Level-Translation Solution
• Voltage Level Translation Product Portfolio
www.ti.com I/O Buffers and Termination
7.4.2 1.5-V SSTL

The KeyStone I DDR3 interface is compatible with the JEDEC 79-3C DDR3 Specification [11]. The I/O
buffers are optimized for use with direct connections to up to nine DDR3 SDRAMs and can be configured
to drive a DDR3 UDIMM module. The DDR3 SSTL interface does not require series terminations on data
lines (end terminations may be required). The 1.5-V SSTL DDR3 interface also supports active ODT (on-
die-terminations).
7.4.3 SerDes Interfaces

There are many SerDes peripherals on the KeyStone I devices. The SerDes interfaces available include
(others may apply): HyperLink, SRIO, PCI, SGMII, and AIF. These serial interfaces all use 8b/10b-
encoded links and SerDes macros.
NOTE: Not all SerDes peripherals are available on all KeyStone I devices. For details, see the
These interfaces use a clock recovery mechanism so that a separate clock is not needed. Each link is a
serial stream with an embedded clock so there are no AC timings or drive strengths as found in the
LVCMOS or SSTL interfaces.
There are several programmable settings for each SerDes interface that affect the electrical signaling. The
most important of these are: transmitter output amplitude, transmitter de-emphasis, and receiver adaptive
equalization. Recommendations for these settings for particular board topologies are provided in SerDes
Implementation Guidelines for KeyStone I Devices. The SerDes interfaces use CML logic. Compatibility to
LVDS signals is possible and is described in Section 7.
7.4.4 LJCB Differential Clock Inputs

The reference clock inputs are connected to a low-jitter clock buffer (LJCB) in KeyStone I devices. This
buffer is designed to interface with LVDS, LVPECL, and HCSL clock input levels using AC coupling
capacitors. The LJCB has internal termination and an internal common mode voltage generator so no
external termination is needed between the AC coupling capacitors and the KeyStone I device. The driver
selected for your design may require termination on the output. For the driver to determine what
termination is necessary, see the device-specific data manual.
Driver AC Coupling KeyStone Device

Termination Capacitors
Driver LJCB
Figure 28. LJCB Clock Inputs
7.5 General Termination Details

The use of terminations in various nets is designed to address one of three major issues:
• Signal overshoot
• Signal undershoot
• Fast slew rates
I/O Buffers and Termination www.ti.com
Depending on the nature of the signal, if an overshoot or undershoot were to occur, the resulting energy
induced may be catastrophic to the connected device (or DSP). Many standards — including SDRAM
standards — have overshoot and undershoot requirements that must be met to ensure device reliability.
The DSP and other devices also incorporate both source and sync current limitations as well as VIL/VIH
and VOL/VOH requirements that must be maintained. Many of these established limitations are controlled by
using terminations.
Clocking terminations many times require special consideration (AC or DC termination styles). In all cases,
proper positioning of the termination and selection of components is critical.
Terminations are also valuable tools when trying to manage reflections. Adding in a series termination on
a particular net can relocate the point of reflection outside of the switching region (although this is better
managed by physical placement).
It is always recommended that single-ended clock, data, and control lines be modeled to establish the
optimum location on the net.
To determine if a termination should be used (data, control, or single-ended clock traces) the following test
should be used:
• If the uni-directional (one way) propagation delay for the trace(s) in question is ≤ 20% of the rise/fall
time (whichever is fastest) or ≥ 80% of the rise/fall time (whichever is slowest) then a termination
should not be needed.
Further determination of whether or not a termination is required can be ascertained using the second
condition:
• A transmission line requires termination (to its intended impedance) if the uni-directional Tpd
(propagation delay) for the respective net is ≥ 50% of the respective rise or fall time fall time (fastest of
the two).
The most common of terminations are for single-ended nets and follow the basic design shown in
Figure 29.
Series Termination
KeyStone Device
Place close to Driver
Driver Receiver
Figure 29. Basic Series Termination
www.ti.com Power Saving Modes
8 Power Saving Modes

The KeyStone I device incorporates advanced power saving methods that can be used when power
consumption is a concern. These methods range from disabling particular peripherals to placing the
device into a hibernation state. This section describes the low-power methods and hibernation states, and
their limitations and configuration methods.
Relevant documentation for Power Savings:
• Device-specific data manual
• KeyStone Architecture Power Sleep Controller (PSC) User's Guide
The KeyStone I family of devices has been designed with four basic modes of operation:
• Off - Power off, no power supplied to the device at all
• Active - Power on, out of reset, and fully functional
• Standby - Power on, not in a fully active state, lower power consumption than active, greater power
consumption than in hibernation modes
• Hibernation - Two specific operational modes intended to significantly reduce power consumption
(includes level 1 and 2)
For additional details on configuration and application of different power saving and operational modes,
see the following sections, the device-specific data manual and application reports.
8.1 Standby Mode

The KeyStone I family of devices includes a standby mode. This mode of operation is intended to provide
maximum readiness from the device. This mode differs from additional power saving modes identified in
Section 8.2 in that standby mode does not incorporate all the advanced power savings features
associated with other modes.
There are several different combinations of standby mode. Different combinations would take into account
such interfaces as SRIO, Ethernet, PCIe, and HyperLink.
During standby mode, the IOs remain in their current state. The IOs are not disabled or placed into reset
state.
In standby mode, the status of each peripheral (if available) is indicated in Table 31. For detailed
descriptions on power saving modes, see the device-specific data manual.
Table 31. Standby Peripheral Status

Peripheral Status Notes
Debug Interface ON
PCIe ON
DDR3 ON
SRIO ON
AIF ON
Accelerator(s) ON For applicable accelerators, see the device-specific data manual
HyperLink ACTIVE
FFTC OFF
Cores INACTIVE For core status during standby mode, see the device-specific application reports and data
manual
SGMII ACTIVE
EMIF ACTIVE
Power Saving Modes www.ti.com
8.2 Hibernation Modes

The KeyStone I device incorporates several hibernation modes that allow power consumption to be
reduced during operation when only partial functionality is required (periods of inactivity). The benefits of
using a hibernation mode include faster wake up times (as compared to cold boot) in addition to the
typical power savings obtained.
Access to the hibernation modes is obtained through writable registers. For the memory map, see the
Each KeyStone I device incorporates different hibernation modes. For additional details, see the device-
specific data manual and application reports.
8.2.1 Hibernation Mode 1

During hibernation mode 1, only the MSMC SRAM memory content is retained. The MSMC MMR is not
retained. Prior to entering this specific hibernation mode, the bootcfg MMR PWRSTATECTL must be
properly configured and set. For the procedure for entering and exiting this hibernation mode, see the
Exiting hibernation mode 1 requires a chip-level reset, which is initiated using an external reset pin,
usually the RESET. The estimated wake-up time from this hibernation mode is less than 100 ms.
8.2.2 Hibernation Mode 2

During hibernation mode 2, the MSMC SRAM memory contents are lost and only information stored in the
DDR3 SDRAM is retained. The MSMC MMR is not retained. After a chip-level reset is asserted, the
MSMC is reset to its default settings.
Exiting hibernation mode 2 requires a chip-level reset, which is initiated using an external reset pin,
usually the RESET. The estimated wake up time from this hibernation mode is greater than 1.0 second.
Table 32 shows, at a high level, one possible combination of hibernation mode configurations. For
additional information, see the device-specific data manual and application report.
Table 32. Hibernation Mode Peripheral Status

Peripheral Mode 1 Status Mode 2 Status Notes
Debug OFF OFF
Interface
PCIe ON ON
DDR3 ON ON
MSMC SRAM ON OFF
SRIO ON ON
AIF ON ON Can be multiple configurations
Accelerator(s) ON OFF For applicable accelerators, see the device-specific data manual
HyperLink OFF OFF
FFTC OFF OFF
Cores INACTIVE INACTIVE For core status during stand by mode, see the device-specific application
report and data manual
SGMII ACTIVE ACTIVE Can be multiple configurations
EMIF ACTIVE ACTIVE
www.ti.com Power Saving Modes
8.3 General Power Saving and Design Techniques

There exist multiple additional low power techniques that can be implemented in a design (depending on
the use case). If the application requires multiple devices and can accommodate partitioning such that
certain device can handle the bulk of the workload across defined periods of time, then the use of
hibernation modes (identified above) is recommended.
In other use cases, entire sections of a board may be capable of powering down – making the application
more green. However, care must be taken not to drive an IO that does not have power.
Selection of components including pull-up and pull-down resistor values can reduce (or increase) current
requirements on the power supplies. Power supply efficiency can be optimized through proper design and
implementation.
The following list provides a general overview of typical power saving techniques. Not all techniques apply
for all applications. For additional information, see the device-specific KeyStone I data manual.
• Generally, unused I/O pins should be configured as outputs and driven low to reduce ground bounce.
• Keep ground leads short, preferably tie individual ground pins directly to the respective ground plane.
• Eliminate pull-up resistors (where possible), or use pull-down resistors (where possible).
• The use of multi-layer PCBs that accommodate multiple separate power and ground planes take
advantage of the intrinsic capacitance of GND-VCC plane topology and thus (when designed correctly)
help to establish a better impedance board while using the intrinsic capacitance inherent in the stack-
up design. A better-matched impedance board will prevent over and under shoots and / or reflections.
• Where possible, create synchronous designs that are not affected by momentarily switching pins.
• Keep traces connecting power pins stretching from power pins to a power plane (or island, or a
decoupling capacitor) should be as wide and as short as possible. This reduces series inductance, and
therefore, reduces transient voltage drops from the power plane to the power pin. Thus, reducing the
possibility of ground bounce and coupled noise.
• Use high quality surface-mount low series resistance (ESR) capacitors to minimize the lead
inductance. The capacitors should have an ESR value as small as possible. Ceramic capacitor should
be used where possible.
• Use separate power planes for analog (PLL) power supplies (where possible).
• All PLL power supply planes should have a ground plane directly next to them in the stackup to
minimize all power-generated noise.
• Place analog or digital components only over their respective ground planes.
• Avoid sharing vias. Connect each ground pin to the ground plane individually.
• The use of the recommended filters will aid in isolating the PLL power from the digital power rails.
• Place as many bypass or decoupling capacitors as possible (minimum is the recommended amount),
these should be placed between CVDD and ground and be as close as possible. Ideally, the
respective CVDD pin should be tied directly to the decoupling capacitor, which, in turn, should be tied
directly to ground. This provides the shortest and lowest inductance path possible. If there must be a
single trace connecting the decoupling capacitor to ground it should be between the CVDD pin and the
decoupling capacitor. If this is the case, then the trace between the device and decoupling capacitor
must be as wide as possible.
• Sequence devices appropriately — especially during power up and power down. This will minimize
crowbar and peak currents, which, in some cases, could be a 50% increase over the initial design.
• Use of larger diameter vias to connect the capacitor pads between power and ground planes aid in
minimizing the overall inductance.
Mechanical www.ti.com
9 Mechanical
9.1 Ball Grid Array (BGA) Layout Guidelines

The BGA footprint and pin escapes can be laid out as defined in Flip Chip Ball Grid Array Package
Reference Guide [15]. If the DDR3 interface is used, there are specific recommendations for the BGA pad
and pin escape vias given in DDR3 Design Requirements for KeyStone Devices .
Given the 0.80 mm-pitch of KeyStone I devices, it is recommended that non-solder-mask-defined (NSMD)
PCB pads be used for mounting the device to the board. With the NSMD method, the opening in the
solder mask is slightly larger than the copper pad, providing a small clearance between the edge of the
solder mask and the pad. (Figure 30).
Solder Mask Bare PCB Substrate
Pad
Figure 30. Non-Solder-Mask Defined (NSMD) PCB Land
While the size control is dependent on copper etching and is not as accurate as the solder mask defined
method, the overall pattern registration is dependent on the copper artwork, which is quite accurate. The
trade-off is between accurate dot placement and accurate dot size. NSMD lands are recommended for
small-pitch BGA packages because more space is left between the copper lands for signal traces.
Dimensioning for the pad and mask are provided in the Flip Chip Ball Grid Array Package Reference
Guide .
9.2 Thermal Considerations

Understanding the thermal requirements for the KeyStone I device along with proper implementation of a
functional thermal management scheme is necessary to assure continued device performance and
reliability. Each application, platform, or system must be designed to ensure that the maximum case
temperature of the device never exceeds the allowable limit. For thermal requirements, limitations, and
operating ranges, see the device-specific data manual. Proper heat dispersion can be obtained using a
variety of methods, including heat sinks.
NOTE: Improper thermal design that results in thermal conditions outside the allowable range will
decrease device reliability and may cause premature failure.
9.2.1 Heat Transfer

There exist several methods for meeting the thermal requirements of the device. The two most common
methods are conduction and convection.
www.ti.com Mechanical
9.2.1.1 Conduction
Conduction heat transfer refers to the conduction or direct contact between two surfaces resulting in a
thermal transfer from one surface (higher temperature) to another (lower temperature). When referring to
the cooling of the device in an application or system, this usually involves the direct contact of a heat sink
(attached or possibly part of the system enclosure). The heat generated by the device is transferred to an
attached heat sink/outer enclosure, thereby, reducing the heat (conducting it away) in the device. In all
conduction methods, the material used, compression forces, and ambient temperatures affect the transfer
rates.
9.2.1.2 Convection
Convection refers to the thermal transfer of heat from one object (DSP) to the ambient environment
typically by means of air movement (fan). Convection or air movement can be pulled or pushed across the
device depending on the application. If a fan is attached to the device the direction is usually to force the
air away from the device where as if the system or application board includes external cooling fans, the air
direction can be either direction (push or pull). Air movement in any enclosure involves a plenum whether
intended or not. It is always best to optimize the air movement to maximize the effects and minimize any
excessive back pressure on the fan (to maximize fan life and minimize noise).
9.3 KeyStone I Power Consumption

The KeyStone I processor can consume varying amounts of power depending heavily on usage,
implementation, topology, component selection, and process variation. For early power estimations, see
the device-specific TI power application brief. For more detailed configurations and more accurate power
figures, see the device-specific power application report and spreadsheet.
NOTE: All of the numbers provided (in product briefs and application reports) assume properly
designed power supplies and proper implementation of the recommended variable core
(SmartReflex) supply.
9.4 System Thermal Analysis

For an overview on performing a thermal analysis and suggestions on system level thermal solutions, see
the Thermal Design Guide for KeyStone Devices.
9.5 Mechanical Compression

Mechanical compression, especially where conduction cooling is concerned (Section 9.2.1.1), is important.
Excessive compression may improve thermal transfer but also increases the risk of damage to the device
or inducing added electrical shorts between BGA balls on the application hardware. Excessive mechanical
compression is typically noted when using BGA sockets or interfacing a heat sink/enclosure in direct
contact with the device. For mechanical compression and BGA assembly, see the device-specific data
manual and all relevant application reports.
Routing Guidelines www.ti.com
Table 33 and Table 34 are provided as assistance for applications in which conductive cooling is used and
a mechanical compression heat sink is incorporated. The following data assumes an absolute maximum
solder ball collapse of 155 µm, 0.8 mm solder ball pitch, 841 pins, and uniform pressure across the entire
device lid, a 23 mm × 23 mm lid, and proper device assembly to the PCB.
Table 33. Maximum Device Compression - Leaded Solder Balls

Maximum Lid Pressure for Leaded Ball - 0.8 mm pitch (841 ball package)
TEMP 5 years 10 years 15 years 20 years
70° C 108.67 psi 90.59 psi 81.57 psi 75.69 psi
90° C 73.20 psi 61.00 psi 54.90 psi 50.83 psi
100° C 61.00 psi 75.69 psi 45.64 psi 42.47 psi
Table 34. Maximum Device Compression - Lead-Free Solder Balls

Maximum Lid Pressure for Lead Free Ball - 0.8 mm pitch (841 ball package)
TEMP 5 years 10 years 15 years 20 years
70° C 544.47 psi 447.32 psi 397.62 psi 365.99 psi
90° C 442.80 psi 363.73 psi 323.06 psi 298.21 psi
100° C 402.14 psi 329.84 psi 293.70 psi 271.10 psi
NOTE: The values provided in Table 33 and Table 34 are Not To Exceed estimates based on
modeling and calculations. Added safety margin should be included to account for variations
in PCB planarity, solder mask, and thermal conductive material between the DSP and
compression heat sink.
10 Routing Guidelines
This section provides routing guidelines for designs using a KeyStone I device. The layout of the PCB is
critical for the proper operation of the device. Some of the information needed to complete a proper layout
of KeyStone I interfaces will be found in other documents, specifically the DDR3 interface and the SerDes
interfaces. In addition, other components on the board, for example power supplies and Ethernet PHYs,
may have additional guidelines in their documentation. It is important to review all the layout guidelines for
all the parts in your design.
10.1 Routing Power Planes

Power supply planes should be used to distribute the voltage from a power supply to the associated power
supply pins on the KeyStone I device and to the bulk and decoupling capacitors. In addition, power planes
should be used to route filtered power supply signals such as the AVDDAx and VDDRx. Each of these
filtered voltages are typically connected to only one pin on the KeyStone I device, so small cut-out planes
may be used to connect the filter to the local decoupling and the power pin.
Each core and I/O supply voltage regulator should be located close to the device (or device array) to
minimize inductance and resistance in the power delivery path. Additional requirements include a separate
power plane for the core, I/O, and ground rails, all bypassed with high-quality low-ESL/ESR capacitors.
For power plane requirements and support of specific power pads on each power supply, see the device-
specific power supply data sheets.
Because each KeyStone I device has its own AVS supply, the power plane for that supply is usually
localized to the area of the device. And, because the KeyStone device may draw a significant amount of
current, the resistivity of the plane must be kept low to avoid a voltage differential between the power
supply and the connections on the device. This voltage differential is minimized by making the copper
planes thicker or larger in area. Be sure to consider both the power planes and the ground planes.
www.ti.com Routing Guidelines
The resistance of the planes is determined by the following formula:

• R = ρ * length/(width * thickness)
where, ρ is the resistivity of copper equal to 1.72E-8 Ω-meters.
PCB layer thickness is normally stated in ounces. One ounce of copper is about 0.012 inches or 30.5E-6
meters thick. The width must be de-rated to account for vias and other obstructions. A 50-mm strip of 1-oz
copper plane de-rated 50% for vias has a resistance of 0.57 mΩ per inch.
The same calculation should be made for all power planes to ensure a proper current path from the supply
to the device is present.
It is highly recommended that 1-oz copper or greater be used for planes used to distribute the
supply voltages and for ground planes in KeyStone I designs.
10.1.1 Routing Filtered Power Planes

As stated above, the filtered voltage must be routed with a sufficient current path to avoid voltage droop
and to minimize noise coupling. Because the pins for the filtered voltages are generally towards the center
of the part, a common mistake is to route this voltage on a 4-mil trace similar to those used for signal
traces. Figure 31 shows the bottom layer and one of the inner layers of a PCB layout for a KeyStone I
device.
Figure 31. Filtered Voltage Plane
The connections to the ferrite and the local decoupling is highlighted in the lower right and the pin for the
KeyStone I device is shown on the upper left. The outline delineates the cut-out on the inner layer creating
the small local plane associated with this filtered voltage. The same calculation shown above can be used
to calculate the resistivity of the plane.
10.2 Clock Signal Routing
10.2.1 Clock and High Performance Transmission Lines - Microstrip versus Stripline
Clock nets (and most other high-performance nets) are typically designed for either a microstrip or stripline
topology. Figure 32 shows the difference between a microstrip and a stripline topology.
Figure 32. Stripline and Microstrip Topologies
The most obvious difference between these two topologies is the location of the respective net (trace). In
a stripline topology (left side) the net or trace is embedded in the PCB and usually between ground or
power layers. In the microstrip topology (right side), the net or trace is on an outer layer and adjacent to a
power or ground plane.
There are several reasons why one (stripline or microstrip) topology would be selected over the other, key
factors determining which topology is used include:
• Interconnection - The number of pins on the interconnecting devices may limit the width of traces on
the outer layers.
• Performance - High speed signals should not be routed on multiple layers, they should generally be
routed on the top layers.
• EMI - Where emissions or the susceptibility to spurious radiation may impact signal integrity, stripline
topologies are recommended.
• Timing - Propagation delays differ between microstrip and stripline topologies, the use of either must
be comprehended for any and all timing-critical nets.
• Routing - The routing of high-speed clock signals is critical to a functional design. Proper return
current paths to minimize switching noise is essential. Routing clocks that are > 250 MHz should be
done using only a microstrip topology as long as they are < 2 inches (50.8 mm) in length.
A detailed discussion of transmission line routing for clocks and critical signals can be found in
Appendix B.
10.2.2 Clock Routing Rules

As a general rule, the following routing guidelines should be followed for all clock signals:
• All high-performance clock signals should be routed on the same layer where possible. If not possible,
then the impact of multiple layers on performance, propagation delays, and signal integrity must be
taken into account.
• Clock nets, if routed internally, must be captured between two planes. Do not have two clock routing
layers adjacent to one another.
• Clock nets involved in synchronous communications must have identical number of vias and be skew-
matched within 5% of the total clock period.
• Vias and terminations (if required) must be positioned in locations such that reflections do not impact
signal quality or induce an unwanted change of state.
• The target impedance for a single-ended clock net is 50 Ω and the PCB impedance must be ± 5% or
47.5 Ω to 52.5 Ω.
• The target impedance for a differential ended clock nets is 100 Ω and the differential PCB impedance
must be ± 5% or 95.0 Ω - 105 Ω.
• All microstrip clock lines must have a ground plane directly adjacent to the clock trace.
• All stripline clock lines must have a ground or power planes directly adjacent (top and bottom) to the
clock trace.
• Terminate any single-ended clock signals.
• Confirm whether or not the differential clock sources require terminations.
• A ground plane must be located directly below all clock sources (oscillators, crystals).
• No digital signals must be routed underneath clock sources.
• Clock nets (source to target) must be as short as possible – emphasis on reflections.
• Complementary differential clock nets must be routed with no more then two vias.
• Escape vias from the device or clock source must be of either via-in-pad or dog-bone type design.
Dog-bone designed escapes must be short enough to ensure any reflections do not impact signal
quality.
• The number of vias used in complementary differential clock pairs should be identical.
• Differential clock signals (complementary nets) must be skew-matched to within a maximum 5 pS of
one another.
10.3 General Routing Guidelines
10.3.1 General Trace Width Requirements

Trace widths are largely dependent on frequency and parasitic coupling requirements for adjacent nets
within the application design. As a general rule of thumb, and given the complexity of many markets using
these high-performance devices, a typical trace width is on the order of 4 mils or 0.1016 mm. Trace widths
of this size typically allow for better routing and improved routing densities.
Spacing to other traces is a function of the stack-up impedance as well as induced isolation to minimize
(eliminate) crosstalk. Most single-ended nets have a width of 4 mils (0.1016 mm) and are spaced a
minimum of 1.5× the trace width from all other traces (2× the trace width is recommended).
10.3.2 General Routing Rules

• Single-ended clock signals should have additional spacing to other nets to avoid crosstalk or coupling.
• Optimal performance should require trace widths to be no closer then 3× the dielectric height (between
trace and adjacent ground or power plane).
• The distance between complementary differential nets must be as close as possible (typically 1× the
trace width).
• Complementary differential nets must be routed in parallel for the entire length (except escapes) with
identical spacing between each complementary net.
• The trace spacing between different differential pairs must be a minimum of 2× the trace width or 2×
the spacing between the complementary parasitic differential pairs.
• Do not incorporate split power and ground planes, all planes should be solid.
• Nets routed as power must be of sufficient size to accommodate the intended power plus unintended
peak power requirements.
• Keep high-speed signals away from the edges of the PCB. If necessary, pin the edges of the PCB to
prevent EMI emissions.
• Organize all signals into net classes prior to routing.
• All clock trace routes must be as straight as possible, minimize or eliminate serpentine routing.
• Eliminate all right angle traces. Chambered traces (if needed) are recommended.
• Determine all constraints prior to routing respective net classes (skew, propagation delays, and so
forth)
• Optimize all single-ended nets, minimize lengths where possible (as long as they do not violate any net
class requirements or violate timing conditions)
• Place decoupling capacitors as close to the target device as possible.
• Place bulk capacitors in close proximity to their respective power supply.
• Use the largest possible vias for connecting decoupling and bulk capacitors to their respective power
and ground planes. If traces to connect the decoupling capacitor to planes are used, then they must be
short and wide.
• Route single-ended nets perpendicular or orthogonal to other single-ended nets to prevent coupling.
• All transmission line conductors should be adjacent to its reference plane.
• All clocks should be routed on a single layer.
• Nets within a particular net class should have a matched number of vias if used (vias are not
recommended if possible).
• Vias and terminations (if required) shall be positioned in locations such that reflections do not impact
signal quality or induce an unwanted change of state.
• All microstrip nets should have a ground plane directly adjacent to the respective trace.
• All stripline nets should have a ground or power plane directly adjacent (top and bottom) to the
respective net where possible. If it is not possible to enclose the respective net within a ground/ground
or ground/power stack-up, then at least one return path must be provided for proper operation.
• Vias and escape vias to and from the device shall be taken into account with regards to timing,
reflections, loading, and so forth.
• It is always recommended that all net classes be modeled for optimal performance.
• Each power or ground pin should be tied to its respective plane individually.
• AC coupling capacitor as well as DC resistor termination placement should take into account
associated parasitics, coupling capacitance, and multipoint reflections. For additional details, see
Section 10.6.
10.4 DDR3 SDRAM Routing

The routing of the DDR3 SDRAM interface is crucial to the successful function of that interface. A detailed
description of the routing requirements for DDR3 PCB layout can be found in the DDR3 Design
Requirements for KeyStone Devices Application Report. These requirements must be carefully followed to
ensure proper operation. Failure to comply with the requirements documented will result in an interface
that is nonfunctional even at lower clock rates.
10.5 SerDes Routing

The routing of each SerDes interface is crucial to the successful function of that interface. A detailed
description of the routing requirements for SerDes PCB layout can be found in the SerDes Implementation
Guidelines for KeyStone I Devices. Each SerDes interface has specific requirements that must be met.
This document describes the routing for the AIF, SRIO, SGMII, HyperLink, and PCIe interfaces. Failure to
comply with the requirements documented will result in an interface that is nonfunctional even at lower
clock rates.
10.6 PCB Material Selection

Proper PCB (printed circuit board) material [also referred to as PWB (printed wiring board)] has a large
impact on the impedance of the stack up and even more impact on the propagation delays of signals.
Table 38 shows the approximate dielectric constant (Er) differences for common PWB materials used
today. Selection of material is paramount to a successful design. Special consideration must be taken if
routing signals on different layers where timing is concerned (except where not allowed by this design
guide).
10.7 PCB Copper Weight

Proper copper thickness (referred to as planes) is important to the overall impedance, thermal
management, and current carrying capabilities of the design. During the examination and review of all
application hardware it is important to consider each of these variables. It is strongly recommended that
1-oz. copper (or greater) be used for planes that distribute the supply voltages and also for ground
planes in KeyStone I designs.
As a general rule of thumb, the inner layers or planes of most PCBs are typically 1-oz copper while the
outer layers are 0.5-oz copper. Outer layers receive solder masking, which in the finished product, is
usually similar to the 1-oz inner copper weight. Table 35 shows basic information regarding copper weight,
thickness, and current carrying capabilities (other factors can positively or negatively affect these
specifications).
Table 35. Copper Weight and Thickness

Cu Weight Cu Thickness (mils) Cu Thickness (mm)
2.0 oz 2.8 0.07112
1.0 oz 1.4 0.03556
½ oz 0.7 0.01778
Simulation and Modeling www.ti.com
Figure 33 shows the relationship between trace width, copper thickness (1 oz shown), and current
carrying capability.
Figure 33. Trace Width to Current Specifications
11 Simulation and Modeling

This section contains specific details pertaining to simulation and modeling.
Relevant documentation for simulation and modeling:
• KeyStone I data manual
• Using IBIS Models for Timing Analysis (SPRA839)
11.1 IBIS Modeling

Texas Instruments offers a fully-validated standard IBIS model (IBIS 5.0 compliant) inclusive of all SerDes
IO buffers. The available model allows for full simulation using industry-standard plug-in tools to such
programs as Hyperlynx®. Because of the high-performance nature of many of the KeyStone I IOs, the
traditional IBIS model will also include embedded IBIS AMI models.
11.1.1 IBIS AMI Modeling

Given the performance requirements of many of today’s input/output buffers, current IBIS models (IBIS 4.x
compliant) are unable to properly characterize and produce results for many IOs operating above several
hundred megahertz. For this reason, the IBIS standard (4.x and previous) was modified (new revision 5.0
released Aug. 2008) incorporating the new AMI model.
AMI models (Algorithmic Modeling Interface) are black boxes that incorporate a unique parameter
definition file that allows for specific data to be interchanged between EDA toolsets and standard models.
The AMI models are not available on the www.ti.com website and require an NDA. Contact your TI FAE
for access to the models.
www.ti.com References
12 References
1. OBSAI Reference Point 4 Specification, Version 1.1: available at http://www.obsai.com/
5. CPRI Specification, Version 4.1 : available at http://www.cpri.info/
6. JEDEC SSTL_18 Specification, (EAI/JESD8-15a), September 2003: available at http://www.jedec.org/
7. SGMII Specification (ENG-46158), Version 1.8, April 27, 2005
8. LVDS Electrical Specification (IEEE-1596.3-1996), 1996: available at https://www.ieee.org/index.html
9. XAUI Electrical Specification (IEEE-802.3ae-2002), 2002: available at https://www.ieee.org/index.html
10. RapidIO Interconnect Part VI: Physical Layer 1×/4× LP-Serial Specification, Version 2.0.1, March 2008
11. JEDEC DDR3 SDRAM Specification (EAI/JESD79-3C): available at http://www.jedec.org/
12. Boundary Scan Test Specification (IEEE-1149.1): available at https://www.ieee.org/index.html
13. AC Coupled Net Test Specification (IEEE-1149.6): available at https://www.ieee.org/index.html
14. I2C Bus Specification (9398 393 40011), Version 2.1, January, 2000
15. Texas Instruments: Flip chip ball grid array package reference guide
16. Texas Instruments: Emulation and trace headers technical reference manual
17. JEDEC 2.5 V ± 0.2 V (Normal Range), and 1.8 V to 2.7 V (Wide Range) Power Supply Voltage and
Interface Standard for Non-Terminated Digital Integrated Circuit Specification (EAI/JESD8-5), October
1995 http://www.jedec.org/download/search/JESD8-5.gif
18. JEDEC Power Supply Voltage and Interface Standard Specification (EAI/JESD8-7), February 1997
http://www.jedec.org/download/search/JESD8-7.gif
19. Texas Instruments:DDR3 Design Requirements for KeyStone Devices
20. Texas Instruments: TMS320C6678 Multicore Fixed and Floating-Point Digital Signal Processor Data
Sheet
21. Texas Instruments: KeyStone architecture Serial RapidIO (SRIO) user's guide
22. Texas Instruments: Keystone architecture DDR3 memory controller user's guide
23. Texas Instruments: KeyStone architecture Gigabit Ethernet (GbE) switch subsystem user's guide
24. Texas Instruments: KeyStone architecture Inter-Integrated Circuit (I2C) user's guide
25. Texas Instruments: KeyStone architecture General-Purpose Input/Output (GPIO) user's guide
26. Texas Instruments: KeyStone architecture Timer64P user's guide
27. Texas Instruments: Using IBIS modes for timing analysis
28. Texas Instruments: TMS320C66x DSP CPU and Instruction set reference guide
29. Texas Instruments: KeyStone architecture Phase-Locked Loop (PLL) user's guide
30. Texas Instruments: KeyStone I architecture Antenna Interface 2 (AIF2) user's guide
31. Texas Instruments: KeyStone architecture DSP bootloader user's guide
32. Texas Instruments: Filtering techniques: isolating analog and digital power supplies in TI's PLL-based
CDC devices
33. UCD9244 Quad PWM System Controller: (http://www.ti.com/product/ucd9244)
34. Texas Instruments: CDCx706/x906 termination and signal integrity guidelines
35. Texas Instruments: Voltage translation between 3.3-V, 2.5-V, 1.8-V, and 1.5-V logic standards with the
TI AVCA164245 and AVCB164245 dual-supply bus-translating transceivers
36. Texas Instruments: Selecting the right level-translation solution
37. Texas Instruments: PCA9306 dual bidirectional I2C bus and SMBus voltage-level translator data sheet
38. Texas Instruments: Thermal design guide for Keystone devices
39. Texas Instruments: AC-coupling between differential LVPECL, LVDS, HSTL, and CML
40. Texas Instruments: DC-coupling between differential LVPECL, LVDS, HSTL, and CM
References www.ti.com
41. Texas Instruments: Power consumption summary for KeyStone C66x devices
42. Texas Instruments: Clocking design guide for KeyStone devices
43. UCD9222 Dual Digital PWM System Controller: (http://www.ti.com/product/ucd9222)
44. Texas Instruments: UCD7242 - digital dual synchronous-buck power driver
45. Texas Instruments: CDCE62002 - four output clock generator/jitter cleaner with integrated dual VCOs
46. Texas Instruments: CDCE62005 - five/ten output clock generator/jitter cleaner with integrated dual
VCOs
47. Texas Instruments: Common trace transmission problems and solutions
48. Texas Instruments: Using xdsprobe with the XDS560 and XDS510
49. Texas Instruments: TCI6630K2L multicore DSP+ARM KeyStone II System-on-Chip (SoC) data
manual
50. Texas Instruments: 66AK2L06 multicore DSP+ARM KeyStone II System-on-Chip (SoC) data manual
51. Texas Instruments: Digital Front End (DFE) user's guide
52. Texas Instruments: KeyStone II architecture Serializer/Deserializer (SerDes) user's guide
53. JESD204B Interface
54. JEDEC Serial Interface for Data Converters
55. Texas Instruments: 66AK2L06 JESD attach to ADC12J4000/DAC38J84 design getting started guide
56. Texas Instruments: Understanding JESD204B Subclasses and deterministic latency
57. Radio Frequency Software Developer Kit (RFSDK)
58. SYS/BIOS and Linux Multicore Software Development Kits (MCSDK) for C66x, C647x, C645x
Processors
59. Texas Instruments: TCI6638K2K multicore DSP+ARM KeyStone II System-on-Chip (SoC) data
manual
60. Texas Instruments: KeyStone architecture Universal Asynchronous Receiver/Transmitter (UART)
user's guide
61. Texas Instruments: KeyStone architecture Serial Peripheral Interface (SPI) user's guide
62. Texas Instruments: SerDes Implementation Guidelines for KeyStone I Devices
Appendix A
This section contains additional and supporting information to be used during the design and integration of
TI’s KeyStone I devices.
A.1 Phase Noise Plots

TI recommends specific clock types to be used in order to meet all clocking internal requirements. The
following phase noise plots provided are for the CDCE62002 and CDCE62005 alternate clock sources.
These clock sources are currently available for purchase and meet the needs of the device.
Figure 34 shows the phase noise data (plot) at 40 MHz. It is representative of both the CDC362002 and
CDCE62005. This input frequency would be coupled to the DDR3 clock input of the device (provided the
SDRAM were rated for 1.6 Gbps and the PLL multiplier was configured correctly).
Figure 34. Phase Noise Plot - CECE62002/5 @ 40 MHz
Phase Noise Plots www.ti.com
Figure 35 shows the phase noise data (plot) at 66.667 MHz. It is representative of both the CDC362002
and CDCE62005. This input frequency would be coupled to the DDR3 clock input of the device (provided
the SDRAM were rated for 1.333 Gbps and the PLL multiplier was configured correctly).
Figure 35. Phase Noise Plot - CECE62002/5 @ 66.667 MHz
www.ti.com Phase Noise Plots
and CDCE62005. This input frequency would be coupled to the ALTCORECLK or PASSCLK clock input
of the device (provided this was the input frequency selected and the PLL multiplier was configured
correctly).
Phase Noise Plots www.ti.com
and CDCE62005. This input frequency is intended to be coupled to the SYSCLK clock input of the device
(provided this was the input frequency selected and the PLL multiplier was configured correctly). When
visually overlaying this plot onto the previous SYSCLK plot, you can see that the maximum allowable
mask input level is below the device.
www.ti.com Phase Noise Plots
and CDCE62005. This input frequency is intended to be coupled to the SRIO_SGMIICLK, PCIe_CLK, or
MCMCLK clock input of the device (provided this was the input frequency selected and the PLL multiplier
was configured correctly). When visually overlaying this plot onto the previous 156.25 MHz plot, you can
see that the maximum allowable mask input level is below the device.
Reflection Calculations www.ti.com
and CDCE62005. This input frequency is recommended as the input clock for the SRIO_SGMIICLK, or
MCMCLK clock into the device or as an alternate for the PCIe_CLK, ALTCORECLK, PASS_CLK, and
DDR3CLK (to increase phase noise margins). Proper device configuration for this input clock frequency is
mandatory. When visually overlaying this plot onto the previous 312.50 MHz plot, you can see that the
maximum allowable mask input level is below the device.
A.2 Reflection Calculations

This section provides designers and engineers with an example of how to calculate, at a first
approximation, the impact of a reflection. It does not take into account the magnitude or potential for
crosstalk to a parallel signal.
Assumptions:
• Dielectric constant (Er) is 4.1
• Signal is routed internally and externally
• AC coupling capacitor is placed 250 mils (6.35 mm) from the load
• Total net length is 3 inches (76.2mm)
• Frequency = 312.50 MHz or 3.2 ns period
• AC coupling capacitor size is 0402 and mounting pads meet IPC requirements
Calculations:
• =85*SQRT((0.475*G13)+0.67) ==> results in the segment Tpd
www.ti.com Reflection Calculations
As a first approximation, each segment was evaluated for Tpd (propagation delay) along a microstrip net.
Each signal was further evaluated for the fundamental and secondary reflection (Table 36). The individual
segments highlighted in yellow denote which segments are the source of the potential reflection.
For the sake of this example, the period of the reflection is 30 ps. However, this may not always be the
case, which is where a simulation is beneficial.
These reflections (highlighted) would occur in the rising or falling edge of the clock periodic wave form. All
others (see the Figure 40 timing analysis) would induce reflections, perturbations, or inflections during a
logic 0 or logic 1 time frame and depending on magnitude could cause a logic transition, data error, or
double clocking.
Reflections can occur at any point where an impedance mismatch occurs. The illustrations provided are
examples only, and are intended to illustrate the many possible permutations in a simple net that can
occur (not all are comprehended in this example). For these reasons, simulation and modeling are always
recommended.
Table 36. Reflections in ps

t0-t1 t1-t2 t2-t3 t3-t4 t4-t5
2.7500 0.0350 0.0180 0.0350 0.2500
378.17 4.8131 2.4753 4.8131 34.379
69 6 4 6 72
reflection 378.17 382.99 385.46 390.27 424.65
nodes 69 54 85 82
t0-t1 t1-t2 t2-t3 t3-ta2 ta2-ta3 ta3-ta4 ta4-ta5
2.7500 0.0350 0.0180 0.0180 0.0180 0.0350 0.2500
378.17 4.8131 2.4753 2.4753 2.4753 4.8131 34.379
69 6 4 4 4 6 72
reflection 378.17 382.99 385.46 387.94 390.41 395.22 429.60
nodes 69 54 07 61 92 89
t0-t1 t1-t2 t2-t3 t3-t4 t4-t5 t5-tb4 tb4-tb5
2.7500 0.0350 0.0180 0.0350 0.2500 0.2500 0.2500
378.17 4.8131 2.4753 4.8131 34.379 34.379 34.379
69 6 4 6 72 72 72
reflection 378.17 382.99 385.46 390.27 424.65 459.03 493.41
nodes 69 54 85 82 8 77
tb0-tb1 tb1-tc0 tc0-tc1 tc1-tc2 tc2-tc3 tc3-tc4 tc4-tc5
2.7500 2.7500 2.7500 0.0350 0.1800 0.0350 0.2500
378.17 378.17 378.17 4.8131 24.753 4.8131 34.379
69 69 69 6 4 6 72
reflection 378.17 756.35 1134.5 1139.3 1164.0 1168.9 1203.2
nodes 69 37 31 44 97 1 9
tb0-tb1 tb1-tc0 tc0-tc1 tc1-tc2 tc2-tc3 tc3-td2 td2-td3 td3-td4 td4-td5
2.7500 2.7500 2.7500 0.0350 0.1800 0.0180 0.0180 0.0350 0.2500
378.17 378.17 378.17 4.8131 24.753 2.4753 2.4753 4.8131 34.379
69 69 69 6 4 4 4 6 72
reflection 378.17 756.35 1134.5 1139.3 1164.0 1166.5 1169.0 1173.8 1208.2
nodes 69 37 31 44 97 73 48 61 41
tb0-tb1 tb1-tc0 tc0-tc1 tc1-tc2 tc2-tc3 tc3-tc4 tc4-tc5 tc5-te4 te4-te5
2.7500 2.7500 2.7500 0.0350 0.1800 0.0350 0.2500 0.2500 0.2500
378.17 378.17 378.17 4.8131 24.753 4.8131 34.379 34.379 34.379
69 69 69 6 4 6 72 72 72
reflection 378.17 756.35 1134.5 1139.3 1164.0 1168.9 1203.2 1237.6 1272.0
nodes 69 37 31 44 97 1 9 7 49
tb0-tb1 tb1-tc0 tc0-tc1 tc1-tc2 tc2-tc3 tc3-td2 td2-td3 td3-td4 td4-td5 td5-tf4 tf4-tf5
2.7500 2.7500 2.7500 0.0350 0.1800 0.0180 0.0180 0.0350 0.2500 0.2500 0.2500
378.17 378.17 378.17 4.8131 24.753 2.4753 2.4753 4.8131 34.379 34.379 34.379
69 69 69 6 4 4 4 6 72 72 72
reflection 378.17 756.35 1134.5 1139.3 1164.0 1166.5 1169.0 1173.8 1208.2 1242.6 1277
nodes 69 37 31 44 97 73 48 61 41 2
te0-te1 te1-te2 te2-te3 te3-tf4 tf4-tg3 tg3-tg4 tg4-tg5
Table 36. Reflections in ps (continued)

t0-t1 t1-t2 t2-t3 t3-t4 t4-t5
2.7500 0.0350 0.0180 0.0350 0.0350 0.0350 0.2500
378.17 4.8131 2.4753 4.8131 4.8131 4.8131 34.379
69 6 4 6 6 6 72
reflection 378.17 382.99 385.46 390.27 395.09 399.90 434.28
nodes 69 54 85 17 49 46
te0-te1 te1-te2 te2-te3 te3-tf2 tf2-tf1 tf1-tf0 tf0-tg1 tg1-tg2 tg2-tg3 tg3-th4 th4-th5
2.7500 0.0350 0.0180 0.0180 0.0350 2.7500 2.7500 0.0350 0.0180 0.0350 0.2500
378.17 4.8131 2.4753 2.4753 4.8131 378.17 378.17 4.8131 2.4753 4.8131 34.379
69 6 4 4 6 69 69 6 4 6 72
reflection 378.17 382.99 385.46 387.94 392.75 770.93 1149.1 1153.9 1156.3 1161.2 1195.5
nodes 69 54 07 39 07 08 21 96 09 89
te0-te1 te1-te2 te2-te3 te3-tf4 tf4-tf3 tf3-tg2 tg2-th1 th1-tj2 tj2-tj3 tj3-tj4 tj4-tj5
2.7500 0.0350 0.0180 0.0350 0.0350 0.0180 0.0350 0.0350 0.0180 0.0350 0.2500
378.17 4.8131 2.4753 4.8131 4.8131 2.4753 4.8131 4.8131 2.4753 4.8131 34.379
69 6 4 6 6 4 6 6 4 6 72
reflection 378.17 382.99 385.46 390.27 395.09 397.56 402.38 407.19 409.66 414.48 448.86
nodes 69 54 85 17 7 02 34 87 19 16
tm1- tm2- tm3- tm4-
tk0-tk1 tk1-tk2 tk2-tk3 tk3-tf4 tk4-tk5 tk5-tl4 tl4-tl3 tl3-tl2 tl2-tl1 tl1-tl0 tl0-tm1 tm2 tm3 tm4 tm5
2.7500 0.0350 0.0180 0.0350 0.2500 0.2500 0.0350 0.0180 0.0350 2.7500 2.7500 0.035 0.018 0.035 0.25
378.17 4.8131 2.4753 4.8131 34.379 34.379 4.8131 2.4753 4.8131 378.17 378.17 4.8131 2.4753 4.8131 34.379
69 6 4 6 72 72 6 4 6 69 69 6 4 6 72
reflection 378.17 382.99 385.46 390.27 424.65 459.03 463.85 466.32 471.13 849.31 1227.4 1232.3 1234.7 1239.5 1273.9
nodes 69 54 85 82 8 11 65 96 65 93 07 82 95 75
0ps 495ps 990ps 1485ps 1
Reflection20
Reflection21
Reflection22
Reflection23
312.50MHz2
Clock Duty Cycle 45/55_2
Reflection24
Reflection25
Reflection26
Reflection27
Reflection28
Reflection29
Reflection30
Reflection31
Reflection32
Reflection33
Reflection34
Reflection35
312.50MHz3
Clock Duty Cycle 45/55_3
Reflection36
Reflection37
Reflection38
Reflection39
Reflection40
Reflection41
Figure 40. Reflections Timing Analysis
Table 37. Possible First 50 Reflection Combinations

Initial Reflections Secondary Reflections
No Rising Edge Falling Edge Rising Edge Falling Edge
1 0.3782 0.4082 0.7564 0.7864
2 0.383 0.413 0.766 0.796
3 0.3855 0.4155 0.7709 0.8009
4 0.3879 0.4179 0.7759 0.8059
5 0.3903 0.4203 0.7806 0.8106
6 0.3904 0.4204 0.7808 0.8108
7 0.3928 0.4228 0.7855 0.8155
8 0.3951 0.4251 0.7902 0.8202
9 0.3952 0.4252 0.7905 0.8205
10 0.3976 0.4276 0.7951 0.8251
11 0.3999 0.4299 0.7998 0.8298
12 0.4024 0.4324 0.8048 0.8348
13 0.4072 0.4372 0.8144 0.8444
14 0.4097 0.4397 0.8193 0.8493
15 0.4145 0.4445 0.829 0.859
16 0.4247 0.4547 0.8493 0.8793
17 0.4296 0.4596 0.8592 0.8892
18 0.4343 0.4643 0.8686 0.8986
19 0.4489 0.4789 0.8977 0.9277
20 0.459 0.489 0.9181 0.9481
21 0.4639 0.4939 0.9277 0.9577
22 0.4663 0.4963 0.9327 0.9627
23 0.4711 0.5011 0.9423 0.9723
24 0.4934 0.5234 0.9868 1.0168
25 0.7564 0.7864 1.5127 1.5427
26 0.7709 0.8009 1.5419 1.5719
27 0.8493 0.8793 1.6986 1.7286
28 1.1345 1.1645 2.2691 2.2991
29 1.1393 1.1693 2.2787 2.3087
30 1.1491 1.1791 2.2982 2.3282
31 1.1539 1.1839 2.3078 2.3378
32 1.1564 1.1864 2.3128 2.3428
33 1.1612 1.1912 2.3224 2.3524
34 1.1641 1.1941 2.3282 2.3582
35 1.1666 1.1966 2.3331 2.3631
36 1.1689 1.1989 2.3378 2.3678
37 1.169 1.199 2.3381 2.3681
38 1.1739 1.2039 2.3477 2.3777
39 1.1956 1.2256 2.3912 2.4212
40 1.2033 1.2333 2.4066 2.4366
41 1.2082 1.2382 2.4165 2.4465
42 1.2275 1.2575 2.455 2.485
43 1.2323 1.2623 2.4646 2.4946
44 1.2348 1.2648 2.4696 2.4996
45 1.2377 1.2677 2.4753 2.5053
46 1.2396 1.2696 2.4792 2.5092
47 1.2426 1.2726 2.4852 2.5152
Table 37. Possible First 50 Reflection Combinations (continued)

Initial Reflections Secondary Reflections
No Rising Edge Falling Edge Rising Edge Falling Edge
48 1.272 1.302 2.5441 2.5741
49 1.274 1.304 2.5479 2.5779
50 1.277 1.307 2.554 2.584
Appendix B
This section provides detailed information on transmission lines and routing clocks.
B.1 Overview
High-frequency signaling requires the use of transmission lines where signals require minimal distortion,
crosstalk, and EMI radiation. They are used for clocks, differential SerDes signaling, and a variety of other
high-speed signaling. Microstrip and stripline are two popular types of electrical transmission lines that can
be fabricated using printed circuit board technology. Clock and most other high performance signaling
generally use one of these transmission line types.
An industry standards body, the IPC, has created standard IPC-2141A (2004) Design Guide for High-
Speed Controlled Impedance Circuit Boards dealing with their construction.
Before proceeding, a bit of background information is helpful. The use of the term ground plane implies a
large area, low-impedance reference plane. In practice, it may be either a ground plane or a power plane,
both of which are assumed to be at zero AC potential.
It should be understood that there are numerous equations, all represented as describing transmission line
design and behavior. It is a safe to question the accuracy of these equations and assume none of them
are perfect, being only good approximations, with accuracy depending upon specifics. The best known
and most widely quoted equations are those in the IPC 2141A standard, but even these come with
application caveats. The source of the formulas herein is the above mentioned standard.
It is helpful to know that there are a number of transmission line geometries:
• Microstrip
• Embedded microstrip
• Symmetric stripline
• Asymmetric stripline
• Wire microstrip
• Wire stripline
• Edge-coupled microstrip
• Edge-coupled stripline
• Broadside coupled stripline
www.ti.com Overview
These transmission line geometries are shown in Figure 41. This document briefly covers only microstrip
and symmetric stripline.
Figure 41. Transmission Line Geometries
There are multiple impedance calculators available to designers, many of which comprehend these
geometries. One such calculator is available at www.eeweb.com/toolbox/microstrip-impedance/.
Contrasting Microstrip and Stripline Transmission Lines www.ti.com
B.2 Contrasting Microstrip and Stripline Transmission Lines

A simplified side-by-side view of microstrip and stripline transmission lines is shown in Figure 42.
Figure 42. Transmission Line Geometries
A number of physical and electrical factors govern whether signaling should use microstrip or stripline
topology. The key factors determining which topology is used include:
• BGA escape paths - A routing escape path may be on an internal or external layer.
• Timing/performance - Microstrip propagation delays are smaller than those of stripline.
• EMI - Susceptibility to and emission of radiated noise is less with stripline
• Loss characteristics - Stripline is less lossy than microstrip
B.2.1 Microstrip Transmission Lines Construction

A cross section of microstrip transmission line is shown in Figure 43. This geometry is known as a surface
microstrip, or simply, microstrip. It is formed by a dielectric (D) (typically air) above a conductor (C), with
another dielectric substrate (B) separating the conductor from a ground plane (A).
Figure 43. Microstrip Transmission Line Construction
The characteristic impedance of a microstrip transmission line depends on a number of factors including,
the width (W) and thickness (T) of the trace (C), and the thickness (H) of dielectric.
For a given PCB laminate and copper weight, note that all parameters will be predetermined except for W,
the width of the signal trace. The trace width is then varied to create the desired characteristic impedance.
Printed circuit boards commonly use 50 Ω and 100 Ω characteristic impedances.
www.ti.com Contrasting Microstrip and Stripline Transmission Lines
Equation 10 can be used to calculate the characteristic impedance of a signal trace of width W and
thickness T, separated by a dielectric with a dielectric constant εr and thickness H (all measurements are
in common dimensions (mils) from a constant-potential plane).
87 é 5.98H ù
Zo (W ) = ln ê ú
er + 1.41 êë (0.8W + T )úû (10)
The characteristic capacitance of a microstrip transmission line can be calculated in terms of pF/inch as
shown in Equation 11.
0.67 (er + 1.41)
Co (pF / in ) =
ln ëé5.98H / (0.8W + T )ûù (11)
For example, a microstrip transmission line constructed with a 1-ounce (T=1.4) copper-clad 10-mil (H) FR-
4 (εr = 4.0) where the trace width (W) is 20-mils wide results in an impedance of about 50 Ω.
For the 75-Ω video standard, the W would be about 8.3 mils. W may easily be adjusted to create other
impedances. The above equations are from the IPC standards. These equations are most accurate
between 50 Ω and 100 Ω, but is less accurate for lower or higher impedances.
A useful guideline pertaining to microstrip PCB impedance is that for a dielectric constant of 4.0 (FR-4), a
W/H ratio of 2/1 results in an impedance of close to 50 Ω (as in the first example, with W = 20 mils). This
is within 5% of the value predicted by Equation 10 (46 Ω) and generally consistent with the accuracy
(>5%) of observed measurements.
B.2.2 Microstrip Propagation Delay

The propagation delay of a microstrip transmission line is somewhere between the speed of an
electromagnetic waves in the substrate, and the speed of electromagnetic waves in air. This is due to the
electromagnetic wave partial propagation in the dielectric substrate, and partial propagation in the air
above it — propagation delay being a hybrid of the two.
The one way propagation delay of the microstrip transmission line can be calculated using Equation 12 to
obtain ns/ft or Equation 13 to obtain ps/in. A reference table for delay in ps/in for various materials can be
found in Table 38.
t pd (ns / ft ) = 1.017 0.475 er + 0.67
(12)
t pd (ps / in ) = 85 0.475 er + 0.67
(13)
Table 38. Microstrip Signal Delay Through Various Materials

Trace Length in Inches Versus Delay Through Microstrip in ps
Er Name tpd(ps/in) 1” 2” 3” 4” 5” 6” 7” 8” 9” 10”
1 Air 90.9540 90.95 181.91 272.86 363.82 454.77 545.72 636.68 727.63 818.59 909.54
2.2 PTFE/Glass 111.3143 111.31 222.63 333.94 445.26 556.57 667.89 779.20 890.51 1001.83 1113.14
2.9 Ceramic 121.6273 121.63 243.25 364.88 486.51 608.14 729.76 851.39 973.02 1094.65 1216.27
3.5 GETEK 129.8165 129.82 259.63 389.45 519.27 649.08 778.90 908.72 1038.53 1168.35 1298.17
4 FR-4 136.2654 136.27 272.53 408.80 545.06 681.33 817.59 953.86 1090.12 1226.39 1362.65
4.1 FR-4 137.5189 137.52 275.04 412.56 550.08 687.59 825.11 962.63 1100.15 1237.67 1375.19
4.2 FR-4 138.761 138.76 277.52 416.28 555.04 693.81 832.57 971.33 1110.09 1248.85 1387.61
4.3 FR-4 139.9922 139.99 279.98 419.98 559.97 699.96 839.95 979.95 1119.94 1259.93 1399.92
4.4 FR-4 141.2126 141.21 282.43 423.64 564.85 706.06 847.28 988.49 1129.70 1270.91 1412.13
4.5 FR-4 142.4226 142.42 284.85 427.27 569.69 712.11 854.54 996.96 1139.38 1281.80 1424.23
Contrasting Microstrip and Stripline Transmission Lines www.ti.com
B.2.3 Symmetric Stripline Transmission Lines

A stripline transmission line is constructed with a signal trace sandwiched between two parallel ground
planes as shown in Figure 44. It is similar to coaxial cable as it is non-dispersive and has no cutoff
frequency. When compared to microstrip transmission lines, it provides better isolation between adjacent
traces and propagation of radiated RF emissions, but has slower propagation speeds.
Figure 44. Stripline Transmission Line Construction
B.2.4 Characteristic Impedance

The width and thickness (W and T) of the trace, the thickness of the dielectric separating the planes (H),
and their relative permittivity (εr) determine the transmission line's characteristic impedance. In the general
case, the trace need not be equally spaced between the ground planes and the dielectric material may be
different above and below the central conductor. In this case, the transmission line is an asymmetric
stripline transmission line.
Equation 14 defines ZO of a symmetric (H = H1 = H2) symmetric stripline transmission line. The accuracy
of the equation is typically on the order of 6%.
60 é 1.9 (S ) ù
Zo (W ) = ln ê ú
er êë (0.8W + T )úû (14)
A useful ballpark estimate for constructing a 50-Ω symmetric stripline transmission line where εr = 4.0
(FR4) is to make the S/W ratio 2 to 2.2. Recognize that this guideline is only an approximation as it
neglects the value of T. It is important to note that it may be difficult to create some transmission lines of a
high impedance while maintaining a trace width suitable for high yield PCB manufacturing.
The characteristic capacitance of a symmetric stripline transmission line in pf/in can be calculated using
Equation 15.
1.41(er )
Co (pf / in ) =
ln éë3.81H / (0.8W + T )ùû (15)
B.2.5 Stripline Propagation Delay

The propagation delay of a symmetric stripline transmission line in ns/ft and ps/inch can be calculated
using Equation 16 and Equation 17, respectively. For a PCB dielectric constant of 4.0 (FR-4), a symmetric
stripline's delay is 2 ns/ft, or 170 ps/inch. A reference table for delay in ps/in for various materials can be
found in Table 39.
t pd (ns / ft ) = 1.017 er
(16)
t pd (ns / in ) = 85 er
(17)
www.ti.com Changes in Transmission Line Direction (Bends)
Table 39. Stripline Signal Delay Through Various Materials

Trace Length in Inches Versus Stripline Delay in ps
Er Name tpd (ps/in) 1” 2” 3” 4” 5” 6” 7” 8” 9” 10”
1 Air 85.00 85.00 170.00 255.00 340.00 425.00 510.00 595.00 680.00 765.00 850.00
2.2 PTFE/Gla 126.08 126.08 252.15 378.23 504.30 630.38 756.45 822.53 1008.60 1134.68 1260.75
ss
2.9 Ceramic 144.75 144.75 289.50 434.25 579.00 723.75 686.50 1013.25 1158.00 1302.75 1447.50
3.5 GETEK 159.02 159.02 318.04 477.06 636.08 795.10 954.12 1113.14 1272.16 1431.18 1590.20
4 FR-4 170.000 170.00 340.00 510.00 680.00 850.00 1020.00 1190.00 1360.00 1530.00 1700.00
4.1 FR-4 172.1119 172.11 344.22 516.34 688.45 860.56 1032.67 1204.78 1376.90 1549.01 1721.12
4.2 FR-4 174.1982 174.20 348.40 522.59 696.79 870.99 1045.19 1219.39 1393.59 1567.78 1741.98
4.3 FR-4 176.2598 176.26 352.52 528.78 705.04 881.30 1057.56 1233.82 1410.08 1586.34 1762.60
4.4 FR-4 178.2975 178.30 356.60 534.89 713.19 891.49 1069.79 1248.08 1426.38 1604.68 1782.98
4.5 FR-4 180.3122 180.31 360.62 540.94 721.25 901.56 1081.87 1262.19 1442.50 1622.81 1803.12
B.3 Changes in Transmission Line Direction (Bends)

Microstrip transmission lines used with PCBs do not always follow a straight line from point A to Point B.
The transmission line often turns one or more times between these points. Abrupt turns in the
transmission line are not permitted as this will cause a significant portion of the wave energy to be
reflected back towards the wave's source, with only part of the wave energy transmitted past the bend.
This effect is generally best minimized by curving the path of the strip in an arc of a radius at least 3× the
strip-width [8].
Another more commonly used method is to use a mitred bend as this consumes a smaller routing area,
where the mitered portion cuts-away a fraction of the diagonal between the inner and outer corners of an
un-mitred bend as shown in Figure 45.
Figure 45. Mitred Bend
Douville and James have experimentally determined the optimum mitre for a wide range of microstrip
geometries. They have presented evidence for cases where 0.25 ≤ W/H ≤ 2.75 and 2.5 ≤ εr ≤ 25 . Their
findings indicate that the optimum percentage mitre can be represent by Equation 18 where M is the
percentage of mitre.
x æ 27 w ö
M = 100 % = ç 52 + 65e - %
d è 20 h ÷ø (18)
They report a voltage standing wave ratio (VSWR) of better than 1.1 (a return better than 26 dB) for any
percentage mitre of D within 4% produced by the formula.
Changes in Transmission Line Direction (Bends) www.ti.com
It is important to note that for a minimum W/H ratio of 0.25, the percentage mitre is 96%, so that the strip
is very nearly cut through. This means that care must be taken to understand the effect the percentage of
miter has on the manufacturability of the PCB. At some point, the manufacturing tolerances may limit the
use of mitring and the use of appropriate trace radii may be required. Note that the propagation delay is
affected by changes in trace direction as the electrical length is somewhat shorter than the physical path-
length of the strip for both the curved and mitred bends.
www.ti.com Revision History
Revision History
NOTE: Page numbers for previous revisions may differ from page numbers in the current version.
Changes from C Revision (August 2013) to D Revision ................................................................................................ Page
• Update was made in Section 1.1. ....................................................................................................... 3

• Update was made in Section 2.2. ....................................................................................................... 6
• Update was made in Section 2.2.1. ..................................................................................................... 7
• Updates were made in Section 2.2.1.1. ................................................................................................ 7
• Updates were made in Section 2.2.4.1. .............................................................................................. 12
• Update was made in Section 2.2.5.3. ................................................................................................. 13
• Update was made in Section 2.3.3. ................................................................................................... 15
• Updates were made in Section 2.4.1. ................................................................................................. 16
• Updates were made in Section 3.3. ................................................................................................... 26
• Updates were made in Section 3.4. ................................................................................................... 27
SPRABI2D – November 2010 – Revised March 2019 Revision History 99

IMPORTANT NOTICE AND DISCLAIMER
TI PROVIDES TECHNICAL AND RELIABILITY DATA (INCLUDING DATASHEETS), DESIGN RESOURCES (INCLUDING REFERENCE
DESIGNS), APPLICATION OR OTHER DESIGN ADVICE, WEB TOOLS, SAFETY INFORMATION, AND OTHER RESOURCES “AS IS”
AND WITH ALL FAULTS, AND DISCLAIMS ALL WARRANTIES, EXPRESS AND IMPLIED, INCLUDING WITHOUT LIMITATION ANY
IMPLIED WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT OF THIRD
PARTY INTELLECTUAL PROPERTY RIGHTS.
These resources are intended for skilled developers designing with TI products. You are solely responsible for (1) selecting the appropriate
TI products for your application, (2) designing, validating and testing your application, and (3) ensuring your application meets applicable
standards, and any other safety, security, or other requirements. These resources are subject to change without notice. TI grants you
permission to use these resources only for development of an application that uses the TI products described in the resource. Other
reproduction and display of these resources is prohibited. No license is granted to any other TI intellectual property right or to any third
party intellectual property right. TI disclaims responsibility for, and you will fully indemnify TI and its representatives against, any claims,
damages, costs, losses, and liabilities arising out of your use of these resources.
TI’s products are provided subject to TI’s Terms of Sale (www.ti.com/legal/termsofsale.html) or other applicable terms available either on
ti.com or provided in conjunction with such TI products. TI’s provision of these resources does not expand or otherwise alter TI’s applicable
warranties or warranty disclaimers for TI products.
Mailing Address: Texas Instruments, Post Office Box 655303, Dallas, Texas 75265
Copyright © 2019, Texas Instruments Incorporated

Sprabi 2 D

Uploaded by

Copyright:

Available Formats

Sprabi 2 D

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Sprabi 2 D

Uploaded by

Copyright:

Available Formats

Application Report

SPRABI2D – November 2010 – Revised March 2019

Hardware design guide for KeyStone™ I devices

20 Emulation Voltage Translation With Buffers ............................................................................ 41

1.1 Before Getting Started

1.2 Design Documentation

1.3 Acronyms Used in This Document

2.1 Determining Your Power Requirements

2.2 Power Rails

Primary Voltage Source

Figure 1. KeyStone I Power Supply Planes (Rails)

Table 1. SmartReflex VID Value Mapping

Table 2. SmartReflex VID Value Mapping for 1400 MHz Devices

2.2.1.2 SmartReflex VCNTL Interface

2.2.1.3 Supported Power Supply Solutions

2.2.1.4 LM10011 and Analog Controller Solutions

CVDD SW VIN Vin

VCNTL0 VIDA SET

Figure 2. LM10011 CVDD Supply Block Diagram

2.2.1.5 UCD92xx Digital Controller Solutions

2.2.1.5.1 UCD92xx Communications

VCNTL0 1B1 1A1 VIDA

Figure 3. VCNTL Level Translation

2.2.1.6 CVDD for Designs With Multiple KeyStone I Devices

2.2.2.1 VDDTn Filters

2.2.5 Other Supplies

2.2.5.1 VTT Termination Supply

2.2.5.2 VREF Supply

Figure 4. DDR3 VREFSSTL Voltage Divider

2.3 Power Supply Filters

2.3.1 Filters Versus Ferrite Beads

Table 3. Power Rail Filter Recommendations

(1) Filter not required if peripheral not used.

2.3.2 AVDDAn Filters

Primary Voltage Source

100 mF 10 mF 4.7 mF 100 nF 10 nF 1 nF 560 pF

Figure 5. AVDDAx Power Supply Filter

2.3.3 VDDRn Filters

Primary Voltage Source

100 mF 10 mF 4.7 mF 100 nF 10 nF 1 nF 560 pF

Fixed 1.5-V 100 nF 100 nF 10 nF

Figure 6. VDDRn Power Supply Filter

2.3.4 VDDTn Filters

Primary Voltage Source

100 mF 10 mF 4.7 mF 100 nF 10 nF 1 nF 560 pF

Fixed Core 100 nF 100 nF 100 nF

100 nF 100 nF 100 nF

Figure 7. VDDTn Power Supply Filter

2.4 Power-Supply Decoupling and Bulk Capacitors

2.4.1 Selecting Bulk Capacitance

Max Impedance to Current Load

0.0500 0.0500 0.0500

15mV 25mV 35mV

Figure 8. ESR Plot for Delta Current to Tolerance #1

Max Impedance to Current Load

15mV 25mV 35mV

Figure 9. ESR Plot for Delta Current to Tolerance #2

Max Impedance to Current Load

15mV 25mV 35mV