DDR3 Design Requirements

Application Report
SPRABI1B—May 2014
DDR3 Design Requirements for KeyStone Devices
High-Performance and Multicore Processors
Abstract
This document provides implementation instructions for the DDR3 interface
incorporated in the Texas Instruments (TI) KeyStone series of DSP devices. The DDR3
interface supports 1600 MT/s and lower memory speeds in a variety of topologies (see
the specific device Data Manual for supported speeds). This document assumes the
user has a familiarization with DRAM implementation concepts and constraints.
Contents
1 Migrating Designs from DDR2 to DDR3 (Features & Comparisons). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1 Topologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.1 Balanced Line Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.1.1 Balanced Line Topology Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.2 Fly-By Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.1.2.1 Balanced Line Topology Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 ECC (Error Correction) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 DDR3 Features & Improvements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3.1 Read Leveling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3.2 Write Leveling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3.3 Pre-fetch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3.4 ZQ Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3.5 Reset Pin Functionality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.6 Additional DDR2 to DDR3 Differences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2 Prerequisites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1 High Speed Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 JEDEC DDR3 Specification – Compatibility & Familiarity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Memory Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4 Memory Speeds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.5 Addressable Memory Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.6 DDR3 SDRAM/UDIMM Memories, Topologies, and Configurations . . . . . . . . . . . . . . . . . . . . . . . 10
2.6.1 Topologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.6.2 Configurations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.6.2.1 Memories – SDRAM Selection Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.7 DRAM Electrical Interface Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.7.1 Slew. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.7.2 Overshoot & Undershoot Specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.7.2.1 Overshoot & Undershoot Example Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.7.3 Typical DDR3 AC & DC Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.7.4 DDR3 Tolerances and Noise – Reference Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3 Package Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.1.1 ×4 SDRAM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.1.2 ×8 SDRAM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.1.3 ×16 SDRAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Please be aware that an important notice concerning availability, standard warranty, and use in critical applications
of Texas Instruments semiconductor products and disclaimers thereto appears at the end of this document.
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 1 of 48
www.ti.com
4
5
6
7
Page 2 of 48
3.1.4 ×32 SDRAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.1.5 ×64 SDRAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Physical Design and Implementation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.1 Electrical Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.1.1 Pin Connectivity & Unused Pins – SDRAM Examples. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.1.2 Pin Connectivity – ECC UDIMM & Non-ECC UDIMM Examples . . . . . . . . . . . . . . . . . . . . . 20
4.2 Signal Terminations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.2.1 External Terminations – When Using Read & Write Leveling . . . . . . . . . . . . . . . . . . . . . . . 21
4.2.2 External Terminations – When Read and Write Leveling is Not Used . . . . . . . . . . . . . . . 21
4.2.3 Internal Termination – On-Die Terminations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.2.4 Active Terminations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.2.5 Passive Terminations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.2.6 Termination Component Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.3 Mechanical Layout and Routing Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.3.1 Routing Considerations – SDRAMs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.3.1.1 Mechanical Layout – SDRAMs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.3.1.2 Stack Up – SDRAMs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.3.1.3 Routing Rules – General Overview (SDRAMs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.3.1.4 Routing Rules – Address and Command Lines (SDRAMs). . . . . . . . . . . . . . . . . . 27
4.3.1.5 Routing Rules – Control Lines (SDRAMs). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.3.1.6 Routing Rules – Data Lines (SDRAMs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.3.1.7 Routing Rules – Clock Lines (SDRAMs). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.3.1.8 Routing Rules – Power (SDRAMs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.3.1.9 Round-Trip Delay Impact on Routing – KeyStone I . . . . . . . . . . . . . . . . . . . . . . . . 30
4.3.1.10 Write Leveling Limit Impact on Routing – KeyStone I . . . . . . . . . . . . . . . . . . . . 32
4.3.2 Routing Considerations – UDIMMs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.3.2.1 Mechanical Layout – UDIMMs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.3.2.2 Stack Up – UDIMMs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.3.2.3 Routing Rules – General Overview (UDIMMs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . .36
4.3.2.4 Routing Rules – Address and Command Lines (UDIMMs). . . . . . . . . . . . . . . . . . 36
4.3.2.5 Routing Rules – Control Lines (UDIMMs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.3.2.6 Routing Rules – Data Lines (UDIMMs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.3.2.7 Routing Rules – Clock Lines (UDIMMs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.3.2.8 Routing Rules – Power (UDIMMs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.3.2.9 Write-Leveling Limit Impact on Routing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.4 Timing Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.5 Impedance Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.5.1 Routing Impedances – KeyStone I Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.5.1.1 Data Group Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.5.1.2 Fly-By Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.5.2 Routing Impedances – KeyStone II Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.5.2.1 Data Group Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.5.2.2 Fly-By Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.5.3 Comparison to JEDEC UDIMM Impedance Recommendations . . . . . . . . . . . . . . . . . . . . . 42
4.6 Switching and Output Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Simulation and Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.1 Simulation and Modeling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .43
5.2 Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.3 Models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.4 TI Commitment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Power. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.1 DDR3 SDRAM Power Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.1.1 Vref Voltage Requirements. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.1.2 VTT Voltage Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.2 DSP DDR3 Power Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.3 DDR3 Power Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
6.4 DSP DDR3 Interface Power Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
6.5 Sequencing – DDR3 and DSP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Disclaimers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
www.ti.com
8 Revision History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
List of Tables
Table 1
Table 2
Table 3
Table 4
Table 5
Table 6
Table 7
Table 8
Table 9
Table 10
Table 11
Table 12
Table 13
Table 14
Table 15
Table 16
Table 17
Table 18
×8 Width DDR3 SDRAM Possible Configurations Supported . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
×16 Width DDR3 SDRAM Possible Configurations Supported. . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
×32 Width DDR3 SDRAM Possible Configurations Supported. . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Discrete SDRAM Configurations with ECC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
DDR3 SDRAM Selection Criteria for KeyStone I and KeyStone II Devices. . . . . . . . . . . . . . . . . . 12
OS & US Requirements for Address and Control Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
OS and US Requirements for CK, CK#, DQ, DQS, DQS#, and DM Lines . . . . . . . . . . . . . . . . . . . . 14
DDR3 Single-Ended Output Levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
DDR3 Differential Output Levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
SDRAM Net Class Routing Rules. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Address and Command Line Numeric Routing Rules. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Control Line Numeric Routing Rules. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Data Lane Numeric Routing Rules. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Data and Data Strobe Byte Lane Grouping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Clock Lane Numeric Routing Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Maximum Round Trip Delay Example - Invert Clock Out Enabled and Disabled . . . . . . . . . 32
Maximum Write Leveling Skew Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Minimum Write Leveling Skew Example. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Figure 1
Figure 2
Figure 3
Figure 4
Figure 5
Figure 6
Figure 7
Figure 8
Figure 9
Figure 10
Figure 11
Figure 12
Figure 13
Figure 14
Figure 15
Figure 16
Figure 17
Typical DDR Balanced Line Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Typical DDR3 Fly-By Architecture. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Control and Address Overshoot & Undershoot Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Data, Clock, Strobe, & Mask Overshoot & Undershoot Requirements. . . . . . . . . . . . . . . . . . . . 14
Signal Overshoot Calculations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Signal Undershoot Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
DSP-to-SDRAM Connection Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
DSP to UDIMM Connection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Interface Topology for Single and Dual Rank. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
DDRCLKOUT and DQ/DQS/# Routing from the DSP to SDRAM(s) . . . . . . . . . . . . . . . . . . . . . . 29
Data Group Impedances During Write Cycles on KeyStone I. . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Data Group Impedances During Read Cycles on KeyStone I . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Fly-By Impedances on KeyStone I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Data Group Impedances During Write Cycles on KeyStone II . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Data Group Impedances During Read Cycles on KeyStone II . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Fly-By Impedances on KeyStone II. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
DDR3 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
List of Figures
Scope
The primary goal of this document is to establish a minimum set of requirements
necessary to help assure functional success in new application designs for Texas
Instruments high performance multiprocessor DSPs incorporating DDR3 memory
interfaces.
Background
Technological advances in memory architecture in both speed and densities for DDR3
require a different mindset when it comes to application implementation and design
compared to the customary and traditional SRAM, DDR, and DDR2 devices.
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 3 of 48
www.ti.com
Related Specifications and Documentation
The following documentation shall be used in conjunction with this design guide to
properly design in and implement a successful DDR3 interface to Texas Instruments
high performance multiprocessor DSPs.
JESD 79-3C
JEDEC DDR3 Standard
JESD 79-3E
Page 4 of 48
SPRUGV8
DDR3 Memory Controller for KeyStone Devices User Guide
By Part #
TMS320TCI66xx Data Manual (appropriate data manual to be used)
SPRABI2:
Hardware Design Guide for KeyStone Devices
TN-41-04
DDR3 Dynamic On-Die Termination; Micron, technical Note
TN-41-06
DDR3 Termination Data Strobe (TDQS); Micron
MO-269D
JEDEC Document: MO (Module Outline) {for DDR3}
SO-007B
JEDEC Document: SO (Socket Outline)
TN-42-02
DDR3 ZQ Calibration; Micron
TN-04-54
High-Speed DRAM Controller Design; Micron,
TN-41-01
Calculating Memory System Power for DDR3; Micron,
TN-41-07
DDR3 Power-Up, Initialization, and Reset; Micron,
TN-41-08
Design Guide for Two DDR3-1066 UDIMM Systems; Micron
JEDEC 21-C
Unbuffered DIMM Design Specification
Pub 95 PS-001A
Connector Performance Standards for Outlines of Solid State Related Products – 240 pin
DDR3 UDIMM
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
1 Migrating Designs from DDR2 to DDR3 (Features & Comparisons)
www.ti.com
1 Migrating Designs from DDR2 to DDR3 (Features & Comparisons)
This section is not intended to present a detailed listing of differences between DDR2
and DDR3 designs, but to provide key insight into specific differences that will have a
positive impact as customers migrate from a DDR2 to a DDR3 platform (based on the
assumption the DDR3 interface is implemented correctly).
1.1 Topologies
In a DDR2 to DDR3 comparison, the single greatest improvement from a topology
standpoint is the change from a Balanced T to a Fly-By architecture. Each architecture
is described briefly below.
1.1.1 Balanced Line Topology
In a traditional DDR2 design, a balanced T style topology is typically recommended (if
not required) for address and control lines (depending on the number of SDRAMs
used). This is generally recommended to balance any delays to each SDRAM device.
The general concept of a balanced line topology is not used in DDR3 implementations
in favor of fly-by topology, which better accommodates the higher-performance
SDRAMs. Figure 1 shows the general concept of the balanced line topology found in a
typical DDR2 design.
Figure 1
Typical DDR Balanced Line Topology
DDR2
DDR2
Address
Address
Control
Control
Data
Data
Data
Control
Address
DDR2 Interface
DSP
1.1.1.1 Balanced Line Topology Issues
The down side to the use of a balanced T line topology for DDR2 designs is that it may
introduce a varying amount of additional skew because of the inclusion of multiple
stubs and stub lengths for each individual net. The addition of multiple loads on
respective address and control nets limits bandwidth. Skews normally encountered
between the address/control and data nets also cause bandwidth limitations.
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 5 of 48
1 Migrating Designs from DDR2 to DDR3 (Features & Comparisons)
www.ti.com
1.1.2 Fly-By Topology
The DDR3 fly-by architecture provides a benefit to layout and routing of control and
address signals. In this topology, each respective signal from the DSP DDR3 controller
is routed sequentially from one SDRAM to the next, thus eliminating reflections
associated with any stub or superfluous traces previously seen in DDR2 designs.
Figure 2 shows a typical DDR3 SDRAM fly-by interface in a single rank topology.
Typical DDR3 Fly-By Architecture
R
DDR3
DDR3
DQS/DQS #
DQS/DQS#
Address
Address
Control
Control
DQS/DQS#/DQM
Address
R
Control
R
Data
Data
CLKIN
CLKIN
DDR3 Interface
DVDD15
DVDD15/2 (VTT)
Figure 2
DSP
Data
DDRCLKOUT
Note 1: DVDD15 is a 1.5-V supply rail common to both the SDRAM and DSP.
Note 2: DVDD15/2 or VTT refers to 750 mV.
Note 3: Differential-ended terminations require dedicated discrete component(s) per complementary net.
1.1.2.1 Balanced Line Topology Issues
The down side to the use of a fly-by topology for DDR3 designs is the induced delay
from the DSP DDR3 controller to the SDRAMs. In fact, the delay is different at each
SDRAM. Correction or compensation for the different controller-to-DRAM lengths is
handled through read and write leveling. Also keep in mind the following:
• Data nets are point-to-point unless designed as a dual rank implementation.
• Clock nets are point-to-multipoint and end-terminated (differently than control,
command, or address nets).
• Control nets are point-to-multipoint and end-terminated to VTT.
• Address/Command nets are point-to-multipoint and are end-terminated to
VTT.
1.2 ECC (Error Correction)
Error correction has not been supported in previous TI DSPs using JEDEC-compliant
SDRAM (DDR or DDR2). Error correction is now supported in TI’s new KeyStone
DSP processor family. Supporting ECC allows for the automatic detection and
correction of single-bit and double-bit errors. ECC software configuration and control
is described in detail in the DDR3 Memory Controller for KeyStone Devices User Guide.
Page 6 of 48
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
1 Migrating Designs from DDR2 to DDR3 (Features & Comparisons)
www.ti.com
1.3 DDR3 Features & Improvements
In addition to TI’s DDR3 controller now supporting ECC, the DSP DDR3 controller
and new DDR3 SDRAMs offer five notable additional features:
• Read leveling
• Write leveling
• Change in pre-fetch size
• ZQ calibration
• A reset pin
These features are described briefly below. Detailed configuration is described in the
DDR3 Memory Controller for KeyStone Devices User Guide.
1.3.1 Read Leveling
The memory controller also automatically corrects for delay skew between SDRAMs
during read leveling. Read leveling takes advantage of values loaded into the SDRAMs
multi-purposed register (MPR). The values loaded into this register are used by the
DSP DDR3 controller to calibrate each signal path relative to skew. Each respective
SDRAM byte is then internally corrected thus improving performance.
1.3.2 Write Leveling
The memory controller automatically corrects for delay skew between SDRAMs during
write leveling. During write leveling, correction for SDRAM skew (the tDQSS, tDSS,
and tDSH) is handled using a programmable DQS delay which aligns the timing
relationship to the clock and strobe signals. During the write-leveling procedure, the
DSP controller delays the DQS until a valid change of state is detected at the SDRAM
clock (CK) signal (see Section 4.3 for additional details).
1.3.3 Pre-fetch
The new DDR3 architecture now supports an eight-bit pre-fetch to improve
back-to-back accesses (DDR2 allowed only a four-bit pre-fetch).
1.3.4 ZQ Calibration
ZQ calibration is intended to control the on-die termination (ODT) values and output
drivers (RTT and RON respectively) of the SDRAM. ZQ calibration is not a controllable
feature from the DSP. It is controlled using a precision ( 1% tolerance) 240- resistor.
The DDR3 SDRAM ZQ calibration cycle is made up of an initial long calibration cycle
(ZQCL) requiring 512 clock cycles to complete (which is why it is typically performed
during the initial boot or reset conditions) and a shorter ZQ calibration period.
The subsequent short (ZQCS) calibration requires only 64 clock cycles and used when
the SDRAM is idle. The periodic short calibrations cycles accommodate minor
variations in temperature and voltage. The short calibration cycle (ZQCS) is designed
to correct for a minimum 0.5% impedance error within the allotted 64 clock cycles.
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 7 of 48
1 Migrating Designs from DDR2 to DDR3 (Features & Comparisons)
www.ti.com
See the selected SDRAM data sheet for the maximum ODT, temperature and voltage
sensitivity values. The ZQ calibration is intended to help minimize PCB impedance
discontinuities between traces and SDRAM drivers.
Note—Texas Instruments requires the use of a dedicated ZQ resistor (240-)
to be connected to each SDRAM ZQ pin (cannot share pins).
1.3.5 Reset Pin Functionality
The new DDR3 architecture also supports a reset pin. This reset pin is designed to allow
the user to clear all data (information) stored the DDR3 SDRAM. The advanced benefit
of this feature is that there is no need to reset each control register separately or restart
(power down and up again) each individual DDR3 SDRAM. By initiating a reset, the
SDRAM will recover in a known good state (if needed).
The reset function of the UDIMM or SDRAM are an active-low (RESET) LVCMOS
input and referenced to VSS. The SDRAM / UDIMM input pin functions rail-to-rail
with a DC HIGH  0.8 × VDD (1.5 V  0.8 = 1.2 V) and DC LOW  0.2 × VDD
(1.5 V × 0.2 V = 0.3 V).
CAUTION—The TI DSP DDR3 controller cannot be held in reset for more than
one hour during the initial power-up. Also, the TI DSP DDR3 controller
cannot be held in reset for more than 5% of its total power-on hours. Exceeding
these limits will cause a gradual reduction in the reliability of the part.
1.3.6 Additional DDR2 to DDR3 Differences
The change from supporting both a single-ended and differential DQS (DDR2) to only
a differential DQS (DDR3) improves noise immunity, and allows for longer signal
paths without compromising signal integrity.
Page 8 of 48
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
2 Prerequisites
www.ti.com
2 Prerequisites
2.1 High Speed Designs
The goal of this document is to make the DDR3 system implementation and
integration easier while reducing the added risk of designing in a high performance
interface. It is still expected that the PCB design work (design, layout, and fabrication)
is performed, supervised, or reviewed by a highly knowledgeable high-speed PCB
designer. This includes a thorough understanding of all high-speed design rules.
Specific areas to avoid include ground plane cuts, incorrect spacing, and signal skew
mis-matches as well as timing violations. The total system should be evaluated for such
areas including power, filtering, termination, crosstalk, and EMI.
2.2 JEDEC DDR3 Specification – Compatibility & Familiarity
The DDR3 interface on KeyStone I devices is designed to be compatible with the
JEDEC JESD79-3C DDR3 specification. The DDR3 interface on KeyStone II devices is
designed to be compatible with the JEDEC JESD79-3E DDR3 specification. It is
assumed that the reader is familiar with these specifications and the basic electrical
operation of the interface. In addition, several memory manufacturers provide detailed
application reports on DDR3 operation.
2.3 Memory Types
Devices from many manufacturers were available at the time this document was
prepared. It is recommended that only high-quality DRAM parts that fully comply with
the latest DDR3 JEDEC specification and are from well-known manufacturers be used.
Incorrect DRAM part or manufacturer selection can cause performance problems that
may not be resolvable within the available software or configuration limitations of the
DSP.
In addition to supporting the DDR3 SDRAMs, the TI KeyStone family of DSPs also
supports SDRAMs including ECC (error correction) or UDIMMs (unbuffered). The
KeyStone family of DSPs also supports DDR3L SDRAMs as they are backwards
compatible to 1.5 V. DDR3L SDRAMs operating at 1.35 V are not supported at this
time.
2.4 Memory Speeds
The KeyStone DSP memory interface currently supports various configurations as
specified in the JEDEC DDR3 standard. KeyStone devices support data rates of DDR3
1600 MT/s and lower. See the device-specific data manual for supported data rates.
2.5 Addressable Memory Space
KeyStone devices can address up to 8GB of contiguous memory. Note that some DDR3
memory interfaces are limited to a smaller memory space such as 2GB. See the specific
device-specific data manual for more information.
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 9 of 48
2 Prerequisites
www.ti.com
2.6 DDR3 SDRAM/UDIMM Memories, Topologies, and Configurations
2.6.1 Topologies
The current DDR3 controller design implementation supports several different DRAM
topologies. The following list describes the known available topologies to be used with
the TI KeyStone DSPs. See the device-specific data manual for a definitive confirmation
on available memory topologies for the DSP before proceeding.
• ×8 (16 is the minimum supported width, two 8 devices are required)
• ×16
• ×32 (with and without ECC)
• ×64 (with and without ECC)
• UDIMM (unbuffered DIMM) [×72 & ×36]
2.6.2 Configurations
The current DDR3 controller design implementation allows multiple DRAM
configurations to be used. The following list describes the known usable
DRAM configurations compatible with TI KeyStone family of DSPs. See the
device-specific data manual for a definitive confirmation on usable memory
configurations and bus widths before proceeding.
The following tables are intended to provide a general overview of the possible DDR3
DRAM topologies usable with the KeyStone DSP. In all cases, only JEDEC-compliant
(JESD79-3C) SDRAMs are supported. For all tables below, the notation * implies
possible support, not a plan of record.
Table 1
Device Width
×8 SDRAM
×8 Width DDR3 SDRAM Possible Configurations Supported
Total Memory / Memory Topology
Rank Width
Total Size
1Gb / (16M × 8 × 8) × 2 SDRAMs
×16
256MB
1Gb / (16M × 8 × 8) × 4 SDRAMs
×32
512MB
4Gb / (16M × 8 × 8) × 8 SDRAMs
×64
1024MB
1Gb / (16M × 8 × 8) × 8 SDRAMs × 2 ranks
×64
2048MB
2Gb / (32M × 8 × 8) × 2 SDRAMs
×16
512MB
2Gb / (32M × 8 × 8) × 4 SDRAMs
×32
1024MB
2Gb / (32M × 8 × 8) × 8 SDRAMs
×64
2048MB
2Gb / (32M × 8 × 8) × 8 SDRAMs × 2 ranks
×64
4096MB
4Gb / (64M × 8 × 8) × 2 SDRAMs
×16
1024MB
4Gb / (64M × 8 × 8) × 4 SDRAMs
×32
2048MB
4Gb / (64M × 8 × 8) × 8 SDRAMs
×64
4096MB
4Gb / (64M × 8 × 8) × 8 SDRAMs × 2 ranks
×64
8192MB
8Gb / (64M × 8 × 8) × 2 SDRAMs
×16
2048MB
8Gb / (64M × 8 × 8) × 4 SDRAMs
×32
4096MB
8Gb / (64M × 8 × 8) × 8 SDRAMs
×64
8192MB
End of Table 1
Page 10 of 48
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
2 Prerequisites
www.ti.com
Table 2
Device Width
×16 SDRAM
×16 Width DDR3 SDRAM Possible Configurations Supported
Total Memory / Memory Topology
Rank Width
Total Size
1Gb / (8M × 16 × 8) × 1 SDRAMs
×16
128MB
1Gb / (8M × 16 × 8) × 2 SDRAMs
×32
256MB
1Gb / (8M × 16 × 8) × 4 SDRAMs
×64
512MB
1Gb / (8M × 16 × 8) × 4 SDRAMs × 2 ranks
×64
1024MB
2Gb / (16M × 16 × 8) × 1 SDRAM
×16
256MB
2Gb / (16M × 16 × 8) × 2 SDRAMs
×32
512MB
2Gb / (16M × 16 × 8) × 4 SDRAMs
×64
1024MB
2Gb / (16M × 16 × 8) × 4 SDRAMs × 2 ranks
×64
2048MB
4Gb / (32M × 16 × 8) × 1 SDRAMs
×16
512MB
4Gb / (32M × 16 × 8) × 2 SDRAMs
×32
1024MB
4Gb / (32M × 16 × 8) × 4 SDRAMs
×64
2048MB
4Gb / (32M × 16 × 8) × 4 SDRAMs × 2 ranks
×64
4096MB
8Gb / (64M × 16 × 8) × 1 SDRAMs
×16
1024MB
8Gb / (64M × 16 × 8) × 2 SDRAMs
×32
2048MB
8Gb / (64M × 16 × 8) × 4 SDRAMs
×64
4096MB
8Gb / (64M × 16 × 8) × 4 SDRAMs × 2 ranks
×64
8192MB
End of Table 2
Table 3
Device Width
×32 SDRAM
×32 Width DDR3 SDRAM Possible Configurations Supported
Total Memory / Memory Topology
Rank Width
Total Size
2Gb / (8M × 32 ×8) × 1 SDRAM
×32
256MB*
4Gb / (8M × 32 × 8) × 2 SDRAMs
×64
512MB*
4Gb / (8M × 32 × 8) × 2 SDRAMs × 2 ranks
×32
256MB*
8Gb / (8M × 32 × 8) × 4 SDRAMs × 2 ranks
×64
512MB*
4Gb / (16M × 32 × 8) × 1 SDRAM
×32
512MB*
8Gb / (16M × 32 × 8) × 2 SDRAM
×64
1024MB*
8Gb / (16M × 32 × 8) × 2 SDRAMs × 2 ranks
×32
1024MB*
16Gb / (16M × 32 × 8) × 4 SDRAMs × 2 ranks
×64
2048MB*
End of Table 3
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 11 of 48
2 Prerequisites
www.ti.com
In addition to the discrete SDRAM devices configurations listed above, the following
ECC configurations are supported.
Table 4
Discrete SDRAM Configurations with ECC
Device Width
Memory rank Topology
Rank Width
Total Size
×8 SDRAM
1Gb (16M × 8 × 8) × 5 SDRAMs
×36
512MB
×16 SDRAM
1
1Gb (16M × 8 × 8) × 9 SDRAMs
×72
1024MB
1Gb (16M × 8 × 8) × 9 SDRAMs × 2 ranks
×72
2048MB
2Gb (32M × 8 × 8) × 5 SDRAMs
×36
1024MB
2Gb (32M × 8 × 8) × 9 SDRAMs
×72
2048MB
2Gb (32M × 8 × 8) × 9 SDRAMs × 2 ranks
×72
4096MB
4Gb (64M × 8 × 8) × 5 SDRAMs
×36
2048MB
4Gb (64M × 8 × 8) × 9 SDRAMs
×72
4096MB
4Gb (64M × 8 × 8) × 9 SDRAMs × 2 ranks
×72
8192MB
8Gb (64M × 8 × 8) × 5 SDRAMs
×36
4096MB
8Gb (64M × 8 × 8) × 9 SDRAMs
×72
8192MB
1Gb (8M × 16 × 8) × 3 SDRAMs
×36
256MB
1Gb (8M × 16 × 8) × 5 SDRAMs
×72
512MB
1Gb (8M × 16 × 8) × 5 SDRAMs × 2 ranks
×72
1024MB
2Gb (16M × 16 × 8) × 3 SDRAMs
×36
512MB
2Gb (16M × 16 × 8) × 5 SDRAMs
×72
1024MB
2Gb (16M × 16 × 8) × 5 SDRAMs × 2 ranks
×72
2048MB
4Gb (32M × 16 × 8) × 3 SDRAMs
×36
1024MB
4Gb (32M × 16 × 8) × 5 SDRAMs
×72
2048MB
4Gb (32M × 16 × 8) × 5 SDRAMs × 2 ranks
×72
4096MB
8Gb (64M × 16 × 8) × 3 SDRAMs
×36
2048MB
8Gb (64M × 16 × 8) × 5 SDRAMs
×72
4096MB
8Gb (64M × 16 × 8) × 5 SDRAMs × 2 ranks
×72
8192MB
End of Table 4
1. The ECC device can be either ×8 or ×16 as long as the number of row and column address bits match for all devices in the
memory array.
2.6.2.1 Memories – SDRAM Selection Criteria
Table 5 shows the DDR3 SDRAM selection criteria necessary for KeyStone I and
KeyStone II devices.
Table 5
DDR3 SDRAM Selection Criteria for KeyStone I and KeyStone II Devices
Description
Min
Max
Unit
Notes
Width
8
16
bit
Minimum supported total bus width is 16 (2- 8)
Depth (Density)
512M
8192M
bit
Definition may also include ranked devices (UDIMM)
Data Rate
800
1600
MT/s
Clock Rate
400
800
MHz
Temperature Range
0
95
°C
VDD
1.425
1.575
V
VDDq
1.425
1.575
V
Latency
5
11
Depends on end-use application
All CAS latencies supported between 5 and 11
JEDEC-compliant DDR3 SDRAM
End of Table 5
Page 12 of 48
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
2 Prerequisites
www.ti.com
DIMM support extends to modules whose SDRAMs fall within the above
specifications. DIMM bus width may be 64 or 72 with ECC support. DIMMs must
be unbuffered (UDIMMS), and may have up to two ranks. KeyStone I devices do not
support address mirroring. KeyStone II devices support UDIMM topologies that have
the address bits in the second rank mirrored.
2.7 DRAM Electrical Interface Requirements
This section briefly defines the electrical interface requirements for using
JEDEC-compliant DDR3 SDRAM with the KeyStone DSP. Additional information
and requirements may exist. Where different or conflicting requirements exist
(between applicable standards, SDRAM and DSP data sheets, and this design guide),
this design guide should take precedence. Details provided in the following subsection
were obtained from TI internal reference material and applicable JEDEC DDR3
SDRAM standards.
2.7.1 Slew
The released JEDEC standard describes in detail skew requirements imposed on the
SDRAM, in particular CK, CK#, DQS, and DQS#. In order to meet these requirements,
loading, SDRAM component selection, and trace routing will have a large impact. See
the modeling and simulation section of this guide for additional information regarding
meeting slew rate requirements.
2.7.2 Overshoot & Undershoot Specifications
Overshoot (OS) and undershoot (US) limitations are defined in the following tables
and figures. A theoretical instantaneous maximum upper and lower amplitude limit of
400 mV is allowed for overshoots and undershoots (assuming zero time is involved).
Because each overshoot and undershoot has a component of time, each applicable
signal must be further independently evaluated. Table 6 shows the limitations as listed
in the current released standard for all address and control signals. (Measurements are
obtained at the pin of the memory DRAM and not the DSP.)
Table 6
OS & US Requirements for Address and Control Lines
Speed
Max pk OS Amplitude (V)
Max pk US Amplitude (V)
Max OS Area Above VDD (V-ns)
Max US Area Below VSS (V-ns)
DDR3-800
0.4
0.4
0.67
0.67
DDR3-1066
0.4
0.4
0.5
0.5
DDR3-1333
0.4
0.4
0.4
0.4
DDR3-1600
0.4
0.4
0.33
0.33
End of Table 6
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 13 of 48
2 Prerequisites
www.ti.com
Figure 3 shows the limitations as listed in the current released standard for all address
and control pins.
Figure 3
Control and Address Overshoot & Undershoot Requirements
Maximum Amplitude
Volts
Overshoot
Vdd
Vss
Undershoot
Maximum Amplitude
Time
Table 7 shows the limitations as listed in the current released standard for all data,
clock, strobe, and mask signals.
Table 7
OS and US Requirements for CK, CK#, DQ, DQS, DQS#, and DM Lines
Speed
Max pk OS Amplitude (V)
Max pk US Amplitude (V)
Max OS Area Above VDD (V-ns)
Max US Area Below VSS (V-ns)
DDR3-800
0.4
0.4
0.25
0.25
DDR3-1066
0.4
0.4
0.19
0.19
DDR3-1333
0.4
0.4
0.15
0.15
DDR3-1600
0.4
0.4
0.13
0.13
End of Table 7
Figure 4 shows the limitations as listed in the current released standard for all data,
clock, strobe, and mask signals.
Figure 4
Data, Clock, Strobe, & Mask Overshoot & Undershoot Requirements
Maximum Amplitude
Volts
Overshoot
Vddq
Vssq
Undershoot
Maximum Amplitude
Time
Page 14 of 48
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
2 Prerequisites
www.ti.com
2.7.2.1 Overshoot & Undershoot Example Calculations
The following provides as an example of the steps necessary to calculate the overshoots
and undershoots for any waveform.
Assumptions: VDD = 1.5 V; VSS = 0.00 V; Vref = 0.75 V (VDD/2)
Overshoot example (Figure 5)
1. Determine amplitude over VDD
2. Determine the duration of the amplitude
3. Calculate the final value
OS = Amplitude × Duration
OS = 180 mV × 1 ns
OS = 0.18 V-ns
4. Compare results to the applicable SDRAM row and column (Section 2.7.2)
paying attention to speed grade and signal type differences. If the SDRAM speed
was DDR3-800 or DDR3-1066 and this were a data, clock, strobe, or mask net, the
overshoot would be acceptable. In the case of data, clock, strobe, or mask
DDR3-1333 and DDR3-1600 speed grades, this level of overshoot is not
acceptable. Note: this level is acceptable for all speed grades of control and
address lines.
Signal Overshoot Calculations
Volts
Figure 5
1.6
1.5
1.4
1.3
1.2
1.1
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0,0
1
2
3 4
nS
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 15 of 48
2 Prerequisites
www.ti.com
Undershoot example (Figure 6)
1. Determine amplitude under VSS
2. Determine the duration of the amplitude
3. Calculate the final value
US = Amplitude × Duration
US = 215 mV × 1.5 ns
US = 0.322 V-ns
4. Compare results to the applicable SDRAM row and column (Section 2.7.2 paying
attention to speed grade and signal type differences). If the SDRAM was
DDR3-800 or DDR3-1066 and this were a data net, this overshoot would be
acceptable. In the case of DDR3-1333 and DDR3-1600, this level of undershoot is
not acceptable. Note: this level is acceptable for control and address lines.
Compare results to the applicable SDRAM row and column (notice speed grade
differences). This example undershoot is unacceptable for all speed grades for all
data, clock, strobe, and mask signals. It is acceptable for all speed grades of
address and control signals.
Figure 6
Signal Undershoot Calculations
0, 0
Volts
0.7
0.6
0.5
0.4
0.3
0.2
0.1
-0.1
-0.2
Page 16 of 48
nS
1
2
3
DDR3 Design Requirements for KeyStone Devices Application Report
4
5
SPRABI1B—May 2014
Submit Documentation Feedback
2 Prerequisites
www.ti.com
2.7.3 Typical DDR3 AC & DC Characteristics
Table 8
DDR3 Single-Ended Output Levels
Symbol
Parameter
DDR3-800, 1066, 1333, and 1600
Unit
VOH(DC)
DC output high measurement level (for IV curve linearity)
0.8 × VDDQ
V
VOM(DC)
DC output mid measurement level (for IV curve linearity)
0.5 × VDDQ
V
VOL(DC)
DC output low measurement level (for IV curve linearity)
0.2 × VDDQ
V
VOH(AC)
AC output high measurement level (for output SR)
VTT + 0.1 × VDDQ
V
VOL(AC)
AC output low measurement level (for output SR)
VTT - 0.1 × VDDQ
V
End of Table 8
Table 9
DDR3 Differential Output Levels
Symbol
Parameter
DDR3-800, 1066, 1333, and 1600
Unit
VOHdiff(AC)
AC differential output high measurement level (for output SR)
+ 0.2 × VDDQ
V
VOLdiff(AC)
AC differential output low measurement level (for output SR)
- 0.2 × VDDQ
V
End of Table 9
2.7.4 DDR3 Tolerances and Noise – Reference Signals
Limitations on DC voltage tolerance and AC noise for all reference voltages is well
defined in the applicable JEDEC standard (pg. 129 of JESD79-3C). Strict conformity to
these limitations is important to help ensure proper functionality of the DDR3 SDRAM
interface. The Vref tolerance is ±1% or VDD/2 ±1%, which equates to
0.7425 V – 0.7575 V. To achieve this tight tolerance, it is recommended (when using a
standard resistor divider network) that better than 1% tolerance components be used.
The alternative would be an active reference voltage source. Proper component
selection and decoupling is critical (see the layout and routing section for additional
details).
It is important to properly design the reference supply voltage to track VDD/VDDq.
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 17 of 48
3 Package Selection
www.ti.com
3 Package Selection
Individual SDRAMs are available in four different packages that support the various
memory device densities, and widths. The following summarizes and describes the
primary differences between package and density for common 8 and 16 SDRAMs.
3.1 Summary
The following packaging subsection summarizes at a high level each DDR3 SDRAM
configuration and pinout. See the manufacturers data sheets for the latest information.
3.1.1 ×4 SDRAM
•
KeyStone devices do not support DDR3 SDRAMs in an 4 configuration.
•
8 devices are compatible in both the 106-pin and 78-pin packages provided the
layout is correct and accounts for the larger package inclusive of the NC (support
ball) pin placements. (Note: if designed for the larger package, there is no
difference in routing required.)
1Gb and 2Gb devices are pin-compatible in the 106-pin and 78-pin packages
The 1Gb, 2Gb, and 4Gb devices are pin-compatible in the 78-pin package (Note:
other packages not available at the time this document was prepared)
1Gb devices, regardless of package, are not compatible between 16 and 8
devices, see the respective data sheets for additional details. (Note: It is possible
to layout for both 16 and 8 devices but this requires additional consideration
on switching signals and net attachments, which vary between packages. As a
general rule, Texas Instruments does not recommend designing the application
board for both topologies [8 & 16]).
The twin-die version of the 8 SDRAM is not compatible with the monolithic
version of the same density.
3.1.2 ×8 SDRAM
•
•
•
•
3.1.3 ×16 SDRAM
•
•
3.1.4 ×32 SDRAM
•
3.1.5 ×64 SDRAM
•
Page 18 of 48
1Gb and 2Gb 16 devices are compatible in the 96-pin package. (Note: Some
pinouts may change across manufacturers – review data sheets before selecting
the DRAMs).
The twin-die version of the 16 SDRAM are not compatible with the monolithic
version of the same density.
At the time this document was prepared, no 32 devices exist. Texas Instruments
cannot ensure functionality or support for 32 SDRAM devices.
At the time this document was prepared, no 64 devices exists. Texas
Instruments cannot ensure functionality or support for 64 SDRAM devices.
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
4 Physical Design and Implementation
www.ti.com
4 Physical Design and Implementation
4.1 Electrical Connections
This section discusses the proper electrical interface between JEDEC-compliant DDR3
SDRAMs and UDIMMs to the Texas Instruments KeyStone DSP family DDR3
controller. These sections also provide details for the electrical pin connectivity
between the DSP DDR3 interface and DDR3 SDRAM or UDIMM. This is not an
all-inclusive listing of available parts as DRAM manufacturers are continuously
developing higher density parts or obsoleting others. It should also be noted that not all
DSPs will have the same bus width – see the device data sheet for details. This section
assumes both 32-bit and 64-bit wide buses are used.
4.1.1 Pin Connectivity & Unused Pins – SDRAM Examples
Correct DDR3 pin connectivity is vital to help insure the performance and reliability of
the DSP/DRAM system. An example set of pin connections from DSP-to-SDRAM are
shown in Figure 7.
Figure 7
DSP-to-SDRAM Connection Example
DVDD15
TI
DM[3:2]
DDRVref
Keystone
DDR3
Controller
Interface
DVDD15
DQS[3:2 ] / DQS#[3:2]
DQ[ 31:16]
BA[2:1]
DDR3
x16
Command/ Control
Vtt
ADD[13:0]
CLK0 / CLK0#
DM[1:0]
DQ[15:0]
DDR3
x16
DQS[1:0 ] / DQS#[1:0]
Pin nomenclatures are identified in detail in the respective DSP and SDRAM data
sheets and should be confirmed before releasing the design to layout, PCB fabrication,
or production. This section does not include details for non-JEDEC compliant
(JESD79-3C) SDRAM components nor does TI recommend the use of any
non-compliant JEDEC SDRAMs.
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 19 of 48
4 Physical Design and Implementation
www.ti.com
4.1.2 Pin Connectivity – ECC UDIMM & Non-ECC UDIMM Examples
This section defines the recommended configuration(s) when using a standard DDR3
UDIMM (unbuffered DIMM) with and without ECC (Error Detection & Correction).
Additional details pertaining to UDIMM implementation can be found in the JEDEC
DDR3 UDIMM standard 21C, the JEDEC UDIMM mechanical standard, MO-269,
and the JEDEC socket standard, SO-007B (latest revision). See Section 4.3.2 for
additional details pertaining to routing requirements.
Correct DDR3 pin connectivity is vital to ensure the performance and reliability of the
DSP/UDIMM system. See the device-specific data manual for pin connectivity.
Pin nomenclatures are identified in detail in the respective DSP and UDIMM data
sheets and should be confirmed before releasing the design to layout, PCB fabrication,
or production. This section does not include specific details for non-JEDEC compliant
UDIMMs nor does TI recommend the use of any non-compliant JEDEC UDIMMs.
Figure 8 shows the basic interconnection between the TI KeyStone DSP and respective
ECC UDIMM.
Figure 8
DSP to UDIMM Connection
DVDD15
DVDD15
DDR3
UDIMM
TI
Command/ Address
CKE / CS# / ODT
DDRVref
Keystone
DDR3
Controller
Interface
Vtt
CLK0 / CLK0#
CLK1 / CLK1#
CB[7:0]
DQS[8:0 ] / DQS#[8:0]
DM[8:0]
DQ[63:0]
Page 20 of 48
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
4 Physical Design and Implementation
www.ti.com
4.2 Signal Terminations
The following subsection describes the two signal termination methods (leveling and
non-leveling) used for DDR3 interfaces, specific termination placement, and the
impact of incorrect termination schemes and component values. Terminations are
placed at the end of the signal path. In the current fly-by architecture, placing the
terminations at the last SDRAM improves the overall signal characteristics, which is an
improvement over previous DDR2 SDRAM topology.
4.2.1 External Terminations – When Using Read & Write Leveling
As a rule of thumb, terminations should be applied to all clock, address, and control
lines. Although all address, control, and command lines should be end-terminated
when using leveling, clocking nets may require different termination values than that
used on the command and address nets. Each respective address and command net
should be end-terminated using a resistor (in the range of 39  to 42 ) and connected
to VTT (preferred value is 39 , 1%). VTT is defined as VDDq/2 or 0.75 V.
The DDR3 clock nets also must be end-terminated. However, instead of
end-terminating the clock nets (DDRCLKOUTP/Nx where x is DDRCLKOUTP/N0 or
DDRCLKOUTP/N1, whichever is used), each net shall be terminated with a series
39-1% resistor to a 0.1-μF capacitor to DVDD15 (VDDq). Figure 2 (and respective
notes) show the required clock termination implementation. All components should
be 1% tolerance or better. VTT should be generated using a resistor divider network
(1% tolerance or better). For proper operation, the VTT termination must track
VDDq/2.
Again, an important point is that the parallel termination should be placed at the last
SDRAM in the fly-by or daisy-chained architecture. Each trace to the respective
termination should be  within 500 mils and the opposite side of the termination
resistor should tie directly to the VTT rail.
External terminations may not be required – the only way to determine if your
topology requires end terminations is to perform complete simulations inclusive of
all topology parasitics.
Note—All VTT terminations should be placed at the end of the transmission
line (net). Incorrect placement will have an impact on performance and
functionality.
4.2.2 External Terminations – When Read and Write Leveling is Not Used
Texas Instruments does not recommend the use of DDR3 without leveling. The use of
DDR3 SDRAMs without read and write leveling places an undue burden and
constraints on the physical design and topology. In essence, any DDR3 design not
making use of read and write leveling becomes nothing more than a DDR2 layout with
all of the negative implications associated with DDR2 and none of the benefits of DDR3
(except possibly speed).
4.2.3 Internal Termination – On-Die Terminations
Prior DSPs supporting DDR2 interfaces did not support on-die terminations (ODT),
whereas the KeyStone family of DSPs now supports ODT. One of the primary
advantages to using DDR3 is the fact that the data lines no longer require series
terminations to optimize signal overshoots and undershoots. The current DDR3
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 21 of 48
4 Physical Design and Implementation
www.ti.com
instantiation allows for a wider range of values. In addition, DDR3 now supports
dynamic ODT, which has enormous benefits in a complex application board topology.
For DDR3, the DSP controller ODT pins (connected to each SDRAM) serve to turn on
or off the SDRAM internal termination. The actual ODT functionality of each SDRAM
is controlled using the mode registers (see the respective SDRAM data sheets for
additional information).
4.2.4 Active Terminations
Active terminations, regardless of configuration, are not required. The combination of
TI’s new DSP controller and the new DDR3 DRAM architecture in the supported
configurations eliminates the need for active terminations.
4.2.5 Passive Terminations
Unlike DDR2, the addition of series passive components is no longer necessary when
using DDR3 (with leveling). The only passive terminations to be used (when using
DDR3 with leveling) are the end-termination pull-up resistors identified in
Section 6.1.2.
4.2.6 Termination Component Selection
All termination components shall be 1% tolerance, component size (form factor) will
be dependent upon power requirements and parasitics. Texas Instruments
recommends 0402 size discrete (passive) components to improve routing, and reduce
parasitic inductance.
Page 22 of 48
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
4 Physical Design and Implementation
www.ti.com
4.3 Mechanical Layout and Routing Considerations
This subsection defines the basic requirements regarding mechanical layout and PCB
routing. These are general guidelines, which, when followed in conjunction with the
remainder of sections in this design guide (and good engineering practices), will help
provide a functional DDR3 interface. In all cases, it is recommended that the DDR3
interface be modeled to verify functionality and performance.
4.3.1 Routing Considerations – SDRAMs
This subsection describes the specific requirements for discrete SDRAM
implementations.
4.3.1.1 Mechanical Layout – SDRAMs
This subsection provides basic information regarding mechanical DDR3  DSP layout
constraints. Included are such topics as routing, stack up, trace lengths, and the use of
net classes. Issues not covered (in any great detail) within this document include
thermal considerations, part density, pick and place issues, and different packages /
footprints.
This document assumes that the user has an above average level of understanding
regarding mechanical layout and design – including the impact of trace width, spacing,
via size, bulk and decoupling capacitance selection, and placement.
4.3.1.2 Stack Up – SDRAMs
Stack up refers to the mechanical layer assembly of the printed circuit board. In all
high-speed designs it is good practice to maintain symmetry between the top half of the
board and the bottom half of the board. Referencing (sandwiching) signal routing
layers containing high speed signals between ground planes reduces EMI problems and
provides a constant and controlled impedance transmission path. DDR3 IO power
planes can also be used as reference layers for address/command/control signals as long
as there are decoupling capacitors distributed across the plane to provide a
low-impedance return path to ground. Power and ground planes should ideally be solid
planes without breaks. Power planes can have breaks or splits outside of the DDR3
routing region.
Proper stack up must also provide the proper characteristic printed circuit board (PCB)
impedance. Most DSP application board systems require a PCB impedance of 50 .
It is recommended that a DDR3 implementation make use of a minimum of four
routing layers: two for address/command/control signals and two for data-group
signals. Data-group routing should be on layers close to the bottom of the board to
minimize stubs. An end-application PCB will likely have a minimum total of 8 layers if
sufficient board area is available, but high-performance boards or boards making use
of an extended peripheral set may consist of 12 or more layers.
4.3.1.3 Routing Rules – General Overview (SDRAMs)
Several key points to remember when routing any signals on the application board:
• Organize the power, ground, and signal planes so that you eliminate or
significantly reduce the number of split / cut planes present in the design (no
splits are allowed under any DDR3 routes).
• It is strongly recommended that all SDRAMs are mounted on the top side of the
PCB alongside the SoC for single rank designs.
• Apply net classes, e.g., group key signals together.
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 23 of 48
4 Physical Design and Implementation
•
•
•
•
•
•
•
•
•
•
•
•
Page 24 of 48
www.ti.com
Maintain an acceptable level of skew across the entire DDR3 interface (by net
class).
Use proper low-pass filtering on the Vref pins.
Follow the fly-by architecture concept for all address, command, clock, and
control lines.
Increase the size of the decoupling capacitor trace width to as large as possible,
keep the stub length as short as possible.
Center-to-center spacing, including serpentine, must be at least 5 W where W is
the trace width. Additional spacing can be added between differential pairs and
other routing groups to minimize crosstalk. Spacing of 4 W can be used, but is not
appropriate for bus speeds over 1066 MT/s.
Maintain a common ground reference for all bypass/decoupling capacitors,
DSPs, and SDRAMs.
Take into account the differences in propagation delays between microstrip
and stripline nets when evaluating timing constraints. All long routes should be
stripline to reduce EMI and timing skew, and any microstrip routed for BGA
breakouts should be as short as possible.
All length-matching is based on an equivalent stripline length. An equivalent
stripline length is defined as the length of a stripline trace that will have the same
delay as the microstrip portion of the route (see the JEDEC UDIMM specification
for more information on velocity compensation).
There can not be any mid-point vias in the design on any data group net. Extra
vias that are located on the data group will negatively impact signal integrity.
It is strongly recommended that all nets be simulated to assure proper design,
performance, and signal integrity.
Routes along the same path and routing segment must have the same number of
vias. Vias can be blind, buried, or HDI microvia for improved SI but are not
required for standard data rates. Similarly, back drilling vias is not required for
standard data rates but can be used to eliminate via stubs.
It is strongly recommended that the routing channels between the DSP to
SDRAM be dedicated solely to the SDRAM interface and that no other signals be
routed in the area. Other signals routed on the same layers must be kept apart
from the DDR3 routes. There must be additional separation of the DDR3 nets of
at least 6W. In addition, these other traces should not be referenced to the DDR3
IO power planes. If other signals must be routed through this area, they need to
be isolated to their own routing plane(s) and shielded from the DDR3 routes by
a solid ground plane.
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
4 Physical Design and Implementation
www.ti.com
Net classes are an important concept when routing high speed signals that incorporate
timing constraints or timing relationships. When routing the DDR3 nets, there are four
basic groups (net classes) to consider – Table 10 shows the recommended net classes:
Table 10
SDRAM Net Class Routing Rules
Net Class
Signals
Notes
Data
DQS[8:0], DQS#[8:0], DQ[n:0], CB[7:0], DM[8:0]
1, 2, 4
Address/Command
BA[2:0], A[n:0], RAS, CAS, WE
2, 3
Control
CS, CKE, ODT
2, 5
Clock
DDRCLKOUTP/N[1:0]
5
Note 1: CB[7:0] refer to ECC devices
Note 2: n refers to some number of lines and is dependent upon device selected
Note 3: Observe relationship between DQ, DQS, DQS#, DDRCLKOUT, and DDRCLKOUT#
Note 4: The data net class will be subdivided into multiple byte lane routing groups each containing a
DQS/DQS# pair and the associated DM and 8 data bits
Note 5: Some nets in class may be used only in dual-rank topologies
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 25 of 48
4 Physical Design and Implementation
www.ti.com
Figure 9 shows the typical interface topology when connecting single-rank and
dual-rank designs. Note that the dual-rank illustration shows only the connections
for byte lane 0, and fly-by connections will still need to include SDRAMs on subsequent
byte lanes.
Figure 9
Interface Topology for Single and Dual Rank
DDR3
DDR3
DQS/DQS #
DQS/DQS#
Address
Address
Address
R
Control
Control
Control
R
R
DQS/DQS#/DQM
Data
Data
CLKIN
CLKIN
DDR3 Interface
DVDD15
DVDD15/2 (VTT)
Single Rank DDR3 Implementation
DSP
Data
DDRCLKOUT
Dual Rank DDR3 Implementation
DDR3
RANK 2
DDR3
RANK 1
DQS/DQS #
DQS /DQS #
Address
Address
Control
Control
Address
R
Control
R
Data
Data
Data
DDR3 Interface
DVDD15
DVDD15/2 (VTT)
DQS/DQS#/DQM
DSP
R
CLKIN
R
DDRCLKOUT0
CLKIN
DDRCLKOUT1
Note: The CLKOUT, CS, ODT, and CKE pins will be unique to each rank in a dual rank design.
Page 26 of 48
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
4 Physical Design and Implementation
www.ti.com
4.3.1.4 Routing Rules – Address and Command Lines (SDRAMs)
The following rules must be followed when routing address/command nets in a DDR3
design:
• 50- (+/- 5%) single-ended impedance required.
• All nets in the address and command fly-by groups must route along the same
path from the controller to each SDRAM sequentially, and then to the VTT
termination.
• All nets in the address and command fly-by groups must be length-matched from
the controller to each SDRAM separately within +/- 20 mils of the clock along the
same route.
• All nets in the address and command fly-by groups must have the same number
of vias in each length-matched segment.
• Address lines cannot be swapped to simplify routing.
• Address and command fly-by groups must have stubs less than 80 mils and be
length-matched within +/- 10 mils.
• All nets in the address and command fly-by groups must route adjacent to a solid
ground plane or a solid DVDD15 power plane with adequate distributed
decoupling to provide high frequency return.
• All nets in the address and command fly-by groups should be routed on
close layers to minimize via skew – these are normally close to the center or upper
layers of the board.
Table 11 shows the numeric routing rules listed above for address/command lines.
Table 11
Address and Command Line Numeric Routing Rules
Rule Number Parameter
Value
Unit
1
Net Impedance (single-ended)
50

2
Skew between fly-by group nets
+/- 20
mils
3
Stub length
< 80
mils
4
Stub skew
+/- 10
mils
4.3.1.5 Routing Rules – Control Lines (SDRAMs)
The following rules must be followed when routing control nets in a DDR3 design:
• 50- (+/- 5%) single-ended impedance required.
• All nets in the control fly-by groups must route along the same path from the
controller to each SDRAM sequentially, and then to the VTT termination.
• All nets in the control fly-by-groups must be length matched from the controller
to each SDRAM separately within +/- 20 mils of the clock along the same route.
• All nets in the control fly-by groups must have the same number of vias in each
length-matched segment.
• Control fly-by-groups must have stubs less than 80 mils and be length-matched
within +/- 10 mils.
• All nets in the control fly-by groups must route adjacent to a solid ground plane
or a solid DVDD15 power plane with adequate distributed decoupling to provide
high frequency return.
• All nets in the control fly-by groups should be routed on close layers to minimize
via skew – these are normally close to the center or upper layers of the board.
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 27 of 48
4 Physical Design and Implementation
www.ti.com
Table 12 shows the numeric routing rules listed above for control lines.
Table 12
Control Line Numeric Routing Rules
Rule Number Parameter
Value
Unit
1
50

Net Impedance (Single-Ended)
2
Skew between fly-by group nets
+/- 20
mils
3
Stub length
< 80
mils
4
Stub skew
+/- 10
mils
4.3.1.6 Routing Rules – Data Lines (SDRAMs)
The following rules must be followed when routing data nets in a DDR3 design:
• 100- (+/- 5%) differential impedance required on data strobe (DQS) pairs.
• DQS/DQS# pairs must be routed differentially.
• 50- (+/- 5%) single-ended impedance required on point-to-point routes.
• All nets in each data group must have same number of vias - maximum of two.
The number per signal within each byte group must match.
• All nets in a single byte-lane group should be routed on the same layer to
eliminate the addition of length-skew from the via barrels.
• All data-group nets must route adjacent to a solid ground plane.
• All data strobe pairs must be length-matched with +/- 1 mils of each other.
• All nets within a single data group must be length-matched with +/- 10 mils.
• Data bits within a byte-lane can be swapped to simplify routing.
• Data group nets are routed point-to-point and do not have VTT terminations.
Table 13 shows the numeric routing rules listed above for data lines.
Table 13
Data Lane Numeric Routing Rules
Rule Number Parameter
Value
Unit
1
Net Impedance (Single-Ended)
50

2
Net Impedance (Differential)
100

3
Skew between DQS pairs
+/- 1
mils
4
Skew between data group nets for given byte lane
+/- 10
mils
Table 14 shows the required data signal byte lane groupings.
Table 14
Data and Data Strobe Byte Lane Grouping
Data Group
Data
Data Strobe
Data Strobe
Data Mask
BYTE LANE 0
DQ[7:0]
DQS0
DQS0#
DM0
BYTE LANE 1
DQ[15:8]
DQS1
DQS1#
DM1
BYTE LANE 2
DQ[23:16]
DQS2
DQS2#
DM2
BYTE LANE 3
DQ[31:24]
DQS3
DQS3#
DM3
BYTE LANE 4
DQ[39:32]
DQS4
DQS4#
DM4
BYTE LANE 5
DQ[47:40]
DQS5
DQS5#
DM5
BYTE LANE 6
DQ[55:48]
DQS6
DQS6#
DM6
BYTE LANE 7
DQ[63:56]
DQS7
DQS7#
DM7
ECC BYTE LANE
CB[7:0]
DQS8
DQS8#
DM8
End of Table 14
Page 28 of 48
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
4 Physical Design and Implementation
www.ti.com
4.3.1.7 Routing Rules – Clock Lines (SDRAMs)
The following rules must be followed when routing clock nets in a DDR3 design:
• CK/CK# pairs must be routed differentially.
• 100- (+/- 5%) differential impedance required on all clock pairs.
• Each CK/CK# pair is managed as a separate routing group.
• Each CK/CK# pair must be routed to all SDRAMs within a single rank.
• All nets in the clock fly-by groups must route along the same path from the
controller to each SDRAM sequentially and then to an AC VTT termination.
• All clock pairs must be length-matched from the controller to each SDRAM
separately within +/- 1 mils of each other.
• All nets in the clock fly-by group must have the same number of vias in each
length-matched segment.
• Clock pair stubs must be less than 40 mils and length-matched within +/-1 mil.
• All clock pair nets must route adjacent to a solid ground plane.
Table 15 shows the numeric routing rules listed above for data lines.
Table 15
Clock Lane Numeric Routing Rules
Rule Number Parameter
Value
Unit
1
Net Impedance (differential)
100

2
Skew between CK/CK# pairs
+/- 1
mils
3
Stub length
<40
mils
4
Stub skew
+/- 1
Figure 10 shows the required DDRCLKOUT and DQ/DQS/# DQ/DQS/DQS#
routing from the DSP to SDRAM(s) for a single-rank topology.
DDRCLKOUT and DQ/DQS/# Routing from the DSP to SDRAM(s)
DQS2p/n
DQS3p/n
SPRABI1B—May 2014
Submit Documentation Feedback
d
DQS3p/n
CLKd
SDRAM
DQS1p/n
DQS0p/n
CLKa
DQS1p/n
b
CLKb
DQS2p/n
c
CLKc
SDRAM
DSP
a
SDRAM
DQS0p/n
CLK
SDRAM
Figure 10
DDR3 Design Requirements for KeyStone Devices Application Report
Page 29 of 48
4 Physical Design and Implementation
www.ti.com
4.3.1.8 Routing Rules – Power (SDRAMs)
When routing the DDR3 SDRAM Vref voltages, it is necessary to decouple them at the
SDRAMs and not at the source. The use of 0.01-μF and 0.1-μF ceramic capacitors (0402
or smaller recommended) should be distributed across the Vref power rail with one
0.01-μF and 0.1-μF ceramic capacitor located at each Vref pin, and one 0.1-μF capacitor
directly at the source. Traces between the decoupling capacitors and Vref pins should
be a minimum of 30 mils (0.762 mm) wide and as short as possible. The Vref pins and
interconnection to decoupling capacitors should maintain a minimum of 15 mils
(0.381 mm) spacing from all other nets. All Vref nets should be routed on the top layer.
Vref pins should be isolated with, or shielded with ground.
When routing the SDRAM VTT power supply, the regulator should be kept close to the
VTT pin on the respective SDRAMs. In most cases, a VTT voltage island will be used,
and it is recommended that the voltage island be placed on the component side signal
layer. There should be a minimum of one 0.1-μF decoupling capacitor close to each
VTT SDRAM pin and a minimum of one 10-μF to 22-μF bulk ceramic (low ESR)
capacitor on the VTT island. The number of VTT bulk capacitors is based on the size
of the island, and topology and loading.
4.3.1.9 Round-Trip Delay Impact on Routing – KeyStone I
The leveling processes in the DDR3 interface impose an upper limit on the maximum
round-trip delay. If this limit is exceeded, the DDR3 interface may fail the leveling
process and data corruption may occur. This limit is sufficiently large that
well-controlled topologies will not likely exceed the limit.
The round-trip delay for a given SDRAM is defined as the sum of two delays. The first
is the longest delay for the clock, command, control, and address groups to that
SDRAM. The second is the delay for the data group to that same SDRAM. This
round-trip delay must be calculated for each byte-lane to each SDRAM device
implemented in the DDR3 memory topology, including SDRAM devices on DIMM.
All of these individual sums must be below the limit to help ensure robust operation.
Internally, the DDR3 controller logic has a theoretical upper limit of four clock cycles.
There are multiple processes that have variation terms that reduce this time window, as
listed below:
• ddrclkoutperiod – period of reference clock the DSP is providing to SDRAM
• tDQSCK - DQS to CK skew limit from SDRAM datasheet – this is stated for
each standard speed grade in the JEDEC DDR3 SDRAM standard
• Invert clock out – delay of half a clock cycle if enabled – option normally used
only in very small memory topologies
• adjustment_sum – sum that accounts for all of the leveling errors, buffer delays,
and jitter terms associated with the circuitry, and the leveling adjustment – upper
limit is defined to be half a clock period plus 225 ps.
The following equation provides an approximation of the maximum round trip delay:
• Case 1: Invert clock out disabled
– round_tripdelay < (4 * ddrclkoutperiod) - tDQSCK - adjustment_sum
– round_tripdelay < (4 * ddrclkoutperiod) - tDQSCK - 0.5 * ddrclkoutperiod 225 ps
– round_tripdelay < (3.5 * ddrclkoutperiod) - tDQSCK - 225 ps
Page 30 of 48
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
4 Physical Design and Implementation
www.ti.com
•
Case 2: Invert clock out enabled (adds an additional half-clock period of delay to
the command delay term)
– round_tripdelay < (4 * ddrclkoutperiod) - tDQSCK - 0.5 * ddrclkoutperiod adjustment_sum
– round_tripdelay < (4 * ddrclkoutperiod) - tDQSCK - 0.5 * ddrclkoutperiod - 0.5 *
drclkoutperiod - 225 ps
– round_tripdelay < (3 * ddrclkoutperiod) - tDQSCK - 225 ps
Based on the previous equations, the following calculations and summary table show
the write leveling skew limitations for both invert clock out enabled and disabled, given
the DDR3-1333 and DDR3-1600 JEDEC SDRAM specification. The first column for
each speed-grade category lists the maximum write leveling skew in picoseconds. The
second column for each lists the maximum write leveling skew in inches assuming a
signal propagation rate of 180 ps/in.
For DDR3-1333:
• ddrclkoutperiod = 1500 ps
• tDQSCK = +/- 255 ps
• margin = 100 ps
Case 1: Invert clock out disabled:
– round_tripdelay < (3.5 * ddrclkoutperiod) - tDQSCK - 225 ps
– round_tripdelay < (3.5 * 1500 ps) - 255 ps - 225 ps
– round_tripdelay < 4770 ps
Case 2: Invert clock out enabled (adds an additional half-clock period of delay to
the command delay term):
– round_tripdelay < (3 * ddrclkoutperiod) - tDQSCK - 225 ps
– round_tripdelay < (3 * 1500 ps) - 255 ps - 225 ps
– round_tripdelay < 4020 ps
For DDR3-1600:
• ddrclkoutperiod = 1250 ps
• tDQSCK = +/-225 ps
• margin = 100 ps
Case 1: Invert clock out disabled
– round_tripdelay < (3.5 * ddrclkoutperiod) - tDQSCK - 225 ps
– round_tripdelay < (3.5 * 1250 ps) - 225 ps - 225 ps
– round_tripdelay < 3925 ps
Case 2: Invert clock out enabled (adds an additional half-clock period of delay to
the command delay term)
– round_tripdelay < (3 * ddrclkoutperiod) - tDQSCK - 225 ps
– round_tripdelay < (3 * 1250 ps) - 225 ps - 225 ps
– round_tripdelay < 3300 ps
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 31 of 48
4 Physical Design and Implementation
www.ti.com
Table 16 shows the round trip delay limitations for both invert clock out enabled and
disabled. The first column for each lists the maximum round-trip delay in picoseconds.
The second column for each lists the maximum routing length in inches assuming a
signal propagation rate of 180 ps/in.
Table 16
Maximum Round Trip Delay Example - Invert Clock Out Enabled and Disabled
Speed Grade
Invert Clock Out Disabled
Invert Clock Out Enabled
DDR3-1333
4770 ps
26.50 in
4020 ps
22.33 in
DDR3-1600
3925 ps
21.81 in
3300 ps
18.33 in
Because this is preliminary guidance and some small margin should be subtracted from
these delays to account for additional terms such as multi-rank delay skew, TI
recommends that the maximum routing lengths be reduced by 10%.
4.3.1.10 Write Leveling Limit Impact on Routing – KeyStone I
The write-leveling process in the DDR3 interface imposes a limit on the maximum and
minimum skew between the command delay and the data delay. If these limits are
exceeded, the DDR3 interface may fail the write leveling process and data corruption
may occur. These limits are sufficiently large that well-controlled topologies will not
likely exceed the limits.
The command delay is defined as delay for the clock, command, control, and address
group signals from the DSP to a given SDRAM. The data delay is the delay for the data
group signals to that same SDRAM. The write-leveling result is effectively the
difference, or skew, between these two delays.
The maximum write-leveling skew is the largest difference between the two delays in
the topology to a single SDRAM. Likewise, the minimum write-leveling skew is the
smallest difference between the two delays in the topology to a single SDRAM.
The write-leveling logic has a theoretical upper limit of 2500 ps. This limit does not
scale with SDRAM data rate. The theoretical upper limit equates to two full clock cycles
when the clock frequency is 800 MHz for DDR3-1600. It is reduced by half a clock
cycle when invert clock out is enabled, as this effectively lengthens the clock by this
amount.
The following set of equations provides an approximation of the maximum and
minimum write-leveling skew allowed:
• ddrclkoutperiod – period of reference clock the DSP is providing to SDRAM
• tWLS – from JEDEC DDR3 SDRAM specification, write-leveling setup time
from rising CK, CK# crossing to rising DQS, DQS# crossing
• tJIT(per, lck) – clock period jitter during DLL locking period
• commanddelay – delay for the clock, command, control and address group signals
from the DSP to a given SDRAM
• datadelay – delay for the data group signals to that same SDRAM
• write_levelingskew – defined as the value commanddelay - datadelay
• margin - additional margin added for preliminary use
Maximum write-leveling skew:
• Case 1: Invert clock disabled
– write_levelingskew < 2500 ps - tWLS - tJIT(per, lck) - margin
Page 32 of 48
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
4 Physical Design and Implementation
www.ti.com
•
Case 2: Invert clock enabled (adds an additional half-clock period of delay to the
command delay term)
– commanddelay + (0.5 *ddrclkoutperiod) - datadelay = write_levelingskew <
2500 ps - tWLS - tJIT(per, lck) - margin
– write_levelingskew < 2500 ps - tWLS - tJIT(per, lck) - (0.5 *ddrclkoutperiod) margin
Minimum write-leveling skew:
• Case 1: Invert clock disabled
– write_levelingskew > tWLS + tJIT(per, lck) + margin
• Case 2: Invert clock enabled (adds an additional half-clock period of delay to the
command delay term)
– commanddelay + (0.5 *ddrclkoutperiod) - datadelay = write_levelingskew > tWLS
+ tJIT(per, lck) + margin
– write_levelingskew > tWLS + tJIT(per, lck) - (0.5 *ddrclkoutperiod) + margin
tWLS and tJIT(per, lck) are standard JEDEC DDR3 SDRAM timing parameters that
can be obtained from the specific datasheet of the SDRAM chosen.
Note—Because this is preliminary guidance, some small margin should be
subtracted from these delays to account for additional terms such as
multi-rank delay skew. TI currently recommends setting the extra margin term
to 100 ps.
Based on the previous equations, the following calculations and summary table shows
the write-leveling skew limitations for both invert clock out enabled and disabled given
the DDR3-1333 and DDR3-1600 JEDEC SDRAM specification. The first column for
each speed-grade category lists the maximum write-leveling skew in picoseconds. The
second column for each lists the maximum write-leveling skew in inches, assuming a
signal propagation rate of 180 ps/in.
For DDR3-1333:
• ddrclkoutperiod = 1500 ps
• tWLS = 195 ps
• tJIT(per, lck) = +/- 70 ps
• margin = 100 ps
Maximum write-leveling skew:
• Case 1: Invert clock disabled
– write_levelingskew < 2500 ps - tWLS - tJIT(per, lck) - margin
– write_levelingskew < 2500 ps - 195 ps - 70 ps -100 ps
– write_levelingskew < 2500 ps - 195 ps - 70 ps -100 ps
– write_levelingskew < 2135 ps
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 33 of 48
4 Physical Design and Implementation
•
www.ti.com
Case 2: Invert clock enabled (adds an additional half-clock period of delay to the
command delay term)
– write_levelingskew < 2500 ps - tWLS - tJIT(per, lck) - (0.5 *ddrclkoutperiod) margin
– write_levelingskew < 2500 ps - 195 ps - 70 ps - (0.5 *1500 ps) - 100 ps
– write_levelingskew < 1385 ps
Minimum write-leveling skew:
• Case 1: Invert Clock disabled
– write_levelingskew > tWLS + tJIT(per, lck) + margin
– write_levelingskew > 195 ps + 70 ps + 100 ps
– write_levelingskew > 365 ps
• Case 2: Invert clock enabled (adds an additional half-clock period of delay to the
command delay term)
– write_levelingskew > tWLS + tJIT(per, lck) - (0.5 *ddrclkoutperiod) + margin
– write_levelingskew > 195 ps + 70 ps - (0.5 *1500 ps) + 100 ps
– write_levelingskew > -385 ps
Note—This minimum write-leveling skew calculation with invert clock enabled
shows how the invert clock mode can be used to correct a small amount of
negative skew between the command and data groups. However, as specified
in Section 4.3.1.7, all topologies should be designed for a positive skew between
the command delay and data delay to avoid this situation.
For DDR3-1600:
• ddrclkoutperiod = 1250 ps
• tWLS = 165 ps
• tJIT(per, lck) = +/- 60 ps
• margin = 100 ps
Maximum write-leveling skew:
• Case 1: Invert clock disabled
– write_levelingskew < 2500 ps - 165 ps - 60 ps -100 ps
– write_levelingskew < 2500 ps - 165 ps - 60 ps -100 ps
– write_levelingskew < 2175 ps
• Case 2: Invert clock enabled (adds an additional half-clock period of delay to the
command delay term)
– write_levelingskew < 2500 ps - 165 ps - 60 ps - (0.5 *1250 ps) - 100 ps
– write_levelingskew < 1610 ps
Minimum write-leveling skew:
• Case 1: Invert clock disabled
– write_levelingskew > 165 ps + 60 ps + 100 ps
– write_levelingskew > 325 ps
Page 34 of 48
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
4 Physical Design and Implementation
www.ti.com
•
Case 2: Invert clock enabled (adds an additional half-clock period of delay to the
command delay term)
– write_levelingskew > 165 ps + 60 ps - (0.5 *1250 ps) + 100 ps
– write_levelingskew > -300 ps
Note—This minimum write-leveling skew calculation with invert clock enabled
shows how the invert clock mode can be used to correct a small amount of
negative skew between the command and data groups. However, as specified
in Section 4.3.1.7, all topologies should be designed for a positive skew between
the command delay and data delay to avoid this situation.
Table 17
Maximum Write Leveling Skew Example
Invert Clock Out State
Disabled
Enabled
Speed Grade
Skew in ps
Skew in inches
Skew in ps
Skew in Inches
DDR3-1333
2135
11.861
1385
7.694
DDR3-1600
2175
12.083
1610
8.944
Table 18
Minimum Write Leveling Skew Example
Invert Clock Out State
Disabled
Enabled
Speed Grade
Skew in ps
Skew in inches
Skew in ps
Skew in Inches
DDR3-1333
365
2.027
-385
2.138
DDR3-1600
325
1.805
-300
1.666
Note—Because this is preliminary guidance and some small margin should be
subtracted or added from these delays to account for additional terms such as
multi-rank delay skew, TI recommends that the maximum routing lengths be
reduced by 10% and the minimum routing lengths be increased by 10%.
4.3.2 Routing Considerations – UDIMMs
4.3.2.1 Mechanical Layout – UDIMMs
When designing the application hardware to use an UDIMM (unbuffered DIMM),
there are several issues to take into account that may differ when using individual
SDRAMs, including power and route lengths. Key issues including stack-up and
power/ground-plane referencing do not differ between the two topologies (SDRAM
versus UDIMM).
4.3.2.2 Stack Up – UDIMMs
The board stack-up should not change regardless of whether you are using an UDIMM
or individual SDRAMs. General guidelines for stack-up under the SDRAMs section
apply. It is worth mentioning that a good-quality connector and proper impedance is
critical to minimize signal reflections.
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 35 of 48
4 Physical Design and Implementation
www.ti.com
4.3.2.3 Routing Rules – General Overview (UDIMMs)
General routing rules between the KeyStone DSP DDR3 interface and UDIMM are the
same as those identified for discrete components (SDRAMs). Note that a single DDR3
interface on any KeyStone DSP cannot currently support multiple DIMMs in a design.
Return ground paths and power-plane decoupling also are critical and should be
evaluated properly.
All aspects of this application note and especially those pertaining to the use and
implementation of UDIMMs assumes that the user has an above average level of
understanding regarding mechanical layout and design – including the impact of trace
width, spacing, via size, and bulk and decoupling capacitance selection and placement.
4.3.2.4 Routing Rules – Address and Command Lines (UDIMMs)
All address and command nets between the DSP and UDIMM must be referenced to a
solid ground or solid power plane — the best option would be a solid ground plane. The
entire address bus and/or command nets can be referenced to the DVDD15 power
plane as long a sufficient distributed decoupling is located near the endpoints and vias
of all routes to provide a low-impedance current return path. All address and command
nets routing from the DSP to the UDIMM socket must be adjacent to the referenced
plane.
The address bus and command nets between the DSP and UDIMM should be routed
away from the data bus and respective nets.
4.3.2.5 Routing Rules – Control Lines (UDIMMs)
All control nets between the DSP and UDIMM must be referenced to a solid ground or
solid power plane — the best option would be a solid ground plane. The control nets
can be referenced to the DVDD15 power plane as long a sufficient distributed
decoupling is located near the endpoints and vias of all control routes to provide a low
impedance current return path. All control nets routing from the DSP to the UDIMM
socket must be adjacent to the referenced plane.
The control nets between the DSP and UDIMM should be routed away from the data
bus and respective nets.
4.3.2.6 Routing Rules – Data Lines (UDIMMs)
All data-group signals must be routed adjacent to a solid ground plane.
If possible, all data-group nets should be routed internally and close to the bottom of
the board. All nets in a single byte-group should be routed on the same layer to
eliminate addition length skew from the via barrels. All data nets should be
skew-matched between data/strobe within the respective byte lane. The byte lanes are
identified in the SDRAM routing rules in Table 14.
4.3.2.7 Routing Rules – Clock Lines (UDIMMs)
All DDR3 clocks between the DSP and respective UDIMM must be routed as
differential pairs. Each differential clock pair must be length-matched to the address,
command, and control signals.
Page 36 of 48
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
4 Physical Design and Implementation
www.ti.com
4.3.2.8 Routing Rules – Power (UDIMMs)
For the UDIMM, there exist three separate power supplies, all derived from a common
rail. The first is the 1.5-V supply that provides power to all the DDR3 UDIMM I/Os.
The second supply is the VREF supply, which must track the VDD15 supply and
establishes a reference voltage for the UDIMM. The last supply is the bus termination
supply (VTT).
Each of the Vref supplies to the UDIMM (VrefCA and VrefDQ) can originate from a
common rail but must be individually decoupled at the UDIMM. VREF must be 50% of
the VDD/VDDQ level and meet the tolerances identified in the respective UDIMM data
manual. The typical method for establishing each of these reference voltages is through
a 1% (or better) resistor divider network. It is important that the VREF (VrefCA and
VrefDQ) voltages track the VDD/VDDQ level across all corners (process, temperature,
and noise). See the applicable data manual for transient (AC & DC) requirements.
When routing the UDIMM VREF voltages, properly decouple them as close to the
socket as possible. The use of 0.01-μF and 0.1-μF ceramic capacitors (0402 or smaller
recommended) should be distributed across the VREF power rail with one 0.01-μF and
one 0.1-μF ceramic capacitor located at each VREF pin and one 0.1-μF capacitor directly
at the source. Traces between the decoupling capacitors and VREF pins should be a
minimum of 0.030 inch (0.762 mm) wide and as short as possible. The VREF pins and
interconnection to decoupling capacitors should maintain a minimum of 0.015 inch
(0.381 mm) spacing from all other nets. All VREF nets should be routed on the top layer.
VREF pins should be isolated with, or shielded with ground.
The UDIMM termination voltage (VTT) must be at a constant level of 0.750 V and
must be capable of sinking a reasonable amount of current while maintaining voltage
regulation. VTT must remain stable at all times for the UDIMM to function properly.
Issues including noise and crosstalk must be eliminated or reduced to a negligible
amount. VTT like VREF must track all variations with respect to VDD/VDDq.
When routing the UDIMM VTT power supply the regulator should be kept as close to
the VTT pin on the respective SDRAMs. In most cases a VTT voltage island will be used
and it is recommended that the voltage island be placed on the component-side signal
layer. There should be a minimum of one 0.1-μF decoupling capacitor close to each
VTT SDRAM pin and a minimum of one 10-μF to 22-μF bulk capacitor on the VTT
island. The number of VTT bulk capacitors is based on the size of island and topology.
4.3.2.9 Write-Leveling Limit Impact on Routing
The write-leveling process in the DDR3 interface imposes an upper and lower limit on
the maximum and minimum skew between the command delay and the data delay. If
this limit is exceeded, the DDR3 interface may fail the write leveling process and data
corruption may occur. This limit is sufficiently large so that well-controlled topologies
will not likely exceed the limit.
The same write-leveling limits in KeyStone I designs outlined for SDRAMs in
Section 4.3.1.10 also apply to UDIMMS. Any calculations performed must take both
board and UDIMM routing into consideration.
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 37 of 48
4 Physical Design and Implementation
www.ti.com
4.4 Timing Considerations
DDR3 requires strict timing relationships between CK (and CK#) and the
address/control lines, and between data and the DQS (and DQS#) lines. The TI DSP
DDR3 interface is designed to comply with the DDR3 JEDEC standard with regards to
timing constraints. See the applicable standard when evaluating timing. Additional
timing considerations or constraints may be included in the respective DSP or SDRAM
data sheets.
4.5 Impedance Considerations
The guidelines below review the impedance considerations for routing and device
configuration.
4.5.1 Routing Impedances – KeyStone I Devices
The DDR3 interface on KeyStone I devices has, in general, been verified to operate up
to 1333 MT/s with up to nine loads on the fly-by nets (address, control, command and
clock). Validation has focused on single-rank memory topologies with only a single
load on the data group signals. Topologies with fewer than nine loads should all operate
at a 1333 MT/s level of performance or below.
4.5.1.1 Data Group Signals
All data-group signals are point-to-point in the validated topologies. The data-group
signals are driven by the KeyStone device on writes and driven by the SDRAM
memories during reads. No external resistors are needed on these routes. The receivers
in both cases (SDRAMs on writes and KeyStone device on reads) will assert on-die
terminations (ODT) at the appropriate times. The following diagrams show the
impedances seen on these nets during write and read cycles.
Figure 11 shows the impedances seen on the nets during a write cycle. During writes,
the output impedance of the KeyStone I device is approximately 45 It is
recommended that the SDRAM be implemented with a 240- RZQ resistor and be
configured to present an ODT of RZQ/6 for an effective termination of 40 
Figure 11
Data Group Impedances During Write Cycles on KeyStone I
DSP
SDRAM

PCB
SDRAM WRITE
Page 38 of 48
DDR3 Design Requirements for KeyStone Devices Application Report
DDR_TERM =
default RZQ /6 =
40 disable
DYN_ODT)
SPRABI1B—May 2014
Submit Documentation Feedback
4 Physical Design and Implementation
www.ti.com
Figure 12 shows the impedances seen on the nets during a read cycle. During reads, it
is recommended that the SDRAM be configured for an effective drive impedance of
RZQ/7 or 34  (assuming RZQ resistor is 240 The on-die termination (ODT)
within the KeyStone I device will have an effective Thevenin impedance of 45 
Figure 12
Data Group Impedances During Read Cycles on KeyStone I
DSP
SDRAM
PCB
ODT = 
SDRAM_DRIVE –
default RZQ /7 = 34
SDRAM READ
4.5.1.2 Fly-By Signals
The fly-by signals include the address, control, command, and clock routing
groups. The fly-by signals consist of the fly-by routing from the DSP, stubs at each
SDRAM, and terminations after the last SDRAM. The address, control, and command
groups will be terminated through a 39.2- resistor to VTT. The clock pairs will be
terminated through 39.2- resistors to a common node connected to a capacitor that
is then connected to VDDQ. The KeyStone I device will present a 45- output
impedance when driving these signals. These relationships are shown in Figure 13.
Figure 13
Fly-By Impedances on KeyStone I
DSP
SDRAM
VTT

PCB

ADDRESS / CONTROL / COMMAND / CLOCK
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 39 of 48
4 Physical Design and Implementation
www.ti.com
4.5.2 Routing Impedances – KeyStone II Devices
The DDR3 interface on KeyStone II devices has, in general, been verified to operate up
to 1600 MT/s with up to nine loads on the fly-by nets (address, control, command, and
clock). Validation has focused on single-rank memory topologies with only a single
load on the data group signals. Topologies with fewer than nine loads should all operate
at a 1600 MT/s level of performance or below.
4.5.2.1 Data Group Signals
All data-group signals are point-to-point in the validated topologies. The data-group
signals are driven by the KeyStone device on writes and driven by the SDRAM
memories during reads. No external resistors are needed on these routes. The receivers
in both cases (SDRAMs on writes and KeyStone device on reads) will assert on-die
terminations (ODT) at the appropriate times. The following diagrams show the
impedances seen on these nets during write and read cycles.
Figure 14 shows the impedances seen on the nets during a write cycle. During writes,
the output impedance of the KeyStone II device is approximately 40 It is
recommended that the SDRAM be implemented with a 240  RZQ resistor and be
configured to present an ODT of RZQ/6 for an effective termination of 60 
Figure 14
Data Group Impedances During Write Cycles on KeyStone II
DSP
SDRAM
ZO –
RZQ/6 =

DDR_TERM
- RZQ/4 =
60disable
DYN_ODT)
Page 40 of 48
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
4 Physical Design and Implementation
www.ti.com
Figure 15 shows the impedances seen on the nets during a read cycle. During reads, it
is recommended that the SDRAM be configured for an effective drive impedance of
RZQ/7 or 34  (assuming RZQ resistor is 240 The ODT within the KeyStone II
device will have an effective Thevenin impedance of 60 
Figure 15
Data Group Impedances During Read Cycles on KeyStone II
DSP
SDRAM
PCB
ODT –
RZQ/4 =
6
SDRAM_DRIVE –
RZQ/7 = 34
SDRAM READ
4.5.2.2 Fly-By Signals
The fly-by signals include the address, control, command, and clock routing
groups. The fly-by signals are composed of the fly-by routing from the DSP, stubs at
each SDRAM, and terminations after the last SDRAM. The address, control, and
command groups will be terminated through a 39.2- resistor to VTT. The clock pairs
will be terminated through 39.2- resistors to a common node connected to a
capacitor that is then connected to VDDQ. The KeyStone II device can be configured
to present either a 40- output impedance when configured for RZQ/6 (recommended
for discrete components) or a 34- output impedance when configured for RZQ/7
(recommended for UDIMMs). These relationships are shown in Figure 16.
Figure 16
Fly-By Impedances on KeyStone II
DSP
SDRAM
VTT
ZO – 34RZQ/
7) for UDIMM and
34 (RZQ/7) or
40 (RZQ/6) for
discrete
SPRABI1B—May 2014
Submit Documentation Feedback
PCB
39.2
ADDRESS / CONTROL / COMMAND / CLOCK
DDR3 Design Requirements for KeyStone Devices Application Report
Page 41 of 48
4 Physical Design and Implementation
www.ti.com
4.5.3 Comparison to JEDEC UDIMM Impedance Recommendations
The JEDEC UDIMM specification provides guidance for UDIMM manufacturers
including recommendations for routing impedances. These recommendations are
similar to the routing impedances discussed above. The JEDEC UDIMM specification
recommends 40- routes on the fly-by nets between the controller and the first
SDRAM and then 60- routes between the SDRAMs and to the terminations.
However, footnotes under Table 39 in the JEDEC UDIMM specification state that
lightly loaded UDIMMs (like the topologies we expect customers to implement with
KeyStone devices) should target 49  rather than 60 . Differential routes can be
constructed from these same single-ended track structures resulting in differential
impedances under 100 .
4.6 Switching and Output Considerations
Signal switching and output swings can have a significant if not catastrophic impact
on signal integrity and EMI compliance. Routing, SDRAM placement, stack up, and
use cases can also attribute to added dynamic switching noise. The concept of dynamic
switching noise and induced ground bounce is usually attributed to improper layout
and stack up, insufficient decoupling, and improper routing. To minimize the output
and switching problems, system modeling is recommended and usually employed.
Page 42 of 48
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
5 Simulation and Modeling
www.ti.com
5 Simulation and Modeling
This section provides a quick overview regarding the importance of simulating and
modeling the DSP and DDR3 interfaces.
5.1 Simulation and Modeling
All high-performance interfaces, especially those operating above 300 MHz, should
always be modeled. Proper simulation and modeling (which must include a complete
application board (PCB) stack up, DSP DDR3 interface, and SDRAMs) is important to
verify and confirm component placement, selection, and signal integrity. In a
high-performance interface, signal stubs, perturbations, non-monotonic waveforms,
inflections, and reflections become significant. Time spent properly modeling the
DDR3 interface regardless of the topology or condition selected will pay off in the long
run.
Figure 17 shows the impact of proper placement and routing of high speed signals.
Figure 17
DDR3 Simulations
yp
y
V
V
1750.0
1550.0
1350.0
1150.0
V
o
l
t
a
g
e
m
V
-
950.0
750.0
550.0
350.0
150.0
-50.0
0.00
500.0
1000.0
1500.0
2000.0
2500.0
Time (ps)
3000.0
3500.0
4000.0
4500.0
5.2 Tools
A variety of software tools exists to give customers the ability to model
high-performance interfaces. Up until recently, the most common simulation tool was
Spice or HSpice. These tools are extremely costly, difficult to use, and vary between
types. Other tools including the IBIS standard and protocol were traditionally avoided
because of their limitation at high frequencies.
5.3 Models
Different simulation models can be found for the DDR3 SDRAMs and will be available
from the SDRAM vendor you have selected. The available models are typically one of
the following three formats: Spice, HSpice, or IBIS. Texas Instruments intends to
provide IBIS models to our customers.
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 43 of 48
5 Simulation and Modeling
www.ti.com
5.4 TI Commitment
Texas Instruments has taken the initiative to provide IBIS models compliant with the
latest 5.0 IBIS standard for the DDR3 interface. The Texas Instruments DSP IBIS model
will be correlated to internal Matlab or Spice models, and functionally against timing
and performance parameters prior to release (TMS).
Page 44 of 48
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
6 Power
www.ti.com
6 Power
This section briefly describes the DDR3 relative power, DSP DDR3 interface relative
power, power assessment, and power sequencing.
6.1 DDR3 SDRAM Power Requirements
There exists three different power supplies for the SDRAM (1.5-V primary IO supply,
Vref, and VTT) and three different power supplies for the DSP (1.1 V, 1.5 V, and Vref).
It is recommended that the KeyStone DSP DDR3 interface and SDRAM share a
common 1.5-V supply rail. A common 1.5-V supply rail (±5% maximum AC/DC
tolerance) simplifies the overall design, and reduces the differential between the two
devices and minimizes the need for additional layers (power) in the end-use
application.
6.1.1 Vref Voltage Requirements
There exist two DDR3 SDRAM Vreference pins: VrefCA and VrefDQ. VrefCA is the
reference voltage for all command, address, and control pins. VrefDQ is the reference
voltage for the data lines. It is not necessary, but typically recommended, that both
Vreference voltages originate from the same supply source. Both Vreference pins must
be derived from VDD/2 (VDDq/2). The recommended Vreference (Vref)
implementation is by using a simple resistor divider with 1% or better accuracy. The
distance between the source voltage through the divider network and to the decoupled
Vreference pins must be short. Each Vreference pin must properly track the VDD/2
(VDDq/2) variations over voltage, noise, and temperature differences. The pk-to-pk
AC and DC noise on the Vreference pins cannot exceed ±2% or 1.5 mV.
6.1.2 VTT Voltage Requirements
The DDR3 SDRAM termination voltage is referred to as VTT and requires a 750-mV
supply. The recommended VTT source is a regulator that is capable of sinking a
sufficient amount of current while at the same time maintaining a tight voltage
tolerance. Like the Vref pins, the distance between the VTT source voltage and SDRAM
pin must be short and decoupled properly. The VTT pin must be kept stable and
properly track the VDD/VDDq variations over voltage, noise, and temperature
differences. The pk-to-pk AC and DC noise on the VTT pin cannot exceed ± 2% or
1.5 mV.
6.2 DSP DDR3 Power Requirements
On the DSP, the AVDDA2 pins are designed to supply power to the internal DDR3
clock PLL. These pins must be connected to a clean 1.8-V supply rail. There exist other
DSP 1.8-V supplies that can be used provided the voltage tolerance is maintained and
a separate filter is used. All DVDD15 pins are designed to supply power to the DSP
DDR3 IO buffers. As with the 1.5-V power pins, these must also be connected to a clean
1.5-V supply rail. The DDR3xVREFSSTL pin(s) provides the reference voltage to the
DSP DDR3 interface. This supply pin is derived from VDD/2 using 1% or better
resistors.
See the respective DSP data sheet and application notes for power supply requirements.
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 45 of 48
6 Power
www.ti.com
6.3 DDR3 Power Estimation
Actual power for each SDRAM is dependent upon many factors. Most DDR3 SDRAM
manufacturers have application notes to aid in estimating the DDR3 power. Power can
be divided into two categories: active and leakage. All SDRAMs contain some element
of each. When designing the power supplies, it is strongly recommended that you take
into account both active and leakage power as well as peak in-rush currents. A suitable
margin should always be added to the final calculation.
As a general rule of thumb, a single DDR3 SDRAM in a 16 (256M Byte) configuration
running at 1.333 MT/s and assuming 50% reads and 50% writes will consume a
maximum total power of 864 mW or 576 mA.
6.4 DSP DDR3 Interface Power Estimation
Excluding the DDR3 SDRAM power discussed above, power estimates for the DSP
DDR3 interface can be obtained using the respective KeyStone DSP power application
note and spreadsheet.
6.5 Sequencing – DDR3 and DSP
The DSP device requires specific power sequencing as well as the DDR3 DRAM(s). In
all cases, see the respective data sheet for verification of power up and power down
sequencing requirements.
In all cases, the DRAM cannot be allowed to drive into the DSP until the DSP has been
fully powered and sequenced properly.
Page 46 of 48
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
7
www.ti.com
Disclaimers
7 Disclaimers
Comments and proposals identified within this document are presented as guidelines.
Texas Instruments can not fully anticipate every possible design permutation,
topology, and application variation and therefore requires each end user to perform
proper due diligence (using good engineering practices and including proper
component selection) when designing for a DDR3 interface. This includes proper
simulation and modeling to verify the recommendations provided function in an
acceptable manner (in the end-use application).
This application guide is not intended to be the sole source or reference design guide.
Many factors not encompassed within this document can force a change in component
values, selection, placement, or termination type.
The possible permutations available due to component, layout, and assembly can also
induce variations not encompassed in the recommendations provided. Voltage rails,
manufacturer parts selection, switching levels should all be verified before committing
the design to production.
It is the end user’s responsibility to always verify the design, and connectivity between
any active or passive component and the targeted DSP. This includes verifying against
individual data sheets and application notes.
Simulation and modeling play an important role in system verification and validation.
This effort and the need to conduct this level of analysis should not be overlooked in
any design – regardless of simplicity or complexity.
This is not a primer for high-speed design. Use of the information provided assumes a
strong understanding of electrical and mechanical design requirements in a
high-performance DSP application environment.
From time to time there exist DSP or application design changes. It is the end user’s
responsibility to verify that the appropriate data sheets and application notes are
followed.
Specifications continuously change and are updated. Verify the specifications have not
changed and that the documents used are correct for the version of silicon you are
designing with.
Prior to final design release it is recommended that a full engineering assessment be
performed by the end user in order to assure functionality.
SPRABI1B—May 2014
Submit Documentation Feedback
DDR3 Design Requirements for KeyStone Devices Application Report
Page 47 of 48
8 Revision History
www.ti.com
8 Revision History
The following table lists changes for revisions.
Revision
Date
Description of Changes
SPRABI1B
May 2014
Added Impedance Calibration Section (Page 1-38)
Modified the DDRCLKOUT and DQ/DQS/# Routing from the DSP to SDRAM(s) figure to show clock fly-by topology
(Page 1-29)
Modified the Typical DDR3 Fly-By Architecture figure to better show fly-by architecture (Page 1-6)
Removed details of specific example topologies. Pin connectivity now only illustrated in Figures 7 and 8 (Page 1-19)
Removed tables of supported memories from Chapter 2 (Page 1-9)
Removed the Appendix. Information overlapped that in JEDEC spec and device data manual. (Page 1-47)
Revised routing rule sections for SDRAMS and UDIMMs (Page 1-23)
Updated the Interface Topology for Single and Dual Rank figure to better show single and dual rank topologies
(Page 1-26)
SPRABI1A
September 2011
SPRABI1
August 2011
Page 48 of 48
Initial Document Release
DDR3 Design Requirements for KeyStone Devices Application Report
SPRABI1B—May 2014
Submit Documentation Feedback
IMPORTANT NOTICE
Texas Instruments Incorporated and its subsidiaries (TI) reserve the right to make corrections, enhancements, improvements and other
changes to its semiconductor products and services per JESD46, latest issue, and to discontinue any product or service per JESD48, latest
issue. Buyers should obtain the latest relevant information before placing orders and should verify that such information is current and
complete. All semiconductor products (also referred to herein as “components”) are sold subject to TI’s terms and conditions of sale
supplied at the time of order acknowledgment.
TI warrants performance of its components to the specifications applicable at the time of sale, in accordance with the warranty in TI’s terms
and conditions of sale of semiconductor products. Testing and other quality control techniques are used to the extent TI deems necessary
to support this warranty. Except where mandated by applicable law, testing of all parameters of each component is not necessarily
performed.
TI assumes no liability for applications assistance or the design of Buyers’ products. Buyers are responsible for their products and
applications using TI components. To minimize the risks associated with Buyers’ products and applications, Buyers should provide
adequate design and operating safeguards.
TI does not warrant or represent that any license, either express or implied, is granted under any patent right, copyright, mask work right, or
other intellectual property right relating to any combination, machine, or process in which TI components or services are used. Information
published by TI regarding third-party products or services does not constitute a license to use such products or services or a warranty or
endorsement thereof. Use of such information may require a license from a third party under the patents or other intellectual property of the
third party, or a license from TI under the patents or other intellectual property of TI.
Reproduction of significant portions of TI information in TI data books or data sheets is permissible only if reproduction is without alteration
and is accompanied by all associated warranties, conditions, limitations, and notices. TI is not responsible or liable for such altered
documentation. Information of third parties may be subject to additional restrictions.
Resale of TI components or services with statements different from or beyond the parameters stated by TI for that component or service
voids all express and any implied warranties for the associated TI component or service and is an unfair and deceptive business practice.
TI is not responsible or liable for any such statements.
Buyer acknowledges and agrees that it is solely responsible for compliance with all legal, regulatory and safety-related requirements
concerning its products, and any use of TI components in its applications, notwithstanding any applications-related information or support
that may be provided by TI. Buyer represents and agrees that it has all the necessary expertise to create and implement safeguards which
anticipate dangerous consequences of failures, monitor failures and their consequences, lessen the likelihood of failures that might cause
harm and take appropriate remedial actions. Buyer will fully indemnify TI and its representatives against any damages arising out of the use
of any TI components in safety-critical applications.
In some cases, TI components may be promoted specifically to facilitate safety-related applications. With such components, TI’s goal is to
help enable customers to design and create their own end-product solutions that meet applicable functional safety standards and
requirements. Nonetheless, such components are subject to these terms.
No TI components are authorized for use in FDA Class III (or similar life-critical medical equipment) unless authorized officers of the parties
have executed a special agreement specifically governing such use.
Only those TI components which TI has specifically designated as military grade or “enhanced plastic” are designed and intended for use in
military/aerospace applications or environments. Buyer acknowledges and agrees that any military or aerospace use of TI components
which have not been so designated is solely at the Buyer's risk, and that Buyer is solely responsible for compliance with all legal and
regulatory requirements in connection with such use.
TI has specifically designated certain components as meeting ISO/TS16949 requirements, mainly for automotive use. In any case of use of
non-designated products, TI will not be responsible for any failure to meet ISO/TS16949.
Products
Applications
Audio
www.ti.com/audio
Automotive and Transportation
www.ti.com/automotive
Amplifiers
amplifier.ti.com
Communications and Telecom
www.ti.com/communications
Data Converters
dataconverter.ti.com
Computers and Peripherals
www.ti.com/computers
DLP® Products
www.dlp.com
Consumer Electronics
www.ti.com/consumer-apps
DSP
dsp.ti.com
Energy and Lighting
www.ti.com/energy
Clocks and Timers
www.ti.com/clocks
Industrial
www.ti.com/industrial
Interface
interface.ti.com
Medical
www.ti.com/medical
Logic
logic.ti.com
Security
www.ti.com/security
Power Mgmt
power.ti.com
Space, Avionics and Defense
www.ti.com/space-avionics-defense
Microcontrollers
microcontroller.ti.com
Video and Imaging
www.ti.com/video
RFID
www.ti-rfid.com
OMAP Applications Processors
www.ti.com/omap
TI E2E Community
e2e.ti.com
Wireless Connectivity
www.ti.com/wirelessconnectivity
Mailing Address: Texas Instruments, Post Office Box 655303, Dallas, Texas 75265
Copyright © 2014, Texas Instruments Incorporated