lattice timingclosure

Achieving Timing Closurein FPGA Designs

Lattice Semiconductor Corporation5555 NE Moore CourtHillsboro, OR 97124(503) 268-8000

November 2008

Achieving Timing Closure in FPGA Designs ii

CopyrightCopyright © 2008 Lattice Semiconductor Corporation.

This document may not, in whole or part, be copied, photocopied, reproduced, translated, or reduced to any electronic medium or machine-readable form without prior written consent from Lattice Semiconductor Corporation.

TrademarksLattice Semiconductor Corporation, L Lattice Semiconductor Corporation (logo), L (stylized), L (design), Lattice (design), LSC, E2CMOS, Extreme Performance, FlashBAK, flexiFlash, flexiMAC, flexiPCS, FreedomChip, GAL, GDX, Generic Array Logic, HDL Explorer, IPexpress, ISP, ispATE, ispClock, ispDOWNLOAD, ispGAL, ispGDS, ispGDX, ispGDXV, ispGDX2, ispGENERATOR, ispJTAG, ispLEVER, ispLeverCORE, ispLSI, ispMACH, ispPAC, ispTRACY, ispTURBO, ispVIRTUAL MACHINE, ispVM, ispXP, ispXPGA, ispXPLD, LatticeEC, LatticeECP, LatticeECP-DSP, LatticeECP2, LatticeECP2M, LatticeMico8, LatticeMico32, LatticeSC, LatticeSCM, LatticeXP, LatticeXP2, MACH, MachXO, MACO, ORCA, PAC, PAC-Designer, PAL, Performance Analyst, PURESPEED, Reveal, Silicon Forest, Speedlocked, Speed Locking, SuperBIG, SuperCOOL, SuperFAST, SuperWIDE, sysCLOCK, sysCONFIG, sysDSP, sysHSI, sysI/O, sysMEM, The Simple Machine for Complex Design, TransFR, UltraMOS, and specific product designations are either registered trademarks or trademarks of Lattice Semiconductor Corporation or its subsidiaries in the United States and/or other countries. ISP, Bringing the Best Together, and More of the Best are service marks of Lattice Semiconductor Corporation.

HyperTransport is a licensed trademark of the HyperTransport Technology Consortium in the U.S. and other jurisdictions.

Other product names used in this publication are for identification purposes only and may be trademarks of their respective companies.

DisclaimersNO WARRANTIES: THE INFORMATION PROVIDED IN THIS DOCUMENT IS “AS IS” WITHOUT ANY EXPRESS OR IMPLIED WARRANTY OF ANY KIND INCLUDING WARRANTIES OF ACCURACY, COMPLETENESS, MERCHANTABILITY, NONINFRINGEMENT OF INTELLECTUAL PROPERTY, OR FITNESS FOR ANY PARTICULAR PURPOSE. IN NO EVENT WILL LATTICE SEMICONDUCTOR CORPORATION (LSC) OR ITS SUPPLIERS BE LIABLE FOR ANY DAMAGES WHATSOEVER (WHETHER DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL, INCLUDING, WITHOUT LIMITATION, DAMAGES FOR LOSS OF PROFITS, BUSINESS INTERRUPTION, OR LOSS OF INFORMATION) ARISING OUT OF THE USE OF OR INABILITY TO USE THE INFORMATION PROVIDED IN THIS DOCUMENT, EVEN IF LSC HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES. BECAUSE SOME JURISDICTIONS PROHIBIT THE EXCLUSION OR LIMITATION OF CERTAIN LIABILITY, SOME OF THE ABOVE LIMITATIONS MAY NOT APPLY TO YOU.

LSC may make changes to these materials, specifications, or information, or to the products described herein, at any time without notice. LSC makes no

Achieving Timing Closure in FPGA Designs iii

commitment to update this documentation. LSC reserves the right to discontinue any product or service without notice and assumes no obligation to correct any errors contained herein or to advise any user of this document of any correction if such be made. LSC recommends its customers obtain the latest version of the relevant information to establish, before ordering, that the information being relied upon is current.

Type Conventions Used in This Document

Convention Meaning or Use

Bold Items in the user interface that you select or click. Text that you type into the user interface.

<Italic> Variables in commands, code syntax, and path names.

Ctrl+L Press the two keys at the same time.

Courier Code examples. Messages, reports, and prompts from the software.

... Omitted material in a line of code.

.

.

.

Omitted lines in code and report examples.

[ ] Optional items in syntax descriptions. In bus specifications, the brackets are required.

( ) Grouped items in syntax descriptions.

{ } Repeatable items in syntax descriptions.

| A choice between items in syntax descriptions.

Achieving Timing Closure in FPGA Designs iv

Achieving Timing Closure in FPGA Designs v

Contents

Achieving Timing Closure in FPGA Designs 1Introduction 1

Learning Objectives 1Time to Complete this Tutorial 2System Requirements 2Windows and Linux/UNIX, Command Line Conventions 2About the Tutorial Design 2The Lattice Timing Closure Methodology 3The Tutorial Process Flow 3Preparing the Command-line Environment 5

Task 1: Perform Logic Synthesis 6Logic Synthesis with Synplify 6

Task 2: Translate the Design 11Create the Output Directories 11Create a Logical Design Database File (.ngd) 12Command-line Summary 13

Task 3: Create Timing and Location Constraints 14Create a Logical Preference File (.lpf) 14Run the Design Planner Spreadsheet View 14Assign a Block Preference 15Assign Clock Preferences 17Assign I/O Timing Preferences 19Create Location Preferences 22Create a Physical Design Database File (.ncd) 23Estimate Baseline Performance Timing 24Command-line Summary 26

Task 4: Establish Baseline Performance 26PAR Session 1: Run Initial Place and Route 26Command-line Summary 31

Task 5: Increase Place and Route Effort 33PAR Session 2: Run PAR with a Higher fMAX Objective 33

Contents

Achieving Timing Closure in FPGA Designs vi

PAR Session 3: Run PAR in Multi-Placement Mode 35PAR Session 4: Run PAR with Multiple Routing Passes 38Command-line Summary 39

Task 6: Floorplan for Performance 40Prepare to Run the Design Planner from the Command Line 41Examine Critical Paths 41Use Design Planning and Grouping 42

Summary 52Recommended References 53Glossary 54Appendix A: Logic Synthesis with Precision RTL 57Appendix B: Command Line Interface Quick Reference 64

Achieving Timing Closure in FPGA Designs 1

Achieving Timing Closure in FPGA Designs

IntroductionThis tutorial demonstrates techniques for optimizing your Lattice FPGA design to meet timing performance objectives. In the tasks of this tutorial, you will use the ispLEVER software to implement an HDL design and see the influence of map and place-and-route preferences, increased placement and routing effort, and floorplanning. You will analyze the primary clock domain of the design and establish a performance baseline. You will then push the system to meet a faster timing objective until it fails, after which you will apply techniques to try to meet timing closure if technically possible.

This tutorial uses the command line interface, which gives you more flexibility for viewing and analyzing tutorial files and allows you to issue custom command sequences. Some of these features are not available from the ispLEVER Project Navigator GUI.

Learning ObjectivesWhen you have completed this tutorial, you should be able to:

Use the recommended methodology to meet timing objectives of your design specification

Use utilization and timing reports to understand device implementations

Influence map, place, and route to increase performance or minimize area utilization

Examine critical paths

Use scripts or batch files to process your design from the command line

Achieving Timing Closure in FPGA Designs Introduction


Use the Design Planner graphical point tool to define timing constraints and floorplan the design

Time to Complete this TutorialThe time to complete this tutorial is about five hours.

System RequirementsThe following software is required to complete the tutorial:

ispLEVER software

The following systems were used to implement the tutorial design and measure program run times.

System: Microsoft Windows XP Version 2002, Service Pack 3

Computer: Intel Pentium processor, 3.4 GHz, 2.0 GB of RAM

Windows and Linux/UNIX, Command Line ConventionsCommand line sequences in the tutorial use the Windows backslash (\) convention to indicate directory paths and delimiters. For Linux or UNIX platforms, substitute forward slashes (/).

About the Tutorial DesignThe tutorial design is a small system-on-chip implementation that utilizes a RISC processor, a 2048x32 on-chip RAM, an SDR SDRAM memory controller, a development interface, and a “traffic cop” module. The module serves as an arbiter of the Wishbone on-chip bus that connects the components. Figure 1 shows a high-level block diagram of the system.

The OpenRISC 1200 processor and Wishbone on-chip bus are both fully documented ASIC or FPGA IP cores that are available for free from the OpenCores organization. While the tutorial does not explore the functionality



of the design, it does demonstrate the timing closure and optimization techniques that can be effectively used within the design.

The Lattice Timing Closure MethodologyThe tutorial follows the methods recommended by Lattice for achieving timing requirements. Figure 2 illustrates the sequence of steps used in this method.

The Tutorial Process FlowThe tutorial tasks examine a few key techniques used by Lattice Applications Engineering to meet timing objectives with ispLEVER programs. The techniques can provide (though not always) improved results. They are presented in an order that provides the quickest implementation with the largest performance benefit.

Figure 6 on page 32 illustrates the process flow that establishes the design’s baseline performance. In this flow you will translate the EDIF netlist, produced by logic synthesis, into the databases that the ispLEVER design tools require for map, place, route, and timing analysis.

Figure 1: Tutorial Design

Note

The OpenRISC 1200 processor and the Wishbone interface are both active Open Source projects available from www.opencores.org and are covered by the GNU Lesser General Public License. They are freely available, freely usable and re-usable open source hardware. The SDR SDRAM Memory Controller is based on Lattice Reference Design 1010 and is covered by the permissions explained in the header section of the source files.



The following overview describes each processor:

edif2ngd – A netlist reader tool that converts an EDIF 2 0 0 format netlist into a binary .ngo file. This file describes the design in terms of components and hierarchy.

ngdbuild – A build utility that combines one or more .ngo files and binds each design component of the design to a Lattice FPGA library primitive. This produces an .ngd logical design database.

map – A design mapper program that converts the logical components of the .ngd into device-specific physical components. This produces an .ncd physical design database.

par – A place and route (PAR) program that assigns the locations of physical components and adds routing to connect them. This produces a placed and routed .ncd file.

Figure 2: Lattice Timing Closure Methodology



trce – A static timing analysis tool that you can run on an .ncd physical design with associated timing preferences to produce a timing report.

Design Planner – A design planning tool with a graphical user interface that helps you visualize design placement, routing, critical paths, and programming.

Preparing the Command-line EnvironmentThis section shows how to set up the Windows or UNIX OS environment to run ispLEVER command line executables.

To set Windows OS environment variables:

1. From the Start Menu, choose Settings > Control Panel > System.

2. Click the Advanced tab, and then click Environment Variables.

3. Select the variable PATH from the User variables for your system or from the System variables, and then click Edit.

4. Add the following path nodes to the Variable Value string:

<drive>\<install_dir>;<drive>\<install_dir>\ispcpld\bin;<drive>\<install_dir>\ispfpga\bin\nt;

where <drive> is the volume in which ispLEVER is installed

and <install_dir> is the ispLEVER installation directory.

For example, given a default installation of ispLEVER v7.2, add the path nodes as follows:

c:\ispTOOLS7_2;c:\ispTOOLS7_2\ispcpld\bin;c:\ispTOOLS7_2\ispFPGA\bin\nt;

5. Click OK to close the Edit System Variable dialog box.

6. Click OK to close the Environment Variables dialog box.

7. Clock OK to close the System Properties dialog box.

Note

If your system includes an installation of Xilinx ISE software, you should ensure that the ispLEVER-related path nodes occur before the Xilinx-related path nodes. Both software systems use similarly named program executables such as ngdbuild, map, par, etc.

Achieving Timing Closure in FPGA Designs Task 1: Perform Logic Synthesis


Task 1: Perform Logic SynthesisThis tutorial gives you three options for synthesis:

Create a netlist using Synplicity Synplify®. If you prefer to use Synplify, proceed with the Task 1 instructions that follow.

Create a netlist using Mentor Graphics Precision® RTL Synthesis. If you prefer to use Precision, use the instructions provided in “Appendix A: Logic Synthesis with Precision RTL” on page 57, and then proceed to Task 2.

Use the EDIF 2 0 0 netlist that is already provided in the timing_closure_tutor folder. If you prefer to use this netlist, skip to “Task 2: Translate the Design” on page 11.

Logic Synthesis with SynplifySynplify synthesis produces a netlist file that uses .edn for the file name extension.

To create a Synplify project and add the source files:

1. Create a new folder, Synplify, inside the timing closure tutorial directory, as follows:

<drive>\<install_dir>\examples\Tutorial\timing_closure_tutor\Synplify

2. From the Start menu, choose Programs > Lattice Semiconductor > Accessories > Synplify Pro for Lattice.

The Synplicity Synplify Pro for Lattice interface opens.

3. Choose File > New, and in the dialog box, select Project File (Project).



4. Under File Location, click the browse button.

5. Click the down arrow at the top, next to Look In, and navigate to

\<install_path>\examples\Tutorial\timing_closure_tutor.

6. Select the Synplify folder and click Choose to return to the New dialog box.

7. In the File Names text box, type ec_fpga_top and click OK.

8. Click the Add File button to open the Select Files to Add to Project dialog box.

9. Click the down arrow in the Look In box at the top and navigate to\<install_path>\examples\Tutorial\timing_closure_tutor\dbg_interface

Select all Verilog HDL (.v) files except dbg_defines.v and timescale.v.

Click Add.

10. Navigate to the ..\timing_closure_tutor\onchip_ram folder.

Select the two Verilog HDL (.v) files.

Click Add.

11. Navigate to the ..\timing_closure_tutor\or1200 folder.

Select all Verilog HDL (.v) files except or1200_defines.v.

Click Add.

12. Navigate to the ..\timing_closure_tutor\sdr_sdram folder.

Select all Verilog HDL (.v) files except sdr_par.v.

Click Add.

13. Navigate to the ..\timing_closure_tutor\uart16550 folder.

Select all Verilog HDL (.v) files except uart_defines.v and timescale.v.

Click Add.



14. Navigate to the ..\timing_closure_tutor folder.

Select the ec_fpga_top.v and tc_top.v files.

Click Add

15. Click OK.

To save the project:

1. In the Synplify main window, expand the Verilog folder.

2. Make sure that the ec_fpga_top.v file is listed last. If it is not, select it and drag it to the bottom of the list.

3. Choose File > Save.

To select the settings:

1. Click the Implementation Options button on the left.

2. Click the Device tab and select the following settings:

Technology: Lattice-ECP

Part: LFECP10E

Package: Q208C

Speed: -4

3. Select the Constraints tab. Select the first option under Frequency (MHz) and type 66 in the text box.

4. Select the Implementation Results tab and verify that ec_fpga_top.edn is listed in the Results File Name box and that ..\Synplify\rev_1 is displayed as the Results directory.

5. Click OK.

6. Right-click the ec_fpga_top project in the main window and choose Compile Only.

The Synplify software compiles the design based on the source files, targeted device, and settings you specified.

To create a new SCOPE file:

1. Choose File > New.

2. Select Constraint File (Scope), and then type ec_fpga_top in the File Names box.

3. Under File Location, verify that the path name is correct for the Synplify folder and click OK.

Note

If you want to preserve the original tutorial design files, copy the timing_closure_tutor directory to another location before proceeding.



4. Click OK to accept the defaults in the Create a New Top-Level SCOPE File dialog box and open the SCOPE editor.

5. In the SCOPE editor, select the Clocks tab on the bottom left.

6. Enable the constraints for clk and jtag_tck by selecting these options in the Enable column. Specify the following in row 1:

Clock: jtag_tck

Frequency (MHz): 10

7. Specify the following in row 2:

Clock: clk

Frequency (MHz): 66

8. Select the Inputs/Outputs tab.

9. Select the options in the Enable column for rows 1 and 2. Specify the following in row 1:

Port: <input default>

Type: input delay

Clock Edge: clk:r

Value: 10

10. Specify the following in row 2:

Port: <output default>

Type: output delay

Clock Edge: clk:r

Value: 10

11. Choose File > Save and save the file as ec_fpga_top_1.sdc. Click Yes to add the file to the project.

12. Close the SCOPE Editor.



To create the netlist and view the performance summary:

1. In the Synplify main window, click Run.

Synplify creates an output EDIF netlist, ec_fpga_top.edn, for the implementation. When the process has completed, the name of the netlist appears in the rev_1 pane to the right.

2. In the rev_1 pane, double-click ec_fpga_top.edn to open the netlist in a new tab.

3. Return to the ec_fpga_top.prj tab, and then double-click ec_fpga_top.srr to open the log file in a new tab.

4. Locate the Performance Summary section and notice the Estimated Frequency for clk.

Performance Summary *******************Worst slack in design: -7.790

Requested Estimated Starting Clock Frequency Frequency ----------------------------------------------clk 66.0 MHz 32.2 MHz jtag_tck 10.0 MHz 41.1 MHz ==============================================

5. Choose File > Exit, and in the dialog box, click Yes to save your changes to the project.

6. Copy the ec_fpga_top.edn file from the Synplify\rev_1 folder and paste it in the \timing_closure_tutor folder.

Note

This process might take several minutes to complete.

Achieving Timing Closure in FPGA Designs Task 2: Translate the Design


Task 2: Translate the DesignIn this task, you will create the output directories for the tutorial and you will translate the EDIF netlist produced by logic synthesis into a logical design database (.ngd).

Create the Output DirectoriesTo prepare for these tasks, you will set up directories for map, and par. These directories will store the map and timing analysis information for numerous iterations.

To make the needed directories using Windows OS:

1. From the Windows desktop, choose Start > Programs > Accessories > Command Prompt to open the Command Prompt window.

2. Type the following command to change the directory to the tutorial design folder:

cd \<install_path>\examples\Tutorial\timing_closure_tutor

3. Type the following commands to make new directories in the tutorial design folder:

md map1\par1 md map1\par2md map1\par3 md map1\par4 md map2

To make the needed directories using X-Windows:

1. Open a terminal.

2. Type the following command to change the directory to the tutorial design folder:

cd <install_path>/examples/Tutorial/timing_closure_tutor

3. Type the following commands to make the map directories:

mkdir map1 map2

4. Type the following commands to make the map1/par directories:

mkdir map1/par1mkdir map1/par2

Note

The exact location of the Command Prompt shortcut might vary from system to system.

The following Command Prompt window properties are recommended for a Screen Area of 1280 by 1024:

Font: Lucinda ConsoleFont Size: 14Layout – Screen Buffer Size: Width 800, Height 600Layout – Window Size: Width 116, Height 53



mkdir map1/par3mkdir map1/par4

Create a Logical Design Database File (.ngd)This procedure implements the two-part process of building the logical design database: the edif2ngd process converts the netlist file to an binary database file (.ngo), and the ngdbuild converts the .ngo to a logical design database (.ngd) file that can be used as input to mapping.

To create an .ngd Logical Design Database File:

1. Type the following command at the command prompt:

edif2ngd

The netlist reader runs and reports copyright information and the basic usage information.

edif2ngd: version ispLever_v72_PROD_Build (35)Copyright (c) 1991-1994 by NeoCAD Inc. All rights reserved. Copyright (c) 1995 AT&T Corp. All rights reserved. Copyright (c) 1995-2001 Lucent Technologies Inc. All rights reserved. Copyright (c) 2001 Agere Systems All rights reserved. Copyright (c) 2002-2008 Lattice Semiconductor Corporation, All rights reserved. MachXO LatticeXP2 LatticeXP LatticeSCM LatticeSC LatticeECP2S LatticeECP2MS LatticeECP2M LatticeECP2 LatticeECP LatticeEC

2. Type the following command to convert the tutorial EDIF netlist into an NGO object file:

edif2ngd –l LatticeECP ec_fpga_top.edf ec_fpga_top.ngo

The netlist reader runs and creates an NGO object file in the local folder.

3. Type the following command to build an .ngd file:

ngdbuild -a LatticeECP –p .\timing_closure_tutor ec_fpga_top.ngo ec_fpga_top.ngd > ngdbuild.log

The build tool runs and creates an .ngd file in the local folder.

Note

If you used Synplify for logic synthesis, the EDIF netlist is named ec_fpga_top.edn.



4. Type the following command to run the ispLEVER Text Editor and open the log file:

synedit ngdbuild.log.

The log file opens in the Text Editor. The many warnings that appear are normal; most of them reflect loadless nets that result from unused carry-out outputs of counters or the unused portions of memory arrays. It is good practice to examine these warnings since the nets and associated logic will be optimized away later.

5. In the Text Editor, choose File > Close to close the log file, but keep the Text Editor open.

Command-line SummaryThe following ispLEVER command lines were issued in this task to process the tutorial design. To quickly create a batch or script file for the tutorial, copy the commands to a text file.

rem ***********************************************************rem ispLEVER Tutorial: Timing Closure of FPGA Designsrem Task 2: Translate the Designremrem ***********************************************************

rem Create an .ngd Logical Database Fileedif2ngd -l LatticeECP ec_fpga_top.edf ec_fpga_top.ngo

ngdbuild -a LatticeECP -p .\timing_closure_tutorec_fpga_top.ngo ec_fpga_top.ngd > ngdbuild.log

Achieving Timing Closure in FPGA Designs Task 3: Create Timing and Location Constraints


Task 3: Create Timing and Location ConstraintsIn this task, you will use the Spreadsheet View of the ispLEVER Design Planner to assign timing constraints that are considered essential for good placement and routing results and that will produce a complete static timing analysis. You will define constraints such as frequency, false paths, and I/O timing.You will then use the design mapper to produce a physical design database (.ncd), and you will run an early static timing analysis to estimate the maximum operating frequency and timing constraints.

Create a Logical Preference File (.lpf)The logical preference file contains all the design constraints that are created after the build database process. Because you are using the command-line interface, you will need to create this file manually.

To create the logical preference file (.lpf):

1. In the Text Editor, choose File > New to create a blank document.

2. Choose File > Save As and make sure that the timing_closure_tutor folder is shown as the targeted directory.

3. In the File Name box, type ec_fpga_top.lpf and click OK.

4. Choose File > Close but keep the Text Editor open.

Logical preferences can be created and modified before and after mapping and after placement and routing. Saving logical preference changes to the .lpf file will ensure that they persist through repeated modifications and PAR iterations.

Run the Design Planner Spreadsheet ViewThe Pre-Map Design Planner provides a convenient spreadsheet view of preferences. The small Design Planner Control window that accompanies the Spreadsheet View allows you to hide and display this view. You should keep this Control window open throughout this task.

To run the Design Planner Spreadsheet View:

1. Type the following command:

flmainappw

The Design Planner Control window opens.

2. Choose File > Open Design and browse to the following directory:

<install_path>\examples\Tutorial\timing_closure_tutor

3. Select ec_fpga_top.ngd, and click Open.

The Device Selector dialog box appears.

4. Select the following options:

Device: LFECP10ESpeed Grade: -4



Package Type: PQFP208Operation Conditions: Commercial

5. Click OK.

Design Planner loads the logical design database (.ngd) of the tutorial design in Spreadsheet View and Package View. Keep the small Design Planner Control window open while you use Design Planner.

Assign a Block PreferenceIn this procedure, you will assign a BLOCK preference to the global reset of the system, rstn. When given a BLOCK preference, the ispLEVER system timing analysis skips the nets, paths, buses, and component pins that are not relevant to the synchronous timing of the design.

To block asynchronous, false paths:

1. Maximize Spreadsheet View, and then choose Preference > Block Preference to open the dialog box.

2. In the Filter text box, type rst* and click Go.

3. Select rstn_c and click Add.

4. Click OK.

Note

If you used Precision RTL for logic synthesis, the reset net is named rstn_int.



A BLOCK preference is defined for global Reset rstn_c and added to the list on the Block sheet in Spreadsheet View.

5. Select the Global tab.

6. Under Preference Value, double-click the box for Block Asynchpaths and choose ON from the menu.

7. Repeat Step 6 for Block Resetpaths and Block InterClock Domain Paths.

8. Choose File > Save to save the BLOCK preferences to the .lpf file.

A BLOCK preference is set for all other potential asynchronous paths such as primary inputs and set/reset signals.



Assign Clock PreferencesIn this procedure, you will assign frequency preferences to the primary clocks of the system, clk and jtag_tck. When given a frequency preference, the ispLEVER system infers a 50% dutycycle and a 1/FREQUENCY period clock. With these objectives, the place-and-route program (PAR) will attempt to optimize the levels and routing of logic between registers to achieve a period that is less than or equal to 1/FREQUENCY. Figure 3 illustrates the principle.

The bubble between REG1 and REG2 represents the logic and routing delay that will be optimized to meet the timing constraint.

Later in this tutorial you will apply the technique of over-constraining clock frequency by 20% while pushing the effort of place and route.

To create frequency preferences:

1. In Spreadsheet View, choose Preference > Period/Frequency to open the dialog box.


Type: FREQUENCYSecond Type: Clock Port

Figure 3: Levels of Logic And Routing Between Registers



The Available Clock Ports display is updated.

3. Select clock port clk. Specify 25 in the Frequency box and select MHz.

4. Click Add.

5. Select clock port jtag_tck. Specify 10 in the Frequency box and select MHz.

6. Click Add, and then click OK.

A FREQUENCY preference is defined for clock port clk and jtag_tck and added to the list on the Period/Frequency sheet. The 25 MHz objective for clk is a very conservative goal for the system based on the estimates reported by logic synthesis in Task 1 of the tutorial.

7. Choose File > Save to save the clock preferences to the .lpf.

Note

The target frequency is based on the estimate reported by logic synthesis.



Assign I/O Timing PreferencesIn this procedure, you will assign I/O timing preferences to the primary I/Os of the system, an SDR SDRAM memory interface. Figure 4 illustrates the input setup principle, which is the amount of time required for the system interface bus, sdr_DQ, to be stable before the arrival of the clock edge. If you do not define this for PAR, the system will assume that the time is the same as the period of the clock. This might be far too pessimistic or optimistic, depending on the PCB trace delays that are involved. In this example, setup time (tSU) is defined as 10 ns, which accounts for both the combinational logic and the external delays of the device.

The clock-to-output (tCO) delay is another recommended I/O timing preference. It defines the requirement for output signal validity after the arrival of the clock. If you do not specify this for PAR and static timing analysis, it will be completely unconstrained. In the tutorial design, sdr_DQ tCO is set to 10 ns, as shown in Figure 5.

To group I/Os:

1. In the Spreadsheet View, choose Preference > Grouping Assignment to open the dialog box.

2. In the Group Name text box, type sdr_DQ.

3. In the Filter box under Available Ports, type sdr_DQ* and click GO.

4. Under Available Ports, select sdr_DQ_0 through sdr_DQ_3. Use the Shift key to select all four.

5. Click the > button to add these ports to the Selected Ports list.

6. Click the Add button to create the new group and add it to the Available Group List.

Figure 4: I/O Timing – Input Setup Example



7. Click OK.

The sdr_DQ group is added to the Group Name list in the bottom left pane of the Spreadsheet View.

8. Choose File > Save to save the timing preferences to the .lpf.

Creating group preferences is a convenient way to refer to collections of objects. In the next procedure, you will define tSU and tCO constraints relative to these groups of ports.

Figure 5: I/O Timing – Clock-to-Output Example



To create I/O timing preferences:

1. Choose Preference > Input_Setup/Clock_To_Out to open the dialog box.


Type: INPUT_SETUPSecond Type: GroupTime: 10 (ns)

The Available Input Groups and Available Clock Ports/Nets lists are updated.

3. Select sdr_DQ from the Available Input Groups list.

4. Select clk from the Available Clock Ports/Nets list.

5. Click Add.

The Existing Preference List is updated.


Type: CLOCK_TO_OUTSecond Type: GroupTime: 10 (ns)

The Available Output Groups and Available Clock Ports/Nets lists are updated.

7. Select sdr_DQ from the Available Output Groups list.

8. Select clk from the Available Clock Ports/Nets list.

9. Click Add, and then click OK.



Key INPUT_SETUP and CLOCK_TO_OUT preferences are defined for the tutorial design and added to the list on the In/Out Clock sheet.

10. Choose File > Save to save the I/O timing preferences to the .lpf file.

Create Location PreferencesIn this procedure, you will establish a set of I/O locations to reflect the typical scenario of a predefined PC board layout.

To create location preferences:

1. In Spreadsheet View, choose the Port Attributes tab.

2. Click the Name column header to sort the pins alphabetically.

3. In the Name column, select sdr_A_0.

4. Press and hold the Shift key, scroll down, and click uart_stx to select a range of ports.

5. Right-click any selected cell and choose Assign Pins from the pop-up menu to open the dialog box.

The Assign Pins dialog box enables you to assign one or more ports to a selection of pin locations based on a set of filters. The right side of the dialog box controls the types of pins that are available based on the criteria you define, such as type, polarity, and order.

6. In the Pin Types panel, clear all pin types except I/O Only.

7. Under “And polarity is,” select Don’t Care.

8. Select Auto Sort. Select Pin Number and Ascending in the “Sort by” panel and clear any other selections.

9. On the bottom left, select every from the Assign Pin drop-down menu.



The list of pins is adjusted based on the filters, and the pins to be assigned appear in bold type. The result is a selection of pins beginning with pin 7 and ascending to pin 58.

10. Click Check Pins to confirm that basic assignment rules are met, and then click OK to close the Information dialog box.

11. Click Assign Pins.

Spreadsheet View is updated with the pin assignments and related banks.

12. Choose File > Save.

13. In the Design Planner Control window, choose File > Exit to close all windows of the Design Planner.

Create a Physical Design Database File (.ncd)The map program uses the .ngd and .lpf files to create the physical database file, a network of device-specific component. It writes any new or modified preferences to the physical preference file (.prf).

To create the .ncd physical database file:

1. Type the following command to run the design mapper on the .ngd and .lpf files:

map -a LatticeECP –p LFECP10E –t PQFP208 –s 4 -m ec_fpga_top.ngd -o map1\map1.ncd ec_fpga_top.lpf > map1\map1.log

The options for MAP are summarized below:

The design mapper converts the logical .ngd into a physical .ncd file and reports the results in the log file.

2. From the Text Editor, choose File Open and from the Files of type list, choose All Files.

3. Navigate to the map1\map1.log file, and click Open.

The Text Editor loads the log file. The Design Summary, shown below, provides a preview of the relative amount of device resources required to accommodate the tutorial design.

Design Summary: Number of registers: 1802 PFU registers: 1781 PIO registers: 21

-a <pmname> Map to the LatticeECP architecture.

-p <device> Map to the specified device.

-t Map to the specified package.

-s Use the speed indicated.

-m Create an .ncd even if the design exceeds the available resources.

-o Output to the specified .ncd file name and directory.



Number of SLICEs: 3391 out of 5120 (66%) SLICEs(logic/ROM): 3375 out of 3840 (88%) SLICEs(logic/ROM/RAM): 16 out of 1280 (1%) As RAM: 16 As Logic/ROM: 0 Number of logic LUT4s: 5035 Number of distributed RAM: 16 (32 LUT4s) Number of ripple logic: 218 (436 LUT4s) Number of shift registers: 0 Total number of LUT4s: 5503 Number of PIO sites used: 35 out of 147 (24%) Number of PIO FIXEDDELAY: 0 Number of 3-state buffers: 0 Number of DQSDLLs: 0 out of 2 (0%) Number of PLLs: 0 out of 4 (0%) Number of block RAMs: 12 out of 30 (40%) Number of GSRs: 1 out of 1 (100%) JTAG used : No Readback used : No Oscillator used : No Startup used : No

4. In the Text Editor, choose File > Close to close the log file, but keep the Text Editor open.

Estimate Baseline Performance TimingIn this procedure, you will run the static timing analysis tool, TRACE, on the pre-route version of the design. This will enable you to determine the worst-case path that will influence the design’s maximum operating frequency. In general, the results from pre-route static timing analysis will be approximately twice as fast as the routed version of the design.

To estimate baseline performance timing:

1. Type the following command to run the static timing analysis tool on the pre-route .ncd file.

trce -c -o map1\map1.twr map1\map1.ncd -v 10 ec_fpga_top.prf

The options for TRACE are summarized below:

For more details on command line options, see “Appendix B: Command Line Interface Quick Reference” on page 64.

Note

Each time you modify preferences in the .lpf file, you must remap the design to write the changes to the physical preference file (.prf).

-c Report unconstrained connections.

-v 10 Report the ten worst-case paths per preference in a full (verbose) timing report.

-o Specify the output file name.



TRACE runs, using the timing constraints you specified earlier, and reports the worst-case paths in the log file, map1.twr.

2. In the Text Editor, choose File > Open. In Files of Type, select All Files, navigate to the map1\map1.twr file, and click Open.

3. Use the Edit > Find command to locate each of the following major sections in the TRACE report:

Preference: FREQUENCY PORT “clk” 25.000000 MHz ;Notice that the ten critical paths are presented in terms of both Logical and Physical details. In general, the longest paths related to clk are between the or1200_operandmuxes module and or1200_top. This review will provide a first hint of how to potentially floorplan the design for performance.

Report SummaryThe Report Summary provides an estimate of the pre-route device timing. It shows the constraint defined by the preferences you set earlier in this task and the estimate generated by static timing analysis. Running TRACE immediately after the translation and design mapping phase gives you an early estimate of the design’s timing characteristics and of what your baseline performance preferences should be. Routing accounts for approximately 70% of the delay in a heavily interconnected design, such as the one in this tutorial. TRACE reports ~ 134 MHz here; therefore, a good first objective is .3(130 MHz) = 40 MHz.

Connections not covered by the preferencesThis section often illustrates unconstrained nets that can, when constrained, result in a more complete and accurate analysis by TRACE.

One design mapping optimization that might improve design performance is register retiming. In the next step, you will enable this map option to see if the design can be adjusted.

4. In the Text Editor, choose File > Close to close the TRACE report, but keep the Text Editor open.

5. At the command prompt, type the following command to run the design mapper with the retime option on the .ngd file:

map -a LatticeECP -p LFECP10E -t PQFP208 -s 4 -m ec_fpga_top.ngd -o map2\map2.ncd ec_fpga_top.lpf -retime > map2\map2.log

The design mapper converts the logical .ngd file into a physical .ncd file and reports the results in the log file.

Note

The results above are based on the EDIF file (ec_fpga_top.edf) supplied with the tutorial. Your results might differ.

Note

With no timing preferences defined, TRACE performs a default enumeration of the worst-case path.

Achieving Timing Closure in FPGA Designs Task 4: Establish Baseline Performance


6. In the Text Editor, choose File > Open. In Files of Type, select Project Log Files, and navigate to the map2\map2.log file.

7. Notice the message regarding retiming near the top.

The tutorial design might exhibit changes or it might not. If retiming occurs, a Retiming Report will summarize any adjustments made by the design mapper. You can confirm any performance benefit by running place and route and static timing analysis on the map2.ncd file. Notice that running TRACE on the pre-route design will not show any benefit from the netlist rebalancing performed by the retiming algorithm.

8. Choose File > Close and leave the Text Editor open.

Command-line SummaryThe following ispLEVER command lines were issued in this task to process the tutorial design. To quickly create a batch or script file for the tutorial, copy the command to a text file.

rem ********************************************************rem ispLEVER Tutorial: Timing Closure of FPGA Designsrem Task 3: Create Timing and Location Constraintsremrem ********************************************************

rem Create an .ncd Physical Database Filemap -a LatticeECP -p LFECP10E -t PQFP208 -s 4 -retime -m

ec_fpga_top.ngd -o map2\map2.ncd ec_fpga_top.lpf > map2\map2.log

rem Estimate Baseline Performance Timing trce -c -o map1\map1.twr map1\map1.ncd -v 10 ec_fpga_top.prftrce -c -o map2\map2.twr map2\map2.ncd -v 10 ec_fpga_top.prf

Task 4: Establish Baseline PerformanceIn this task, you will use the ispLEVER Place & Route program and the static timing analysis program TRACE to establish the tutorial design’s baseline performance. You will compare the timing preferences defined earlier against the implementation results and determine which objectives were met or not met.

PAR Session 1: Run Initial Place and RouteIn this session you will run place and route with a frequency goal of 25 MHz for the system clock, and then you will run static timing analysis with the current speed grade (-4).

To run place and route and static timing analysis:

1. At the command prompt, type the following command to run PAR:

par -d 1 -i 6 -l 5 -n 1 -w map1\map1.ncd map1\par1\par1.ncd ec_fpga_top.prf



The following list summarizes the options for this PAR session:

For more details on command-line options, see “Appendix B: Command Line Interface Quick Reference” on page 64.

PAR uses the location and timing constraints you specified earlier and creates a placed and routed version of the tutorial .ncd file at map1\par1\par1.ncd.

2. From the Text Editor, choose File > Open. In Files of Type, select All Files, navigate to the map1\par1\par1.par file, and click Open.

Near the beginning, the PAR report describes the physical resources, signals, and unique connections required to accommodate the tutorial design.

Device utilization summary:

GSR 1/1 100% used IOLOGIC 21/288 7% used PIO (prelim) 35/288 12% used 35/147 23% bonded EBR 12/30 40% used MULT36 1/5 20% used SLICE 3391/5120 66% used

Number of Signals: 7279Number of Connections: 23442

3. Page down further in the PAR report and notice the routing assignments made by PAR for high fan-out signals such as clocks (clk load), set/resets (sr load), and clock enables (ce load).

-d 1 Run one delay-based cleanup pass of the router.

-i 6 Run six iterations of the router.

-l 5 Run the placer at the maximum effort level.

-n 1 Run one iteration of the placer.

-w Overwrite any existing files.

Note

This process might take 10-20 minutes.

Note

The utilization summary above is based on the EDIF netlist supplied with the tutorial (ec_fpga_top.edf). Your results might differ.



The primary and secondary clocking resources of the LatticeECP/EC architecture are used in these cases to provide the best performance.

Global Clock Resources: CLK_PIN : 3 out of 4 (75%) PLL : 0 out of 4 (0%) DCS : 0 out of 8 (0%)

Quadrants All (TL, TR, BL, BR) - Global Clocks: PRIMARY "clk_c" from comp "clk" on CLK_PIN site "180 (PT25A)", clk load = 964 PRIMARY "jtag_tck_c" from comp "jtag_tck" on CLK_PIN site "81 (PB25A)", clk load = 125 SECONDARY "or1200_top/or1200_cpu/N_660_i" from F0 on comp "or1200_top/or1200_cpu/SLICE_1892" on site "R29C27C", clk load = 0, ce load = 37, sr load = 0 SECONDARY "or1200_top/or1200_cpu/wb_freeze_i" from F1 on comp "or1200_top/or1200_cpu/SLICE_1928" on site "R11C26B", clk load = 0, ce load = 97, sr load = 0 SECONDARY "or1200_top/or1200_cpu/or1200_if/un1_flushpipe_2_iZ0Z_0" from F0 on comp "or1200_top/or1200_cpu/or1200_if/SLICE_2326" on site "R29C25A", clk load = 0, ce load = 34, sr load = 0 SECONDARY "jtag_trst_c" from comp "jtag_trst" on CLK_PIN site "139 (PR18A)", clk load = 0, ce load = 0, sr load = 121

PRIMARY : 2 out of 4 (50%) DCS : 0 out of 2 (0%) SECONDARY: 4 out of 4 (100%)


5. At the command prompt, type the following command to run TRACE:

trce -v 1 -o map1\par1\par1.twr map1\par1\par1.ncd ec_fpga_top.prf

The static timing analysis program runs using the timing constraints that you specified earlier. It creates a timing report file at map1\par1\par1.twr.

6. From the Text Editor, choose File > Open. In Files of Type, select All Files and navigate to the map1\par1\par1.twr file.

Note

If you used Precision for logic synthesis in Task 1, internal signals such as clk will appear as clk_int instead of clk_c.



The Report Summary provides a report of the post-route device timing. Given the worst-case path between registers, the maximum operating frequency is ~28 MHz.

7. Locate the timing details for the FREQUENCY PORT clk preference and notice the following details about the report:

The Logical Details section describes the worst case critical path that is limiting the speed of signal clk. The originating synchronous component in the example below is the onchip_ram_top, and the destination synchronous component is at the or1200_except module of the or1200 RISC processor. This information illustrates a path that crosses hierarchical boundaries, which can guide possible floorplanning of the design for better performance.

The Physical Path Details section describes the same path in terms of LatticeECP/EC slice/PFU or IOLOGIC details. Each site along the path is described using the syntax:

R<row#>C<col#>.<pin_name> to R<row#>C<col#>.<pin_name>

The type of delay contributed by each element along the path appears in the list.

Careful examination of the PAR and TRACE reports will help you decide how to improve performance with the least amount of time and effort.

Did PAR apply primary and secondary routing resources to high-fan out signals? If not and the resources are available, then additional preferences such as USE PRIMARY might help.

How does routing relative to logic contribute to the total delay? If routing accounts for more than 80%, then the design could benefit from additional placement and routing iterations. It might also benefit from careful manual floorplanning.

Is a low number of critical paths limiting fMAX? If so, selective grouping of slices might lower routing distances.

Report Summary------------------------------------------------------------------------------------------Preference | Constraint| Actual|Levels---------------------------------------------------------------------------- | | |FREQUENCY PORT "clk" 25.000000 MHz ; | 25.000 MHz| 28.406 MHz| 9 | | |FREQUENCY PORT "jtag_tck" 10.000000 MHz | | |; | 10.000 MHz| 33.740 MHz| 8 | | |INPUT_SETUP GROUP "sdr_DQ" 10.000000 ns | | |CLKPORT "clk" ; | 10.000 ns| 3.313 ns| 1 | | |CLOCK_TO_OUT GROUP "sdr_DQ" 10.000000 | | |ns CLKPORT "clk" ; | 10.000 ns| 7.621 ns| 2 | | |----------------------------------------------------------------------------



================================================================================Preference: FREQUENCY PORT "clk" 25.000000 MHz ; 4096 items scored, 0 timing errors detected.--------------------------------------------------------------------------------

Passed: The following path meets requirements by 2.398ns

Logical Details: Cell type Pin type Cell/ASIC name (clock net +/-)

Source: FF Q onchip_ram_top/ack_weZ0 (from clk_c -) Destination: FF Data in or1200_top/or1200_cpu/or1200_except/eearZ0Z_29 (to clk_c +)

Delay: 17.462ns (16.0% logic, 84.0% route), 9 logic levels.

Constraint Details:

17.462ns physical path delay onchip_ram_top/SLICE_391 to or1200_top/or1200_cpu/or1200_except/SLICE_487 meets 20.000ns delay constraint less 0.000ns skew and 0.140ns DIN_SET requirement (totaling 19.860ns) by 2.398ns

Physical Path Details:

Name Fanout Delay (ns) Site ResourceREG_DEL --- 0.476 R23C35B.CLK to R23C35B.Q0 onchip_ram_top/SLICE_391 (from clk_c)ROUTE 6 1.137 R23C35B.Q0 to R18C35D.D0 onchip_ram_top_ack_weCTOF_DEL --- 0.289 R18C35D.D0 to R18C35D.F0 tc_top/t18_ch_upper/SLICE_1440ROUTE 11 1.820 R18C35D.F0 to R23C32B.A0 tc_top_t18_ch_upper_i4_out_0_0_0_1CTOF_DEL --- 0.289 R23C32B.A0 to R23C32B.F0 or1200_top/SLICE_1875ROUTE 15 1.028 R23C32B.F0 to R24C33B.C1 or1200_top/dcpu_ack_o_0xCTOF_DEL --- 0.289 R24C33B.C1 to R24C33B.F1 or1200_top/or1200_dc_top/SLICE_1866ROUTE 8 1.705 R24C33B.F1 to R14C34B.D1 or1200_top/dcpu_ack_dcCTOF_DEL --- 0.289 R14C34B.D1 to R14C34B.F1 or1200_top/or1200_cpu/or1200_freeze/SLICE_1940ROUTE 1 1.549 R14C34B.F1 to R11C26B.C1 or1200_top/or1200_cpu/or1200_freeze/wb_freezeZ0Z_1CTOF_DEL --- 0.289 R11C26B.C1 to R11C26B.F1 or1200_top/or1200_cpu/SLICE_1928ROUTE 139 5.149 R11C26B.F1 to R13C28A.D0 or1200_top/or1200_cpu/wb_freeze_iCTOF_DEL --- 0.289 R13C28A.D0 to R13C28A.F0 or1200_top/or1200_cpu/or1200_except/SLICE_1935ROUTE 2 1.631 R13C28A.F0 to R13C28A.C1 or1200_top/or1200_cpu/or1200_except/N_528CTOF_DEL --- 0.289 R13C28A.C1 to R13C28A.F1 or1200_top/or1200_cpu/or1200_except/SLICE_1935ROUTE 1 0.655 R13C28A.F1 to R15C28A.C1 or1200_top/or1200_cpu/or1200_except/N_640_0Z0Z_1CTOF_DEL --- 0.289 R15C28A.C1 to R15C28A.F1 or1200_top/or1200_cpu/or1200_except/SLICE_487ROUTE 1 0.000 R15C28A.F1 to R15C28A.DI1 or1200_top/or1200_cpu/or1200_except/N_803_iZ0 (to clk_c) -------- 17.462 (16.0% logic, 84.0% route), 9 logic levels.




Figure 6 illustrates the data files produced and processes that you have executed so far to implement the tutorial design.

In the next task, you will perform the timing closure steps shown in Figure 2 on page 4 in an attempt to maximize the performance of the clock port clk.

Command-line SummaryThe following ispLEVER commands were issued in this task to process the tutorial design. To quickly create a batch or script file for the tutorial, copy the commands to a text file.

rem ***********************************************************rem ispLEVER Tutorial: Timing Closure of FPGA Designsrem Task 4: Establish Baseline Performanceremrem ***********************************************************

rem PAR Session 1: Initial PAR par -d 1 -i 6 -l 5 -n 1 -w map1\map1.ncd map1\par1\par1.ncd

ec_fpga_top.prftrce -v 1 -o map1\par1\par1.twr map1\par1\par1.ncd

ec_fpga_top.prf

Note

Your results might vary.



Figure 6: Tutorial Process Flow – Establishing Baseline Performance

Achieving Timing Closure in FPGA Designs Task 5: Increase Place and Route Effort


Task 5: Increase Place and Route EffortIn this task, you will use the command-line interface to run the place and route program PAR and the static timing analysis program TRACE. In a series of three PAR sessions, you will attempt to maximize fMAX of input clock clk using the following key techniques:

Increase the frequency preference

Apply five different “seeds” with the PAR placer algorithm

Run ten routing passes on the best result of the multi-placement technique

It is important to remember that although the tutorial design might not exhibit significant benefit from these procedures, your own design can significantly improve. These PAR techniques are the ones most commonly used by Lattice Applications Engineering, and they will result in the most benefit in the least amount of time.

PAR Session 2: Run PAR with a Higher fMAX ObjectiveIn this PAR session, you will rerun place and route with a higher FREQUENCY preference objective (40 MHz), which will push PAR harder to exceed the baseline found at 25 MHz. The processes are shown in Figure 7. An increase of 20% or more in target frequency is often an effective technique to use with PAR. The session will also demonstrate how to examine the performance of the fastest (-5) speed grade LFECP10 device.

Figure 7: Tutorial Process Flow – Changing Target Clock Frequency and Speed Grade



To run place and route with the higher fMAX:

1. From the Text Editor, choose File > Open. In Files of Type, select Logic Preference File (.lpf).

2. Navigate to the timing_closure_tutor folder, select ec_fpga_top.lpf, and click Open.

3. Modify the FREQUENCY PORT “clk” preference. Add the PAR_ADJ parameter as shown below.

FREQUENCY PORT “CLK” 25.000000 MHz PAR_ADJ 15;

The preference keyword PAR_ADJ uses the simple addition formula of 25 + 15 = 40 MHz to direct PAR while running TRACE on the 25MHz objective.

4. Choose File > Save, and then choose File > Close and keep the Text Editor open.

5. At the command prompt, type the following command to remap the design and apply the modified preference to the physical preference file:

map -a LatticeECP -p LFECP10E -t PQFP208 -s 4 -m ec_fpga_top.ngd -o map1\map1.ncd ec_fpga_top.lpf

6. Type the following command to run PAR:

par -d 1 -i 6 -l 5 -n 1 -w map1\map1.ncd map1\par2\par2.ncd ec_fpga_top.prf

The place and route program reruns the design with the higher target frequency and creates a placed and routed version of the tutorial .ncd at map1\par2\par2.ncd.

To run static timing analysis:

1. Type the following command to run TRACE:


The static timing analysis program runs, using the timing constraints you specified earlier, and creates a report file at map1\par2\par2.twr.

2. Type the following command to run TRACE using the -sp (speed grade) option:

trce -sp 5 -v 1 -o map1\par2\par2sp5.twr map1\par2\par2.ncd ec_fpga_top.prf

The static timing analysis program runs, using the fastest (-5) speed grade, and creates a report file at map1\par2\par2sp5.twr.

3. From the Text Editor, choose File > Open. In Files of Type, select All Files and navigate to the map1\par2 directory.

4. Select the par2.twr and par2sp5.twr files and click Open.

Note




5. Choose Window > Tile Horizontal to compare the reports, and examine the Report Summary section in each of these timing report files.

The table below illustrates the differences between the last three static timing analysis results and the relative gains in fMAX.

6. Choose File > Close All and keep the Text Editor open.

PAR Session 3: Run PAR in Multi-Placement ModeThe majority of delays associated with timing paths in FPGA designs are routing delays. In this PAR session, you will apply a technique for discovering a placement of components in the device floorplan that minimizes routing distances. The multi-placement mode of PAR allows up to 99 unique placement seeds that can be attempted for each run.

In this session, you will rerun place and route in the multi-placement mode. You will then run static timing analysis with the five resultant .ncd output files, as shown in Figure 8.

F (max) % Gain

par1.twr clk=28.41 MHz Baseline

par2.twr clk=40.24 MHz 41.64%

par2sp5.twr clk=48.08 MHz 69.24%

Note

Your results might vary from the tutorial. As an exercise, calculate the % gain, if any, between PAR sessions.

Figure 8: Tutorial Process Flow – Multiple Placement Cost Tables



To run PAR and TRACE:

1. Type the following command to run PAR:

par -d 1 -i 3 -l 5 -n 5 -s 5 -t 2 -w map1\map1.ncd map1\par3\par3.dir ec_fpga_top.prf

The options for this PAR session are summarized below:

The additional controls produce five routed .ncd file results instead of the default of one. Each result uses a different cost table entry—seed 2 through 6—to guide the placer algorithm. Since the earlier runs of PAR used seed 1, the -t switch sets the new starting point at seed 2.

The place and route program runs, using the location and timing constraints you specified earlier. It creates a placed and routed version of the tutorial .ncd at map1\par3\par3.dir. Notice the naming convention of the result files created by PAR: 5_2.ncd through 5_6.ncd. The number 5 indicates placer effort, and 2 through 6 indicates each seed.

Since routing delay is the major cause of delay in most FPGA designs, a useful PAR strategy is to discover a placement of components in the device floorplan that minimizes routing distances. The multi-placement mode of PAR allows up to 99 unique placement seeds that can be attempted for each run. Notice the low number of routing passes (-i 3) applied for this run. This speeds the overall PAR runtime in this mode.

Note

This session will take about one hour to complete using a 1.6 GHz PC with 500 MB of RAM.

If the session is taking longer than you wish, use Ctrl+C in the command window. After a moment, PAR will be interrupted and you will be given the following options:

Continue processing and ignore the interrupt.

Exit the program normally at the next check point. This will result in saving the best results so far after concluding the current processing.

Exit the program immediately.

During a long multi-placement run, you might want to use the second option, which will allow you to save the immediate results for analysis.

-d 1 Run one delay-based cleanup pass of the router.

-i 3 Run three iterations of the router.

-l 5 Run the placer at the maximum effort level.

-n 5 Run five iterations of the placer.

-s 5 Save five of the results from the run.

-t 2 Start at the second placer cost table entry.




2. Type the following command to run TRACE on each result of the multi-placement mode:

for /L %n in (2,1,6) do trce -v 1 -o map1\par3\par3.dir\5_%n.twr map1\par3\par3.dir\5_%n.ncd ec_fpga_top.prf

The static timing analysis program runs, using the timing constraints you specified earlier, and creates report files 5_2.twr through 5_6.twr at map1\par3\par3.dir.

3. Type the following command to quickly examine the TRACE report files and determine the best result:

find “MHz” map1\par3\par3.dir\*.twr

Though multi-placement can be the most time-consuming mode for running PAR, it can also be the most effective in improving results. The following table illustrates the differences between the baseline performance and PAR Session 3.

The results are less than those from PAR Session 2 (40.24 MHz). The tutorial uses the best result of the multi-placement session to demonstrate the design flow. In practice, you might need to run more multi-placement seeds to find a better result. As an exercise, try additional PAR runs using the placer table flag (-t) to specify a starting point.

Note

For more information on the for command, see Windows online Help. Equivalent commands are available from the UNIX command line. Use the %% variable to carry out the for command within a batch program.

Note

For more information on the find command, see Windows online Help. Equivalent commands are available from the UNIX command line.

F (max) % Gain


5_2.twr clk=37.21 MHz +30.98%

5_3.twr clk=38.38 MHz +35.09%

5_4.twr clk=36.59 MHz +28.79%

5_5.twr clk=38.41 MHz +35.20%

5_6.twr clk=37.06 MHz +30.45%

Note

Your results might differ from those shown in the tutorial. As an exercise, calculate the % gain, if any, between placements.



PAR Session 4: Run PAR with Multiple Routing PassesIn this PAR session, you will rerun the place and route program using the route-only mode. You will run ten routing iterations and five clean-up passes in an effort to improve the timing results using the -4 speed grade, as shown in Figure 9

The route-only (or reentrant) mode of PAR is also often useful on tough routing jobs where a lower number of passes has not resulted in a fully routed design. .

To run PAR and TRACE:

1. Type the following command to run place and route:

par -c 0 -d 5 -i 10 -k -p -w map1\par3\par3.dir\5_5.ncd map1\par4\par4.ncd ec_fpga_top.prf

The options for this PAR session are summarized below:

The place and route process runs, using the location and timing constraints you specified earlier, and creates a placed and routed version of the tutorial .ncd at map1\par4\par4.ncd.

Figure 9: Tutorial Process Flow – PAR with Multiple Routing Passes

-c 0 Run no cost-based cleanup passes of the router.

-d 5 Run five delay-based cleanup passes of the router.

-i 10 Run ten iterations of the router

-k -p Skip placement and begin routing, leaving the existing routing in place.


Note




2. Type the following commands to run TRACE:


The static timing analysis program runs, using the timing constraints you specified earlier, and creates a report file at map1\par4\par4.twr.

3. From the Text Editor, choose File > Open and navigate to the map1\par4\par4.twr file.

4. Examine the Report Summary section. The following table illustrates the differences between the baseline performance and PAR Session 4.

5. Choose File > Close and keep the Text Editor open.

Command-line SummaryThe following ispLEVER command lines were issued in this task to process the tutorial design. To quickly create a batch or script file for the tutorial, copy the commands to a text file.

rem **********************************************************rem ispLEVER Tutorial: Timing Closure of FPGA Designsrem Task 5: Increasing Place and Route Effortrem ***********************************************************

rem PAR Session 2: Running PAR with a higher Fmaxobjective

par -d 1 -i 6 -l 5 -n 1 -w map1\map1.ncd map1\par2\par2.ncdec_fpga_top.prf

trce -v 1 -o map1\par2\par2.twr map1\par2\par2.ncdec_fpga_top.prf

trce -sp 5 -v 1 -o map1\par2\par2sp5.twrmap1\par2\par2.ncd ec_fpga_top.prf

rem PAR Session 3: Running PAR in multi-placement modepar -d 1 -i 3 -l 5 -n 5 -s 5 -t 2 -w map1\map1.ncd

map1\par3\par3.dir ec_fpga_top.prffor /L %%n in (2,1,6) do trce -v 1 -o

map1\par3\par3.dir\5_%%n.twr map1\par3\par3.dir\5_%%n.ncdec_fpga_top.prf

rem PAR Session 4: Running PAR with multiple routingpasses

par -c 0 -d 5 -i 10 -k -p -w map1\par3\par3.dir\5_5.ncd map1\par4\par4.ncd ec_fpga_top.prf trce -v 1 -o map1\par4\par4.twr map1\par4\par4.ncd

ec_fpga_top.prf

F (max) % Gain


par4.twr clk=38.41 MHz +35.20%

Note

Your results might differ from those shown in the tutorial. As an exercise, calculate the% gain, if any, between PAR sessions.

Achieving Timing Closure in FPGA Designs Task 6: Floorplan for Performance


Task 6: Floorplan for Performance

The performance of some FPGA designs can be improved with careful floorplanning. Floorplanning constraints override the automatic placement algorithms of the place and route program, PAR. The ispLEVER software provides flexible floorplanning constraints that support the placement of individual logic blocks to device sites as well as the grouping of logic blocks to minimize interconnect length.

The ispLEVER universal group (UGROUP) preference enables you to group Slices and embedded blocks, such as sysMEM EBR and sysDSP, and constrain them to a particular area of the device floorplan. Slice-only groups can optionally float and be anchored by PAR automatically. Group constraints can be captured in two ways in the ispLEVER software: as preferences or HDL attributes.

In this task you will use the Design Planner user interface to organize major modules of the tutorial design and produce a UGROUP of Slices.

Lattice recommends the following technique for developing floorplan constraints:

1. Examine the TRACE report and look for obvious candidates for improving timing.

If a few (< 5) critical paths are limiting fMAX, consider grouping the blocks along those paths using the universal group (UGROUP) preference. An important consideration is whether the critical path is really a true path or just a physical connection. If you can determine that the path is false, you can then safely treat it as a timing exception and block it using a preference. Use the Path Tracer utility of the Design Planner to quickly create a UGROUP of blocks along a critical path.

2. If it appears that the design can benefit from design planning on a module-by-module basis, create dedicated areas of the device floorplan.

Consider using REGION and UGROUP preferences, as well as bounding box (BBox), with the Design Planner. Rerun PAR and TRACE to confirm performance gains.

3. If the design benefits from floorplanning, create an HDL-based grouping attribute.

Rewrite UGROUP preferences in terms of a UGROUP or HGROUP HDL attribute so that future HDL revisions will infer UGROUP preferences automatically.

4. Consider higher level floorplanning through logic synthesis.

Note

Before you attempt any floorplanning strategy, you should be aware of the stability of preferences that you define. The logic blocks that a preference refers to could easily be made obsolete from revision to revision of the design’s HDL source code or EDIF netlist.



The stability of group preferences are illustrated in the following table.

Prepare to Run the Design Planner from the Command LineThe Design Planner requires a common prefix name between the .ngd, .ncd, and .prf files. Use the following commands to create this organization from the files generated earlier in the tutorial:

md fpcd fpcopy ..\ec_fpga_top.ngdcopy ..\ec_fpga_top.lpfcopy ..\ec_fpga_top.prfcopy ..\map1\par4\par4.ncd ec_fpga_top.ncd

Examine Critical Paths

To report more critical paths:

1. Type the following command to produce a timing report of the 25 most critical paths.

trce -v 25 -o ec_fpga_top.twr ec_fpga_top.ncd ec_fpga_top.prf

2. Type the following command to quickly examine the TRACE report and determine the best results:

find “Delay:” *.twr

3. Scroll through the list of Delay values from the TRACE report.

The first 25 are related to the clock port clk, and all are greater than 12 ns. This indicates that there is no clear candidate for floorplanning.

4. From the Text Editor, choose File Open. In Files of Type, select All Files, navigate to the fp directory, and select ec_fpga_top.twr.

5. Scroll through the critical paths reported.

In this implementation from Synplify, a couple of different submodules emerge along the critical path: or1200_freeze and or1200_except.

Stability Type Source

Most Universal Group (UGROUP) or Hierarchical group (HGROUP) HDL attributes

HDL

More Universal Groups (UGROUPs) written in terms of pre-map (logical) blocks

.lpf

Least Placement Groups (PGROUPs) written in terms of post-map (physical) elements

.prf



A portion of the ec_fpga_top.twr is reproduced below.

Notice the ROUTE names that begin with or1200_top/or1200_cpu. This is a branch of the Verilog module for the OpenRISC 1200 portion of the tutorial design, as shown in Figure 1 on page 3.

6. Choose File > Close and keep the Text Editor open.

Use Design Planning and GroupingDesign planning and grouping can in some cases improve a design’s performance. This strategy tends to work best with datapath style logic.

In this next procedure, you will attempt to guide placement based on a portion of the or1200_top module. You will examine branches of the design hierarchy and critical paths using the Design Planner GUI.

REG_DEL --- 0.476 R14C10C.CLK to R14C10C.Q0 onchip_ram_top/SLICE_391 (from clk_c)ROUTE 6 0.626 R14C10C.Q0 to R14C9B.D0 onchip_ram_top_ack_weCTOF_DEL --- 0.289 R14C9B.D0 to R14C9B.F0 SLICE_1726ROUTE 2 0.366 R14C9B.F0 to R14C9B.C1 tc_top_xi4_wb_ack_oCTOF_DEL --- 0.289 R14C9B.C1 to R14C9B.F1 SLICE_1726ROUTE 2 0.366 R14C9B.F1 to R14C9C.C0 or1200_top/or1200_dc_top/dcpu_ack_o_0_LZ0Z5CTOF_DEL --- 0.289 R14C9C.C0 to R14C9C.F0 or1200_top/SLICE_1876ROUTE 15 1.394 R14C9C.F0 to R11C10D.C1 or1200_top/dcpu_ack_o_0x_0CTOF_DEL --- 0.289 R11C10D.C1 to R11C10D.F1 or1200_top/or1200_dc_top/SLICE_1866ROUTE 8 0.602 R11C10D.F1 to R11C11A.D1 or1200_top/dcpu_ack_dcCTOF_DEL --- 0.289 R11C11A.D1 to R11C11A.F1 or1200_top/or1200_cpu/or1200_except/SLICE_2502ROUTE 1 1.064 R11C11A.F1 to R14C12B.C1 or1200_top/or1200_cpu/or1200_except/ex_pc_6_i_o2_2_m5_i_m7_i_LZ0Z6CTOF_DEL --- 0.289 R14C12B.C1 to R14C12B.F1 or1200_top/or1200_cpu/or1200_except/SLICE_1887ROUTE 23 1.657 R14C12B.F1 to R11C17B.C1 or1200_top/or1200_cpu/or1200_except/ex_pc_6_i_o2_2_m5_i_m7_i_LZ0Z10CTOF_DEL --- 0.289 R11C17B.C1 to R11C17B.F1 or1200_top/or1200_cpu/or1200_except/SLICE_548ROUTE 8 1.116 R11C17B.F1 to R11C22B.C0 or1200_top/or1200_cpu/flushpipe_r21CTOF_DEL --- 0.289 R11C22B.C0 to R11C22B.F0 or1200_top/or1200_cpu/SLICE_1892ROUTE 37 2.847 R11C22B.F0 to R12C13D.CE or1200_top/or1200_cpu/N_660_i (to clk_c)



To examine a critical path:

1. Type the following command to run the ispLEVER post-map Design Planner:

flmainappw -inp ec_fpga_top.ncd -dir <install>\examples\Tutorial\timing_closure_tutor\fp -prj ec_fpga_top -lpf ec_fpga_top.lpf -prf ec_fpga_top.prf

The Design Planner application opens, showing the Design Planner Control. It then opens the Floorplan View, Post-Mapped View, Package View, and Pre-Mapped View.

2. Maximize the Floorplan View window. Press F6 to fit the floorplan layout in the maximized window, and then choose Tools > Path Tracer.

3. From the Path Tracer, click Query.

4. In the Timing Query dialog box, select FREQUENCY PORT “clk” 25.000000 MHz0 PAR_ADJ 15 from the list and click Query.

Tip

The post-map Design Planner command line requires a drive specification with the -dir switch. For example,

-dir c:\ispTOOLS7_2\examples\tutorial\timing_closure_tutor\fp

As an alternative, you can simply type the flmainappw command and open the ec_fpga_top.ncd file directly from the Design Planner GUI. You can then open the different views from the View menu.

Note

If any of the views do not open, choose View > Default Window Layout from the Design Planner Control window.



The Path Tracer displays a list of combinatorial paths.

5. Click the highlight color of the topmost row to open the Change Display Color dialog box.

6. Select the white color, which will provide the best contrast, and click OK.



7. With the top row still selected, click Locate, and then select the Floorplan View window if it is not visible.

The related slices along the path are highlighted with the color white in the Floorplan View, and a flywire indicates the logical connection.

8. From Floorplan View, click the Physical View button on the toolbar.

The Physical View opens and displays the related path and the physical routes related to the connection.

Note

The appearance of the Floorplan View layout on your computer might vary from the one shown.



To better view the delay path, choose Layer, and turn off Routes.

9. Return to Floorplan View and click the Show Delay Path button to turn off the path display.

Note

If the related path is not displayed, choose Layer and make sure that Delay Path is selected. The appearance of the Physical View on your computer might vary from the one shown.



To examine the design hierarchy:

1. From the Design Planner Control choose View > Pre-Mapped View.

2. Expand the Instances folder, and then scroll down to the or1200_top (LOGIC: 2293) folder and expand it.

Notice the LOGIC: 2293 summary of the Slice utilization within this branch.

3. Expand the Instances folder within the or1200_top (LOGIC: 2293) folder.

4. Locate the or1200_cpu (LOGIC: 2018) folder and expand the Instances folder within it.

5. Locate the or1200_except (LOGIC: 308) folder, right-click it, and choose Locate > Floorplan View.



The selected elements are highlighted in Floorplan View.

6. From Floorplan View choose View > Highlight > Light Blue.

The physical elements related to the or1200_except submodule of the design are highlighted in light blue.

7. In Pre-Mapped View, right-click the or1200_freeze (LOGIC: 29) folder and choose Locate > Floorplan View.

The selected elements are highlighted in Floorplan View.

8. From Floorplan View choose View > Highlight > Green.

The physical elements related to the or1200_freeze submodule of the design are highlighted in green.

With this view, you can now determine where your original logic has been placed, the relative amount of logic required for each branch, and where the critical path is relative to these resources.

In the next procedure, you will create a user group (UGROUP) in an effort to improve the performance of the modules. The best organization comes from careful analysis of your HDL design. Many times the RTL views provided by your synthesis tool will help you visualize datapaths and modules that are good candidates for grouping.

To create a logical group (UGROUP):

1. In Pre-Mapped View, use Ctrl+Click to select the following modules from the or1200_cpu (LOGIC: 2018) folder:

or1200_except (LOGIC: 308)



or1200_freeze (LOGIC: 29)

2. Right-click the selected modules and choose new UGROUP.

3. In the New UGROUP dialog box select the BBox option.

4. Specify 10 for Height and 15 for Width and click OK.

Expand Ugroup in the Group/Module/Region pane of Pre-Mapped View and you will see the newly created Group_0.

Group_0 instructs PAR to place all the logic from the selected modules into a 15 x 10 area (150 PFUs = 600 slices) of the floorplan. One or more logic components will be assigned to a slice by the design mapper (map). The 600 slices should be more than enough to accommodate the 337 logic components reported in the Pre-Mapped Logical Netlist View (or1200_except (LOGIC: 308) + or1200_freeze (LOGIC: 29).

Lattice recommends that you add a bounding box (BBox) attribute with 20% additional area beyond the minimum required to accommodate the logic. You can estimate the necessary resources based on the notation (LOGIC: n) next to each module. With no BBox attribute defined, PAR will pack the group into the smallest square boundary (height=width) required to accommodate the physical resources. If embedded memory is part of a group, you should ensure that the group is anchored and overlaps enough sysMEM blocks of the device.

Tip

A quick way to help you visualize and calculate the size and height/width ratio for a UGROUP BBox is to create a group in Floorplan View.

Right-click an empty space in Floorplan View and choose Create GROUP from the pop-up menu. Draw the area by dragging your mouse.

Right-click the new group and choose Group Property. Review the GROUP definition. The size and number of resources encompassed by the BBox are reported. Edit the BBox as needed, and optionally, the anchor point. The new group is listed in Pre-Mapped View.

In Pre-Mapped View, drag the modules from the design tree over to the new UGROUP.



5. In Pre-Mapped View, choose File > Save.

The Design Planner saves the new UGROUP to the logical preference file, ec_fpga_top.lpf. It will be written to the physical preference file with the next mapping process.

When you issue the Save command in any Design Planner window, the Save icon is dimmed in all windows, including the Design Planner Control.

6. From the Design Planner Control, choose File > Exit.

The Design Planner Control closes all windows.

To rerun map and place & route with the new module group:

1. At the command prompt, type the following command to run the design mapper on the .ngd file:

map -a LatticeECP –p LFECP10E –t PQFP208 –s 4 –m ec_fpga_top.ngd -o fp.ncd ec_fpga_top.lpf

The mapper runs the design and applies the changes.

2. From the Text Editor choose File > Open. In the Files of type box, choose Preference Language File (.prf), select the ec_fpga_top.prf file, and click Open.

3. Notice the group_0 placement group (PGROUP) produced by the design mapper based on the UGROUP you created:

PGROUP "Group_0" BBOX 10 15 DEVSIZECOMP "or1200_top/or1200_cpu/SLICE_405"COMP "or1200_top/or1200_cpu/or1200_except/SLICE_473"COMP "or1200_top/or1200_cpu/or1200_except/SLICE_474"

:

4. Close the Text Editor.

5. At the command prompt, type the following command to run PAR:

par -d 1 -i 6 -l 5 -n 1 -w fp.ncd ec_fpga_top.ncd ec_fpga_top.prf

The place and route program runs, using the location and timing constraints you specified earlier, and creates a placed and routed version of the tutorial .ncd at ec_fpga_top.ncd.

6. Type the following command to produce a timing report of the 25 most critical paths:

trce -v 25 -o ec_fpga_top_fp.twr ec_fpga_top.ncd ec_fpga_top.prf

7. Type the following command to quickly examine the TRACE report and determine the best result:

find “Delay:” *.twr

8. Scroll through the list of Delay values from the TRACE report.

Note




A new set of critical paths appears, and many paths are over 12 ns. In this case, the grouping constraint has not improved performance. Additional grouping and anchoring might be required in this design to arrive at a floorplan that is superior to what the automatic placement tools can produce.

If you wish, experiment with the Group_0 BBox boundary. Set an anchor point for Group_0 to see how position and shape might influence timing. You can use the Design Planner to make changes to the UGROUP (logical). When you adjust a UGROUP, remember to rerun MAP and PAR.

If you are able to discover a better result, you might wish to migrate the UGROUP into your HDL source code. A portion of the Verilog source for the or1200_cpu module (or1200_cpu.v) is shown below for your reference. Precision RTL syntax appears as a comment following the module instance (//pragma); Synplify syntax appears before the ending semicolon of the module instance (/* synthesis…*).

//// Instantiation of load/store unit//or1200_lsu or1200_lsu(

.addrbase(operand_a),

.addrofs(lsu_addrofs),

.lsu_op(lsu_op),

.lsu_datain(operand_b),

.lsu_dataout(lsu_dataout),

.lsu_stall(lsu_stall),

.lsu_unstall(lsu_unstall), .du_stall(du_stall),

.except_align(except_align),

.except_dtlbmiss(except_dtlbmiss),

.except_dmmufault(except_dmmufault),

.except_dbuserr(except_dbuserr),

.dcpu_adr_o(dcpu_adr_o),

.dcpu_cycstb_o(dcpu_cycstb_o),

.dcpu_we_o(dcpu_we_o),

.dcpu_sel_o(dcpu_sel_o),

.dcpu_tag_o(dcpu_tag_o),

.dcpu_dat_o(dcpu_dat_o),

.dcpu_dat_i(dcpu_dat_i),

.dcpu_ack_i(dcpu_ack_i),

.dcpu_rty_i(dcpu_rty_i),

.dcpu_err_i(dcpu_err_i),

.dcpu_tag_i(dcpu_tag_i)) /* synthesis UGroup=”group_0” PBBox=”10,15” */;

//pragma attribute or1200_lsu UGroup group_0//pragma attribute or1200_lsu PBBox 10,15

Achieving Timing Closure in FPGA Designs Summary


Summary In this tutorial you have learned how to do the following:

Use the methodology recommended by Lattice to meet the timing objectives of your design specification.

Review and understand device implementations reports such as logic utilization and static timing results.

Use a combination of preferences and controls to influence map, place, and route to increase performance.

Examine critical paths and logic utilization from the perspective of the device floorplan.

Use scripts or batch files to process your design from the command line.

Use the ispLEVER Design Planner graphical point tool to define timing constraints and floorplan the design.

Use both logical and physical preferences to establish timing and location constraints.

Achieving Timing Closure in FPGA Designs Recommended References


Recommended ReferencesYou will find more useful information about timing closure in the ispLEVER online Help and the Lattice FPGA Design Guide. The following recommended references are available with the ispLEVER software.

Lattice FPGA Design Guide

HDL Synthesis Coding Guidelines

Strategies for Timing Closure

ispLEVER Online Help > FPGA and Crossover Design

Running FPGA Tools from the Command Line

Design Implementation

Design Verification

LatticeECP/EC sysIO Usage Guide.

LatticeSC PURESPEED I/O Usage Guide.

Design Planner Online Help

Achieving Timing Closure in FPGA Designs Glossary


Glossary

The following are the terms and concepts that you should understand in order to use this tutorial most effectively.

bank An organization of LatticeECP/EC/XP sysIO buffers arranged around the periphery of the device.

clock tree or clock spine A clock distribution network in the LatticeECP/EC/XP device architecture used for high-fanout signals.

clock-to-out delay (tCO) The time it takes for a signal to go from the primary input pin through the clock of the flip-flops or gate of latches to the primary output pin.

critical path The path through a circuit that determines the maximum speed at which the circuit can operate. The critical path is a signal in a section of combinatorial logic that limits the speed of the logic. Storage elements begin and end a critical path, which might include I/O pads.

edif2ngd The ispLEVER translation program that converts EDIF 2 0 0 format netlists into the Lattice NGO object format.

embedded block RAM (EBR) A large dedicated fast memory block in the LatticeECP/EC/XP device architecture.

fMAX The symbol for the maximum operating frequency expressed in hertz units. Clock frequency is the number of complete clock cycles that occur per unit of time. For example, a 50- MHz clock signal will cycle 50 million times per second. The maximum frequency is a function of the longest combinatorial delay between registered elements.

HGROUP An attribute in the HDL code that groups blocks of logic or library components in an FPGA design. HGROUP attributes are expanded into UGROUPs in the Build Database process and can be viewed in the Design Planner. When modified by the Design Planner, they are written to the logical preference file. These groups are then translated by the design mapper into PGROUP preferences in terms of physical components (slices) for the placement and routing process. The design mapper includes a hierarchy identifier for each PGROUP preference, which enforces strict hierarchical control.

IOLOGIC A representation of PIO logic within a PIC block that is generated by the ispLEVER Design Planner tool.

Look-up table (LUT) An architectural element within a LatticeECP/EC/XP PFU for implementing combinational logic or memory.

Manhattan routing A representation of the physical routing interconnect among slices in which the routing is rectilinear.

MAP The ispLEVER design mapper program.



.ncd file A binary format FPGA post-map physical design database file generated by the map program or Map Design process of the Project Navigator. The .ncd file includes mapped and potentially placement and route information.

.ngd file A binary format FPGA pre-map logical design database file generated by the EDIF2.NGD/NGDBuild programs or by the Build Database process in the Project Navigator. The .ngd file represents the logical design information of the ASCII EDIF netlist.

ngdbuild The ispLEVER program that merges NGO object files into a single logical database (.ngd).

PAR The ispLEVER place and route program.

PGROUP (Placement Group) A preference that establishes a partition of physical components. It is typically used to guide the placer algorithm of PAR.

programmable function without RAM/ROM (PFF) A PFU that does not contain a RAM.

programmable function unit (PFU) A block within the LatticeECP/EC device that implements combinational logic, memory, and registers. The core of the architecture consists of PFU blocks that can be programmed to perform logic, arithmetic, distributed RAM, and distributed ROM functions.

programmable I/O cell (PIO) An architectural element within an FPGA that handles actual input-output functions. It includes the I/O registers, tristate registers, and control multiplexers of a PIC.

programmable interface cell (PIC) A cell within a LatticeECP/EC device architecture that contains a group of two PIOs. The edges of the architecture consist of PIC blocks that contain two PIOs connected to sysIO buffers. PICs provide high-speed I/O registers and buffering to support a variety of signal interface standards.

slice An architectural element within an FPGA consisting of two LUT4 lookup tables that feed two registers (programmed to be in FF or latch mode), and some associated logic that allows the LUTs to be combined to perform functions such as LUT5, LUT6, LUT7 and LUT8. It also includes control logic to perform set/reset functions (programmable as synchronous/asynchronous), clock select, chip-select and wider RAM/ROM functions. The registers in the slice can be configured for positive or negative and edge or level clocks. There are four interconnected slices per PFU block. Each slice in a PFU is capable of four modes of operation: logic, ripple, RAM, and ROM. Each slice in the PFF is capable of all modes except RAM.

TRACE (trce) The ispLEVER static timing analysis program.

UGROUP (Universal Group) An attribute that establishes a partition of logical components within the EDIF netlist. The UGROUP attribute is used for grouping blocks of logic or library components in different hierarchies or no hierarchy. UGROUPs are converted into PGROUPs (placement groups) by



the mapping process (map) and are typically used to guide the placer algorithm of PAR.

Achieving Timing Closure in FPGA Designs Appendix A: Logic Synthesis with Precision RTL


Appendix A: Logic Synthesis with Precision RTL

Use the following procedure to synthesize the design using Precision RTL synthesis.

To create a Precision RTL project:

1. From the Start menu, choose Programs > Lattice Semiconductor > Accessories > Precision RTL Synthesis.

The Mentor Graphics Precision Transcript Window opens.

2. Choose File > New Project.

3. In the New Project dialog box, specify ec_fpga_top as the project name.

4. Click the browse button and navigate to <drive>\<install_dir>\examples\Tutorial\timing_closure_tutor\PrecisionRTL.

5. Click OK to close the browse dialog box, and click OK again to save the new project.

To add the input files to the project:

1. Make sure that the Design Center tab is selected, and then add the dbg_interface files as follows:

Right-click the Input Files folder and choose Add Input Files.

Navigate to the ..\timing_closure_tutor\dbg_interface folder.

Select all Verilog HDL (.v) files except dbg_defines.v and timescale.v.



Click Open.

2. Add the onchip_ram files:

Right-click the Input File folder and choose Add Input Files.

Navigate to the ..\timing_closure_tutor\onchip_ram folder.

Select the two Verilog HDL (.v) files.

Click Open.

3. Add the or1200 files:


Navigate to the ..\timing_closure_tutor\or1200 folder.

Select all Verilog HDL (.v) files except or1200_defines.v.

Click Open.

4. Add the sdr_sdram files:


Navigate to the ..\timing_closure_tutor\sdr_sdram folder.

Select all Verilog HDL (.v) files except sdr_par.v.

Click Open.

5. Add the uart16550 files:


Navigate to the ..\timing_closure_tutor\uart16550 folder.

Select all Verilog HDL (.v) files except timescale.v and uart_defines.v.

Click Open.

6. Add the remaining files:


Navigate to the ..\timing_closure_tutor folder.

Select the ec_fpga_top.v and tc_top.v files.

Click Open.

To specify the input file search path for the design files:

1. Choose Tools > Set Options.



2. In the Options dialog box, select Input in the left pane. In the right pane, click the open folder button next to Input File Search Path.

3. In the dialog box, select the timing_closure_tutor folder and click OK.

The folder is added to the Input File Search Path list.

4. Repeat Steps 2 and 3 for the following sub-folders:

dbg_interface

onchip_ram

or1200

sdr_sdram

uart16550

5. Click OK to accept the sub-folder list.

These search paths enable Precision RTL to locate files referred to by the Verilog ‘include compiler directives within sub-modules of the SoC design.



To set up and compile the design:

1. In the Design pane on the left, click Setup Design to open the Project Settings dialog box.

2. Double-click Lattice in the Technology pane, and select Lattice ECP.

3. Select the following settings:

Device: LFECP10E

Speed Grade: 4

Package: PQFP208

Design Frequency – Set Frequency: 66

I/O Constraints – Set Delay – Input Delay: 10

I/O Constraints – Set Delay – Output Delay: 10

4. Click OK.

5. In the Design pane, click Compile.

The Precision software reads the input files and compiles the design based on the source files, targeted device, and settings.



To set constraints and synthesize the design:

1. Select the Design Center tab.

2. In the Design Hierarchy pane, expand the Clocks folder.

3. Perform the following steps:

Right-click clk and select Set Clock Constraints to open the Clock Constraints: Port-clk dialog box.

Specify 15.15 for Period and click OK.

Right-click jtag_tck and select Set Clock Constraints.

Specify 100 for Period and click OK.



4. Expand the Ports folder, and then perform the following steps:

Right-click the Inputs folder and select Set Input Constraints to open the Port Constraints: All Ports dialog box.

Specify 10 for Input Delay and click OK.

Right-click the Outputs folder and select Set Output Constraints to open the Port Constraints: All Ports dialog box.

Specify 10 for Output Delay and click OK.

5. In the Design Pane, click the Synthesize button.

Precision optimizes the design and creates an EDIF netlist, ec_fpga_top.edf, for the implementation.

To save the netlist file

1. From the Project Files pane of the Design Center tab, double-click ec_fpga_top.edf to open the netlist in the text editor.

2. Choose File > Save As, navigate to the ..\timing_closure_tutor folder, and click Save.

3. Click Yes in the message box to replace the EDIF file that already exists.

Note

If you want to preserve the original tutorial design files, copy the timing_closure_tutor directory to another location before proceeding.



To examine the timing report:

1. Click the Design Analysis button in the left pane, and then click Report Timing to open the ec_fpga_top_timing.rep file.

2. Locate the Clock Frequency Report section and examine the Min Period (Frequency) estimated by Precision.

Clock Frequency Report

Domain Clock Name Min Period (Freq)------ ---------- -----------------ClockDomain0 jtag_tck 69.842 (14.318 MHz) ClockDomain1 clk 36.732 (27.224 MHz)

3. Choose File > Exit.

Achieving Timing Closure in FPGA Designs Appendix B: Command Line Interface Quick Reference


Appendix B: Command Line Interface Quick Reference

This appendix describes the command line usage for key ispLEVER program executables and interactive GUI tools. The first section presents program executables in the typical order of execution as shown in Figure 7

Table 1: ispLEVER Processors and Usage

Design Tool Usage

edif2ngd – EDIF netlist translator

Usage: edif2ngd [-r] {-l <libname>} [-nopropwarn] [-mc] <ediffile> [<ngofile>]

where:-r Remove LOC props from the design -l <libname> Design is built from one of the following libraries: LatticeEC LatticeECP LatticeECP2 LatticeECP2M LatticeSC LatticeSCM LatticeXP MachXO-d Design is translated for a specific device.

Device is a combination of the Device Family, Logic Capacity, and Supply Voltage portions of the data sheet Part Number Description. For example: “LFEC3E”<edif_file> EDIF 2 0 0 format file. <ngo_file> Output '.ngo' file. Default is <infile>.ngo.

ngdbuild – build utility

Usage: ngdbuild [-a|-v <arch>] {-p <searchpath>} [-mc [<modulename>]] [-assemble [{<modulename>}]] <ngo_file[.ngo]>[<ngd_file[.ngd]>]

where:-a architecture Expand design for the given architectures: LatticeEC LatticeECP LatticeECP2 LatticeECP2M LatticeSC LatticeSCM LatticeXP MachXO -p search_path Add "search_path" to the list of directories to search when resolving NGO/NMC file references



map – design mapper

Usage: map [-a <pmname>] [-m] [-p <devicename>] [-t <typename>] [-s <speed>] [-swl [<swlfactor:0,10>]] [-g <guidefile[.ncd]>] [-hier] [-pe]

[-c [<packfactor:0,100>]] [-pr <oprffile[.prf]>] [-td [<tdfile[.ncd]>]]

[-o <outfile[.ncd]>] <infile[.ngd]> [<prffile[.prf]>] [-retime]

where: -a pmname Map design to one of the following architectures:

LatticeEC LatticeECP LatticeECP2 LatticeECP2M LatticeSC LatticeSCM LatticeXP MachXO) -c Without packfactor, map as dense as possible. With packfactor, map to given percentage of PFUs. -f Use command line arguments from file. -help Print usage (this message). -hier Use hierarchy information when packing. -m Produce oversized NCDs. -o ncd_file Optional output .ncd file. -p device_name Map design to device name. The device name is a combination of the Device Family, Logic Capacity, and Supply Voltage portions of the data sheet Part Number Description. For example, “LFEC3E”.-pr pref_file Optional output preference file. -s speed Speed grade to use: 3 (slowest) 4 5 (fastest)-t package_name Package name to use: TQFP100 TQFP144 PQFP208 FPBGA256 FPBGA388 FPBGA484 FPBGA672 FPBGA900 FPBGA896 FPBGA1152 FPBGA1517 FPBGA1704-retime Perform register retiming.


Design Tool Usage



par – place and route program

Usage: par [-k] [-p] [-r] [-i <routepasses:0,2000>] [-c <costpasses:0,20>] [-e|-d <delaypasses:0,100>] [-exp 0|1][-u <routecost:1,5>] [-q] [-g <guidefile[.ncd]>] [-gv] [-x] [-a][-l <level:1,5>] [-mf <matchingfactor:0,100>] [-n <iterations:0,100>] [-t <costtable:1,100>] [-sp <speedgrade>] [-s <savebest:1,100>] [-w] [-y] [-m <nodelistfile>] [-exp <explorerstring>] [-mc] [-assemble] [-pe] <infile[.ncd]> <outfile> {<preffile[.prf]>}

where:-a Turn on automatic level skipping for constructive placement. Begins at "-l <1-5>" and automatically skips levels based on completion criteria. The "-n <n>" switch controls the maximum number of iterations per level. Note: "-n <n>", when used with "-a" will be the maximum number

of iterations at each level -- not for the total run.-assemble (Modular Design) Assemble top-level design and the implemented modules into one design. -c Run n cleanup passes of the router. Default: 1.-d Run n delay based cleanup passes of the router, n >= 0.

Default: 0.-e Run n delay based cleanup passes of the router if there are 0

unrouted, n >= 0. Default: 0.-exp Give shorter routes priority (-ex 1). Default: 0-f Read par command line arguments & switches from file.


Design Tool Usage



par – place and route program

-g Use a guide file to place and route against. 1. Keep matching block names. 2. Keep matching netnames/pins. -gv List matched comps/macros/signals in guided par report file-i = Run n iterations of the router. Default: Run until router

decides it will not complete without diverging. -k Skip constructive placement. Run optimize placement, and then

enter the router.Note: Use -k -p to do reentrant routing.

-l Effort Level. Level 5 is maximum effort. Default: 5. -m Multi task par run. File "<node list file>" contains a list of

node names to run the jobs on. -mc (Modular Design) Compile each module. -mf Matching factor. Default: 100 -n Iterations at this level. Use "-n 0" to run until fully routed. See Note under "-a" option. Default: 1. -p Don't run placement. -pe Error out if there are any Preference errors. -q Placer will attempt to minimize overall clock skew. -r Don't run router. -s Save "n" best results for this run. Default: Save All. -sp Change speed grade. Default: Keep Current speed grade. -t Start at this placer cost table entry. Default: 1. -u Use this router cost table entry. Valid range varies by device.

Default: 1.-w Overwrite. Allows overwrite of an existing file (including input file). If specified output is a directory, allows files in directory to be overwritten. -x Ignore Timing preferences in preference file. -y Create .dly file and the delay statistics at end of .par file <infile> Name of input NCD file. <outfile> Name of output NCD file or output directory. Use format "<outfile>.ncd" or "<outfile>.dir". <preffile> Name of preference file.

trce – static timing analysis

Usage: trce [-e|-v [<limit:0,32000>]] [-sp <grade>] [-hld] [-sm <report>][-min] [-max] [-c] [-nmp] [-p] [-o <report[.twr]>] <design[.ncd]>{<preference[.prf]>}

where:<design[.ncd]> Design input file (no default)<preference[.prf]> Optional preference input file(default design.prf)-c Print connections not covered by the preference file in the verbose mode-e [<limit>] Produce detailed error report for timing preferences, optionally limited to the specified number of items-f <cmd file> Read trce command line arguments & switches from file-hld Perform hold time checking-o <report[.twr]> Optional report output file (default design.twr)-p Print loops in the verbose mode-sm <stampfile> Generate stamp output files


Design Tool Usage



trce – static timing analysis

-sp <speed_grade> Enable default speed to be overridden-v [<limit>] Produce verbose timing report for timing preferences, optionally limited to the specified number of items

bitgen – bitstream generator

Usage: bitgen [-help <arch>] [-J w <infile1.ncd> {<infile2.ncd>}] [-J r] <infile[.ncd]> [<outfile>] [<prffile[.prf]>]

Where:-help-J w-J r

Valid architectures are:or3t00MachXOLatticeECLatticeECPLatticeECP2LatticeECP2MLatticeSCLatticeSCMLatticeXPMachXO

promgen – PROM file generator

Usage: promgen [-b] [-p mcs|exo|tek] [-s <size>] [-t <size>] [-o <outfile>]{-u <hexaddr> <file> {<file>}} {-d <hexaddr> <file> {<file>}} {-n <file>{<file>}}

Where: -b Byte wide mode (no bit mirroring) -s <size> Prom size in K bytes (must be power of 2) -t <size> Split the into multiple prom files with user-specified size. The resulting files are named as <file>00.<ext>, <file>01.<ext>, <file>02.<ext>, ..., etc. Cannot be used with switch -s, -n, or -d. -p <format> PROM format (mcs, exo, or tek) -o <file> Output prom file name (default matches first .bit file) -u <hexaddr> <file> {<file>} Load .bit or .rbt file(s) up from address. Multiple bit or .rbt files are daisychained to form a single prom load. -d <hexaddr> <file> {<file>} Load .bit or .rbt file(s) down from address. Multiple .bit or .rbt files are daisychained to form a single prom load.-n <file> {<file>} Load .bit or .rbt file(s) up or down starting from the next address following previous load. Multiple .bit or .rbt files are daisychained to form a single prom load. Must follow a -u, -d, or -n option.


Design Tool Usage



ldbanno – Timing simulation model writer

Usage: ldbanno [-w] [-pre <prfx>] [-neg] [-sup] [-sp <grade>] [-min] [-dis[<del>]] [-sig <sigfile>] [-i] [-n <type>] [-l <libtype>] [-p <prffile>] [-o <netfile>] [<ncdfile>]

where:-w Overwrite the output file(s) -sp <grade> Override speed grade for back annotation -pre <prfx> Prefix to add to module name to make them unique for multi-chip simulation -min Override speed grade to minimum timing for hold check -dis <del> Distribute routing delays by splitting the signal and inserting buffers. <del> is the maximum delay (in ps) between each buffer (1000ps by default) -neg Negative setup/hold delay support. Without this option, all negative numbers are set to 0 in SDF-sup Suppress Custom Design MACO blocks SDF output. <sigfile> Contains names for the signals to be split. If not given along with the -dis option, all signals will be processed-i Create a buffer for each block input that has interconnection delay (only valid for Verilog) <type> Netlist type to write out <libtype> Library element type to use <prffile> Preference file <netfile> The name of the output netlist file. The extension on this file will change depending on which type of netlist is being written. Use -h <type>, where <type> is the output netlist type, for more specific information. <ncdfile> Input file '.ncd'

Use ldbanno -h <type> for detailed netlist and annotation options.

Valid netlist types are: Verilog (generic Verilog format w/ SDF delay file) VHDL (generic VHDL format w/ SDF delay file)

Valid library types are: orca - Default for VHDL, Verilog

ibisgen – IBIS model writer

Usage: ibisgen <padfile> <ibisfile>

Write out design IBIS models.


Design Tool Usage



Table 2: ispLEVER GUI-based Design Tools

Design Tool Usage

EPIC - Device Editor epic

Design Planner flmainappw [@responsefile] [-inp <NGDfile] [-dir <directory>] [-prj <projectfile>] [-a <arch>] [-p part] [-t <package>] [-lpf <logical_preference_file>]

flmainapp [@responsefile] [-inp <NCDfile>] [-dir <directory>] [-prj <projectfile>] [-a <arch>] [-p part] [-t <package>] [-lpf <logical_preference_file>] [-prf <physical_preference_file>]

where:-inp loads the specified premap or post-map database file.-dir specifies the path and directory of the database file.-prj specifies the name of the project.-a specifies the device architecture to which the design will be mapped.-p specifies the device part number.-t specifies the device package.-lpf specifies the logical preference file.-prf specifies the physical preference file.-msg specifies the process stage for the database design file.

Module/IP Manager ispmg

Power Calculator powerest

Schematic Editor schem [filename]

Symbol Editor sym [filename]

Text Editor synedit [filename]

Waveform Stimulus Editor wet [filename]

lattice timingclosure

Documents

twr map1par3par3

ncd map1par3par3

twr map1par2par2

twr map1par4par4

ncd map1par2par2

ncd map1par4par4

twr map1par1par1

ncd map1par1par1