Optimization of a FPGA design requires a multi-dimensional approach that meets the design goals while reducing area, critical path delay, power consumption, and runtime. The Intel^® Quartus^® Prime software includes advisors to address each of these issues. By implementing the advisor's suggestions, you can reduce the time spent on design iterations.

In DSE II, an exploration point is a collection of Analysis & Synthesis, Fitter, and placement settings, and a group of exploration points is a design exploration. A design exploration can also include different fitter seeds.

DSE II compiles the design using the settings corresponding to each exploration point. When the compilation finishes, DSE II evaluates the performance data against an optimization goal that you specify. You can direct the DSE II to optimize for timing, area, or power.

DSE II provides a collection of predefined exploration spaces that focus on what you want to optimize. Additionally, you can define a set of compilation seeds. The number of explorations points is the number of seeds multiplied by the number of exploration modes.

In the DSE GUI, you specify these settings in the Exploration page.

This chapter describes how you can use the Intel^® Quartus^® Prime Netlist Viewers to analyze and debug your designs.

As FPGA designs grow in size and complexity, the ability to analyze, debug, optimize, and constrain your design is critical. With today’s advanced designs, several design engineers are involved in coding and synthesizing different design blocks, making it difficult to analyze and debug the design. The Intel^® Quartus^® Prime RTL Viewer and Technology Map Viewer provide powerful ways to view your initial and fully mapped synthesis results during the debugging, optimization, and constraint entry processes.

Click the link in the preprocessor process box to go to the Settings > Compilation Process Settings page where you can turn on the Run Netlist Viewers preprocessing during compilation option. If you turn this option on, the preprocessing becomes part of the full project compilation flow and the Netlist Viewer opens immediately without displaying the preprocessing dialog box.

Before the Netlist Viewer can run the preprocessor stage, you must compile your design:

• To open the RTL Viewer first perform Analysis and Elaboration.
• To open the Technology Map Viewer (Post-Fitting) or the Technology Map Viewer (Post‑Mapping), first perform Analysis and Synthesis.

The Technology Map Viewer shows the hierarchy of atom primitives (such as device logic cells and I/O ports) in the design. For supported device families, you can also view internal registers and look-up tables (LUTs) inside logic cells (LCELLs), and registers in I/O atom primitives.

Where possible, the Intel^® Quartus^® Prime software maintains the port names of each hierarchy throughout synthesis. However, the software may change or remove port names from the design. For example, the software removes ports that are unconnected or driven by GND or V_CC during synthesis. If a port name changes, the software assigns a related user logic name in the design or a generic port name such as IN1 or OUT1.

You can view Intel^® Quartus^® Prime technology-mapped results after synthesis, fitting, or timing analysis. To run the Technology Map Viewer for an Intel^® Quartus^® Prime project, on the Processing menu, point to Start and click Start Analysis & Synthesis to synthesize and map the design to the target technology. At this stage, the Technology Map Viewer shows the same post-mapping netlist as the Technology Map Viewer (Post‑Mapping). You can also perform a full compilation, or any process that includes the synthesis stage in the compilation flow.

For designs that completed the Fitter stage, the Technology Map Viewer shows how the Fitter changed the netlist through physical synthesis optimizations, while the Technology Map Viewer (Post‑Mapping) shows the post-mapping netlist. If you have completed the Timing Analysis stage, you can locate timing paths from the Timing Analyzer report in the Technology Map Viewer.

To open the Technology Map Viewer, click Tools > Netlist Viewers > Technology Map Viewer (Post-Fitting) or Technology Map Viewer (Post Mapping).

The RTL Viewer and Technology Map Viewer each consist of these main parts:

• The Netlist Navigator pane—displays a representation of the project hierarchy.
• The Find pane—allows you to find and locate specific design elements in the schematic view.
• The Properties pane displays the properties of the selected block when you select Properties from the shortcut menu.
• The schematic view—displays a graphical representation of the internal structure of the design.

Figure 3. RTL Viewer

Netlist Viewers also contain a toolbar that provides tools to use in the schematic view.

• Use the Back and Forward buttons to switch between schematic views. You can go forward only if you have not made any changes to the view since going back. These commands do not undo an action, such as selecting a node. The Netlist Viewer caches up to ten actions including filtering, hierarchy navigation, netlist navigation, and zoom actions.
• The Refresh button to restore the schematic view and optimizes the layout. Refresh does not reload the database if you change the design and recompile.
• Click the Find button opens and closes the Find pane.
• Click the Selection Tool and Zoom Tool buttons to alternate between the selection mode and zoom mode.
• Click the Fit in Page button resets the schematic view to encompass the entire design.
• Use the Hand Tool to change the focus of the viewer without changing the perspective.
• Click the Area Selection Tool to drag a selection box around ports, pins, and nodes in an area.
• Click the Netlist Navigator button to open or close the Netlist Navigator pane.
• Click the Color Settings button to open the Colors pane where you can customize the Netlist Viewer color scheme.
• Click the Display Settings button to open the Display pane where you can specify the following settings:
  • Show full name or Show only <n> characters. You can specify this separately for Node name, Port name, Pin name, or Bus name.
  • Turn Show timing info on or off.
  • Turn Show node type on or off.
  • Turn Show constant value on or off.
  • Turn Show flat nets on or off.
  Figure 4. Display Settings

  
• The Bird's Eye View button opens the Bird's Eye View window which displays a miniature version of the design and allows you to navigate within the design and adjust the magnification in the schematic view quickly.
• The Show/Hide Instance Pins button can alternate the display of instance pins not displayed by functions such as cross-probing between a Netlist Viewer and Timing Analyzer. You can also use this button to hide unconnected instance pins when filtering a node results in large numbers of unconnected or unused pins. The Netlist Viewer hides Instance pins by default.
• If the Netlist Viewer display encompasses several pages, the Show Netlist on One Page button resizes the netlist view to a single page. This action can make netlist tracing easier.

You can have only one RTL Viewer, one Technology Map Viewer (Post-Fitting), and one Technology Map Viewer (Post-Mapping) window open at the same time, although each window can show multiple pages, each with multiple tabs. For example, you cannot have two RTL Viewer windows open at the same time.

The RTL Viewer and Technology Map Viewer attempt to display schematic in a single page view by default. If the schematic crosses over to several pages, you can highlight a net and use connectors to trace the signal in a single page.

From 19.1 release, when you click on a port (in the schematic) or the signal name (in the Connectivity pane), the path is selected. Each node keeps only a dependency path. This dependency path is cleared when you recompile, close the project, select another port, run a new locate, or left-click on a non-port space.

When you select a port or a signal name of the multiple selection, the dependency path is updated only if the port is used. If the port is unused, the dependency path selection is cleared. However, the Resource Property Viewer does not display the dependency path selection for multiple node selections. It serves the objective of single path tracing (with only one selection color).

You can select one or more hierarchy boxes, nodes, state nodes, or state transition arcs that interest you in the Netlist Viewer and locate the corresponding items in another applicable Intel^® Quartus^® Prime software window. You can then view and make changes or assignments in the appropriate editor or floorplan.

To locate an item from the Netlist Viewer in another window, right-click the items of interest in the schematic or state diagram, point to Locate, and click the appropriate command. The following commands are available:

• Locate in Assignment Editor
• Locate in Pin Planner
• Locate in Chip Planner
• Locate in Resource Property Editor
• Locate in Technology Map Viewer
• Locate in RTL Viewer
• Locate in Design File

The options available for locating an item depend on the type of node and whether it exists after placement and routing. If a command is enabled in the menu, it is available for the selected node. You can use the Locate in Assignment Editor command for all nodes, but assignments might be ignored during placement and routing if they are applied to nodes that do not exist after synthesis.

The Netlist Viewer automatically opens another window for the appropriate editor or floorplan and highlights the selected node or net in the newly opened window. You can switch back to the Netlist Viewer by selecting it in the Window menu or by closing, minimizing, or moving the new window.

You can access a range of global synthesis and physical synthesis optimization settings from the Compiler Settings page:

The following sections describe the physical synthesis optimizations available in the Intel^® Quartus^® Prime software, and how they can help improve performance and fitting for the selected device.

In addition, designs targeting newer device families (with smaller process geometry) do not always present the slowest circuit performance at the highest operating temperature. The temperature at which the circuit is slowest depends on the selected device, the design, and the compilation results. The Intel^® Quartus^® Prime software manages this new dependency by providing newer device families with three different timing corners—Slow 85°C corner, Slow 0°C corner, and Fast 0°C corner. For other device families, two timing corners are available—Fast 0°C and Slow 85°C corner.

The Optimize multi-corner timing option directs the Fitter to meet timing requirements at all process corners and operating conditions. The resulting design implementation is more robust across process, temperature, and voltage variations. This option is on by default, and increases compilation time by approximately 10%.

When this option is off, the Fitter optimizes designs considering only slow-corner delays from the slow-corner timing model (slowest manufactured device for a given speed grade, operating in low-voltage conditions).

The slack of a path determines its criticality; slack appears in the timing analysis report, which you can generate using the Timing Analyzer.

Design analysis for timing closure is a fundamental requirement for optimal performance in highly complex designs. The analytical capability of the Chip Planner helps you close timing on complex designs.

A critical chain reports the design paths that limit further register retiming optimization. The Intel^® Quartus^® Prime Pro Edition software provides the Hyper-Retimer critical chain reports to help you improve design performance. You can focus on higher level optimization, because the Hyper-Retimer uses Hyper-Registers to evenly balance slacks on all the registers in a critical chain.

For more information about improving design performance using the Hyper-Retimer critical chain reports, refer to the Interpreting Critical Chain Reports topic in the Intel^® Stratix^® 10 High-Performance Design Handbook.

Most designs that fail timing start out with other problems that the Fitter reports as warning messages during compilation. Determine what causes a warning message, and whether to fix or ignore the warning.

After reviewing the warning messages, review the informational messages. Take note of anything unexpected, for example, unconnected ports, ignored constraints, missing files, and assumptions or optimizations that the software made.

The figure shows an example of inefficient use of a global clock. The highlighted line has a single fan-out from a global clock.

If you assign these resources to a Regional Clock, the Global Clock becomes available for another signal. You can ignore signals with an empty value in the Global Line Name column as the signal uses dedicated routing, and not a clock buffer.

The Non-Global High Fan-Out Signals report lists the highest fan-out nodes not routed on global signals.

Reset and enable signals appear at the top of the list.

If there is routing congestion in the design, and there are high fan-out non-global nodes in the congested area, consider using global or regional signals to fan-out the nodes, or duplicate the high fan-out registers so that each of the duplicates can have fewer fan-outs.

Use the Chip Planner to locate high fan-out nodes, to report routing congestion, and to determine whether the alternatives are viable.

In the Timing Analyzer Report Timing dialog box, enable the Report panel name and Show routing options, and click Report Timing.

The Extra Fitter Information tab shows a miniature floorplan with the path highlighted. The Extra Fitter Information tab is not available for Intel^® Stratix^® 10 devices.

You can also locate the path in the Chip Planner to examine routing congestion, and to view whether nodes in a path are placed close together or far apart.

• CLK_CTRL_Gn—for Global driver
• CLk_CTRL_Rn—for Regional driver

Buffers to access the global networks are located in the center of each side of the device. Buffering to route a core logic signal on a global signal network causes insertion delay. Tradeoffs to consider for global and non-global routing are source location, insertion delay, fan-out, distance a signal travels, and possible congestion if the signal is demoted to local routing.

set_multicycle_path -from <generating_register> -setup -end 2 

One typical cause is math precision error. For example, 10Mhz/3 = 33.33 ns per period. In three cycles, the time is 99.999 ns vs 100.000 ns. Setting a maximum delay can provide an appropriate setup relationship.

Another cause of failure are paths that must be false by design intent, such as:

• Asynchronous paths handled through FIFOs, or
• Slow asynchronous paths that rely on handshaking for data that remain available for multiple clock cycles.

To prevent the Fitter from having to meet unnecessarily restrictive timing requirements, consider adding false or multicycle path statements.

• If paths that fail timing begin or end in shift registers, consider disabling the Auto Shift Register Replacement option. Do not convert registers that are intended for pipelining.
• For shift registers that are converted to a chain, evaluate area/speed trade off of implementing in RAM or logic cells.
• If a design is close to full, you can save area by shifting register conversion to RAM, benefiting non-critical clock domains. You can change the settings from the default AUTO to OFF globally, or on a register or hierarchy basis.

Timing failure can occur when the I/O interface at the top of the device connects to logic driven by a regional clock which is in one quadrant of the device, and placement restrictions force long paths to and from I/Os to logic across quadrants.

Use a different type of clock source to drive the logic - global, which covers the whole device, or dual-regional which covers half the device. Alternatively, you can reduce the frequency of the I/O interface to accommodate the long path delays. You can also redesign the pinout of the device to place all the specified I/Os adjacent to the regional clock quadrant. This issue can happen when register locations are restricted, such as with Logic Lock regions, clocking resources, or hard blocks (memories, DSPs, IPs).

The Extra Fitter Information tab in the Timing Analyzer timing report informs you when placement is restricted for nodes in a path. The Extra Fitter Information tab is not available for Intel^® Stratix^® 10 devices.

To reach timing closure, a well written RTL can be more effective than changing your compilation settings. Seed sweeping can also be useful if the timing failure is very small, and the design has already been optimized for performance improvements and is close to final release. Additionally, seed sweeping can be used for evaluating changes to compilation settings. Compilation results vary due to the random nature of fitter algorithms. If a compilation setting change produces lower average performance, undo the change.

Sometimes, settings or constraints can cause more problems than they fix. When significant changes to the RTL or design architecture have been made, compile periodically with default settings and without Logic Lock regions, and re-evaluate paths that fail timing.

Partitioning often does not help timing closure, and must be done at the beginning of the design process. Adding partitions can increase logic utilization if it prevents cross-boundary optimizations, making timing closure harder and increasing compile times.

You can view a list of ignored constraints in the Timing Analyzer GUI by clicking Reports > Report Ignored Constraints or by typing the following command to generate a list of ignored timing constraints:

report_sdc -ignored -panel_name "Ignored Constraints"

Analyze any constraints that the Intel^® Quartus^® Prime software ignores. If necessary, correct the constraints and recompile your design before proceeding with design optimization.

You can view a list of ignored assignment in the Ignored Assignment Report generated by the Fitter.

The I/O paths that do not meet the required timing performance are reported as having negative slack and are highlighted in red in the Timing Analyzer Report pane. In cases where you do not apply an explicit I/O timing constraint to an I/O pin, the Intel^® Quartus^® Prime timing analysis software still reports the Actual number, which is the timing number that must be met for that timing parameter when the device runs in your system.

If any clock domains have failing paths (highlighted in red in the Report pane), right-click the clock name listed in the Clocks Summary pane and select Report Timing to get more details.

When you select a path in the Summary of Paths tab, the path detail pane displays all the path information. The Extra Fitter Information tab offers visual representation of the path location on the physical device. This can reveal whether the timing failure is distance related, due to the source and destination node being too close or too far. The Extra Fitter Information tab is not available for Intel^® Stratix^® 10 devices.

The Data Path tab displays the Data Arrival Path and the Data Required Path. You can determine the path segments contributing the most to the timing violations with the incremental information. The Waveform tab shows the signals in the time domain, and plots the slack between arrival data and required data.

The Technology Map Viewer provides schematic, technology-mapped, representations of the design netlist, and can help you to assess which areas in a design can benefit from reducing the number of logic levels. To locate a timing path in one of the viewers, right-click a path in the timing report, point to Locate Path, and select Locate in Technology Map Viewer. You can also investigate the physical layout of a path in detail with the Chip Planner.

• Focus on improving the paths that show the worst slack. The Fitter works hardest on paths with the worst slack. If you fix these paths, the Fitter might be able to improve the other failing timing paths in the design.
• Check for nodes that appear in many failing paths. These nodes are at the top of the list in a timing report panel, along with their minimum slacks. Look for paths that have common source registers, destination registers, or common intermediate combinational nodes. In some cases, the registers are not identical, but are part of the same bus.
• In the timing analysis report panels, click the From or To column headings to sort the paths by source or destination registers. If you see common nodes, these nodes indicate areas of your design that might be improved through source code changes or Intel^® Quartus^® Prime optimization settings. Constraining the placement for just one of the paths might decrease the timing performance for other paths by moving the common node further away in the device.

For details about the number and types of global routing resources available, refer to the relevant device handbook.

Check the global signal utilization in your design to ensure that the appropriate signals have been placed on the global routing resources. In the Compilation Report, open the Fitter report and click Resource Section. Analyze the Global & Other Fast Signals and Non-Global High Fan-out Signals reports to determine whether any changes are required.

You might be able to reduce skew for high fan-out signals by placing them on global routing resources. Conversely, you can reduce the insertion delay of low fan-out signals by removing them from global routing resources. Doing so can improve clock enable timing and control signal recovery/removal timing, but increases clock skew. Use the Global Signal setting in the Assignment Editor to control global routing resources.

The Timing Optimization Advisor guides you in making settings that optimize your design to meet your timing requirements. To run the Timing Optimization Advisor click Tools > Advisors > Timing Optimization Advisor. This advisor describes many of the suggestions made in this section.

When you open the Timing Optimization Advisor after compilation, you can find recommendations to improve the timing performance of your design. If suggestions in these advisors contradict each other, evaluate these options and choose the settings that best suit the given requirements.

The example shows the Timing Optimization Advisor after compiling a design that meets its frequency requirements, but requires setting changes to improve the timing.

When you expand one of the categories in the Timing Optimization Advisor, such as Maximum Frequency (fmax) or I/O Timing (tsu, tco, tpd), the recommendations appear in stages. These stages show the order in which to apply the recommended settings.

The first stage contains the options that are easiest to change, make the least drastic changes to your design optimization, and have the least effect on compilation time.

Icons indicate whether each recommended setting has been made in the current project. In the figure, the checkmark icons in the list of recommendations for Stage 1 indicates recommendations that are already implemented. The warning icons indicate recommendations that are not followed for this compilation. The information icons indicate general suggestions. For these entries, the advisor does not report whether these recommendations were followed, but instead explains how you can achieve better performance. For a legend that provides more information for each icon, refer to the “How to use” page in the Timing Optimization Advisor.

Each recommendation provides a link to the appropriate location in the Intel^® Quartus^® Prime GUI where you can change the settings. For example, consider the Synthesis Netlist Optimizations page of the Settings dialog box or the Global Signals category in the Assignment Editor. This approach provides the most control over which settings are made and helps you learn about the settings in the software. When available, you can also use the Correct the Settings button to automatically make the suggested change to global settings.

For some entries in the Timing Optimization Advisor, a button allows you to further analyze your design and see more information. The advisor provides a table with the clocks in the design, indicating whether they have been assigned a timing constraint.

When you turn on Optimize Hold Timing in the Advanced Fitter Settings dialog box, the Intel^® Quartus^® Prime software adds delay to paths to ensure that your design meets the minimum delay requirements. If you select I/O Paths and Minimum TPD Paths, the Fitter works to meet the following criteria:

• Hold times (t_H) from the device input pins to the registers
• Minimum delays from I/O pins to I/O registers or from I/O registers to I/O pins
• Minimum clock-to-out time (t_CO) from registers to output pins

If you select All Paths, the Fitter also works to meet hold requirements from registers to registers, as highlighted in blue in the figure, in which a derived clock generated with logic causes a hold time problem on another register.

However, if your design still has internal hold time violations between registers, you can manually add delays by instantiating LCELL primitives, or by making changes to your design, such as using a clock enable signal instead of a derived or gated clock.

This option is useful if routing resources are resulting in no-fit errors, and you want to reduce routing wire use.

The table lists the settings for the Fitter Aggressive Routability Optimizations logic option.

If the fast I/O setting is on, the register is always placed in the I/O element. If the fast I/O setting is off, the register is never placed in the I/O element. This is true even if the Optimize IOC Register Placement for Timing option is turned on. If there is no fast I/O assignment, the Intel^® Quartus^® Prime software determines whether to place registers in I/O elements if the Optimize IOC Register Placement for Timing option is turned on.

You can also use the four fast I/O options (Fast Input Register, Fast Output Register, Fast Output Enable Register, and Fast OCT Register) to override the location of a register that is in a Logic Lock region and force it into an I/O cell. If you apply this assignment to a register that feeds multiple pins, the Fitter duplicates the register and places it in all relevant I/O elements.

For more information about the Fast Input Register option, Fast Output Register option, Fast Output Enable Register option, and Fast OCT (on-chip termination) Register option, refer to Intel^® Quartus^® Prime Help.

The Intel^® Quartus^® Prime software automatically adjusts the applicable programmable delays to help meet timing requirements. For detailed information about the effect of these options, refer to the device family handbook or data sheet.

After you have made a programmable delay assignment and compiled the design, you can view the implemented delay values for every delay chain and every I/O pin in the Delay Chain Summary section of the Compilation Report.

You can assign programmable delay options to supported nodes with the Assignment Editor. You can also view and modify the delay chain setting for the target device with the Chip Planner and Resource Property Editor. When you use the Resource Property Editor to make changes after performing a full compilation, recompiling the entire design is not necessary; you can save changes directly to the netlist. Because these changes are made directly to the netlist, the changes are not made again automatically when you recompile the design. The change management features allow you to reapply the changes on subsequent compilations.

Although the programmable delays in newer devices are user-controllable, Intel recommends their use for advanced users only. However, the Intel^® Quartus^® Prime software might use the programmable delays internally during the Fitter phase.

For details about the programmable delay logic options available for Intel devices, refer to the following Intel^® Quartus^® Prime Help topics:

You can achieve the same type of effect in certain devices by using the programmable delay called Input Delay from Dual Purpose Clock Pin to Fan-Out Destinations.

Intel devices have a variety of hierarchical clock structures. These include dedicated global clock networks, regional clock networks, fast regional clock networks, and periphery clock networks. The available resources differ between the various Intel device families.

For the number of clocking resources available in your target device, refer to the appropriate device handbook.

Global clock networks, regional clock networks, and periphery clock networks have an additional level of clock hierarchy known as spine clocks. Spine clocks drive the final row and column clocks to their registers; thus, the clock to every register in the chip is reached through spine clocks. Spine clocks are not directly user controllable.

To reduce these spine clock errors, constrain your design to use your regional clock resources better:

• If your design does not use Logic Lock regions, or if the Logic Lock regions are not aligned to your clock region boundaries, create additional Logic Lock regions and further constrain your logic.
• If Periphery features ignore Logic Lock region assignment, possibly because the global promotion process is not functioning properly. To ensure that the global promotion process uses the correct locations, assign specific pins to the I/Os using these periphery features.
• By default, some Intel^® FPGA IP functions apply a global signal assignment with a value of dual-regional clock. If you constrain your logic to a regional clock region and set the global signal assignment to Regional instead of Dual-Regional, you can reduce clock resource contention.

Coding style affects the performance of a design to a greater extent than other changes in settings. Always evaluate the code and make sure to use synchronous design practices.

Before performing design optimizations, understand the structure of the design as well as the effects of techniques in different types of logic. Techniques that do not benefit the logic structure can decrease performance.

Be aware of the number of logic levels needed to implement your logic while you are coding. Too many levels of logic between registers might result in critical paths failing timing. Try restructuring the design to use pipelining or more efficient coding techniques. Also, try limiting high fan-out signals in the source code. When possible, duplicate and pipeline control signals. Make sure the duplicate registers are protected by a preserve attribute, to avoid merging during synthesis.

If the critical path in your design involves memory or DSP functions, check whether you have code blocks in your design that describe memory or functions that are not being inferred and placed in dedicated logic. You might be able to modify your source code to cause these functions to be placed into high-performance dedicated memory or resources in the target device. When using RAM/DSP blocks, enable the optional input and output registers.

Ensure that your state machines are recognized as state machine logic and optimized appropriately in your synthesis tool. State machines that are recognized are generally optimized better than if the synthesis tool treats them as generic logic. In the Intel^® Quartus^® Prime software, you can check the State Machine report under Analysis & Synthesis in the Compilation Report. This report provides details, including state encoding for each state machine that was recognized during compilation. If your state machine is not recognized, you might have to change your source code to enable it to be recognized.

During the synthesis stage of the Intel^® Quartus^® Prime compilation, physical synthesis optimizations operate either on the output from another EDA synthesis tool, or as an intermediate step in synthesis. These optimizations modify the synthesis netlist to improve either area or speed, depending on the technique and effort level you select.

To view and modify the synthesis netlist optimization options, click Assignments > Settings > Compiler Settings > Advanced Settings (Fitter).

If you use a third-party EDA synthesis tool and want to determine if the Intel^® Quartus^® Prime software can remap the circuit to improve performance, use the Perform WYSIWYG Primitive Resynthesis option. This option directs the Intel^® Quartus^® Prime software to un-map the LEs in an atom netlist to logic gates, and then map the gates back to Intel-specific primitives. Intel-specific primitives enable the Fitter to remap the circuits using architecture-specific techniques.

The Intel^® Quartus^® Prime technology mapper optimizes the design to achieve maximum speed performance, minimum area usage, or balances high performance and minimal logic usage, according to the setting of the Optimization Technique option. Set this option to Speed or Balanced.

During the Fitter stage of the Intel^® Quartus^® Prime compilation, physical synthesis optimizations make placement-specific changes to the netlist that improve speed performance results for the specific Intel device.

Identify the default optimization targets of your Synthesis tool, and set your device and timing constraints accordingly. For example, if you do not specify a target frequency, some synthesis tools optimize for area.

You can specify logic options for specific modules in your design with the Assignment Editor while leaving the default Optimization Technique setting at Balanced (for the best trade-off between area and speed for certain device families) or Area (if area is an important concern). You can also use the Speed Optimization Technique for Clock Domains option in the Assignment Editor to specify that all combinational logic in or between the specified clock domains are optimized for speed.

To achieve best performance with push-button compilation, follow the recommendations in the following sections for other synthesis settings. You can use DSE II to experiment with different Intel^® Quartus^® Prime synthesis options to optimize your design for the best performance.

Duplicating the sources of these types of globally-influential signals can help to disperse them across many hops, or even across many clock cycles, and focus more on local transfers.

For example, by duplicating a high fan-out signal in the form of a tree of registers, you can disperse the signal over several clock cycles. As the signal progresses down the tree, it progressively feeds more into local copies of the original registers, such that any individual register's destinations are well-localized and its influence on register optimization is minimal. The key to this optimization is to determine how to assign the original signal’s fan-outs among the duplicates. If any individual register requires driving a large distance, the benefit of the tree can be removed.

You can manually create a register tree and group the endpoints in the RTL by leveraging your system-level knowledge about how best to disperse the signal throughout your design, but it can be time consuming and have a widespread impact. For more information about manually creating a register tree, refer to Manual Register Duplication.

You can create register trees automatically in one of the following ways.

• Estimated Physical Proximity
• Hierarchical Proximity

Each method has its own methodology to determine the number of duplicates to create and how to assign the fan-outs between the duplicates.

Synthesis tools support options or attributes that specify the maximum fan-out of a register. When using Intel^® Quartus^® Prime synthesis, you can set the Maximum Fan-Out logic option in the Assignment Editor to control the number of destinations for a node so that the fan-out count does not exceed a specified value. You can also use the maxfan attribute in your HDL code. The software duplicates the node as required to achieve the specified maximum fan-out.

Logic duplication using Maximum Fan-Out assignments normally increases resource utilization, and can potentially increase compilation time, depending on the placement and the total resource usage within the selected device.

The improvement in timing performance that results from Maximum Fan-Out assignments is design-specific. This is because when you use the Maximum Fan-Out assignment, the Fitter duplicates the source logic to limit the fan-out, but does not control the destinations that each of the duplicated sources drive. Therefore, it is possible for duplicated source logic to be driving logic located all around the device. To avoid this situation, you can use the Manual Logic Duplication logic option.

If you are using Maximum Fan-Out assignments, benchmark your design with and without these assignments to evaluate whether they give the expected improvement in timing performance. Use the assignments only when you get improved results.

You can manually duplicate registers in the Intel^® Quartus^® Prime software regardless of the synthesis tool used. To duplicate a register, apply the Manual Logic Duplication logic option to the register with the Assignment Editor.

set_instance_assignment -name DUPLICATE_REGISTER -to <register_name> <num_duplicates>

where,

• register_name is the register to duplicate. To create a register tree from a chain, create a unique assignment for each register in the chain. DUPLICATE_REGISTER assignments are processed in the appropriate order if they apply to registers that drive each other in a chain.
• num_duplicates is the number of duplicates of the register to create (including the original). If the original signal has M fan-out, the average fan-outs of the duplicates are M/N but any individual duplicate may have more or fewer, at the discretion of the algorithm.

The DUPLICATE_REGISTER assignment is processed during the Fitter stage. It is necessary to create the duplicates and assign fan-outs between the duplicates based on early estimates of physical proximity to maximize the amount of time spent optimizing the design post-duplication. However, this renders fine-grained assignment decisions imprecise. The DUPLICATE_REGISTER assignment is best used when the number of duplicates is small (under 100) and the groups created are coarse-grained enough to allow for flexibility during optimization after the duplicates are created.

The Fitter Duplication Summary panel of the Fit report details the DUPLICATE_REGISTER assignments picked up by Intel^® Quartus^® Prime Pro Edition. It also summarizes any registered signal with greater than 1000 fan-outs, as they could be reasonable candidates for DUPLICATE_REGISTER assignments in future.

set_instance_assignment -name DUPLICATE_HIERARCHY_DEPTH -to <register_name> <num_levels>

where,

• register_name is the last register in a chain that fans out to multiple hierarchies. To create a register tree, ensure that there are sufficient simple registers behind the node and those simple registers are automatically pulled into the tree.
• num_levels corresponds to the upper bound of the number of registers that exist in the chain to use for duplicating down the hierarchies.

The DUPLICATE_HIERARCHY_DEPTH assignment is processed during the Synthesis stage. It is common for high-fanout signals to go through a pipeline of registers and drive into a sub-hierarchy of modules. For example, a system-wide reset can be propagated over several clock cycles and driven into many modules across the design. In several scenarios, it is useful to take advantage of the structure of this sub-hierarchy to infer the structure of the register tree to be created, such that endpoints within similar hierarchies are assigned the same copy of the signal, and branches in the design hierarchy dictates where to place branches in the register tree.

• Turn on register balancing or retiming
• Turn on register pipelining
• Turn off resource sharing

These options can increase performance, but typically increase the resource utilization of your design.

Changes in the design impact performance between compilations. This random variation is inherent in placement and routing algorithms—it is impossible to try all seeds and get the absolute best result.

If a change in optimization settings marginally affects the register-to-register timing or number of failing paths, you cannot always be certain that your change caused the improvement or degradation, or whether it is due to random effects in the Fitter. If your design is still changing, running a seed sweep (compiling your design with multiple seeds) determines whether the average result improved after an optimization change, and whether a setting that increases compilation time has benefits worth the increased time, such as with physical synthesis settings. The sweep also shows the amount of random variation to expect for your design.

If your design is finalized you can compile your design with different seeds to obtain one optimal result. However, if you subsequently make any changes to your design, you might need to perform seed sweep again.

Click Assignments > Compiler Settings to control the initial placement with the seed. You can use the DSE II to perform a seed sweep easily.

To specify a Fitter seed use the following Tcl command :

set_global_assignment -name SEED <value>

Location assignments are less flexible for the Intel^® Quartus^® Prime Fitter than Logic Lock assignments. Additionally, if you are familiar with your design, you can enter location constraints in a way that produces better results.

You can use the Intel^® Quartus^® Prime software to analyze the average MTBF due to metastability when a design synchronizes asynchronous signals and to optimize the design to improve the MTBF. These metastability features are supported only for designs constrained with the Timing Analyzer, and for select device families.

If the MTBF of your design is low, refer to the Metastability Optimization section in the Timing Optimization Advisor, which suggests various settings that can help optimize your design in terms of metastability.

This chapter describes how to enable metastability analysis and identify the register synchronization chains in your design, provides details about metastability reports, and provides additional guidelines for managing metastability.

In traditional FPGA timing closure flows, the starting point for most design analysis is the critical path. Due to the nature of Intel^® Stratix^® 10 devices and the availability of the Hyper Retimer, it is best to start you timing closure activities from the Retiming Limit Report. You want to give the tool as many optimization opportunities as possible before having to look into more time intensive and potentially manual timing closure techniques.

The Retiming Limit Details report specifies:

• Clock Transfer: Clock domain, or the clock domain transfer for which the critical chain applies
• Limiting Reason: Design conditions which prevent further optimizations from happening.
• Critical Chain Details: Timing paths associated with the timing restrictions.

When running Fast Forward compilation, the Compiler removes signals from registers to allow mobility within the netlist for subsequent retiming. Fast Forward compilation generates design-specific timing closure recommendations, and predicts maximum performance with removal of all timing restrictions.

After you complete Fast Forward explorations, you can determine which recommendations to implement to provide the most benefit. Implement appropriate recommendations in your RTL, and recompile the design to achieve the performance levels that Fast Forward reports.

The Fast Forward Details Report provides the following information:

After implementing timing closure recommendations in your design, you can rerun the Retime stage to obtain the predictive performance gains.

You can continue exploring performance and implementing RTL changes to your code until you reach the desired performance target. Once you have completed all the modifications you want to do, continue your timing closure activities with the traditional techniques explained in this document.

For more information about implementing Fast Forward timing closure recommendations in your design, refer to the Implement Fast Forward Recommendations section of the Intel^® Stratix^® 10 High-Performance Design Handbook

Transfers between external interfaces (for example, high-speed I/O or serial interfaces) and the FPGA often require routing many connections with tight setup and hold timing requirements. When this option is turned on, the Fitter performs P2C placement and routing decisions before those for core placement and routing. This reserves the necessary resources to ensure that your design achieves its timing requirements and avoids routing congestion for transfers with external interfaces.

This option is available as a global assignment, or can be applied to specific instances within your design.

The following section described how to view various design elements in the Chip Planner.

Some reasons to use empty Logic Lock regions are:

• Preliminary floorplanning.
• Complex incremental builds.
• Team based design and interconnect logic.
• Confining logic placements.

Since Logic Lock regions do not reserve any routing resources, the Fitter may use the area for routing purposes.

Use the Core Only attribute for empty Logic Lock regions. When you include periphery resources in empty regions, you restrict the periphery component placement, which can result in a no fit design. After you name the empty region, you can perform the same manipulations as with any populated Logic Lock Region.

Figure 59. Logic Placed Outside of an Empty region

The figure shows an empty Logic Lock region and the logic around it. However, some IOs, HSSIO, and PLLs are in the empty region. This placement happens because the output port connects to the IO, and the IO is always part of the root_partition (top-level partition).