Intel® Quartus® Prime Pro Edition User Guide: Design Optimization

ID 683641
Date 10/04/2021
Public

A newer version of this document is available. Customers should click here to go to the newest version.

Document Table of Contents

5.4.4.3. Evaluate Wires Added for Hold

During routing the Fitter may add wire between register paths to increase delay to meet hold time requirements. The Fitter reports how much routing delay was added in the Estimated Delay Added for Hold Timing report. Excessive additional wire can indicate an error with the constraint. The cause of such errors is typically incorrect multicycle transfers between multi-rate clocks, and between different clock networks.

Review the specific register paths in the Estimated Delay Added for Hold Timing report to determine whether the Fitter adds excessive wire to meet hold timing.

Figure 17. Estimated Delay Added for Hold Timing Report

An example of an incorrect constraint which can cause the router to add wire for hold requirements is when there is data transfer from 1x to 2x clocks. Assume the design intent is to allow two cycles per transfer. Data can arrive any time in the two destination clock cycles by adding a multicycle setup constraint as shown in the example:

set_multicycle_path -from 1x -to 2x -setup -end 2

The timing requirement is relaxed by one 2x clock cycle, as shown in the black line in the waveform in the figure.

Figure 18. Timing Requirement Relaxed Waveform
The default hold requirement, shown with the dashed blue line, can force the router to add wire to guarantee that data is delayed by one cycle. To correct the hold requirement, add a multicycle constraint with a hold option.
set_multicycle_path -from 1x -to 2x -setup -end 2
set_multicycle_path -from 1x -to 2x -hold -end 1

The orange dashed line in the figure above represents the hold relationship, and no extra wire is required to delay the data.

The router can also add wire for hold timing requirements when data transfers in the same clock domain, but between clock branches that use different buffering. Transferring between clock network types happens more often between the periphery and the core. The following figure shows data is coming into a device, a periphery clock drives the source register, and a global clock drives the destination register. A global clock buffer has larger insertion delay than a periphery clock buffer. The clock delay to the destination register is much larger than to the source register, hence extra delay is necessary on the data path to ensure that it meets its hold requirement.

Figure 19. Clock Delay

To identify cases where a path has different clock network types, review the path in the Timing Analyzer, and check nodes along the source and destination clock paths. Also, check the source and destination clock frequencies to see whether they are the same, or multiples, and whether there are multicycle exceptions on the paths. Finally, ensure that all cross-domain paths that are false by intent have an associated false path exception.

If you suspect that routing is added to fix real hold problems, you can disable the Optimize hold timing advanced Fitter setting (Assignments > Settings > Compiler Settings > Advanced Settings (Fitter) > Optimize hold timing). Recompile the design with Optimize hold timing disabled, and then rerun timing analysis to identify and correct any paths that fail hold time requirements.

Note: Disable the Optimize hold timing option only when debugging your design. Ensure to enable the option (default state) during normal compiles. Wire added for hold is a normal part of timing optimization during routing and is not always a problem.