You can define and assess the maximum frequency (FMAX) with a design that runs on a given architecture and speed grade by iteratively increasing the target clock frequency and re-running both synthesis and implementation until small setup slack violations (WNS < 0) are reported by timing analysis on the fully routed design. AMD recommends using the Default or PerformanceOptimized synthesis directives along with the Explore implementation directives and strategy to get the best achievable FMAX. In some cases, alternate strategies can show higher FMAX depending on the size of the design and the nature of the critical logic paths. For the implementation results with small setup violations, the maximum frequency is computed as follows:
- FMAX (MHz) = max(1000/(Ti - WNSi))
- Ti is the target clock period (ns) used during the implementation run "i"
- WNSi is the worst negative slack (ns) of the target clock used during the implementation run "i"
Additional important considerations:
- Using overly tight clock periods can lead to automatic effort reduction in the Vivado Implementation tools to avoid high compilation time due to unrealistic target and large timing violations. Use reasonably tight clock constraints instead.
- For designs with multiple clocks, you must proportionally decrease all synchronous clock periods until one of them starts failing timing after implementation (preferably the fastest clock or the clock with the most timing paths).
For a given design implementation, the maximum operating frequency on hardware across temperature and voltage ranges supported by the target device speed grade is defined by 1000/(T - WNS), with WNS positive or negative. When operating under nominal temperature and voltage conditions, typically in a lab environment, it is usually possible to operate the design at a slightly higher frequency.