Reading and Interpreting a Complexity Report - 2023.2 English

Vivado Design Suite User Guide: Design Analysis and Closure Techniques (UG906)

Document ID
Release Date
2023.2 English

The Complexity Characteristics table from the previous example shows the Rent exponent and average fanout for each level of hierarchy below the top level. The typical ranges to consider when reviewing these metrics are the following:

  • Rent exponent:
    • Between 0.0 and 0.65: The complexity is considered low to normal and does not highlight any potential problems.
    • Between 0.65 and 0.85: The complexity is considered high, especially when the total number of instances is above 25k.
    • Above 0.85: The complexity is very high, and if the number of instances is also high, the design can potentially fail during implementation.
  • Average fanout:
    • Below 4: It is considered normal.
    • Between 4 and 5: The implementation tools can show difficulty to place the design without congestion. In the case of a SSI device, if the total number of instances is above 100k, the placer can have problems finding a placement solution that fits in 1 SLR or that is spread over 2 SLRs.
    • Above 5: The design can potentially fail during implementation.

You must treat high Rent exponents and/or high average fanouts for larger modules with higher importance. Smaller modules, especially under 10k total instances, can have higher Rent exponent and average fanout, and yet be simple to place and route successfully. For this reason, the Total Instances column must always be reviewed along with the Rent exponent and average fanout.

The complexity characteristics might not always predict routing congestion. Other factors such as I/O location constraints, floorplanning, and macro primitive location in the target device can limit the placement solution space and introduce congestion. The effect of such constraints is better analyzed by the congestion reports available after placement.

Other items to consider when interpreting the Complexity Characteristics table:

  • A higher percentage of LUT6s in a module usually increases the average fanout and potentially the Rent exponent.
  • A high number of RAMB and DSPs can increase the Rent exponent because these primitives have a large amount of connectivity.
  • The hierarchical instances with higher Rent exponents or higher average fanouts are not always a problem because the placer operates on a flat netlist and can break these instances into easier groups of logic to place. This report provides an indication of where a netlist problem can possibly exist if a module stands out clearly.
  • Hierarchical instances with lower Rent exponents can lead to congestion when combined with other items such as a large dataflow passing through the module when placed.

When a large module exhibits a high Rent exponent and/or average fanout that is causing congestion and timing issues, consider the following actions:

  • Reduce the connectivity of the module. Preserving the hierarchy to prevent cross-boundary optimization in synthesis can reduce the use of LUT6s and consequently reduce the netlist density.
  • Reduce the number of LUT inputs at synthesis.
  • Try to disable LUT combining in synthesis.
  • Use a Congestion Strategy during Implementation or SpreadLogic placement directive that can potentially help to relieve congestion. If the design is targeting an SSI Device, consider trying several SSI placement directives.
  • Use simple floorplanning at the SLR level for SSI devices, or at the clock region level in general, to keep congested groups of logic separate, or to guide global placement towards a solution similar to a previously found good placement.
  • Leverage QoR suggestions and Intelligent Design Runs to automatically solve congestion issues.