models module#

Delay models the measurements need: the FPGA’s package and the board’s vias.

Package delay#

UG863’s LPDDR4/4x timing table says “Include adaptive SoC package delays for all skew calculations”, and its “Required Memory Routing Guidelines for All Interfaces” says “Include package delay in routing constraints when determining signal trace lengths unless otherwise specified”, using “the midpoint/average between the minimum and maximum values”. package_pin_models() reads a package delay CSV in the shape of Vivado’s I/O ports export (UG899 Appendix A) and gives each FPGA pin with a delay a TerminatingPinModel at that midpoint, so JITX adds the pin’s package delay to every path that ends on it, delay limits included (constraint says how the length rule reads that).

The design reads SYNTHETIC_CSV, whose values are synthetic estimates, not AMD data. The memory’s pins get no model: UG863’s table asks only for the adaptive SoC’s delays.

Via delay#

Every via of the reference substrate carries a first-order delay model for each pair of copper layers it joins (via_models()), so a route through any of them is measured. The models carry no loss, so a loss limit measures the trace alone.

read_package_delays(path)[source]#

Midpoint package delay in seconds for each ball a Vivado-style I/O ports CSV gives a delay, read by column name. Comment lines start with #; a row with a blank delay (an export lists every pin) is skipped.

Return type:

dict[str, float]

Parameters:

path (Path)

package_pin_models(fpga, path)[source]#

A TerminatingPinModel at the midpoint delay, and no loss, for each of the FPGA’s balls in the CSV at path.

Return type:

list[TerminatingPinModel]

Parameters: