Top Ten Things I wish I never had needed to learn about LVDS_25_DCI:
1) Parallel DCI input standards in Virtex2 continuously modulate the input termination offset voltage unless you enable bitgen's FreezeDCI option 2) With FreezeDCI on, the entire bottom half of 2V40, 2V80, and any CS144 packages are unavailable for LVDS_25_DCI inputs (this includes half the global clock inputs to the chip) due to DCI unavailability in banks having only ALT_VRP/N pins 3) With FreezeDCI on, dual purpose config pins cannot be used as LVDS_25_DCI inputs 4) 5.2i S/W doesn't catch illegal pin assignments due to #2 and #3 5) With FreezeDCI on, input terminator accuracy for 2R values degrades to +/-20% 6) With FreezeDCI on, each bank will have a (different) random input offset voltage due to split terminator 2R variations 7) LVDS_25_DCI terminator overhead power per input pair far exceeds the theoretical 62.5 mW number published in Answer Record 15633 8) With FreezeDCI on, worst case VCCO power overhead per LVDS_25_DCI input pair approaches 100 mW 9) With FreezeDCI on, worst case DCI VRP/N VCCO power overhead per I/O bank approaches 200 mW10) 5.2i Xpower incorrectly assigns DCI power to the 1.5V VCCINT supply, and it doesn't use the worst case DCI power numbers
11) V2 Power Estimator spreadsheet doesn't support LVDS_25_DCI, but if you fake it by using two single ended DCI 2R split terminated inputs per actual LVDS pair, it also uses the wildly optimistic power numbers12) LVDS_25_DCI IBIS models don't work in HyperLynx
13) Massive 8pf IBIS C_COMP input capacitance value for the V2 LVDS inputs requires external back termination and/or input matching scheme to achieve reasonable signaling when driving FPGA inputs from a modern high speed LVDS driverInteresting Answer Database Search Keywords:
FreezeDCI LVDS AND DCI AND termination DCI AND power IBIS AND Hyperlynx ( in answer archive )
Suggestions to Xilinx:
- Have somebody document the plethora of V2 DCI hardware and software problems ('challenges'? 'features'?) in one place ( a detailed application note? ) ASAP.
- Hiding the FPGA IOB/CLB/FF/interconnect power consumption numbers within an encrypted spreadsheet and buggy SW makes it impossible to cross-check the resulting power calculations.
- Please take a look at page 145 of the ORCA-4 datasheet ("Package Parasitics"): there, in human readable form, is a usable package model that can be simulated in any SPICE.
- Also note that the ORCA-4 IBIS C_COMP value for the general purpose LVDS inputs is a much more reasonable 2 pf. - Real differential LVDS input terminators are quite wonderful (no VCCO power hit, no split terminator DC offset problems).
Making them available (LXXX_25_DT) only in the V2Pro, and not in the Spartan3, is an exceptionally HUGE mistake.
Brian