How to connect secondary power of level shifters

[titan73]

1. How is it possible with pdn to have 3rd power straps in the grid?
   In the floorplan script I use set_voltage_domain -power VDDD1V2 -ground GNDD. I'd like to add a secondary power VDDD for the level shifter but only 2 PG nets can be specified.
2. Can the router connect the level shifter secondary power pins to the VDDD straps?

[maliberty]

@gadfort suggestions?

[gadfort]

We do something like that for power switches, but I don't think we have any suggest for level shifters (I'm sure we dont). With a testcase and description of that is expected I can take a look and see what is needed.

[rovinski]

Generally, this is what an sroute should handle. I believe in most cases, the receiving power domain should perform an sroute from the grid to the location(s) of the level shifter(s). If there's a large number of level shifters that make this less practical, then I'm not sure what the best solution would be.

[gadfort]

That is an excellent point, the level shifters would also need to be placed before this is possible.

[titan73]

There are 278 level shifters in my design. Is sroute usable in that case?

[maliberty]

We have a very basic sroute you could try if that is appropriate.

[titan73]

Is there an example on how to use sroute?

[titan73]

I'd like something like this:

The grid is created with 3 nets: VDDD1V2, VDDD and GNDD. The VDDD1V2 and GNDD straps connect to the power rails in metal1. However PDN supports only 2 nets, 1 for power and 1 for ground.
The primary power/ground of the level shifter are VDDD1V2 and GNDD. The secondary power should be treated as a signal and routed during the routing step. The route starts from the secondary power pin and connects to the VDDD strap.

To achieve that in OR:

• PDN have to supports more that 2 PG nets
• The router should be able to route secondary power as a signal and connects it to the related strap

@gadfort Do you still need a test case? If so I'll upload one to the PII sftp server.

[gadfort]

@titan73 unfortunately I did not write sroute, so you will need to check with @maliberty about that.

[maliberty]

It was contributed by UMich and it has light documentation and one unit test (pdn/test/sroute_test.tcl). I believe it was focused on the problem of connecting a power ring to a gate so I'm not sure how well it generalizes.

[titan73]

I took a look on the command and its test. Indeed the documentation is light. I couldn't figure out what this function really does.
Here are some observations:
- OR crashes when the optional options -outerNet, -ongrid, -metalspaces and -metalwidths are omitted
- A power ring is needed to use the command
- -net option specifies both the power net and the gate pin. It works because the test has the same name for both
- Omitting -fixed_vias is similar to -fixed_vias {}
- I couldn't figure out what -outerNet and -layers option does after tweaking values
- -metalspaces does not have any effect after tweaking values
- the first segment starting from the pin has a width of the pin width/height and not the one specified by -metalwidths

The gate pin in the test is in met3 which is not very common for a standard cell. I edited the test def to change the pin layer to met1 to match a classical standard cell and I updated the command with:

add_sroute_connect \
  -net "VIN" \
  -outerNet "VDD" \
  -layers {met1 met4} \
  -cut_pitch {200 200} \
  -fixed_vias {M1M2_PR_M M2M3_PR_M M3M4_PR_M} \
  -metalwidths {1000 1000} \
  -metalspaces {800} \
  -ongrid {met1 met2 met3 met4} \
  -insts "temp_analog_1.a_header_0 temp_analog_1.a_header_1 temp_analog_1.a_header_2"

The horizontal routing uses the same layer as the gate pin causing short with a met1 rail.

It seems this command has been written specifically for the design the test is based on and I don't see any use outside this case.

[titan73]

Created #8770.