CTS on ihp-sg13g2

[mole99]

Hello,

is there a reason why the ihp-sg13g2 platform does not explicitly specify the clock buffers for CTS?

When I run any design through ORFS, only buf_1/buf_2 are used for the clock tree even though stronger buffers are available (buf_4/buf_8/buf_16).

I would label this as a bug, as stronger clock buffers are probably more efficient in large clock trees.

Thank you!

[maliberty]

The goal is to automate the buffer selection process. Users can override that if necessary but the hope is that it isn't. Are you observing this on a particular design?

[mole99]

That sounds like a great feature!

Yes, I've used my own design ttihp0p2 and only buf_1/buf_2 are used there.
Here is the project, to be placed under flow/platforms/ihp-sg13g2/: ttihp0p2-one-sprite-pony.zip

On the other hand, because of your comment I tried a different design, riscv32i from the examples. There, only buf_2/buf_4 are used for driving the clock tree, but inv_1/inv_2/inv_4 and buf_1/buf_4/buf_8/buf_16 are used as clock loads, of which buf_16 is a relatively large cell.

Maybe the algorithm works correctly, but it feels a bit strange.

Here's the netlist for ttihp0p2: 6_final_ttihp0p2-one-sprite-pony.zip

Here's the netlist for riscv32i: 6_final_riscv32i.zip