I think you mean function RepairSetup::addWireAndBuffer(). Yes, if we had a distributed RC or a pie RC, we can compute Elmore delay using downstream cap. The code seems to be focusing on a pre-route case without extracted RC. If we have the full distributed RC available, we can cut the RC network at the desired buffer location to do more accurate estimation and para update. This will require significant code change though.

I think you mean function RepairSetup::addWireAndBuffer().

Yes, that's what I meant, thanks.

Yes, if we had a distributed RC or a pie RC, we can compute Elmore delay using downstream cap. The code seems to be focusing on a pre-route case without extracted RC.

At that place within the rebuffering code, we can use the same RC estimates which estimate_parasitics -placement is using. E.g. to implement the R*C/2 + R*C_downstream model:

diff --git a/src/rsz/src/Rebuffer.cc b/src/rsz/src/Rebuffer.cc
index a7ed71319..dd8310f0f 100644
--- a/src/rsz/src/Rebuffer.cc
+++ b/src/rsz/src/Rebuffer.cc
@@ -330,7 +330,7 @@ BufferedNetSeq RepairSetup::addWireAndBuffer(const BufferedNetSeq& Z,
     bnet_wire->wireRC(corner, resizer_, layer_res, layer_cap);
     double wire_res = wire_length * layer_res;
     double wire_cap = wire_length * layer_cap;
-    double wire_delay = wire_res * wire_cap;
+    double wire_delay = wire_res * (wire_cap / 2 + p->cap());
     BufferedNetPtr z = make_shared<BufferedNet>(
         BufferedNetType::wire, wire_end, wire_layer, p, corner, resizer_);
     // account for wire load