Well I don't know exactly, but when we talk about dimesions in extraction, it implies how the capacitance will be extracted.
Say when we say a simple extraction, it might just consider overlap caps.
While a 3 D extraction, implies cap extraction in all direction, i.e. overlap caps,
fringing caps, side caps etc...
we just consider two interconnect lines (parallel) on metal 1
in 2.5D : when extract, the mutual and self capacitance is proportional with the length because
in 2.5D simulation we think the length is infinitely long and calculate the unit cap. and Res.
so in 2.5D, the accrucy is less than 3D, but higher then 2D
To my knowledge, there is nothing like 2.5D in mathematics.
Let me make it clear:
In 3D extraction the captable contains the capacitance values extracted from all the directions. Eg:fringe capacitance,edge cap,area cap,cross over cap,coupling cap.
In 2D we have only area cap and crossover cap and coupling cap.
But in 2.5D it can have 2D cap values and only few 3D values i.e.., it may miss either edge cap or fringe cap.
that is why 3D is more accurate.