Power of fast VLSI models is insensitive to wires' thinness

VLSI f-models which allow the switching time to decrease to f(D) when the length of all wires is restricted by D are called 'fast' if the decrease is slightly superlinear. The fast models are so strong and robust that their computational power cannot be increased by and combination of the following: (1) making zero the width of each wire of length d, except for its log d segment, thus eliminating layout and area considerations; (2) allowing wires to transmit log d bits simultaneously; (3) making the switching time f(d) of each node depend only on the length d of its own input wires, thus enabling small subcircuits to run faster; (4) changing f while preserving Sigma /sub k/ 1/f(k); (5) enabling the nodes to change connections arbitrarily in the run time. The authors construct a kind of operating system link server (linx, for short) that simulates all these powers online. The condition of superlinearity cannot be weakened.<
>