In the 1960s, IBM engineer E.F. Rent noticed something weird while tallying pin counts on their computer modules: no matter what logic block he looked at, the number of external pins followed a power law of the number of internal gates. He never published it, but Landman and Russo formalized it in 1971, and it's been the most stubbornly accurate empirical law in digital design ever since.

The rule: T = k · G^p, where T is the number of terminals (pins/wires crossing the boundary), G is the number of gates inside, k is a constant near 1–4, and p is the "Rent exponent" — typically 0.5 to 0.75 for real logic.

Why it matters: Rent's Rule predicts wiring demand. If you double the gates in a block, you don't get double the I/O — you get 2^p times the I/O. With p = 0.6, doubling gates means 1.52× more pins. This sub-linear scaling is why chips can pack billions of transistors behind a few thousand pins.

Concrete example — a microprocessor die: A modern CPU core has roughly 100 million gates and around 2000 signal pins (excluding power/ground). Plug that in: 2000 = k · (10^8)^p. With k = 3, that gives p ≈ 0.35 — unusually low, because CPUs are highly locally-connected. Memory arrays have even lower p (near 0.1), while random logic like crossbars pushes p toward 0.75. A truly random interconnect would have p = 1.0.

Practical consequence — floorplanning and hierarchy: Rent's Rule tells you how many wires must cross every hierarchical boundary. When you partition a design, the tool estimates required routing tracks per boundary using p. If you exceed the available tracks, you get congestion, detours, and timing failures. A high-p block (crossbar switch, register file) needs to be placed where the metal stack has bandwidth to feed it.

Rule of thumb: For any digital block you're designing, estimate pin count as T ≈ 4 · G^0.6. A 10,000-gate block wants ~250 pins. A 1M-gate subsystem wants ~4,000. If your architecture demands more, you have a communication-bound design; if far fewer, you probably have deep pipelines with narrow interfaces (good for timing, bad for throughput).

Rent's Rule also predicts wire length distributions, which drives the RC delay budget in place-and-route — the same p exponent shows up in Donath's wire-length estimator, linking floorplan quality directly back to this 60-year-old empirical observation.