The Thermodynamics of Copying Information: Accuracy, Not Erasure, Sets the Price

Copying a bit, unlike erasing it, has no Landauer floor: copying a known bit onto a blank target is logically reversible, and the minimum copy energy is set by accuracy. In an antipodal discrimination model against Gaussian thermal noise the minimum is 1.31e-19 J per bit at an error probability of 1e-15 at 300 K. DNA replication copies at 8.30e-20 J per bit with a final error rate of 1e-10 per base (2.00004 ATP equivalents per base, including proofreading and mismatch-repair costs), within a factor of 1.92 of the accuracy floor per copied symbol. Silicon copying costs 4.50e-11 J per bit (DRAM to DRAM with SECDED error correction), 2.17e-10 J per bit (solid-state drive to drive) and 8.52e-9 J per bit (hard disk to hard disk), that is 3.45e8 to 7.03e10 times the floor, because a silicon bit is a macroscopic object. Model calculations are labelled SIMULATED, NOT MEASURED. The deposit includes the paper PDF, the computation script (compute.py) that produces every number in the paper, and its output (results.json).