The Dissipative Decoder: Thermodynamic Cost Bounds on the Serial Decoding Throughput Basin — and Why Silicon Escapes Them

Serial decoding systems—ribosome, cognition, language, and artificial intelligence—converge to 3–6 bits per processing event. This study models serial decoding as maximizing net information return—bits gained minus discrimination cost—and shows that the size of the optimal alphabet is set by how discrimination cost scales. Two cost regimes are identified. In Regime A (alphabet-bound), biological pairwise molecular recognition makes discrimination expensive and quadratic in alphabet size; the net-return optimum M* = (αγ ln 2)−1/α then falls near M = 20 amino acids for the estimated biological cost parameters, placing biology inside the 3–6 bit basin. In Regime B (capacity-bound), silicon discrimination is cheap and scales sub-linearly (α < 1; a preliminary log-log fit across an energy benchmark gives a sub-linear exponent, with the full dataset to be released), so the same objective places the optimum at a far larger alphabet and silicon is not confined to the basin. The optimum is set by the discrimination-cost structure, not a parameter-free constant. The ribosome sits essentially on its informational (rate-distortion) floor—zero slack in bits—but thermodynamically it dissipates roughly 25× the Landauer minimum (~80 kT to encode ~4.39 bits vs. a ~3 kT floor); a 7-billion-parameter transformer operates approximately 109 above its Landauer floor. A temperature prediction tested across 29 organisms yields partial correlation r = −0.451 (p = 0.014) between amino acid entropy and growth temperature, controlling for GC-content. The basin is a speed limit for pairwise discriminators; silicon escapes it entirely. The two-regime framework resolves why AI lands in the biological throughput basin despite having no thermodynamic constraint on vocabulary size.