A Thermodynamically Favoured Molecular Computer: Robust, Fast, Renewable, Scalable

Like life, computers are out-of-equilibrium.^1,2 Thermodynamically favoured error states are thwarted by energetically-costly processes such as kinetic proofreading of biological polymers, error-correcting codes in computer data storage, and redundancy in molecular programming. Decades of theoretical work shows that unlike life thermodynamic computers can operate by drifting naturally to equilibrium.^3,4 Similar ideas underlie machine learning models⁵ and search algorithms such as simulated annealing,⁶ although executed on non-equilibrium architectures at enormous energy cost.⁷ Physically implementing thermodynamically favoured computation is a decades-long challenge that could reduce dependence on fuel-consuming error-correction and precise kinetic control. Here, we demonstrate a thermodynamically favoured Scaffolded DNA Computer (SDC) on ten programs including MULTIPLICATION-by-3, D<sc>ivision</sc>-by-2, 8-bit P<sc>arity</sc>-detection, and A<sc>ddition</sc> of up to 25-bit numbers. SDC algorithms have simple experimental protocols, can be reused dozens of times and small instances run in under a minute. The SDC is grounded in mathematical, physical and computer science principles that explain why the output is thermodynamically favoured, why no error-correction nor precise step-by-step kinetic control are required and how it is programmable and scalable. This work creates a new way to think about equilibrium computation in all manner of synthetic systems.