Maynooth’s DNA Computer Reaches Its Answer by Cooling Down

Researchers at Maynooth University in Ireland have built a DNA computer that multiplies, divides and adds by letting molecules settle into their lowest-energy arrangement, according to a paper published in Nature on 16 September 2026. The team ran ten programs on the system, including multiplication by three, division by two, 8-bit parity detection and addition of 25-bit numbers, which the authors describe as a 100-bit computation.

The paper, “A thermodynamically favoured molecular computer”, has Tristan Stérin and Abeer Eshra as joint first authors, with Constantine Evans, Janet Adio and Damien Woods of the university’s Hamilton Institute among the co-authors. The recipe is almost domestic. Water, salt, short DNA strands and a longer DNA scaffold go into a test tube, which is heated and then cooled. The university’s release says the system needs no continuous electricity supply, only a little heat to start the computation.

Why the right answer is the stable one

The authors call the design a Scaffolded DNA Computer. A single scaffold strand carries a row of unique binding positions. Compute tiles attach to those positions and to their neighbours, and in this implementation each tile carries three bits of program and data on each side. A program and its input are simply a chosen subset of tiles mixed with the scaffold. A mismatch between neighbouring tiles counts as an algorithmic error and carries an energy cost, so the mixture tends to swap the wrong tile out.

The paper sets out three conditions that make the correct output energetically favourable: compute tiles are supplied in excess, typically ten times the scaffold concentration, so every position gets filled; correct bindings between neighbours are favoured over errors; and each bond is weak enough to come apart and re-form, which allows errors to be corrected. The abstract argues that this is why the system does not need separate error correction or precise kinetic control, and why it should be programmable and scalable.

Thirty seconds for 10 plus 3, fourteen hours for millions

Speed is where the claims need the most care. The university says a calculation of 10 plus 3 took 30 seconds, while adding numbers in the range of about 11 million to 34 million took up to 14 hours. The paper’s abstract says small instances run in under a minute and that the programs can be reused dozens of times; the release says the system performed up to 25 different calculations in a row. Eshra did not pretend it competes with a laptop.

That ranking against other DNA computers is the team’s own comparison. The release offers no independent benchmark across DNA computing systems to support it.

From origami to arithmetic

The work sits in a lineage that Nature’s research briefing on the paper traces through Erik Winfree’s 1998 thesis on algorithmic self-assembly of DNA and Paul Rothemund’s 2006 Nature paper on DNA origami. The Maynooth group’s wider programme is funded by a €4 million European Innovation Council grant for DISCO, a project on DNA-based storage and computation led by Woods, according to the university. It points to long-term data storage and, eventually, molecular systems that could work inside cells to detect disease as possible uses.

The honest reading is that this matters as a design principle rather than as a machine. A computer that is correct because correctness is the stable state is a different idea from one that has to be steered through each reaction step,. What the paper does not show is a problem on which a molecular computer beats silicon, or a demonstration far beyond 100 bits, and Woods is candid about that: “This is blue skies science. We don’t know where the future is going to take us.”