Biocomputing and DNA computing

DNA stores information at a density no engineered medium approaches, because the information sits in the covalent structure of a molecule about two nanometres wide. A gram of DNA can in principle hold on the order of hundreds of petabytes, and dry DNA kept cool and dark remains readable after centuries — sequences have been recovered from permafrost specimens tens of thousands of years old. Those two properties, density and longevity, are the entire case for the technology. Everything else is a disadvantage.

Writing is the bottleneck

Reading DNA has become cheap; writing it has not. Phosphoramidite synthesis adds one base per chemical cycle at a stepwise yield below unity, so the fraction of full-length product falls geometrically with length. That is why storage schemes chunk data into oligonucleotides of roughly a hundred to a few hundred bases, each carrying an index, rather than writing one long strand. Cost per base has fallen far more slowly than sequencing cost, and the gap — many orders of magnitude between synthesis and any conventional write operation — is the reason DNA storage is not competitive for anything but cold archival data that is written once and read rarely. Enzymatic synthesis using template-independent polymerases is the main route being pursued to change this; it is not yet a settled result.

The error model has two sources

Synthesis produces substitutions and, characteristically, insertions and deletions from failed or double couplings; homopolymer runs are worse. Sequencing adds its own errors, different in kind by platform. Indels are the difficult class, because they shift the reading frame of the encoded payload and break the byte alignment that classical block codes assume. Practical systems therefore combine constrained encodings — avoiding long homopolymers and extreme GC content, which are hard to synthesise and hard to sequence — with outer erasure codes across many oligos plus inner codes within each, and rely on reading each fragment many times so that a consensus can be formed. This works: published demonstrations have recovered files intact from pools with substantial per-base error and large fractions of sequences missing altogether.

Random access and the destructive read

A DNA pool is not addressable. Retrieval works by PCR with primers unique to a file, which amplifies the wanted subset out of the mixture — so the address space is limited by how many primer pairs can coexist without cross-hybridising, and every access consumes some of the sample and requires re-amplification. Latency is set by chemistry and sequencing runtime, measured in hours, not milliseconds.

Where it is not competitive

Against magnetic tape, DNA loses on cost per byte written, on write throughput, and on access time by very large margins; it wins on volumetric density and on shelf life without periodic migration. Tape libraries require re-copying every decade or so as formats and media age, and that recurring cost is the honest comparison point. DNA is a candidate for archives that are enormous, rarely read, and expected to outlive several generations of hardware. It is not a candidate for a working store, and claims that it will replace conventional storage generally ignore the write-cost gap.

Related work in genetic circuits — logic implemented in transcription and translation inside living cells — is a separate strand, useful where the computation must happen where the cell is, not where speed matters.