Skip to main content
Be Bitwise

Information Storage in DNA

DNA & Molecular Computing · Frontiers track

8 min read

In this module

Nature's Storage Medium

So far we've covered information stored in silicon: charges in DRAM capacitors, magnetic orientations on platters, trapped electrons in flash cells. The densities are staggering by 1980s standards. A terabyte SSD fits in your palm.

Nature has been doing better for about three billion years. DNA — deoxyribonucleic acid — is the molecule that holds the genetic instructions for every living thing we know about, and as a storage medium it's frankly absurd. A single gram of DNA can theoretically hold around 215 petabytes of data. The whole public internet, several times over, in something the size of a sugar packet.

This lesson covers how to coax digital data into DNA, the encoding tricks that keep things readable, and where the technology actually sits today.

The DNA Molecule

A DNA molecule is a double-stranded helix composed of four types of nucleotide bases:

  • Adenine (A)
  • Thymine (T)
  • Guanine (G)
  • Cytosine (C)

Pairing follows strict rules: A binds T, G binds C. This Watson-Crick complementarity is what makes DNA both a stable archive and a copy-on-demand medium.

Think of a DNA strand as a string over a four-letter alphabet — ATGCCTAAGGT and so on. Each base sits roughly 0.34 nm along the backbone. Four possible values per position means log₂(4) = 2 bits per base.

text
5' ─ A─T─G─C─C─T─A─A─G─G─T ─ 3'
│ │ │ │ │ │ │ │ │ │ │
3' ─ T─A─C─G─G─A─T─T─C─C─A ─ 5'
Each base pair: ~0.34 nm
Each base: 2 bits of information
1 gram of DNA: ~2.15 × 10^17 bytes
DNA's Information Density

DNA holds about 10^19 bits per cubic centimetre. A modern SSD manages roughly 10^12 bits per cm³. That puts DNA about ten million times denser than the best solid-state storage on the market. The reason is simply that the bits live at the scale of single chemical bonds; there's nothing more compact short of running individual atoms as memory cells.

Encoding Digital Data in DNA

The simplest encoding maps pairs of bits to nucleotides:

text
00 → A
01 → T
10 → G
11 → C

Under this scheme the byte 0x4D (01001101 in binary) becomes T G C T.

The naive mapping won't survive contact with the lab. Synthesis and sequencing have their own biases and failure modes, and a real encoding has to plan for them:

Avoiding Homopolymers

Long runs of the same base (AAAAAAA) trip up both the synthesisers writing the strand and the sequencers reading it. Was that five As, or six? Sequencers genuinely struggle to count. Sensible encodings cap the run length so the question never comes up.

GC Content Balance

Strands skewed too far in GC content tend to fold back on themselves into hairpins or G-quadruplexes, which mess up both synthesis and sequencing. Production encodings keep GC content between 40% and 60% to avoid the molecular acrobatics.

Error-Correcting Codes

Every step — synthesis, storage, sequencing — introduces errors: insertions (extra bases), deletions (missing bases), substitutions (wrong base). Real DNA storage systems wrap the payload in error-correcting codes so corruption at any of those stages doesn't sink the file.

The systems that have actually worked at scale lean on fountain codes or Reed-Solomon variants:

  1. Break the data into small blocks
  2. Add redundancy (error-correction parity data)
  3. Encode each block as a DNA sequence (an oligo)
  4. Add an index sequence to each oligo so blocks can be reassembled in order
  5. Synthesize thousands of copies of each oligo for redundancy

A Practical Encoding: The Fountain Code Approach

The landmark 2017 work by Erlich and Zielinski at Columbia University used a fountain code (Luby Transform code) to encode data:

  1. The input file is divided into small data blocks
  2. The fountain code generates encoded droplets by XORing random subsets of data blocks
  3. Each droplet is mapped to a DNA sequence with balanced GC content and no long homopolymers
  4. An index and checksum are added to each sequence
  5. The sequences are synthesized as DNA oligos

To read the data back:

  1. The DNA is sequenced (multiple copies of each oligo are read)
  2. Majority voting corrects individual read errors
  3. The fountain code decoder recovers the original data blocks

The result hit 1.57 bits per nucleotide — close to the 2-bit theoretical ceiling — with perfect recovery of the original payload, projecting out to that headline figure of 215 petabytes per gram.

The DNA Storage Workflow

A complete DNA data storage cycle involves three steps. Here's the whole round trip first, with where each kind of damage creeps in and what the encoding does about it:

Every stage corrupts something; the redundancy added at the start is what pays for it at the end.

Writing (DNA Synthesis)

Synthesis means building DNA strands chemically, one base at a time. The dominant approach is oligonucleotide array synthesis: thousands of short sequences (oligos, generally 100–300 bases each) grown in parallel across a chip-like surface, addressed individually by the synthesiser.

Current synthesis technology:

  • Oligo length: 100-300 bases
  • Error rate: ~1 error per 200 bases (0.5%)
  • Cost: ~$0.001 per base (declining)
  • Speed: hours per synthesis run

A single megabyte demands millions of oligos, which today takes hours of wet-lab time and runs into the thousands of dollars.

Storage (Preservation)

Once synthesized, DNA can be stored in several ways:

  • In solution: dissolved in buffer at -20°C. Easy, but shelf life suffers at higher temperatures.
  • Lyophilised (freeze-dried): a dry powder that holds its integrity for decades at room temperature.
  • Encapsulated in silica: DNA sealed inside glass beads, essentially mimicking fossilisation. Estimated to last thousands of years, even when stored warm.

The proof of DNA's stubbornness is in the wild: readable strands have been recovered from woolly mammoths frozen for ten millennia, and from human remains older still.

Durability Beyond Any Electronic Medium

No electronic medium comes close on durability. Hard drives degrade in 5–10 years. Flash cells leak charge over decades. Archival tape gives you maybe 30 years on a good day. Properly stored DNA outlasts the lot, which is why archival storage is the obvious first market: government records, cultural heritage, scientific datasets — anything that has to outlive the storage technology itself without endless format-migration cycles.

Reading (DNA Sequencing)

Reading data back means sequencing the strands to recover the base order. A few competing technologies dominate the field:

Next-generation sequencing (NGS): Illumina-style platforms sequence millions of short fragments in parallel. Better than 99.9% accuracy, but a single run takes hours to days.

Nanopore sequencing: Oxford Nanopore's MinION threads a strand through a protein pore and watches the electrical signature each base produces. Minutes to hours, USB-powered, around 95% per-read accuracy. The portability is the headline feature.

Random access: One of DNA's nicer tricks is that you don't have to read the whole archive to retrieve one file. Give each file a unique pair of primer sequences, and PCR will selectively amplify just those strands out of a soup of millions. Functionally, it's a random-access read on a molecular library.

Current DNA Storage Milestones

The field has progressed rapidly from concept to demonstration:

2012: Church et al. stored a 659 KB book in DNA, establishing the basic encoding approach.

2017: Erlich and Zielinski achieved near-theoretical-maximum information density (1.57 bits/nucleotide) using fountain codes.

2018: Microsoft and the University of Washington stored 200 MB of data and demonstrated random access retrieval via PCR.

2019: The DNA Data Storage Alliance was formed by major technology companies (Microsoft, Western Digital, Twist Bioscience) to develop industry standards.

2021: Microsoft demonstrated the first fully automated DNA storage system - writing, storing, and reading data without manual intervention.

2023-2024: Multiple research groups demonstrated DNA storage in chip-based microfluidic systems, moving towards miniaturized, scalable storage devices.

DNA vs. Electronic Storage: A Comparison

PropertySSD (Flash)HDD (Magnetic)DNA
Density~10^12 bits/cm³~10^11 bits/cm³~10^19 bits/cm³
Durability5-10 years5-10 years1,000+ years
Write speedGB/s100s MB/sKB/s (current)
Read speedGB/s100s MB/sMB/s (current)
Write cost per GB~$0.10~$0.03~$1,000,000
Energy to maintainContinuous powerContinuous powerNone (passive)
Random access~0.1 ms~5 msMinutes (PCR)

DNA wins density and durability by orders of magnitude. It loses cost and speed by the same. Where DNA storage sits today is roughly where electronic storage sat in the 1950s: proven, demonstrated, nowhere near commodity. The whole question is how quickly synthesis and sequencing prices keep falling.

The Path Forward

DNA data storage is most likely to first be adopted for cold archival storage - data that is written once, stored for a very long time, and rarely (if ever) read. Examples include:

  • Government archives and legal records
  • Cultural heritage preservation (museum collections, historical documents)
  • Scientific datasets (genomic databases, astronomical surveys)
  • Enterprise data compliance (regulations requiring data retention for decades)

In all of these, the brutal write cost is amortised across decades or centuries of storage, and the zero-power, zero-migration footprint flips the total-cost-of-ownership calculation in DNA's favour.

The next lesson moves DNA from passive shelf to active substrate — how molecular reactions can act as logic gates and crunch real computational problems.