A DNA-based memory device uses 100x less power, Penn State team reports

The electricity bill for running AI has moved from an engineering footnote to a board-level concern at the largest cloud companies, and a measurable slice of that bill goes to memory: the chips that hold the weights a model needs to answer one question, and the storage that archives every prompt and completion. A Penn State team has now built an experimental device that attacks the memory problem from an unusual angle. They combined synthetic DNA with a crystalline semiconductor called perovskite to make a bio-hybrid memristor, a memory resistor that holds its state without continuous power.

According to the researchers, the device draws roughly 1% of the power that conventional memory would need for the same task. That 100x figure is the team's own characterization of a single in-vitro device, not a benchmarked product result, and the work has not yet been replicated outside the Penn State lab. The team has filed a patent application, and the paper appears in Advanced Functional Materials, a peer-reviewed Wiley journal. ScienceDaily and other outlets carried the same university release.

A memristor is a resistor that remembers. Apply a voltage, and the resistance shifts in proportion to how much charge has passed through; remove the power, and the new resistance holds. That property lets a memristor store a bit in its physical state rather than in a circuit that must be refreshed, which is why the device class has drawn interest as a low-power memory candidate for years. The Penn State advance is in the materials, not the underlying physics. The team engineered short synthetic DNA sequences tailored for electronic requirements and paired them with a perovskite, a class of cheap, versatile crystalline semiconductors already used in solar cells and lasers. DNA handles the charge-storage state; perovskite handles the electronic interface.

Biology and silicon have historically not mixed well. Most prior attempts to use DNA for data storage have relied on enzymatic sequencing to read back the bits, a slow, lab-bound process. Putting DNA inside an electronic device, where it can be written and read with standard semiconductor tooling, is the engineering problem Kavya S. Keremane, a postdoctoral researcher in materials science and engineering at Penn State and co-corresponding author, and her collaborators tried to solve. A Penn State release describes the work as borrowing biology to power next-generation data storage.

DNA's theoretical information density is roughly 215 million gigabytes per gram, a property of the molecule itself, while the 100x power figure is the Penn State team's measurement of their own single device. Conflating the two makes the result look more capable than the published paper supports.

DNA data storage is not new as a research program. Academic and industry labs have been writing bits into synthesized DNA for over a decade, and Microsoft has shipped exploratory reads from DNA archives. The Penn State contribution is the bio-hybrid interface, getting synthetic DNA into a device that an electronic system can actually address, and pairing it with perovskite, a material class already in volume production for solar cells. The 100x claim will need independent replication on working arrays before anyone can price it against DRAM or flash on a per-bit basis.

The next milestone is replication outside Penn State and a path from a single device to a small array, the standard hurdle for any new memory class. The team says work continues on scaling; the published result is the materials platform, not yet a chip.