"Reducing AI Power Consumption with 'DNA × Semiconductors'? A New Type of Memory Fusing Biological and Electronic Elements"

As generative AI rapidly spreads, there is a growing problem becoming more serious behind the scenes.

Electricity.

AI does not function solely based on the performance of the model itself. It repeatedly reads large amounts of data from memory, sends it to the processing unit, and writes back the calculation results. This process is repeated an enormous number of times.

Therefore, in next-generation computers, it is just as important to "accelerate calculations" as it is to "store and transfer data with minimal power."

Amidst this, a study has emerged that might fundamentally change computer memory.

The material being used is none other than DNA.

A research team from Pennsylvania State University and others have developed a new memory device that operates at very low voltage by combining artificially designed synthetic DNA with a semiconductor material called "perovskite."

In this study, published in Advanced Functional Materials in 2026, DNA is used not merely as a "molecule that records life information" but as a "functional material" for constructing electronic circuits.

Biology and semiconductor engineering.

These two fields, which have mostly developed separately until now, are beginning to intersect seriously.

DNA has been called the "ultimate storage medium"

When people hear DNA, many think of genetic information.

It is a molecule that stores biological information through sequences of four bases: A, T, G, and C.

And DNA has another remarkable feature.

Its information density is extremely high.

According to Pennsylvania State University's announcement, theoretically, 1 gram of DNA could potentially store about 215 million GB of data.

This is an information density on a completely different level from current SSDs and HDDs.

For this reason, research on converting digital data into DNA base sequences for long-term storage, known as "DNA storage," has been advancing worldwide.

However, there is one important point to understand about this research.

The device developed this time is slightly different from the typical "DNA storage" that stores photos and videos as sequences of A, T, G, and C within DNA.

It utilizes DNA itself as a material for electronic devices.

The research team focused on the property that DNA can be artificially designed.

"Synthetic DNA" designed for electronic circuits, not "natural DNA"

Natural DNA extracted from organisms is very long.

According to the researchers, it tends to tangle like "wet spaghetti" when handled.

This property is difficult to manage when you want to precisely position electronic components at the nanometer scale.

Therefore, the research team used short synthetic DNA.

The paper uses synthetic DNA 22 bases long, incorporating silver nanoparticles.

Short DNA can be easily designed to fit the purpose in terms of length, base sequence, and molecular structure.

In other words, DNA is treated as

"something extracted from organisms and used"

rather than

"a nanomaterial designed on a computer and manufactured with the necessary properties."

Furthermore, the research team added silver nanoparticles to this DNA.

This is a process known as doping.

This adjusted the electrical properties of the DNA layer, making it usable as a path for electron flow.

Combined with this synthetic DNA containing silver nanoparticles was a quasi-2D structure halide perovskite.

Perovskite is a material also noted for use in solar cells, and it is being researched as a next-generation memory material due to its controllable electrical properties.

Interestingly, neither DNA alone nor perovskite alone achieved the same performance.

The combination of these two disparate materials resulted in stable current paths and memory characteristics.

This is truly a "bio-hybrid" electronic device.

The developed device is a "memristor"

The device studied this time is called a "memristor."

The name combines Memory and Resistor, and in Japanese, it is translated as "memory resistor" or similar.

A regular resistor is an electronic component that determines how much current flows.

In contrast, a memristor changes its resistance state based on the current that has flowed in the past and can retain that state.

In other words, the electronic component itself can "remember what kind of electrical signals it received before."

This is a major reason why AI researchers are paying attention to it.

In current general-purpose computers, the processor responsible for computation and the memory that stores data are separate.

When CPUs or GPUs perform calculations,

they read data from memory
↓
send it to the processor
↓
perform calculations
↓
and return the results to memory again

This process occurs repeatedly.

In large-scale AI computations, this "data transport task" itself consumes time and power.

With memristors, there is the potential to perform memory and computation in close proximity or even on the same element.

This leads to next-generation computer technologies known as "in-memory computing" and "neuromorphic computing."

Low voltage below 0.1 volts

The particularly noteworthy figure in this study is "below 0.1V."

According to the paper, the developed memristor could switch resistance states at a very low voltage of less than 0.1V.

Furthermore, the research team reported a low power density of 0.01W/cm².

Durability tests showed about 1000 operation cycles, and the electrical difference between the ON and OFF states exceeded 100,000 times.

It maintained a high ON/OFF ratio for over six weeks even under room temperature and atmospheric conditions.

In perovskite-based devices, stability is a major challenge, so the point that "adding DNA improved not only performance but also stability" is an important part of this research.

The university's announcement also explained that it consumes "100 times less power" compared to conventional technology, and this figure was widely reported through outlets like ScienceDaily.

However, this needs to be read carefully.

The same university announcement also contains a statement that "the same memory function can be achieved with one-tenth the power."

Moreover, the central points highlighted in the abstract of the peer-reviewed paper are the operating voltage below 0.1V, power density of 0.01W/cm², durability, and retention characteristics.

Therefore, it is premature to interpret it as "current SSDs and DRAMs can be reduced to one-hundredth of the power."

The "100 times" figure should be seen as a comparative value indicating the impact of the research results, and it is necessary to see which existing technology and conditions it is compared with.

Why it pairs well with AI

The reason this research is not just about a new type of memory is its compatibility with AI.

Currently, neural networks are widely used in AI.

Artificial neural networks are inspired by the neural circuits of the brain, but the current mainstream AI chips do not calculate in the same way as the human brain.

In the brain, synapses, which are the connections between neurons, are involved in information processing and memory.

In contrast, in general computers, memory devices and processing units are separated.

This is where neuromorphic computing is being researched.

If the resistance value of a memristor is likened to the "strength of the connection" of an artificial synapse, it may be possible to execute part of the massive matrix calculations directly within the memory.

What becomes important here is power consumption.

If you increase the number of artificial synapses to millions, billions, or beyond, it will not be practical if each requires a large amount of power.

This is why the DNA-perovskite type memristor, which can change states at very low voltage, is attracting attention.

In the future,

AI accelerators
edge AI
autonomous sensors
robots
wearable devices
neuromorphic processors

and other devices that process large amounts of information with limited power could potentially apply this technology.

"DNA can store large amounts of data" is slightly misleading

It is important to address a particularly easy-to-misunderstand part of this news.

The statement "DNA can store 215 million GB per gram"

and

"the DNA memristor created this time"

are directly connected to the idea that

"a semiconductor that can store 215 million GB per gram has been completed"

is not accurate.

In DNA storage, digital information is written directly into the sequence of bases A, T, G, and C.

In this device, synthetic DNA is combined with silver nanoparticles and used as a material to control charge transport and resistance switching.

The essence of the research is

not "storing files within DNA,"

but rather

"incorporating DNA as a programmable electronic material into semiconductors."

This may hold greater significance in the long term.

Until now, semiconductor materials have primarily consisted of silicon, oxides, metals, and various compounds.

If biomolecules like DNA are seriously added to this, a new world of electronic materials that can be designed at the molecular level may open up.

On social media, it is perceived as "science fiction becoming reality"

This research is being introduced on social media in the context of "biology meets electronics."

On X, posts emphasize the point of creating ultra-low power memory from synthetic DNA and semiconductors, along with the article title from ScienceDaily.

On LinkedIn, Penn State Research introduces the research results as "next-generation technology for low-energy, high-capacity data storage."

Other technology-related accounts also describe it in future-oriented terms such as

"Nature has stored information in DNA for billions of years,"

"Research has finally emerged to connect DNA to electronic circuits,"

and "It could change the power efficiency of AI hardware."

Particularly on social media, the keywords "DNA," "100 times power saving," and "AI" easily gather reactions.

The fusion of seemingly distant fields like life sciences and computers is perceived as news that evokes an SF-like future.

On the other hand, "DNA won't be in your PC tomorrow"

However, a closer look at social media introductions reveals that it's not all about expectations.

Technical posts on LinkedIn, for example, introduce this research while clearly stating that it is "still at the proof of concept stage" and "not yet a commercial product."

This is an important perspective.

The approximately 1000 durability cycles reported in the paper are meaningful for confirming performance as a research device but cannot be directly compared to the conditions required for actual computer memory.

The same goes for the stability over six weeks.

There is a significant gap between the achievement of "more stable than before" in research and the claim of "usable as a commercial semiconductor that operates for years."

Furthermore, considering mass production,

issues such as whether DNA materials can be uniformly formed,

whether the distribution of silver nanoparticles can be controlled,

whether the same performance can be achieved with a large number of memory cells,

whether the yield of the manufacturing process can be secured,

whether it can be integrated with existing semiconductor manufacturing processes,

and whether performance can be maintained over a long period

must be resolved.

While it is gathering great expectations as the "computer of the future" on social media, what researchers and engineers truly focus on is how far these challenges can be overcome from now on.

It should be noted that the SNS posts confirmed this time are a part of the publicly searchable posts and are not statistically analyzed across the entire SNS. Therefore, it should be viewed as a representative reception seen in public posts, not as "the majority of SNS users think this way."##HTML