Single-Electron Memory: Potential Uses in AI Chips
Artificial intelligence has created a strange problem for the technology industry.
Processors are becoming extraordinarily powerful, yet moving and storing the data those processors need is increasingly becoming a bottleneck.
Modern AI systems constantly transfer enormous amounts of information between processors and different types of memory. Every transfer takes time and consumes electricity. As AI models become larger and data centers become more powerful, those small inefficiencies multiply across billions of operations.
Researchers around the world are therefore searching for a better form of memory.
A recent breakthrough from China may point toward one possible answer.
Researchers at Fudan University have demonstrated a single-electron memory device capable of storing one bit of information using just one electron at room temperature.
The research, published in Science in July 2026, uses an atomically thin two-dimensional semiconductor structure. Fudan University says the device produced a storage window of approximately 0.5 volts from a single electron while retaining the information without continuous power.
That is significant because one electron represents the theoretical minimum amount of charge that can be used to represent a bit.
The technology has been nicknamed Guiyi, roughly meaning “return to one,” reflecting the idea of reducing information storage to a single electron.
But the breakthrough is not important simply because the device is incredibly small.
It attacks several problems that are becoming increasingly important in the AI era:
memory density, speed, energy consumption and data movement.
If researchers can eventually manufacture this technology reliably at scale, it could contribute to a new generation of memory systems capable of storing much more information while consuming less energy.
There is a very large “if” attached to that statement.
A successful laboratory device is very different from a commercially manufactured semiconductor chip.
But the research provides an important glimpse into how radically computing hardware may need to change as conventional semiconductor technologies approach physical limits.
Here are seven reasons single-electron memory deserves attention.
1. Researchers Have Reached the Smallest Possible Unit of Charge Storage
The easiest way to understand the breakthrough is to think about how electronic memory stores information.
Computers ultimately represent information using bits:
0 or 1.
Modern memory devices use electrical states to distinguish between those values.
But those states normally involve many electrons.
Engineers have spent decades shrinking semiconductor devices so more information can fit into smaller areas.
Eventually, however, shrinking encounters a fundamental problem.
Matter is made of particles.
You cannot indefinitely divide electrical charge into smaller pieces.
An electron is an elementary unit of charge.
That makes one electron per bit a theoretical limit for charge-based information storage.
Researchers have pursued single-electron devices for decades, but operating them reliably under practical conditions has been extremely difficult.
Earlier experiments produced signals that were too weak or unstable for useful memory.
Fudan University compares the challenge to detecting the effect of a single drop of water entering a large reservoir.
The new device takes a different approach.
The research team used the atomic-scale properties of two-dimensional semiconductor materials and developed a planar device architecture designed to control individual electrons.
According to Fudan, injecting only one electron generated a 0.5-volt memory window, while the stored state remained stable at approximately 27°C.
That is particularly important because previous single-electron demonstrations often required extremely low temperatures or produced signals too weak for practical operation.
The researchers therefore did more than simply detect an electron.
They demonstrated room-temperature non-volatile single-electron storage.
Non-volatile means the information remains stored even after power is removed.
That puts the concept closer to flash memory than temporary working memory such as DRAM.
The scientific importance is clear.
For charge-based memory, researchers have effectively reached the smallest possible carrier of stored information.
The engineering challenge now becomes turning that physical limit into something manufacturers can use.
2. Two-Dimensional Materials Could Help Chips Keep Shrinking
The breakthrough also demonstrates why two-dimensional materials are attracting so much attention in semiconductor research.
Modern silicon transistors have become astonishingly small.
But shrinking conventional devices becomes increasingly difficult because electrical behavior changes at tiny scales.
Leakage, heat and quantum effects become harder to control.
Two-dimensional materials offer a different approach.
They can be only a few atoms—or even one atomic layer—thick.
That extreme thinness allows engineers to manipulate electrical properties at dimensions difficult to achieve using conventional bulk materials.
Fudan’s memory architecture uses these atomic-scale characteristics to confine and manipulate electrons.
The researchers also incorporated graphene, another two-dimensional material famous for its unusual electrical properties.
Graphene allows electrons to move with very low resistance, making it useful in designs where fast charge transport matters.
The result is not simply a smaller version of today’s flash memory.
It represents a different device architecture designed around the physics of atomically thin materials.
This could become increasingly important as the semiconductor industry approaches the limits of traditional scaling.
For decades, computing improved partly because manufacturers repeatedly made transistors smaller.
More transistors could fit onto a chip.
Performance increased.
Cost per computation declined.
But continuing that pattern is becoming extraordinarily difficult and expensive.
Future improvements may therefore depend more heavily on:
new materials,
new memory architectures,
advanced packaging,
3D integration,
specialized processors,
and different ways of moving data.
Single-electron memory fits directly into this wider transition.
Instead of simply shrinking the old architecture, researchers are asking whether the architecture itself should change.
3. AI Has Turned Memory Into a Major Computing Bottleneck
The timing of this breakthrough matters because artificial intelligence has changed what computers need from memory.
Modern AI accelerators can perform enormous numbers of calculations.
But processors cannot calculate with data they do not have.
Information must constantly move between storage, memory and computing units.
That movement creates what engineers often call the memory wall.
Processor performance can increase faster than the speed at which data reaches the processor.
Imagine building an extraordinarily powerful factory but delivering raw materials through a narrow road.
Making the factory faster eventually stops helping because it spends too much time waiting for supplies.
Something similar happens in computing.
AI makes the problem particularly severe.
Large models involve enormous numbers of parameters and intermediate calculations. Serving millions of AI users means repeatedly moving vast quantities of information.
That is why high-bandwidth memory has become so important to modern AI accelerators.
It is also why the AI memory shortage has become an important part of the broader AI hardware story.
Single-electron memory does not replace high-bandwidth memory tomorrow.
The technologies serve different purposes and exist at radically different stages of maturity.
But the Fudan research demonstrates how scientists are rethinking memory around the demands of future computing.
Fudan argues that improving memory could reduce the delays and energy costs associated with moving data between storage and computing units.
This is important because simply building faster processors cannot solve every AI performance problem.
The entire computing system needs to improve.
That includes:
processors,
memory,
networking,
storage,
cooling,
and software.
Our analysis of the rise of AI factories shows how modern AI infrastructure increasingly behaves like one integrated machine.
Memory is one of the most critical parts of that machine.
4. Single-Electron Memory Could Dramatically Reduce Energy Use
The AI boom is creating another problem:
electricity consumption.
AI data centers contain huge numbers of processors working continuously.
But processors are not responsible for all of the energy consumption.
Moving data also requires power.
Every time information travels between different levels of memory and processing hardware, energy is consumed.
At the scale of modern computing, those transfers matter enormously.
Single-electron memory attacks the problem at an unusually fundamental level.
If information can be represented using less electrical charge, the theoretical amount of energy required for storage operations can also decline.
The Fudan researchers say being able to store one bit by changing the state of a single electron could significantly reduce power consumption while supporting much higher storage density.
That does not mean a future single-electron memory chip would consume virtually no electricity.
A real semiconductor system contains far more than memory cells.
Circuits must read and write information.
Controllers manage operations.
Signals move across interconnects.
Errors need to be handled.
Packaging consumes space and energy.
But improving the efficiency of the fundamental storage element can still create large benefits when multiplied across billions or trillions of operations.
Energy efficiency is becoming increasingly valuable because the AI infrastructure spending boom is running into physical constraints.
AI companies need more than GPUs.
They need electricity.
They need grid connections.
They need cooling.
They need data-center capacity.
A memory architecture capable of reducing energy per operation could therefore become valuable even if its primary advantage were not raw speed.
In the future of computing, performance per watt may matter almost as much as absolute performance.
5. Non-Volatile Memory Could Change How Computers Are Designed
Another important characteristic of the Fudan device is non-volatility.
Today’s computers use several layers of memory because different technologies offer different advantages.
DRAM is fast and useful as working memory, but it is volatile.
Turn off the power and the stored information disappears.
NAND flash preserves information without electricity and provides enormous storage capacity, but it operates differently and is generally slower.
Processors also use extremely fast caches located closer to computing cores.
This creates a hierarchy:
cache → working memory → storage.
Data constantly moves between these layers.
Each movement adds latency and consumes energy.
Researchers have long searched for memory technologies capable of combining more of the desirable properties of today’s separate systems:
high speed,
high density,
low power,
long retention,
and endurance.
No technology perfectly combines all of them.
Fudan’s single-electron device is interesting because it combines extremely small charge storage with non-volatile behavior at room temperature.
The same research team has been working toward this goal through several related breakthroughs.
In April 2025, it reported a two-dimensional flash-memory prototype with programming speeds of approximately 400 picoseconds.
Later that year, the team demonstrated a full-featured 2D flash chip integrated with mature silicon CMOS technology. Fudan reported a memory-cell yield of 94.3% in that prototype.
Fudan University’s earlier 2D flash-chip breakthrough
The new single-electron result builds on that broader research program.
That progression matters.
It shows researchers are not studying one isolated quantum effect.
They are trying to move 2D memory from individual devices toward integrated systems.
If future non-volatile memory became fast enough to operate closer to processors, computing architectures could potentially reduce some of the expensive movement between separate storage layers.
That could be especially useful for edge AI devices, where energy and physical space are limited.
6. Future Phones and Computers Could Run Larger AI Models Locally
Most discussion around AI hardware focuses on giant data centers.
But advanced memory could also change devices we carry every day.
AI is increasingly moving onto:
smartphones,
laptops,
vehicles,
robots,
wearable devices,
and industrial equipment.
Running AI locally has several advantages.
It can reduce latency.
It can allow some functions to work without an internet connection.
Sensitive information can remain on the device.
Cloud-computing costs can fall.
But local AI faces a major limitation:
hardware resources.
A smartphone cannot contain the same memory and computing capacity as a hyperscale data center.
That makes efficiency extremely important.
Higher-density memory could allow devices to store more model data in the same physical area.
Lower energy consumption could extend battery life.
Faster memory could help processors access model information more efficiently.
Fudan researchers have suggested that future versions of their technology could eventually support larger local AI models and longer contextual memory in devices, although those possibilities remain long-term goals rather than commercial capabilities today.
This distinction is important.
The current breakthrough does not mean next year’s smartphone will contain single-electron flash memory.
Moving from a research device to billions of reliable memory cells is an enormous challenge.
But the direction fits a broader technology trend.
AI is moving from centralized cloud systems toward a combination of cloud and edge computing.
The more intelligence moves onto local devices, the more valuable dense, efficient memory becomes.
That creates a potential market far beyond data centers.
7. Manufacturing Will Determine Whether the Breakthrough Actually Matters
This is the most important reality check.
Semiconductor laboratories regularly produce extraordinary breakthroughs.
Most never become commercial products.
A memory device can demonstrate excellent properties under controlled conditions and still fail because manufacturers cannot produce billions of identical cells economically.
Modern semiconductor manufacturing demands astonishing reliability.
A commercial memory chip may contain billions of storage elements.
If too many behave differently, the product becomes unusable.
Manufacturers therefore need:
consistent materials,
extremely high yields,
long-term reliability,
repeatable electrical behavior,
competitive costs,
and compatibility with existing fabrication processes.
This is where single-electron memory faces its biggest challenge.
The Fudan team says its broader 2D-memory work has already demonstrated integration with conventional silicon CMOS processes, which is encouraging.
But scaling a single-electron architecture remains a much larger test.
Researchers need to demonstrate that huge arrays of these devices can behave consistently.
They need to determine endurance.
They need to test long-term retention.
They need to establish manufacturing yields.
They need to integrate control electronics.
And ultimately, the economics need to compete with an enormous existing memory industry.
Fudan says it plans to push the technology toward engineering and large-scale chip integration. The university has discussed commercialization efforts over the next several years.
That ambition deserves attention, but commercialization timelines for new semiconductor technologies frequently slip.
The history of chips is filled with promising devices that worked scientifically but proved too difficult or expensive to manufacture.
That is why the correct interpretation of this breakthrough is not:
“Flash memory has just been replaced.”
It is:
“Researchers have demonstrated a new physical approach that could become important if it can be scaled.”
The difference matters.
Could Single-Electron Memory Replace NAND and DRAM?
Not anytime soon.
Today’s memory industry operates at enormous scale.
NAND flash manufacturers produce highly advanced 3D architectures containing many stacked layers.
DRAM has benefited from decades of manufacturing optimization.
High-bandwidth memory is rapidly expanding because AI accelerators need extraordinary data throughput.
New memory technologies therefore face a difficult benchmark.
They cannot simply be interesting.
They must eventually offer enough improvement in:
cost,
speed,
density,
power,
reliability,
or integration
to justify changing mature manufacturing ecosystems.
Single-electron memory may initially find specialized applications rather than immediately replacing every existing memory type.
Alternatively, some of the underlying 2D technologies could be incorporated into hybrid architectures.
The semiconductor industry often evolves this way.
New technologies rarely replace everything at once.
They enter where their particular advantages matter most.
Why This Breakthrough Matters for the Semiconductor Race
The research also has geopolitical significance.
Semiconductors have become strategic technology.
The United States, China, South Korea, Taiwan, Japan and Europe are investing heavily in chips because computing capability affects AI, manufacturing, communications, defense and economic competitiveness.
China faces restrictions on access to some advanced semiconductor technologies.
That gives Chinese researchers a particularly strong incentive to develop alternative architectures and domestic intellectual property.
Single-electron memory does not eliminate China’s broader semiconductor challenges.
But original breakthroughs in new materials can create opportunities to compete in areas where the technology landscape is not yet mature.
This mirrors what we see across the broader global race for AI leadership.
The competition is no longer only about who has the best AI model.
It increasingly involves:
chips,
memory,
energy,
data centers,
advanced materials,
and manufacturing capacity.
The countries capable of controlling more of that technology stack could gain an important economic advantage.
What Happens Next?
The next phase will be less glamorous than the original scientific breakthrough.
Researchers need to turn a successful device into increasingly large arrays.
Then they need to integrate those arrays with conventional electronics.
Manufacturing yields must improve.
Endurance and retention need extensive testing.
Costs need to become competitive.
Industry partners need to decide whether the architecture solves a problem valuable enough to justify investment.
Those milestones could take years.
But Fudan’s research program already shows a notable progression—from ultrafast 2D flash devices to integrated prototypes and now room-temperature single-electron storage.
That makes the technology worth following.
The most important future announcement will not necessarily be another record.
It will be evidence that the device can be manufactured reliably at meaningful scale.
FAQs
What is single-electron memory?
Single-electron memory stores a bit of information by controlling the state of an individual electron rather than relying on much larger amounts of electrical charge.
What did Fudan University researchers achieve?
Researchers demonstrated room-temperature, non-volatile storage using a single electron in a two-dimensional semiconductor device. Fudan reports a memory window of approximately 0.5 volts.
Why is one electron important?
An electron is an elementary carrier of electrical charge. Using one electron to represent one bit reaches the theoretical charge-storage limit for this type of memory.
Is single-electron memory available commercially?
No. The technology remains at the research and development stage and must overcome major manufacturing and scaling challenges before commercial adoption.
Could it make AI faster?
Potentially. Denser, faster and more energy-efficient memory could reduce some of the data-movement bottlenecks affecting AI computing, but real-world benefits depend on successful large-scale integration.
Will it replace current flash memory?
There is currently no evidence that it will replace NAND or other mainstream memory soon. Existing technologies have mature, enormous manufacturing ecosystems.
The Light Span Perspective
The most remarkable thing about single-electron memory is not that researchers created another faster chip.
It is that they reached a fundamental physical boundary.
For charge-based storage, you cannot use half an electron.
One electron is the limit.
That makes the Fudan experiment scientifically fascinating.
But the AI era makes it economically interesting too.
Computing is increasingly constrained not simply by how quickly processors calculate, but by how efficiently enormous amounts of information can be stored and moved.
AI accelerators need faster memory.
Data centers need lower energy consumption.
Edge devices need more intelligence inside limited power and space.
Every one of those trends increases the value of better memory.
Single-electron storage offers an extreme vision of what that future might look like: information represented using the smallest possible amount of charge in atomically thin devices.
But semiconductor history teaches an equally important lesson.
A breakthrough in physics is not automatically a breakthrough in manufacturing.
The hardest stage may be starting now.
Researchers must prove that single-electron devices can be manufactured consistently across enormous arrays, survive repeated operation and compete economically with some of the most sophisticated manufacturing technologies humans have ever created.
If they cannot, Guiyi may remain an extraordinary scientific achievement.
If they can, the consequences could extend far beyond flash storage.
The technology could help reshape how computers divide processing and memory, how much energy AI systems consume and how much intelligence can operate locally inside everyday devices.
That is why this breakthrough deserves attention without exaggeration.
Researchers have reached the physical limit of charge-based information storage.
The next challenge is turning that limit into a technology the world can actually manufacture.
Continue reading more
