Fudan University researchers in Shanghai have demonstrated a working two-dimensional (2D) flash memory chip that stores data using just one electron — collapsing a figure that stands at roughly 200,000 electrons in today's most advanced DRAM chips from Samsung and SK Hynix. The breakthrough, published on July 16 in the journal Science, could fundamentally reshape how artificial intelligence workloads are handled on low-power devices.

The device and what it does

The chip, named Guiyi, was developed by microelectronics professor Zhou Peng and his colleagues at Fudan University. It traps and reads a solitary electron at room temperature — a capability long considered the theoretical ceiling of semiconductor miniaturisation. Critically, Guiyi amplifies the faint signal produced by a single electron from just tens of millivolts to a robust 0.5 volts, representing a tenfold improvement over any previous single-electron memory attempt.

Why it matters for AI

Current DRAM technology requires approximately 200,000 electrons per bit to maintain reliable data storage — a design constraint that drives both the physical size and energy consumption of memory chips. By reducing that count to one, the Guiyi device reaches the absolute theoretical floor for charge-based storage. According to the research team, the most immediate real-world impact would be enabling large language models to run on mobile phones using minimal power, without the AI losing context during extended conversations.

The competitive backdrop

The announcement arrives as global chipmakers race to overcome the memory bandwidth constraints that throttle AI inference at the edge. Samsung and SK Hynix have invested heavily in high-bandwidth memory (HBM) stacks to feed data-hungry AI accelerators, but those solutions are power-intensive and expensive. A viable single-electron memory could render that architectural approach less dominant over time. The research was supported by the National Natural Science Foundation of China, signalling state-level priority for the work.

What's next

Moving from a laboratory demonstration to a manufacturable process node remains a substantial engineering challenge. The 2D material platform underpinning Guiyi must prove compatible with existing semiconductor fabrication lines before commercial timelines can be discussed. Nonetheless, publication in Science — one of the world's most selective peer-reviewed journals — lends the findings significant credibility and will likely accelerate follow-on research globally. Analysts and chipmakers will be watching closely to see whether Fudan University's team can demonstrate yield and endurance metrics consistent with mass production.