A Little Gold Makes Atom-Thin Computer Memory Readable

Original colored-pencil illustration of semiconductor researcher probing gold-contacted atom-thin memory chip
BitcoinVersus.tech

Semiconductors | November 22, 2026. A memory device made from two layers of an atom-thin material now produces a much clearer electrical readout after University of Wisconsin–Madison engineers changed its metal contacts. Their design pairs gold with heavily doped two-dimensional silicon, addressing a persistent obstacle in an experimental technology called sliding ferroelectric memory. The university reported the work November 10, following publication in ACS Nano on October 20.

Researchers Haolin You and Chen Shao, working with electrical engineering professor Ying Wang, tested a few dozen devices rather than a finished memory array. Their reported electrical contrast improvement reached 100,000-fold (105). That result is a laboratory readout metric, not proof of a 100,000-fold gain in storage capacity, computer speed, or energy efficiency.

Background video: Professor Shu Ping Lau discusses ferroelectricity in two-dimensional bilayers in a 2024 FLEET Centre seminar. The lecture covers related physics, not Wisconsin’s 2026 experiment.

How two sheets of material can remember a bit

A conventional dynamic RAM cell stores electrical charge in a capacitor and periodically refreshes it. A ferroelectric device instead stores information in a material whose electric polarization can remain in one of two states after the switching voltage is removed. Electrons tunneling across a thin barrier can encounter different resistance depending on the stored state, allowing a readout.

Many ferroelectric designs reverse polarization by moving ions within a crystal, potentially stressing chemical bonds. The Wisconsin team investigated a different mechanism: neighboring sheets of rhombohedral molybdenum disulfide, or 3R-MoS₂, shift slightly against each other. Their relative position changes the polarization without requiring the same bond-breaking mechanism. The underlying sliding process may be extraordinarily fast, but the team did not demonstrate a complete commercial memory operating at a picosecond clock rate.

UW–Madison researchers Chen Shao, Haolin You and Ying Wang with laboratory equipment
From left: researchers Chen Shao, Haolin You and Ying Wang. Photo: Joel Hallberg / University of Wisconsin–Madison. The photograph shows the research team, not an image of the memory cell’s internal atomic structure.

The hard part was reading the stored state

Sliding the layers changes their electrical state, yet the resulting signal can be too weak to distinguish reliably. Contacts determine how electrons enter and leave the ultrathin junction. After exploring different combinations, the researchers found that gold on one side and heavily doped two-dimensional silicon on the other produced much stronger contrast. The university describes the resulting signal separation as comparable to a range used in established memory devices.

Contact engineering matters because a memory bit is useful only when the system can write it, retain it, and read it repeatedly despite noise. Researchers still must measure large-array uniformity, switching endurance, retention, fabrication yield, variability, and behavior near other devices. A promising isolated junction is not yet an economical DRAM replacement.

Related technology, not the Wisconsin test: a June 2026 YTN Science report on a different ferroelectric tunnel-junction design for low-power AI computing. Its materials and results must not be treated as measurements from the Wisconsin team.

Why AI hardware engineers are watching

Moving information repeatedly between a processor and separate memory consumes both time and electrical energy. Researchers hope compact nonvolatile junctions may eventually support memory closer to computation, including neuromorphic architectures. A potential manufacturing advantage is the use of processes compatible with existing cleanroom equipment, but commercial wafer-scale production remains unproven.

Other experimental paths illustrate the breadth of the race: subnanometer hafnia films, improved gate oxides for 2D chips, Brown’s multiferroic materials research, and MIT’s cryogenic magnetic memory device. The operating temperatures, switching physics, and maturity levels differ; their performance figures cannot be ranked as though they measured the same workload.

What happens next: Wisconsin’s team plans wafer-scale process studies and tests of how neighboring devices influence one another. Until a complete array is benchmarked against established memory on identical measures, the defensible conclusion is narrower and compelling: researchers made an atom-thin, sliding-polarization memory junction readable using practical contact engineering.

Primary sources: University of Wisconsin–Madison research report, November 10, 2026; ACS Nano research paper, October 20, 2026. The cited video materials are educational context and do not independently verify the new device’s measurements.

We volunteer daily to ensure the credibility of the information on this platform is Verifiably True. If you would like to support to help further secure the integrity of our research initiative, please scan the QR code below.

BitcoinVersus.Tech donation QR code

Leave a Reply