Ferroelectric films lose performance as they are made thinner, which has constrained ferroelectric field-effect transistors for compute-in-memory chips just as the rest of the industry pushes towards atomic dimensions. A team led from the College of Chemistry and Molecular Engineering at Peking University reported a material that resists this: alpha-phase bismuth selenite, a high-dielectric-constant van der Waals ferroelectric oxide whose ferroelectricity holds down to a single monolayer.
Fabrication is what makes it usable. The group developed an in-situ oxidation method compatible with back-end-of-line processing at 400 degrees Celsius or below, enabling uniform wafer-scale growth of ultrathin films, including ferroelectric and semiconductor heterostructures. This is reported as the first controlled wafer-scale fabrication of such films, and the temperature ceiling matters because back-end processing must not damage the transistors already built beneath.
Device metrics are strong across several axes simultaneously. The transistors operate at 0.8 volts with a 20 nanosecond write speed, endurance above 1.5 trillion cycles, ten-year retention, five-bit multilevel storage and energy consumption of 2.8 femtojoules per bit per square micrometre. The authors state these results exceed current industrial hafnia-based ferroelectric systems. The team also demonstrated dynamically reconfigurable logic-in-memory circuits operating below one volt, within the range compatible with standard complementary logic.
The scientific backstory is notable for its length. The group identified this compound as the native oxide of the semiconductor bismuth oxyselenide in 2018 and predicted its ferroelectricity then, with experimental confirmation arriving only after roughly eight years of work. That interval between theoretical prediction and controlled wafer-scale synthesis is a reminder of how much of materials progress consists of growth chemistry rather than discovery, and of how rarely such sustained single-material programmes are funded. The study was published in Science.