The Demonstration of Scalable-HZO/ZrO2 FeFET With Large Memory Window of 2.3V for 3Bit-Per-Cell, Immediate Read After Write, High Endurance of 10⁹ Cycles, and the High Accuracy of 92% for Machine Learning

In this study, we show that using <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{HZO} / \text{ZrO}_{2}$</tex> as a ferroelectric layer in FeFETs outperforms HZO alone. The thick HZO layer forms a low-defect-density <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{ZrO}_{2}$</tex> layer, enhancing the ON/OFF ratio, memory window, and reducing leakage. Interface defects for the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$5 / 3(5 ~\text{nm} / 3 ~\text{nm})$</tex>, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$7 / 3(7 ~\text{nm} / 3 ~\text{nm})$</tex>, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$9 / 3(9 ~\text{nm} / 3 ~\text{nm})$</tex> structures were analyzed, revealing optimal performance at <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$9 / 3$</tex> due to uniform domain growth and low defect density. The <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$9 / 3$</tex> structure achieved faster polarization switching and high write speed. Despite a smaller ON/OFF ratio, it exhibited excellent Gmax/Gmin linearity, crucial for neural networks. Machine learning tests demonstrated 92 % accuracy in 20 epochs, highlighting its AI application potential.