Physics-Informed Data-Driven KNN Methodology of Modeling NCFETs for Superior Circuit Performance

This article presents a novel, fully data-driven, nonparametric methodology for accurately modeling double-gate fully depleted silicon-on-insulator (DG-FDSOI) negative capacitance field-effect transistors (NCFETs) using <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$K$</tex-math> </inline-formula>-nearest neighbors (KNNs) regression, the first such application in the context of ferroelectric (FE) transistor modeling. Unlike conventional Landau–Khalatnikov (LK) equation-based approaches that rely on parameter fitting and assumptions about material behavior, our technique directly learns from experimental polarization–electric (P–E) field data of doped hafnia, enabling robust, accurate, and physically consistent modeling of FE properties across varying conditions. By integrating this AI-trained material model into TCAD simulations and Verilog-AMS models, we realize a seamless pipeline from experimental material data to circuit-level EDA implementation. The proposed framework yields unprecedented device-level improvements, including record-low subthreshold swing, enhanced <sc xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">on</small>-currents, and superior <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${I}_{\text {on}}$</tex-math> </inline-formula>/<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${I}_{\text {off}}$</tex-math> </inline-formula> ratios, all while maintaining electrostatic control and power efficiency. Furthermore, extensive benchmarking of NCFET circuits demonstrates marked gains in intrinsic gain, linearity, output resistance, and energy efficiency compared to standard CMOS at the same technology node and bias conditions. This work marks the first unified demonstration of a physics-informed, machine-learned transistor model that is experimentally validated, scalable, and circuit-compatible. It bridges a critical gap between nanoscale FE physics and system-level circuit design and establishes a new paradigm for modeling emerging device technologies through AI, making it highly relevant and impactful for next-generation low-power semiconductor platforms.