Quantitative scanning capacitance microscopy enabled by machine learning and electrostatic modeling

Scanning capacitance microscopy (SCM) provides nanoscale access to local capacitance and carrier concentration in semiconductor devices, yet its quantitative interpretation is often obscured by probe geometry, parasitic capacitance, and signal noise. Here, we establish an integrated experimental–computational framework that combines dynamic SCM spectroscopy, machine learning (ML)-assisted analysis, and finite-element simulations to quantitatively resolve carrier behavior over a wide carrier-concentration range. Full dC/dV − V spectra are recorded at each spatial pixel, and principal component analysis with k-means clustering is employed to denoise and segment regions with distinct electronic characteristics. Finite-element modeling systematically elucidates the nonmonotonic effects of tip radius, oxide thickness, and carrier concentration, reproducing experimental trends and revealing nonmonotonic dC/dV responses driven by tip–sample electrostatics. The combined analysis establishes a unified quantitative description of SCM contrast and demonstrates that integrating ML with physical modeling enables reliable, high-resolution free-carrier characterization in advanced semiconductor architectures.