- Accepted Paper
Highly nonlinear I-V response in doped silicon-on-insulator devices: Working principle
Phys. Rev. Applied - Accepted 16 November, 2026
DOI: https://doi.org/10.1103/yyld-pcdd
Phys. Rev. Applied - Accepted 16 November, 2026
DOI: https://doi.org/10.1103/yyld-pcdd
Multiterminal devices fabricated on P-doped silicon-on-insulator (SOI) substrate have been implemented as Synaptic highly Nonlinear Elements (SYNEs) for Kolmogorov-Arnold networks (KANs) in advanced artificial intelligence applications. While these devices exhibit the highly tunable transfer characteristics required for such physics-aware computing, especially demonstrating the presence of pronounced negative differential resistance (NDR) regions, the physical mechanisms driving their complex, room temperature nonlinearity has not been investigated. In this work, we elucidate the physical origins of this tunable nonlinearity by combining multielectrode current-voltage characterizations, charge transport measurements, compositional analysis, and technology computer aided design (TCAD) simulations. Combined experimental and theoretical analyses reveal that the device response is intrinsically governed by the specific architecture of the SOI substrate. Interface trap states at the top Si/SiO2 interface strongly modify the thickness of the conducting channel (Heff), rendering it significantly narrower than the physical thickness of the Si device layer (HSOI). Furthermore, applying a control voltage at a single electrode induces self-polarization of the floating back-gate, which fundamentally alters the electrostatic depletion beneath the contacts. Crucially, we demonstrate that the observed nonlinearity cannot be explained on the basis of fixed effective resistances between the contacts but rather relies on their dynamic modulation. To validate our physical insights, we introduce a complementary equivalent circuit model based on standard CMOS transistors, successfully proving that the complex NDR behavior of this single multiterminal device can be replicated through fundamental circuit-level mechanisms. The agreement between experimental data, TCAD simulations, and the analytical model establishes a comprehensive physical framework for SOI-based SYNEs, providing critical design guidelines for the realization of reproducible, scalable, and CMOS-compatible nonlinear computing hardware.
If the author has provided any supplemental materials with this article they will be available upon publication of the version of record.