A ferroelectric-ionic-trapping transistor for low power and secure neuromorphic computing

Abstract

Abstract The convergence of artificial intelligence and pervasive data analytics has created an urgent demand for energy-efficient and secure computing hardware. Neuromorphic synaptic devices emulate the human brain with high parallelism and 3D connectivity to reduce power consumption, yet they lack intrinsic mechanisms to protect stored information from malicious read-out. Here we report a ferroelectric–ionic–trapping field-effect transistor (FITFET) that integrates ferroelectric polarization, oxygen-vacancy migration, and charge trapping to enable both synaptic and secure functionality. The FITFET operates in two programmable regimes: a plain mode combining fast, low-power ferroelectric switching with analog weight modulation, and a secure mode in which the memory window collapses under specific read biases, concealing stored states and suppressing read attacks. This reversible, voltage-controlled transition provides a hardware-native data protection mechanism. Multiscale analyses demonstrate reliable switching between learning and concealment, while system-level simulations show reduced model inversion attacks with minimal accuracy loss.