Abstract: To address the high energy consumption and low efficiency caused by the data transfer bottleneck between processors and memory in von Neumann architectures, this study explores a neuromorphic computing solution based on ferroelectric tunnel junctions (FTJs). FTJs exhibit the essential physical characteristics for neuromorphic computing due to their unique polarization switching behavior, where the statistical properties of polarization reversal directly determine device performance. Using barium titanate-based FTJ memory as a model system, this work investigates the regulatory effects of mechanical boundary conditions on polarization switching statistics. The results demonstrate that the non-uniform stress distribution induced by lattice mismatch between the top/bottom electrodes and the ferroelectric layer significantly modulates the polarization switching dynamics, a finding critical for optimizing neuromorphic computing devices. Through self-developed phase-field simulations combined with non-uniform field theory, we reveal that FTJs exhibit faster polarization switching speeds and higher inhomogeneity compared to unclamped pristine ferroelectric thin films, leading to substantially enhanced neuromorphic computing performance. This study confirms the beneficial role of non-uniform stress engineering in improving FTJ-based neuromorphic computing applications.