Ischemic heart disease diagnosis via magnetocardiography (MCG) requires highly sensitive yet compact magnetometers. While search coil magnetometers (SCMs) offer a cost-effective and portable alternative to superconducting devices, traditional designs rely on uniformly-wound structures optimized for homogeneous magnetic fields. This structure is inefficient for the strongly gradient, dipole-like cardiac magnetic field, thereby limiting the signal-to-noise ratio (SNR). To address this, we propose a global optimization framework tailored for non-uniform cardiac fields. We introduce the structural efficiency as an innovative performance indicator to evaluate spatial flux utilization. Furthermore, we develop a heuristic algorithm combining Softmax parameterization with a two-stage multi-start search to escape local optima in high-dimensional, tightly constrained parameter spaces. The resulting SCM design features a non-uniform, segmented winding distribution that precisely matches the spatial magnetic field gradient, maximizing flux interception while minimizing thermal noise. Experimental validation of the prototype confirmed a sensitivity of 2.37 nV/pT and a low noise floor of 470 fT/√Hz at 10 Hz, aligning closely with theoretical models. By applying the experimental deviation to the validated model, the system is projected to achieve an estimated sensitivity of 9.9 μV/pT and a noise level of 290 fT/√Hz at 10 Hz in gradient cardiac fields. Furthermore, simulation and analytical results demonstrate that the optimized non-uniform structure provides a 60% SNR improvement over conventional uniform designs during differential measurements. This work provides a robust design methodology for high-performance SCMs, facilitating the clinical translation of multi-channel MCG arrays.