Dejun Ba

This study proposes a fault-tolerant low-authority injection (LAI) current-sharing strategy for parallel DC-DC converters. A dual-operational-amplifier architecture is adopted, and a resistor–diode injection interface is introduced to limit the influence of the current-sharing loop on the primary voltage-regulation loop. A piecewise injection model and a simplified small-signal model are developed. A three-module 36 V/5 V half-bridge prototype is experimentally evaluated. At total output currents of 30 A and 90 A, the maximum current-sharing errors are approximately 6.0% and 6.4%, respectively. The converter also maintains stable operation during an intentional current-sharing-bus short-circuit test. The proposed LAI method provides a simple hardware-oriented solution that achieves effective current redistribution while restricting sharing-loop authority and improving fault tolerance.

High-efficiency and high-power-density point-of-load (POL) converters are critical for data center power supplies. Although hybrid resonant switched-capacitor (ReSC) converters can substantially reduce the volume of passive components, they often suffer from severe efficiency degradation when the switching frequency mismatches the resonant frequency due to component tolerances. To address this challenge, this paper proposes a parameter-mismatch insensitive cascaded POL converter by integrating a BUCK stage with a cascaded voltage divider (CVD). By introducing an auxiliary resonant branch, a multi-resonant operation is established, enabling the residual inductor energy caused by component variations to be transferred to the output during the dead time with virtually eliminated hard-switching losses. Consequently, precise matching between the switching frequency and the resonant frequency is no longer mandatory. A 12 V-to-1 V/30 A GaN-based prototype was developed to validate the theoretical analysis. Experimental results demonstrate that the proposed converter maintains high efficiency under a ±10% component variation and achieves robust voltage regulation and fast transient response, making it highly suitable for high-current data center applications.

Inhomogeneous magnetic field distributions in high-frequency planar transformers frequently cause severe localized thermal hotspots and elevated leakage inductance. Traditional interleaved winding designs rely heavily on empirical trial-and-error, which becomes computationally prohibitive for multi-layer parallel structures due to the factorial “curse of dimensionality.” To address this bottleneck, this paper proposes a universal, data-driven optimization methodology. First, a quantitative one-dimensional prefix-sum model is established to correlate winding arrangements with spatial magnetomotive force (MMF) distributions, effectively simplifying the electromagnetic evaluation. Subsequently, a customized Genetic Algorithm (GA) framework, featuring physical-constraint-preserving operators such as Order Crossover (OX), is introduced to efficiently navigate the high-dimensional discrete search space. Using an extreme 26-layer complex parallel winding configuration (Np:Ns = 9:2) as a primary case study, the proposed GA method effectively bypasses over 1.5 million permutations, converging to the global optimum within 100 generations. The optimized structure achieves profound peak-shaving, drastically reducing both the peak MMF and total uncoupled magnetic energy area. This methodology provides a systematic, computationally lightweight EDA solution that fundamentally replaces empirical trial-and-error in the design of high-frequency magnetic components.

This paper simplifies indirect resonant switched-capacitor (ReSC) converters using the state-space average model method. The operation principles of the 4:1 and 5:1 ReSC converters derived from the Dickson (4:1) circuit are analyzed, and the corresponding state-space average matrices are derived based on their equivalent circuits. The resonant inductor of the specific resonant branch is eliminated by analyzing the composition of the state-variable matrix, thereby obtaining the simplified topologies of 4:1 and 5:1 indirect ReSC converters. The simplified topologies are simulated and experimentally verified. The results prove the correctness of the state-space average modeling method and the effectiveness of the simplified topologies.