Exploration and Design of Input Current Switched Capacitor DC-DC Converters for Point-of-Load Data Center Power Delivery

This thesis investigates the concept of the input-current switched-capacitor (ICSC) converter as an alternative point-of-load power delivery architecture for data center applications. In contrast to conventional two-stage converter systems, the proposed topology places an inductive converter at the input stage and a switched-capacitor (SC) converter at the output stage. The approach is motivated by the assumption that regulating the inductive stage as a constant current source enables pulse-width modulation control of the SC stage while exploiting the high switching frequency capability of SC converters to improve transient response and reduce the size of magnetic components. The work first reviews the state of the art and introduces the theoretical foundations of switched-capacitor converters, buck converters, and control concepts for switching power converters. Based on these foundations, the operating principles and topologies of the ICSC converter are analyzed, including charge transfer mechanisms, current amplification, and equivalent output resistance. An averaged model of the complete converter system is derived to describe the dynamic behavior of the proposed architecture. A control concept based on the averaged model is developed and evaluated through simulations. The analysis reveals that the transient behavior of the ICSC stage has only limited influence on the overall system dynamics at high current levels. Instead, the system dynamics are primarily constrained by the control bandwidth of the inductive stage due to the coupling between both converter stages. Consequently, the expected improvement in transient performance through the fast SC stage cannot be fully realized. To validate the theoretical analysis, a hardware prototype is developed and experimentally investigated. Measurements are compared with simulations and the derived averaged model, including parasitic effects. The results show that assumptions made during modeling, particularly the assumption of constant input current, do not fully capture the practical converter behavior. Furthermore, non-ideal effects such as dead time influence converter performance and efficiency. Experimental validation of the control concept is limited by practical issues in the current sensing circuitry of the prototype.