A General-Regression-Neural-Network Based 5V-to-48V Three-Level Buck/Boost Power Converter with 40dB PSRR 90%-Efficiency for SSD Power Loss Protection

Solid-state drives (SSDs) are widely used in data centers, workstations and PCs to replace hard drives for their high data rate, low energy consumption and compact size. To protect the SSD’s cache buffers and mapping table during unpredicted power loss, power loss protection (PLP) circuits are required as a backup power supply to power the SSD for 10s of milliseconds. As shown in Fig. 1, when the SSD is powered by V
IN
, the buck converter in the PLP is in stand-by mode while the boost converter charges an energy-storage capacitor C
STORGE
, to V
SC
,(12V) and maintains this voltage during the standby mode [1]. Once a sudden power loss is detected, the buck converter is activated as the uninterrupted backup power supply to continue powering the SSD by converting V
SC
into the output voltage $V_{LOAD}$ (5V) for the SSD. To ensure enough PLP time under low V
SC
, large-volume tantalum capacitors or supercapacitors are commonly used for PLP circuits, which are expensive and bulky [2]. When PLP circuits are in backup operation, V
SC
across C
STORGE
monotonically decays from $V_{SC,MAX}$ to 5V over several milliseconds which poses challenges to the line regulation of the buck converter [1], [2]. Another challenge in conventional buck converter design is that its transient response is fundamentally limited by its slow feedback loop with a power-hungry error amplifier and bulky compensation network. In addition, conventional buck converters based on silicon switches lack energy efficiency when dealing with high V
SC
and high switching frequencies due to low figures-of-merit of the silicon power switches [3]. This is worse when a two-level buck converter operates under a high V
SC
since there are excessive voltage stresses across each of the power switches [4]. In summary, for PLP circuits, it is highly desirable to develop buck/boost power converters with fast transient response, high energy efficiency, compact size, and low cost.