Hybrid Switched-Capacitor Architecture and Magnetics for Vertical Power Delivery

Pre-FPO Presentation
Date
Feb 22, 2024, 3:00 pm4:00 pm
Location
EQUAD B327

Speaker

Details

Event Description

The increasing demands of high-performance computing, recently propelled by increased adoption of artificial intelligence, require power electronics to be energy efficient with high power densities that can respond quickly to load transients. Vertical power delivery can significantly reduce the interconnect resistance and is a promising approach to improving the end-to-end efficiency for power conversion to high density loads (e.g., microprocessors and LEDs) while reducing the overall footprint occupied by the voltage regulator.

This thesis introduces the challenges of vertical power delivery for high current data center applications. Architectures for vertical power delivery must achieve high efficiency, high power density, and high control bandwidth, which are critical to their thermal performance, signal integrity, and packaging. This thesis presents a hybrid switched-capacitor magnetics architecture that addresses the necessary tradeoffs for vertical power delivery – the Linearly-Extendable Group-Operated Point-of-Load (LEGO-PoL) architecture – together with multiple critical techniques to achieve high performance. The LEGO-PoL architecture uses a hybrid switched-capacitor topology together with coupled magnetics and vertical-stacked packaging to deliver high output currents within a small area with low height. By merging multiple magnetic components into one with coupled magnetics, benefits in size reduction, performance improvement, and control bandwidth are achieved. This thesis quantifies the benefits of coupled magnetics, introduces geometries for vertical power delivery coupled inductors, and presents the optimization of these inductors for low height and high performance. The LEGO-PoL architecture along with the development of vertical coupled magnetics paves the way for vertical power delivery as a promising approach to solving the tradeoffs of performance, size, and speed and meeting the growing energy demands of future computing applications.
 

Adviser: Minjie Chen