This talk presents next-generation solid-state transformers (SSTs), including advanced two-stage architectures and a capacitor-free single-stage SST that eliminates bulky DC-link components. The proposed designs and control strategies significantly improve efficiency, power density, and system reliability. We will also showcase developed 800V/48V and 48V/1V transformerless DC-DC converters tailored for AI data center power delivery. In addition, AI-enhanced predictive control enables fast dynamic response and precise voltage regulation under highly variable workloads. The overall approach provides a scalable and practical pathway toward more efficient, high-density, and sustainable data center power infrastructures.

This talk presents a scalable multiscale structuring approach to enhance phase-change processes for improved energy efficiency across diverse engineering systems. By tailoring surface features from micro- to nano-scales, interfacial phenomena such as boiling, condensation, and frosting can be significantly intensified and controlled. These enhancements enable performance gains in applications ranging from two-phase cooling devices and battery thermal regulation to refrigerant-based air-conditioning and dehumidification systems. Emphasis is placed on scalable fabrication strategies that bridge fundamental transport mechanisms with real-world implementation. The work demonstrates a pathway to reduce energy consumption and improve efficiency across a broad spectrum of phase-change-driven technologies through multiscale structure designs.

Data centers in Singapore currently consume about 7% of the nation’s total electricity, and this demand is rising rapidly with the growth of cloud computing and AI workloads. Addressing their energy use and carbon footprint has become a pressing challenge. Conventional facilities rely heavily on air-cooling systems, which are increasingly inefficient at handling rising thermal loads.

We propose an innovative spray cooling technique that significantly improves heat removal by leveraging convection, evaporation, and boiling. A non-conductive fluid is directly sprayed onto CPUs and other server components, enabling efficient thermal transfer at the source. The vaporized coolant is then condensed and recycled within a closed-loop system integrated into the rack. By using ambient-temperature water in the condensers, the system eliminates the need for energy-intensive chillers.

Our approach has the potential to reduce annual energy costs by up to 26%. In laboratory testing, our prototype achieved a power usage effectiveness (PUE) as low as 1.08 and supported rack densities of up to 23 kW/m3, demonstrating both high efficiency and scalability.

As AI pushes the limits of today’s infrastructure, this panel explores cross-layer innovations, from semiconductor processes and silicon photonics to system-level integration, that will determine whether hardware can keep up with AI scaling.