Robert Pilawa-Podgurski

Powering the AI Revolution: A New Voltage Regulator Enabling Direct 48-V-to-1-V Vertical Power Delivery for Next-Generation Ultra-High-Power Processors

High-performance processors are crucial for technologies like AI and autonomous vehicles, but their power usage is soaring, approaching 1000 W. This proposal seeks to make use of innovative tools and techniques designed in Pilawa-Podgurski’s lab to develop a hybrid switched-capacitor voltage regulation module for ultra-high-power processors that can achieve a 2x reduction in power conversion losses and a 4x reduction in physical size compared to state-of-the-art commercial products. This could reduce overall data center power usage by almost ten percent, benefiting both the environment and businesses.

Fueling the AI Revolution: An Energy Efficient Approach

Updated February 10, 2025

By: Niki Borghei

The AI revolution is here, bringing with it an extraordinary range of new technologies, from advanced chatbots to diagnostic tools. However, this progress comes at a cost. It’s consuming an immense amount of energy.

But Faculty Fellow Robert Pilawa-Podgurski and Innovation Fellow Yicheng Zhu are working on a solution. Their AIC-Bakar Award project is focused on enhancing the energy efficiency of processor power delivery to support the ongoing AI revolution.

With the rapid growth of generative AI, the demand for high-performance processors has surged, pushing power consumption beyond 1000 W. This trend has led to significant energy inefficiencies and thermal management challenges, particularly in large-scale data centers. 

This AIC-Bakar project aims to tackle these issues by developing innovative hybrid switched-capacitor voltage regulation modules for ultra-high-power processors. These modules are designed to reduce power conversion losses by 2x and cut the physical size by 4x compared to current commercial solutions. These advancements have the potential to lower overall data center power consumption by nearly 10%, delivering both environmental and business benefits through improved energy efficiency and cost savings.

Q: So how does this technology work?

Yicheng Zhu: Imagine your computer or phone has a tiny power manager inside it that ensures the processor (the brain of the device) gets both the correct voltage and the right amount of electricity to work efficiently. Now, think about the massive computers used for artificial intelligence (AI) and data centers—they require far more power, and their power managers have to work much harder. These power managers, called voltage regulators, waste some energy as heat and take up a lot of space, making everything less efficient and more expensive to operate.

Our project is like creating a super-efficient, ultra-compact power manager for these large-scale systems. We’re using a new circuit topology called hybrid switched-capacitor power converters, which allows us to use the densest energy storage components and the highest-performance power semiconductor devices in the most effective way possible. Think of it as building a more streamlined power delivery system that wastes less energy and fits into a much smaller space. By making these power managers twice as efficient and four times smaller, we can help data centers reduce their overall electricity usage by almost 10%. This not only saves money but also reduces the environmental impact of running these powerful computers.

Q: What made you interested in this approach?

A: Our interest in this area of research was sparked by the intersection of two transformative trends: the exponential growth of generative AI and high-performance computing, and the urgent need for sustainable energy solutions. As the demand for generative AI and high-performance computing surges, data centers have become the fastest-growing electricity consumers on the power grid. Improving the energy efficiency of data centers is one of the most pressing challenges in energy and sustainability. We are motivated by the potential to create meaningful economic, societal, and environmental benefits through this work and contribute to a more sustainable future.

Q: What inspired you to pursue entrepreneurship?

A: Our interest in entrepreneurship stems from a vision to solve real-world problems through creative and disruptive academic research. We are excited by the opportunity to turn innovative ideas into impactful solutions that drive positive societal change and create value for others. We believe that transforming the high-performance power converter prototypes developed in our laboratories into commercial products—particularly for application in data centers—would significantly enhance the economic, societal, and environmental impact of our work.

Q: How is the AIC-Bakar fund supporting your work?

A: The AIC-Bakar fund we received will be used to support a postdoctoral researcher and partially fund a graduate student researcher for academic research. Additionally, it will cover the purchase of circuit components, lab equipment, and printed circuit boards for hardware development. The funding is critical to advance the very promising initial results to a solution that addresses the full set of power delivery challenges of next-generation GPUs.

Q: When will this technology likely be available on the market?

A: We are working with major data center providers to make sure that the solution addresses the needs of their next-generation products, and we hope to see this technology in commercial systems in a 2-3-year time frame.