Power-hungry AI models and data centres are driving up global energy consumption, raising urgent concerns around sustainability. Addressing this challenge, researchers from Cornell University, NY USA, have unveiled a breakthrough innovation that lowers the energy required to power AI systems – paving the way for greener, more sustainable digital infrastructure.
Presented at the 2025 International Conference on Field-Programmable Logic and Applications (FPL) in Leiden, Netherlands, the research earned a prestigious Best Paper Award. The team’s solution centres on rethinking the hardware architecture of Field-Programmable Gate Arrays (FPGAs) – versatile chips already widely deployed in AI, cloud computing, wireless communication and digital devices.
This advancement, named “Double Duty”, has the potential to reduce AI’s carbon footprint significantly, while maintaining the performance gains needed for next-generation digital construction, smart cities, and other AI-driven sectors.
Why AI’s energy demands pose a sustainability challenge
As AI models grow in complexity – particularly deep learning and natural language processing – the computational power required to train and deploy them increases exponentially.
- Data centres now account for an estimated 1–2% of global electricity use, a figure expected to rise sharply as AI adoption accelerates.
- AI infrastructure, including GPUs, CPUs and FPGAs, requires intensive energy input, contributing to higher carbon emissions.
- Sustainability in AI is becoming a priority across industries, with governments, enterprises and researchers seeking ways to optimise digital infrastructure.
For the construction and built environment sectors, where digital twins, generative design and predictive analytics rely on AI, energy efficiency is vital for aligning with net-zero carbon targets.
The role of FPGAs in AI acceleration
What Are FPGAs?
A Field-Programmable Gate Array (FPGA) is a type of integrated circuit that can be reprogrammed after manufacturing to perform specific tasks. Unlike fixed-function chips, FPGAs offer flexibility and adaptability, making them invaluable for rapidly evolving fields such as:
- AI acceleration – supporting deep learning and neural network workloads.
- Cloud computing – optimising workloads and improving throughput.
- Telecommunications – powering base stations and network infrastructure.
- Healthcare technologies – used in devices such as ultrasound scanners and CAT machines.
- Consumer electronics – embedded in everyday appliances like washing machines.
“FPGAs are everywhere – from network cards and communication base stations to ultrasound machines, CAT scans, and even washing machines,” said co-author Mohamed Abdelfattah, assistant professor at Cornell Tech. “AI is coming to all of these devices, and this architecture helps make that transition more efficient.”
Why AI needs FPGA innovation
Traditional FPGA designs limit performance for AI tasks due to how their logic blocks are structured. These blocks contain:
- Lookup Tables (LUTs): capable of carrying out diverse logical operations.
- Adder chains: optimised for rapid arithmetic functions such as addition, central to image recognition and natural language processing.
Currently, adder chains are only accessible through LUTs, restricting efficiency. For AI workloads that are arithmetic-heavy, this creates unnecessary bottlenecks.
Double duty: A game-changer for AI hardware
Cornell University’s research introduces a new architecture called Double Duty, designed to unlock the full potential of FPGAs.
How It Works
- Independent operations: Double Duty enables LUTs and adder chains to operate independently and simultaneously.
- Resource optimisation: The design maximises existing resources without requiring additional hardware.
- Parallel performance: Logic blocks can handle more computations at once, reducing time and energy expenditure.
Results from testing
- 20% reduction in space required for certain AI tasks.
- 10% improvement in performance across a large suite of circuits.
- Lower energy usage, as fewer chips are needed to complete the same workload.
By making small architectural changes, the team demonstrated that FPGAs can play to their strengths as specialised AI accelerators, rather than mimicking generic processors.
Looking ahead: building sustainable AI infrastructure
The Cornell research highlights the critical role of hardware innovation in ensuring AI’s future is sustainable. As governments and industries seek to balance technological progress with environmental responsibility, energy-efficient AI hardware will be a cornerstone of:
- Sustainable data centres with reduced electricity demands.
- AI adoption in energy-conscious sectors such as healthcare, construction and transportation.
- Next-generation FPGA design that underpins everything from smart appliances to national digital infrastructure.
AI’s rapid expansion is reshaping the digital construction sector and beyond, but sustainability remains a pressing concern. Cornell University’s Double Duty FPGA architecture demonstrates how incremental hardware changes can deliver significant efficiency gains – reducing energy use, cutting carbon emissions and supporting scalable AI deployment.
This breakthrough could represent a crucial step forward in building a smarter, greener future.
Read next: How AI, data, and climate reshape what we build next – Build in Digital
Are you a building professional? Sign up for a FREE MEMBERSHIP to upload news stories, post job vacancies, and connect with colleagues on our secure social feed.