AI Hardware: Lower Energy Consumption
- A new innovation from Cornell University promises to significantly reduce the energy consumption of artificial intelligence systems, addressing growing sustainability concerns within the tech industry.
- As artificial intelligence models grow in complexity and capability, their energy demands have skyrocketed.This poses a substantial challenge to data centers and AI infrastructure,contributing to increased carbon footprints...
- Researchers at Cornell Tech and Cornell Engineering are tackling this problem by focusing on the underlying hardware that powers AI.
“`html
Cornell Researchers Win Award for Energy-Efficient AI Chip Design
Table of Contents
A new innovation from Cornell University promises to significantly reduce the energy consumption of artificial intelligence systems, addressing growing sustainability concerns within the tech industry.
The Challenge: AI’s Growing Energy demand
As artificial intelligence models grow in complexity and capability, their energy demands have skyrocketed.This poses a substantial challenge to data centers and AI infrastructure,contributing to increased carbon footprints and operational costs. According to a 2023 report by the International Energy Agency, data centers already account for approximately 1-1.5% of global electricity use,and AI is expected to dramatically increase this figure (“Data Centres and Data Transmission Networks”).
Rethinking AI Hardware with FPGAs
Researchers at Cornell Tech and Cornell Engineering are tackling this problem by focusing on the underlying hardware that powers AI. Their work centers on Field-programmable Gate Arrays (FPGAs), a type of computer chip known for its adaptability. Unlike Request-Specific Integrated Circuits (ASICs), which are designed for a single purpose, FPGAs can be reprogrammed after manufacturing, making them ideal for the rapidly evolving field of AI.
“FPGAs are everywhere – from network cards and communication base stations to ultrasound machines, CAT scans, and even washing machines,” said Mohamed Abdelfattah, assistant professor at Cornell Tech (Cornell Tech profile). “AI is coming to all of thes devices, and this architecture helps make that transition more efficient.”
How FPGAs Work: Logic Blocks and LUTs
Inside an FPGA chip, computation is performed by logic blocks. These blocks contain Lookup Tables (LUTs), which are fundamental components capable of executing a wide range of logical operations. The Cornell team’s innovation lies in optimizing how these LUTs are utilized to perform AI calculations, reducing the energy required for each operation.
Award-Winning Research and Key Findings
The research team received a Best Paper Award at the 2025 international Conference on Field-Programmable Logic and Applications (FPL), held in Leiden, Netherlands, from September 1 to 5, 2025. The award recognizes the potential of their work to significantly impact the field of AI hardware.
While specific performance metrics haven’t been publicly released beyond the conference presentation, the team’s approach focuses on minimizing data movement within the chip, a major source of energy waste. By optimizing the arrangement and utilization of LUTs, they aim to perform more computations per watt.
The Broader Implications for AI and Sustainability
The development of energy-efficient AI hardware has far-reaching implications. Reducing the energy