While AI (artificial intelligence) is rapidly transforming industries, its progress is increasingly constrained by the hardware it depends on. As traditional chip improvements slow and energy demands rise, researchers are rethinking how computing systems are designed. At Arizona State University, new efforts are underway to develop adaptable hardware that can keep pace with the evolving needs of AI applications.
Aman Arora, an assistant professor in the School of Computing and Augmented Intelligence, is leading research on reconfigurable computing. His work focuses on FPGAs (field-programmable gate arrays), which are flexible chips that can be reprogrammed after manufacturing to create faster, more efficient AI systems tailored to specific tasks.
This approach addresses key limitations of traditional hardware, such as GPUs (graphics processing units), which were not originally designed for modern AI workloads and can struggle with realtime, energy-efficient processing outside of data centers.
Here is how this can help:
- Enable faster, realtime AI performance by reducing computational overhead and customizing hardware to specific tasks.
- Improve energy efficiency and sustainability by minimizing unnecessary data movement and extending hardware lifespans.
- Expand AI capabilities beyond data centers to edge devices such as sensors, medical tools, and autonomous systems.
Looking to the future, innovations in adaptable hardware are poised to redefine how AI systems are built and deployed. By codesigning hardware and software, researchers are moving toward more efficient, flexible, and scalable technologies, which will pave the way for AI to operate seamlessly in real-world environments and unlock new possibilities across science, healthcare, and industry.