Arturo enabled near-memory, AI-supported 3D integrated circuits

Published: 2025-11-27

Arturo Prieto defends his PhD thesis Friday, December 12th, in E:1406 at 09:15. Title of thesis: 3D Integration Technology and Near-Memory Computing for Edge AI. Link to thesis in Lund University Research Portal Zoom link. Zoom ID:...

Arturo Prieto defends his PhD thesis Friday, December 12th, in E:1406 at 09:15.

Title of thesis: 3D Integration Technology and Near-Memory Computing for Edge AI.

Link to thesis in Lund University Research Portal

Zoom link.
Zoom ID: 66847927874.

Describe your research in a popular science way
Digital applications, led by the growing popularity of artificial intelligence (AI), have increased its presence in the world today and enable new possibilities. We can see that AI has reached many aspects of society, allowing us to find answers and inspiration, offering personalized content on social media, or accurate image processing for diagnosis in healthcare treatments. However, the energy consumption of electronic devices processing AI applications is more dominant and has greater relevance in the world today. At the same time, data privacy is a concern, so keeping the processing on local devices avoids privacy breaks, but demands more processing capabilities from battery-powered devices. Specialized hardware needs to be designed to efficiently process the digital applications of the future. From the technology perspective, microchip integration has conventionally followed a miniaturization strategy to increase density. Moreover, from a hardware design point of view, memory and computation are conventionally separated. In this thesis, 3D integration is considered as an alternative to miniaturization to improve area efficiency, and near-memory computation is proposed as a way to bring data storage and processing closer to each other.

What is the most fascinating or interesting with your thesis subject?
The technology in our hands, smartphones and electronics devices, allows us to expand our capabilities in communication, access to information, or entertainment. The most fascinating part of my thesis subject is the evaluation of techniques for improving the circuits that serve as the foundation of a digitally connected world.

Do you believe some results from your research will be applied in practice eventually? And if so, how / how?
Yes, I believe that the work we have done presents suitable solutions to improve the limitations of hardware platforms. The continuous development of digital applications processing vast amounts of data demands energy efficient hardware that can benefit from the design techniques proposed in this thesis work.

Link to the article Arturo enabled near-memory, AI-supported 3D integrated circuits

Masoud Nouripayam improved chip architecture

Published: 2025-11-10

Title of thesis: Near-Memory Computing Architectures for Scalable Edge AI Applications. Link to thesis in Lund University Research Portal Defence: Friday, November 21st, in E:1406 at 09:15. Zoom link. Zoom ID: 69455644174. Describe your...

Title of thesis: Near-Memory Computing Architectures for Scalable Edge AI Applications.

Link to thesis in Lund University Research Portal

Defence:  Friday, November 21st, in E:1406 at 09:15.

Zoom link.
Zoom ID: 69455644174.

Describe your research in a popular science way

Energy, efficiency, and environmental impact are among the most critical concerns guiding every step of technological advancement. Artificial intelligence is expanding into nearly every aspect of daily life and requires access to and processing of huge amounts of data, which leads to increasingly resource-demanding workloads. Deploying AI on battery-powered devices with limited energy and memory makes it essential to address energy efficiency directly at the chip level and rethink conventional computing architectures instead of relying on traditional approaches that only attempt to mitigate performance and energy bottlenecks. My research focuses on enhancing future AI devices to deliver higher performance with lower energy demand, avoiding the waste of time and power in conventional computing architectures. In conventional systems, data is constantly shuttled between processors and memory, which results in unnecessary energy consumption and performance loss. The developed approach enables memory to perform parts of the computation close to where the data already resides. This data-centric computing concept, known as near-memory computing, can make everything from small sensors to advanced edge AI systems significantly more efficient.

What made you want to pursue a PhD?

I was very eager to understand the underlying principles and truly learn how computing systems work. I have always been ambitious to shape technology and to contribute meaningfully to the driving force that pushes it forward. Pursuing a PhD gave me the opportunity to explore ideas in depth and work on innovations that can make a real impact in the future.

What is the most fascinating or interesting with your thesis subject?

What fascinates me most is how relatively small architectural changes near memory can lead to dramatic improvements in both performance and energy efficiency. Even subtle shifts in where and how computations are carried out can unlock significant system-level benefits. This demonstrates that meaningful innovation does not always depend on inventing entirely new technologies. In many cases, the key lies in rethinking and reengineering existing technologies in smarter, more efficient ways, allowing us to push the boundaries of what current hardware can achieve.

Do you believe some results from your research will be applied in practice eventually? And if so, how / how?

In the past few years, near-memory computing has started to gain significant traction, and researchers from different disciplines have contributed to realizing and improving this concept even further. I believe that in the near future, commercial products will be equipped with data-centric hardware architectures in one form or another. This shift will enable more efficient AI processing directly on edge devices, reducing energy consumption and latency while improving overall system performance.

Link to the article Masoud Nouripayam improved chip architecture