PFI-RP: Resilient and Energy-Efficient Memory Chips for Enhanced Mobile AI and Personalized Machine Learning

Stanford University, Stanford CA

Investigators

Abstract

The broader impact of this Partnerships for Innovation - Research Partnerships (PFI-RP) project extends across scientific, technological, and societal realms. By focusing on the development of a resilient, fast-switching, and energy-efficient magnetoresistive random access memory (MRAM), this research aims to revolutionize the landscape of mobile artificial intelligence (AI). Unlike current memory options, this innovation is like a supercharged engine for your smartphone or smartwatch, making them smarter and much longer lasting with just one charge. This means not just faster processing, but also a smaller, personalized AI right in your device. The research tackles thermal challenges, design intricacies and nanofabrication to make this technology work seamlessly, with the goal of creating jobs and preparing students for the future. Think of it as upgrading the brain of your gadgets. Once developed, this AI chip could be used by companies like Apple or Tesla to enable brand-new applications. The demand for this kind of tech is huge and growing, and it could lead to a major breakthrough in how we use AI in our daily lives, while making our devices and privacy securely protected. The project will also empower a diverse group of students with crucial skills for innovation and semiconductor fabrication. The proposed project aims to tackle critical technical hurdles in the development of the next generation of MRAM materials and devices, focusing on enhancing thermal robustness, achieving fast and enduring switching, and ensuring sustainable energy efficiency. With a specific focus on applications in mobile AI, the research seeks innovative solutions that address challenges currently limiting the widespread adoption of MRAM technologies. In the evolving landscape of mobile computing, where AI algorithms promise transformative user experiences, data security and privacy concerns are paramount. Traditional AI models, trained extensively in the cloud, pose risks of compromising personal user information. Moreover, their deployment on mobile devices often results in suboptimal performance for certain user groups. This project will advance the state of the art in mobile AI by leveraging nonvolatile MRAM-based associative memories at the hardware level and embeddings-based neural networks at the software level. By doing so, the research envisions a future where AI applications on mobile devices are not only efficient (preserving battery power) but also secure (keeping private data locally on your own device), ensuring a more inclusive and responsive experience for users of all kinds of AI tools. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.