Grant: Discovery Grants Program - Individual
The University of British Columbia
$48,000.00
Novel storage and memory systems for AI applications
EF 2024-25
Résumé
Artificial intelligence (AI) has made a significant impact on society by automating tasks and improving efficiency. AI shows promise in critical areas such as healthcare and medical diagnosis, environmental sustainability, food security, education and personalized learning, disaster response and management, crime prevention, and public safety, among many others.
However, large AI models come with significant financial and environmental costs. Data centers used for AI operations contribute substantially to carbon emissions, with global data center emissions projected to reach 5.5% of the total by 2025. AI models require expensive and environmentally taxing hardware components, including processors, memory, and storage.
Our research program aims to design new memory and storage systems and develop algorithms that can make AI platforms operate more efficiently and sustainably. The program will focus on three specific target areas:
1. Innovative Memory Management Strategies: AI systems rely on extensive data sets. Efficient training and use of AI models hinge on having the right data readily available when needed. Predicting which data will be required imminently will enable us to fetch the necessary subset of data close to the processing unit while keeping the rest in more cost-effective remote memory.
2. Accelerator-Centric Storage Architectures: Training and utilizing AI models involve specialized hardware processing units called accelerators. In conventional systems, accelerators do not have direct access to storage and data for training and inference. They often rely on an intermediary processor, the CPU (central processing unit), causing delays and energy wastage. Our program will develop and evaluate storage systems that can be accessed directly by accelerators, eliminating the need for an intermediary. This will result in more streamlined and energy-efficient AI systems.
3. Practical Processing-in-Memory Systems: In recent years, our research group has gained expertise in practical processing-in-memory systems. We have shown their potential to accelerate inference in AI recommendation systems by nearly an order of magnitude. However, we have also identified limitations that restrict their application to only a few specific scenarios. Our program aims to create a practical processing-in-memory system that overcomes these constraints.
Our research program will provide training to highly qualified personnel in the fields of operating systems, hardware/software co-design, and machine learning. This education will equip individuals with the skills needed to advance the development of more efficient and sustainable AI systems.
Programme
Discovery Grants Program - Individual
Bénéficiaire
Emplacement
British Columbia, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data
Tel que publié
Application id
RGPIN-2024-04816
Name nom
Fedorova, Alexandra
Department département
ECE
Organization id
2
Institution établissement
The University of British Columbia
Province en
British Columbia
Province fr
Colombie-Britannique
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
48000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1507
Committee name en
Computer Science
Committee name fr
Sciences informatiques
Area of application code
801
Area of application group en
Information and communication services
Area of application group fr
Services d'information et de télécommunications
Area of application en
Computer software
Area of application fr
Logiciel
Research subject code
2720
Research subject group en
Information technology
Research subject group fr
Technologie de l'information
Research subject en
Computer systems software
Research subject fr
Logiciel de systèmes informatiques
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Novel storage and memory systems for AI applications
Keywords
processing in memory; processing in storage; accelerator-centric computing; smart memory; smart storage; virtual memory; tiered memory; memory management; data prefetching
Application summary
Artificial intelligence (AI) has made a significant impact on society by automating tasks and improving efficiency. AI shows promise in critical areas such as healthcare and medical diagnosis, environmental sustainability, food security, education and personalized learning, disaster response and management, crime prevention, and public safety, among many others.
However, large AI models come with significant financial and environmental costs. Data centers used for AI operations contribute substantially to carbon emissions, with global data center emissions projected to reach 5.5% of the total by 2025. AI models require expensive and environmentally taxing hardware components, including processors, memory, and storage.
Our research program aims to design new memory and storage systems and develop algorithms that can make AI platforms operate more efficiently and sustainably. The program will focus on three specific target areas:
1. Innovative Memory Management Strategies: AI systems rely on extensive data sets. Efficient training and use of AI models hinge on having the right data readily available when needed. Predicting which data will be required imminently will enable us to fetch the necessary subset of data close to the processing unit while keeping the rest in more cost-effective remote memory.
2. Accelerator-Centric Storage Architectures: Training and utilizing AI models involve specialized hardware processing units called accelerators. In conventional systems, accelerators do not have direct access to storage and data for training and inference. They often rely on an intermediary processor, the CPU (central processing unit), causing delays and energy wastage. Our program will develop and evaluate storage systems that can be accessed directly by accelerators, eliminating the need for an intermediary. This will result in more streamlined and energy-efficient AI systems.
3. Practical Processing-in-Memory Systems: In recent years, our research group has gained expertise in practical processing-in-memory systems. We have shown their potential to accelerate inference in AI recommendation systems by nearly an order of magnitude. However, we have also identified limitations that restrict their application to only a few specific scenarios. Our program aims to create a practical processing-in-memory system that overcomes these constraints.
Our research program will provide training to highly qualified personnel in the fields of operating systems, hardware/software co-design, and machine learning. This education will equip individuals with the skills needed to advance the development of more efficient and sustainable AI systems.