Grant: Discovery Grants Program - Individual
University of Waterloo
$53,000.00
Rethinking Configurable Logic Architectures in the age of AI
EF 2024-25
Résumé
The field of chip design and computer hardware is undergoing a renaissance with the emergence of compute-intensive, and transformational AI (Artificial Intelligence) and ML (Machine Learning) technologies. The impacts of AI/ML technology affects a wide swath of business including banking and finance, sales and marketing, healthcare, media, among others. Just over the past few years, we have seen companies like OpenAI release generative models like GPT3 (2020) and GPT4 (2023) with a 1000x increase in model size (170B to 176T parameters), and a 100x increase in compute (~300M to ~20G PetaFLOPS). To address this burgeoning demand in computational requirements, we need to rethink hardware design with two objectives: (1) keeping the computing architectures configurable, efficient, flexible, and scalable to adapt to continuously varying workload characteristics, and (2) developing rapid implementation tools to allow AI/ML developers to quickly deploy their models to silicon. Modern systems typically use a variety of solutions including GPUs (Graphics Processing Units), FPGAs (Field-Programmable Gate Arrays), and custom ASICs (Application-Specific Integrated Circuits). In the proposed research program, the long term vision is to explore a novel class of computing devices called CGRAs (Coarse-Grained Reconfigurable Arrays). CGRAs strike a balance between the performance, energy efficiency, and compute density of ASICs with the architecture flexibility of FPGAs and the ease of programmability of GPUs while targeting modern AI/ML workloads.
CGRAs have remained an academic curiosity with limited silicon density, and slow compilation runtimes. The aim is to lift these architectures into the future with a focus on customizability for AI/ML workloads, and a careful co-design of the mapping tools and architecture to allow for quick implementation and compilation. First, the project will look at the compute, storage, and communication demands of a range of AI/ML workloads, and then propose architecture models that allow us to explore various implementation strategies. Next, the project will investigate the structure of mapping tasks and evaluate algorithms that enable quick compilation to these architectures. In the next five years, the research intention is to rethink CGRA architecture for AI/ML workloads, and develop compilation/mapping algorithms to enable fast implementation.
The research conducted under this program is of vital interest to the Canadian hardware industry. Canadian startups and multinational companies such as Groq, Cerebras, AMD, Intel, Untether AI, Tenstorrent, and Microsoft have a growing appetite for hardware developers that can navigate the design space of AI/ML computations, and also develop tools and compilers to target these emerging devices. The HQP trained in this program will meet the demand for hardware engineers, and software tool developers at these companies, and several past HQP from the PI's group are placed there.
Programme
Discovery Grants Program - Individual
Bénéficiaire
Emplacement
Ontario, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data
Tel que publié
Application id
RGPIN-2024-05731
Name nom
Kapre, Nachiket
Department département
Electrical and Computer Engineering
Organization id
33
Institution établissement
University of Waterloo
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
53000
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
1510
Committee name en
Electrical and Computer Engineering
Committee name fr
Génie électrique et informatique
Area of application code
513
Area of application group en
Manufacturing processes and products
Area of application group fr
Procédés et produits industriels
Area of application en
Electrical and electronic machinery and equipment (including computer hardware)
Area of application fr
Machinerie et équipement électriques et électroniques (y compris matériel informatique)
Research subject code
2504
Research subject group en
Electrical and electronic engineering
Research subject group fr
Génie électrique et électronique
Research subject en
Integrated circuits
Research subject fr
Circuits intégrés
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
Rethinking Configurable Logic Architectures in the age of AI
Keywords
reconfigurable computing; machine learning; high-speed design; hardware design; accelerators; coarse-grained architectures
Application summary
The field of chip design and computer hardware is undergoing a renaissance with the emergence of compute-intensive, and transformational AI (Artificial Intelligence) and ML (Machine Learning) technologies. The impacts of AI/ML technology affects a wide swath of business including banking and finance, sales and marketing, healthcare, media, among others. Just over the past few years, we have seen companies like OpenAI release generative models like GPT3 (2020) and GPT4 (2023) with a 1000x increase in model size (170B to 176T parameters), and a 100x increase in compute (~300M to ~20G PetaFLOPS). To address this burgeoning demand in computational requirements, we need to rethink hardware design with two objectives: (1) keeping the computing architectures configurable, efficient, flexible, and scalable to adapt to continuously varying workload characteristics, and (2) developing rapid implementation tools to allow AI/ML developers to quickly deploy their models to silicon. Modern systems typically use a variety of solutions including GPUs (Graphics Processing Units), FPGAs (Field-Programmable Gate Arrays), and custom ASICs (Application-Specific Integrated Circuits). In the proposed research program, the long term vision is to explore a novel class of computing devices called CGRAs (Coarse-Grained Reconfigurable Arrays). CGRAs strike a balance between the performance, energy efficiency, and compute density of ASICs with the architecture flexibility of FPGAs and the ease of programmability of GPUs while targeting modern AI/ML workloads.
CGRAs have remained an academic curiosity with limited silicon density, and slow compilation runtimes. The aim is to lift these architectures into the future with a focus on customizability for AI/ML workloads, and a careful co-design of the mapping tools and architecture to allow for quick implementation and compilation. First, the project will look at the compute, storage, and communication demands of a range of AI/ML workloads, and then propose architecture models that allow us to explore various implementation strategies. Next, the project will investigate the structure of mapping tasks and evaluate algorithms that enable quick compilation to these architectures. In the next five years, the research intention is to rethink CGRA architecture for AI/ML workloads, and develop compilation/mapping algorithms to enable fast implementation.
The research conducted under this program is of vital interest to the Canadian hardware industry. Canadian startups and multinational companies such as Groq, Cerebras, AMD, Intel, Untether AI, Tenstorrent, and Microsoft have a growing appetite for hardware developers that can navigate the design space of AI/ML computations, and also develop tools and compilers to target these emerging devices. The HQP trained in this program will meet the demand for hardware engineers, and software tool developers at these companies, and several past HQP from the PI's group are placed there.