Grant: Discovery Launch Supplement
Institut national de la recherche scientifique
$12,500.00
Model-Specific Deep Learning Hardware Accelerators for Edge Computing
FY 2024-25
Summary
Deep learning (DL) is fundamentally transforming various facets of our lives, revolutionizing applications that span from autonomous vehicles to the prediction of serious diseases. Nonetheless, a significant hurdle in the deployment of deep learning methods lies in their extensive computational and data complexities. In real-time applications, the inference of DL models requires substantial processing speed. Furthermore, in battery-powered Edge devices, the energy consumption associated with DL computations becomes a critical concern. To meet the rapidly growing demands for deploying deep learning across an array of emerging energy- and resource-constrained applications, more efficient computer architectures are imperative.
In our research program, we propose an original approach for crafting highly efficient deep learning hardware accelerators. Unlike existing domain-specific DL accelerators designed to support a range of DL algorithms and models, our approach centers on the precise customization of the dataflow and microarchitecture of the accelerator to match a specific target DL model. This model-specific design approach aims to attain maximum efficiency by tailoring and optimizing the dataflow and microarchitecture precisely to the computational demands of the given target model. To expedite this design process, we will employ computer-aided automation, which will engage in design space exploration to identify the optimal dataflow and microarchitecture configurations. Furthermore, we will integrate accuracy-aware model optimization techniques, such as quantization and pruning, into the design process to further enhance the final efficiency. The highly customized hardware accelerators from this approach are poised to outperform current domain-specific solutions, offering notably improved efficiency.
Success in this research program will empower the cost-effective integration of deep learning technology across an extensive spectrum of energy- and resource-constrained Edge applications. From intelligent biomedical implants to small-footprint wireless sensors, this breakthrough promises to reshape the landscape of possibilities. By doing so, it can significantly influence the industrial sector, propelling the emergence of a new generation of smart miniature Edge devices. Taking a broader perspective, the proposed model-specific acceleration has the potential to democratize deep learning technology, rendering it more attainable and cost-effective for a wide array of applications.
The interdisciplinary and hands-on nature of our proposed research, coupled with its undeniable scientific significance and extensive network spanning multiple academic institutions and industries, creates a fertile ground for cultivating highly qualified personnel (HQP). Here, they will thrive, expanding their expertise and gaining invaluable practical skills in computer architecture design, advanced machine learning, and the exciting realm of Edge computing.
Program
Discovery Launch Supplement
Location
Quebec, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data
As published
Application id
DGECR-2024-00147
Name nom
Vakili, Shervin
Department département
Énergie Matériaux Télécommunications Research Centre
Organization id
56
Institution établissement
Institut national de la recherche scientifique
Province en
Québec
Province fr
Québec
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
12500
Program id
DGECR
Program name en
Discovery Launch Supplement
Program name fr
Supplément Tremplin vers la découverte
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
2719
Research subject group en
Information technology
Research subject group fr
Technologie de l'information
Research subject en
Computer architecture and design
Research subject fr
Architecture d'ordinateur et conception
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
Model-Specific Deep Learning Hardware Accelerators for Edge Computing
Keywords
high-performance computer architecture; machine learning hardware; edge computing; microarchitecture design; approximate computing; pruning of deep neural networks
Application summary
Deep learning (DL) is fundamentally transforming various facets of our lives, revolutionizing applications that span from autonomous vehicles to the prediction of serious diseases. Nonetheless, a significant hurdle in the deployment of deep learning methods lies in their extensive computational and data complexities. In real-time applications, the inference of DL models requires substantial processing speed. Furthermore, in battery-powered Edge devices, the energy consumption associated with DL computations becomes a critical concern. To meet the rapidly growing demands for deploying deep learning across an array of emerging energy- and resource-constrained applications, more efficient computer architectures are imperative.
In our research program, we propose an original approach for crafting highly efficient deep learning hardware accelerators. Unlike existing domain-specific DL accelerators designed to support a range of DL algorithms and models, our approach centers on the precise customization of the dataflow and microarchitecture of the accelerator to match a specific target DL model. This model-specific design approach aims to attain maximum efficiency by tailoring and optimizing the dataflow and microarchitecture precisely to the computational demands of the given target model. To expedite this design process, we will employ computer-aided automation, which will engage in design space exploration to identify the optimal dataflow and microarchitecture configurations. Furthermore, we will integrate accuracy-aware model optimization techniques, such as quantization and pruning, into the design process to further enhance the final efficiency. The highly customized hardware accelerators from this approach are poised to outperform current domain-specific solutions, offering notably improved efficiency.
Success in this research program will empower the cost-effective integration of deep learning technology across an extensive spectrum of energy- and resource-constrained Edge applications. From intelligent biomedical implants to small-footprint wireless sensors, this breakthrough promises to reshape the landscape of possibilities. By doing so, it can significantly influence the industrial sector, propelling the emergence of a new generation of smart miniature Edge devices. Taking a broader perspective, the proposed model-specific acceleration has the potential to democratize deep learning technology, rendering it more attainable and cost-effective for a wide array of applications.
The interdisciplinary and hands-on nature of our proposed research, coupled with its undeniable scientific significance and extensive network spanning multiple academic institutions and industries, creates a fertile ground for cultivating highly qualified personnel (HQP). Here, they will thrive, expanding their expertise and gaining invaluable practical skills in computer architecture design, advanced machine learning, and the exciting realm of Edge computing.