Canada Spends Logo

Explorez notre analyse du budget 2026

Grant: Alliance Grants

McGill University

$78,740.00

CTT-based neuromorphic architecture

EF 2023-24

Résumé
Machine learning applications are rapidly taking center-stage in edge devices and in large systems, covering many use cases including automotive, health care, robotics, computing, and Internet of things. All modern digital computers use classic von Neumann architectures, where compute and memory are separated, requiring continuous communication of data between the memory and computational elements. This approach limits efficiency and leads to a significant increase in power and latency. In this research, a compute in memory approach is employed based on charge trap transistors (CTTs) to design neuromorphic hardware. An array of CTTs, modeling synaptic weights, is integrated in an analog non von Neumann architecture to form a spiking neural network. The following will be investigated as part of this project: (1) characterization of CTTs and CTT arrays - to characterize CTTs, a physics-based model will be developed and simulated. The model will accommodate for the trapping phenomenon within high-k dielectric materials, thermal instability, charge leakage, and various levels of applied voltages. The models will be compared to experimental data from fabricated devices. (2) Design of CTT-based neuron architectures and circuits - building on previous research, circuits will be designed to enable the spiking functionality of the neurons at low energy. The developed circuits will be scalable to accommodate integration of many-neuron systems. A test circuit will be designed, fabricated, and tested to verify the functionality and performance of the proposed approach. (3) Modeling learning architectures to represent the CTT-based neuromorphic hardware - training of synaptic weights is currently performed offline. To ensure accurate training and correct weight update within the CTT-based synapses, the neuromorphic hardware will be modeled within the offline training algorithm to accommodate for hardware-related process, voltages, and temperature variations.
Programme
Alliance Grants
Bénéficiaire
Emplacement
Quebec, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data

Tel que publié

Application id
586551-2023
Name nom
Vaisband, Boris B
Department département
Electrical and Computer Engineering
Organization id
61
Institution établissement
McGill University
Province en
Québec
Province fr
Québec
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2023
Competition year année de concours
2023
Award amount
78740
Program id
ALLRP
Program name en
Alliance Grants
Program name fr
Subventions Alliance
Group en
RESEARCH PARTNERSHIPS
Group fr
PARTENARIATS DE RECHERCHE
Committee code
96
Committee name en
RPP Internal Decision Cttee
Committee name fr
Cté de décision interne - PPR
Area of application code
306
Area of application group en
Energy resources
Area of application group fr
Ressources d'énergie
Area of application en
Energy efficiency
Area of application fr
Efficacité énergétique
Research subject code
2800
Research subject group en
Artificial intelligence
Research subject group fr
Artificial intelligence (computer vision, use2603)
Research subject en
Intelligence artificielle
Research subject fr
Intelligence artificielle (vision artificielle, utiliser 2603)
Application title
CTT-based neuromorphic architecture
Keyword
Analog circuits; Artificial intelligence (AI); Brain modeling; Charge-trap transistor (CTT); Compute-in-memory (CIM); Neural networks (NNs); Neuromorphic architecture; VLSI
Application summary
Machine learning applications are rapidly taking center-stage in edge devices and in large systems, covering many use cases including automotive, health care, robotics, computing, and Internet of things. All modern digital computers use classic von Neumann architectures, where compute and memory are separated, requiring continuous communication of data between the memory and computational elements. This approach limits efficiency and leads to a significant increase in power and latency. In this research, a compute in memory approach is employed based on charge trap transistors (CTTs) to design neuromorphic hardware. An array of CTTs, modeling synaptic weights, is integrated in an analog non von Neumann architecture to form a spiking neural network. The following will be investigated as part of this project: (1) characterization of CTTs and CTT arrays - to characterize CTTs, a physics-based model will be developed and simulated. The model will accommodate for the trapping phenomenon within high-k dielectric materials, thermal instability, charge leakage, and various levels of applied voltages. The models will be compared to experimental data from fabricated devices. (2) Design of CTT-based neuron architectures and circuits - building on previous research, circuits will be designed to enable the spiking functionality of the neurons at low energy. The developed circuits will be scalable to accommodate integration of many-neuron systems. A test circuit will be designed, fabricated, and tested to verify the functionality and performance of the proposed approach. (3) Modeling learning architectures to represent the CTT-based neuromorphic hardware - training of synaptic weights is currently performed offline. To ensure accurate training and correct weight update within the CTT-based synapses, the neuromorphic hardware will be modeled within the offline training algorithm to accommodate for hardware-related process, voltages, and temperature variations.