CMD30 FisMat2023 - Submission - View

Abstract title: Computing with Physical Systems based Oscillatory Neural Networks – Materials, Devices and Circuit Design Overview
Submitting author: Aida Todri-Sanial
Affiliation: Eindhoven University of Technology
Affiliation Address: Prof. Aida Todri-Sanial Head of NanoComputing Research Lab Electrical Engineering Department, Eindhoven University of Technology, NL
Country: Netherlands
Other authors and affiliations:
Oscillatory neural networks (ONNs) are a promising neuromorphic computing paradigm that harnesses the nonlinear dynamics of physical systems based on a network of coupled oscillators.  ONNs are a promising neuromorphic computing paradigm for AI at the edge. ONNs [1-10] are networks of coupled oscillators using their natural synchronization behavior to compute. In the framework of the EU H2020 NeurONN project, we have explored ONN architectures, learning algorithms, and applications to showcase ONN advantages for edge computing. We developed a fully-connected ONN in digital to explore ONN edge applications and on-chip learning capabilities. Additionally, we have also investigated analog design ONN implementation and recently, we have designed and taped out a 16-oscillator node ONN in a 65nm technology node ASIC chip with fully connected topology for solving combinatorial optimization problems. In the framework of the Horizon EU PHASTRAC project, we are exploring novel devices such as phase change materials, VO2 for oscillators and bilayer MO/HfO2 ReRAM for synaptic devices that co-integration can enable a fully analog computing platform. An overview of the design and challenges based on novel materials, devices and circuit design implementation for ONNs will be presented. References:1.         M. Abernot, et al Two-Layered Oscillatory Neural Networks with Analog Feedforward Majority Gate for Image Edge Detection Application. ISCAS 2023  ⟨hal-04007951⟩2. C. Delacour, at al, A Mixed-Signal Oscillatory Neural Network for Scalable Analog Computations in Phase Domain. 2023. ⟨hal-03961010⟩3.  C. Delacour, et al "Energy-Performance Assessment of Oscillatory Neural Networks Based on VO2 Devices for Future Edge AI Computing," doi: 10.1109/TNNLS.2023.3238473.
4.  M. Abernot et al., "Oscillatory Neural Networks for Obstacle Avoidance on Mobile Surveillance Robot E4," doi: 10.1109/IJCNN55064.2022.9891923
5.  S. Carapezzi, et al, "Simulation Toolchain for Neuromorphic Oscillatory Neural Networks Based on Beyond-CMOS VO2 Devices," doi: 10.1109/FLEPS53764.2022.9781525
6.  A. Todri-Sanial et al., "How Frequency Injection Locking Can Train Oscillatory Neural Networks to Compute in Phase," doi: 10.1109/TNNLS.2021.3107771
7. S. Carapezzi et al., "Advanced Design Methods From Materials and Devices to Circuits for Brain-Inspired Oscillatory Neural Networks for Edge Computing," doi: 10.1109/JETCAS.2021.3128756
8. C. Delacour et al., "Oscillatory Neural Networks for Edge AI Computing," doi: 10.1109/ISVLSI51109.2021.00066
9. E. Corti, et al., "Frequency Injection Locking-Controlled Oscillations for Synchronized Operations in VO2 Crossbar Devices,"  doi: 10.1109/DRC52342.2021.9467129
10.S. Carapezzi et al., "Multi-Scale Modeling and Simulation Flow for Oscillatory Neural Networks for Edge Computing,"  doi: 10.1109/NEWCAS50681.2021.9462761