Phastrac Executive Summary
Artificial intelligence is now widely deployed with various applications covering all aspects of our society. Sensing the environment around us is fundamental to the next generation of AI where devices will be capable of perception and reasoning. It is estimated that the next wave of AI revolution will come from next-generation analog-driven industrial electronics connected to the physical world via embedded sensors In the PHASTRAC project, we are proposing a novel neuro-inspired computing architecture where information is encoded in the “phase” of coupled oscillating neurons or oscillatory neural networks (ONN).
PHASTRAC will showcase a novel analog physical computing paradigm based on oscillatory neural networks that are implemented with innovative material devices such as phase change insulator-metal transition devices based on VO2 and novel bilayer MO/HfO2-based analog memristors (where MO stands for metal oxide), also referred as MO/HfO2 resistive random-access memories (RRAM). Neurons are emulated by their oscillatory behavior via phase change VO2 devices and synapses are emulated via coupling weights among neurons via MO/HfO2 RRAM. Describing neurons as oscillators allow phase-based analog computing, where the information is encoded on the phase difference among oscillators rather than the signal amplitude, thus enabling ultra-low power computation. Additionally, synaptic coupling via analog MO/HfO2 RRAM provides analog resistive switching, good retention memory and tunability to emulate synaptic behavior for learning rules. The proposed computing paradigm is based on analog computing of oscillatory neural networks that can be applied for associative memory applications. Thus, PHASTRAC will develop a novel and alternative energy efficient neuromorphic computing paradigm based on energy efficient devices and architectures.
PHASTRAC aims to develop the first-ever analog sensing to computing ONN hardware platform (targeting two demonstrators) and complete with an ONN design methodology toolbox covering aspects from ONN architecture design to algorithms in order to facilitate adoption, testing and experimentation of ONN demonstrator chips by all potential users to unleash the potential of analog sensing to compute ONN technology.
WP1: Materials and Devices Processing for ONN Fabrication | Leader: IBM Research Zurich
In the work package, the goal is to develop the fabrication and process steps for the artificial neurons and synapses. Novel computing devices and circuits for an alternative computing paradigm such as insulator-metal transition VO2 devices for emulating oscillating neurons and bilayer MO/HfO2 memristor devices for emulating synaptic weights, will be developed.
WP2: Device to Architecture Design for Novel ONN Computing Paradigm | Leader: TU Eindhoven
The objective of this work package is to develop modeling and design methods for exploring innovative materials for energy efficient computing. Partners will develop models and simulations for investigating the novel devices’ behavior and their impact on ONN circuits. Ultimately, design space exploration will be performed with novel devices for analog ONN architecture design.
WP3: Multi-Modal Dynamical Learning | Leader: PPKE
In this work package, partners will develop ONN templates for a number of sensory functions, dynamic learning methodology and their circuit implementation for processing sensory information. The focus will be also given to learning algorithms suitable for ONNs to process multiple analog signals varying in frequency and their hardware implementation.
WP4: Applications and Demonstrators | Leader: BMW
The objectives of this work package are to develop demonstrators to investigate analog sensing to computing paradigm with ONNs and also investigate multi-modal dynamical learning for robotics and intelligent interior for autonomous vehicle applications.
WP5: Dissemination and Exploitation | Leader: IBM Research
This work package aims to ensure that the PHASTRAC project has a high impact. To ensure this, we provide effective communication to the public and scientific community, disseminate to appropriate stakeholder groups, to obtain inputs, feedback and requirements from potential stakeholders (industry, end users) for the final integrated system. We will also organize two workshops during the PHASTRAC project.
WP6: Project Management | Leader: TU Eindhoven
The objective of this work package is to ensure that the project meets its objectives within budget and scheduled timescales. Tasks will include monitoring project progress, tracking deliverables and reporting back to the Consortium.