Hybrid CMOS/memristive neuromorphic systems for data analytics
In this project we will bring together competences in complementary fields to lay the foundations for building a new generation of computing systems that can be applied to data analytics and pattern recognition tasks to extract knowledge from the “big data” available within the consortium partners, available from on-line services, and produced in real time by multiple types of sensors. The project will focus on three main activities (sub-projects):
1. Memristive devices (fabrication, characterization and modeling)
2. Neuromorphic computing architectures (hybrid CMOS/memristive circuit design and software modeling for learning and classification)
3. System integration and applications (integration of memory devices into neuromorphic architectures and application to data analytics)
These activities will be carried out with strong interactions among the project partners: in (1), EPFL will develop TaOx-based memristive devices and IBM will study their device physics and develop accurate models for circuit simulations; in (2) INI will design, jointly with EPFL, CMOS spiking neural network architectures that allow the integration of the TaO x -based memristive devices as synapse elements, and develop jointly with IBM large-scale behavioral model simulations for testing machine-learning models for pattern recognition and classification; in (3) EPFL and INI will post process the prototype CMOS chips developed in (2) to integrate the memristive devices onto them, and IBM and INI will test these devices at the system level to validate the large-scale behavioral simulations, and apply them to data analytics tasks.