link to homepage

Navigation and service

PhD position on Neuromorphic computing: The MAMMAL Learning Machine - A Hierarchical Memory Approach Based on Resistive Switching Memory for Neo-Cortex Inspired Machine Learning of Temporal and Spatial Series

Advertising institute: PGI-10 - JARA-Institute Energy-efficient information technology (Green IT)
Reference number: D051/2018, Electronic Engineering

Neuromorphic computing is becoming increasingly important to solve cognitive tasks in applications such as image recognition, automatic translation, large data analysis, and autonomous driving. Its concept relies on brain-inspired artificial neural networks, that are trained rather than programmed. However, today’s successful so-called Deep-Learning systems are only running on software and are still implemented on classical von Neumann computers, which makes them very energy inefficient compared to the brain. Furthermore, their need for massive training data and supervised learning indicates that our brains function very differently.

Hence, the need for a better understanding of the brain, together with improved hardware to mimic that brain functionality. The proposed research work will aim at combining both. It will investigate in more detail Neuroscience based alternative “algorithms” for neuromorphic computing, based on the so-called Hierarchical Temporary Memory concept. The hardware implementation of these algorithms will be realized on resistive switching ReRAM devices, which constitute a new type of emerging memristive devices. The intrinsic variability of these devices offers the potential to store probabilities, and by that to make circuits that not only recognize but also predict.

As this research directly links neuroscience insights with emerging device properties, it spans over two different institutes: the Computational and Systems Neuroscience Institute (INM-6, Prof. Markus Diesmann) and the JARA-Institute Green IT (PGI-10, Prof.’s Rainer Waser, Matthias Wuttig, and Tobias Noll). The role of the INM-6 institute is to provide insights in network structure and connectivity, role of prediction/comparison, and data representation in the brain and, eventually, simulation of the “learning system” elements using platforms as NEST. The role of PGI-10 is on learning system and algorithm development, as well as on the optimization of different types of RS devices (Redox-based resistive RAM as well as Phase Change memory devices).

Type of work
The work consist of building and programming of a prototype system around a ReRAM array, on which different learning algorithms will be implemented and evaluated.

Preferred profile of candidate: Electronic Engineer

Interesting read: “On Intelligence”, author Jeff Hawkins

Forschungszentrum Jülich aims to employ more women in this area and therefore particularly welcomes applications from women. We also welcome applications from disabled persons.

Contact person:
Prof. R. Waser, Dr. Dirk Wouters

Forschungszentrum Jülich
Peter Grünberg Institute
JARA-Institut Energy-efficient information technology Green IT (PGI-10)
Jülich, Germany