Computing with Silicon Neurons
29 January 2014
Scientists use artificial nerve cells for complex data processing tasks
Photo: Kirchhoff-Institut für Physik
Scientists from Berlin and Heidelberg have succeeded in using artificial nerve cells for complex tasks in parallel data processing. These silicon neurons deploy a “neuromorphic” chip that allows them to classify different types of data, thus enabling them to recognize handwritten numbers or distinguish plant species based on their flowers. Researchers from the Institute of Biology/Neurobiology of the Freie Universität Berlin, the Bernstein Center for Computational Neuroscience Berlin and the Kirchhoff Institute for Physics of Heidelberg University participated in the project, whose results were published in the journal PNAS.
As in the past, most computer programmes still process data serially. For their work, the research team refined a new technology that is based on parallel data processing. Through what is known as neuromorphic computing, the silicon neurons perform the computations on special computer chips. These artificial nerve cells are linked together similar to cells in the human brain. When the cell network is fed data, all the silicon neurons work in parallel to solve the problem. The precise nature of their connections determines how the data is processed. The researchers have now developed a special network – a neuromorphic “programme” – for this chip that solves a fundamental computing problem: classifying data with different features.
In designing the network architecture, the researchers found their inspiration in the olfactory nervous system of insects, which is by nature optimised for highly parallel processing of the complex chemical world. The scientists used a chip with neurons made of silicon that was developed at the Kirchhoff Institute for Physics of Heidelberg University. Computer programmes that can classify data are used in a variety of technical devices, such as smartphones. The neuromorphic network chip could also be used in supercomputers that are built on the model of the human brain. The research was conducted by Dr. Michael Schmuker and Prof. Dr. Martin Paul Nawrot Freie Universität Berlin/Bernstein Center Berlin) as well as Thomas Pfeil (Heidelberg University), a researcher and doctoral candidate in the Electronic Vision(s) working group of the Kirchhoff Institute for Physics.
Original publication:
M. Schmuker, T. Pfeil & M.P. Nawrot (2014): A neuromorphic network for generic multivariate data classification. PNAS, published ahead of print January 27, doi:10.1073/pnas.1303053111
