 |
The Local Cluster Neural Network (LCNN) chip |
|
Description
The LCNN chip is low cost neural network hardware device developed specifically for real-time control applications. Because it is an analog computing device it achieves a high degree of parallel computation in a smaller size and at lower power consumption than a digital equivalent. Furthermore, it simplifies electronic control systems because it does not require analog to digital (A/D) and digital to analog (D/A) conversion on inputs and outputs.
The LCNN chip implements the neural network architecture developed at QUT by S.Geva and J.Sitte. Local Cluster networks are feed forward neural nets that have much in common with Radial Basis Function neural nets but are more general. They can be used as classifiers, function approximators and as generalised fuzzy inference engines.
|
T. Koerner designed, fabricated and tested the LCNN analog VLSI chip at the Heinz Nixdorf Institute (HNI) of the University of Paderborn under the direction of Prof. U. Rückert in collaboration with J. Sitte (QUT). The current version of the chip has 6 analog inputs, 8 local clusters and one output. Multiple chips can be combined into networks with larger numbers of local clusters. The current implementation uses digital storage for the network weight coefficients. Network training will typically occur in a chip-in-the-loop arrangement with a computer that carries out the calculations for the weight adaptation algorithm. For fuzzy control applications the precalculated weights for multidimensional membership functions may be downloaded onto the chip |
LCNN chip features
Support
The LCNN chip is supported by feedforward neural net simulation software and a development board with associated control software for Windows NT/2000 platforms.
Contacts:
| Dr. Joaquin Sitte School of Computing Science Faculty of Information Technology Queensland University of Technology GPO Box 2434, Brisbane, Q 4001 Australia Phone +61 7 3864 2925 Fax +61 7 3864 1801 e-mail j.sitte@qut.edu.au http://www.fit.qut.edu.au/~sitte |
Prof. Dr. Ing. Ulrich Rückert Heinz Nixdorf Institut Fachgebiet Schaltungstechnik Heinz Nixdorf Institut Universität-GH Paderborn Fürstenallee 11, 33102 Paderborn Germany Phone +49 5251 606346 Fax. +49 5251 606351 e-mailrueckert@hni.uni-paderborn.de http://hni.uni-paderborn.de/sct |
References
T. Koerner, U. Rueckert and J Sitte, Local cluster neural net analog {VLSI} design, Neurocomputing, 19 (1998) 185-197.
S. Geva, K. Malmstrom and J. Sitte, Local Cluster Neural Net: Architecture, Training and Applications, Neurocomputing 20 (1998) 35 - 56.
S. Geva and J. Sitte, A Constructive Method for Multivariate Function Approximation by Multi-Layer Perceptrons, IEEE Transactions on Neural Networks 3 (1992) 621-623.
T. Koerner, J. Sitte and U. Rueckert, An Analog Local Cluster Neural Net for a 3V supply, Proceedings of the 7th International Conference on Microelectronics for Neural, Fuzzy and Bio-inspired Systems, Granada, Spain (1999), pp. 292- 298.
T. Koerner, U. Rueckert and J. Sitte, An Analog Current Mode {VLSI} Local Cluster Neural Net, Proceedings of the 6th International Conference on Microelectronics for Neural Networks . Evolutionary \& Fuzzy Systems, Dresden, Germany (1997) pp. 257 -- 262.
