A massively parallel neurocomputer with a reconfigurable arithmetical unit
This paper presents a massively parallel neurocomputer system which is mainly based on a new reconfigurable arithmetical unit optimized for the simulation of neural networks. The system offers a very high performance for all typical neural network operations combined with a high flexibility to adapt the available hardware resources to the requirements of a user-selected neural network model. The main system features are the support of many different bitlengths, a high memory bandwidth, a good scalability and a dynamic reconfigurability.
Unable to display preview. Download preview PDF.
- Avellana, N., Carrabina, J., Lisa, F., Reyes, M., and Valderrama, E. “Unidad aritmetica con paralelismo configurable para emular redes neuronales. In IX Congreso de diseno de Circuitos Integrados (Gran Canarias, Spain, 1994).Google Scholar
- Avellana, N., Carrabina, J., Rabaneda, L., and Valderrama, E. Experience on the development of a neuroemulator system. In Fifth Eurochip Workshop on VLSI Training (Dresden, Germany, 1994).Google Scholar
- Hammerstrom, D. A VLSI Architecture for High-Performance, Low-Cost, On-chip Learning. In Proc. IJCNN (San Diego, 1990), pp. 537–543.Google Scholar
- Holt, J., and Hwang, J. Finite precision error analysis of neural networks hardware implementations. IEEE Transactions on Computers 42 (1993), 281–290.Google Scholar
- Morgan, N., Beck, J., Kohn, P., Bilmes, J., Allman, E., and Beer, J. The Ring Array Processor: A Multiprocessing Peripheral for Connectionist Applications. Journal of Parallel and Distributed Computing 14 (1992), 248–259.Google Scholar
- Riedmiller, M., and Braun, H. A direct adaptive method for faster backpropagation learning: The RPROP algorithm. In IEEE International Conference on Neural Networks (ICNN) (1993), H. Ruspini, Ed., pp. 586–591.Google Scholar
- Tollenaere, T. SuperSAB: Fast adaptive backpropagation with good scaling properties. Neural Networks 3, 5 (1990), 561–573.Google Scholar
- Viredaz, M. MANTRA I: An SIMD Processor Array for Neural Computation. In Euro-ARCH '93 (1993), P. Spies, Ed., Springer-Verlag, pp. 99–110.Google Scholar