This Article 
 Bibliographic References 
 Add to: 
Arithmetic Unit Design for Neural Accelerators: Cost Performance Issues
October 1995 (vol. 44 no. 10)
pp. 1256-1260

Abstract—Arithmetic unit design is a key issue when supporting the computational requirements of neural networks. However, there is little quantitative evidence from the study of existing neural accelerators to help choose between arithmetic constructs. This paper presents an assessment of the cost-performance trade-offs between arithmetic constructs for linear neural accelerators.

[1] S.Y. Kung and J.N. Hwang,“Digital VLSI architectures for neural networks,” Proc. IEEE Int’l Symp. Circuits and Systems, pp. 445-448, 1989.
[2] M. Pacheco and P. Treleaven,“A RISC architecture to support neural network simulation,” Proc. Int’l Workshop Artificial Neural Networks, pp. 482-488,Spain, Springer-Verlag, June 1993.
[3] D. Naylor,S. Jones,D. Myers,, and J. Vincent,“Design and application of a real time neural network based image processing system,” to be published in Int’l J. Neural Systems, Word Scientific Publishing Co.
[4] K. Asanovic,J. Beck,T. Callahan,J. Feldman,B. Irissou,B. Kingsbury,P. Kohn,J. Lazzaro,N. Morgan,B. Stoutamire,, and J. Wawrzynek,“CNS-1 architecture specification,” Technical Report TR-93-201, Univ. of California, Berkeley, ICSI, Apr. 1993.
[5] R.W. Means and L. Lisenbee,“Extensible linear floating point SIMD neurocomputer array processor,” Proc. IEEE Int’l Joint Conf. Neural Networks, vol. 1, pp. 587-592, 1991.
[6] T. Baker,“Artificial neural network and image processing using the adaptive solution’s architecture,” SPIE Vol. 1452 Image Processing Algorithms and Techniques II, pp. 502-511, 1991.
[7] S.R. Jones,K. Sammut,, and J. Hunter,“Learning in linear systolic neural network engines: Analysis and implementation,” IEEE Trans. Neural Networks, vol. 4, pp. 584-593, July 1994.
[8] J. Hunter and S. Jones,“Register requirements for linear array neural network emulators,” IEE Electronics Letters, vol. 28, no. 19, pp. 1,804-1,805, Sept.10, 1992.
[9] K. Shoemake and T. Duff, "Matrix Animation and Polar Decomposition," Graphics Interface 92, Canadian Information Processing Soc., Toronto, Ontario, Canada, 1992, pp. 258-264.

Index Terms:
Neural networks, linear array accelerators, arithmetic constructs, instruction set measurements, cost/performance trade-offs.
S.r. Jones, K.m. Sammut, "Arithmetic Unit Design for Neural Accelerators: Cost Performance Issues," IEEE Transactions on Computers, vol. 44, no. 10, pp. 1256-1260, Oct. 1995, doi:10.1109/12.467702
Usage of this product signifies your acceptance of the Terms of Use.