DocumentCode :
2324413
Title :
A neural network integrated circuit supporting programmable exponent and mantissa
Author :
Brown, H. ; Cross, D. ; Wuerz, D., Jr. ; Liou, J.
Author_Institution :
Dept. of Electr. Eng., Univ. of Central Florida, Orlando, FL, USA
fYear :
1990
fDate :
13-16 May 1990
Abstract :
A neural network architecture is presented utilizing small (8-b) word sizes for data and weights while maintaining flexibility in the dynamic range achieved by using adjustable exponent and mantissa field. Design issues regarding tradeoff between small and large precision number systems are reviewed. Simulation results on network performance are included. In small precision systems (8-b), a fixed dynamic may limit the system capability. For certain applications it is desirable to have a flexible dynamic range implementable with a variable size of exponent and mantissa
Keywords :
digital arithmetic; neural nets; parallel architectures; 8 bits; dynamic range; mantissa field; network performance; neural network integrated circuit; precision number systems; programmable exponent; system capability; word sizes; Adders; Computer architecture; Computer networks; Concurrent computing; Decoding; Dynamic range; Logic; Neural networks; Particle separators; Registers;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Custom Integrated Circuits Conference, 1990., Proceedings of the IEEE 1990
Conference_Location :
Boston, MA
Type :
conf
DOI :
10.1109/CICC.1990.124804
Filename :
124804
Link To Document :
بازگشت