论文信息 - Implementation of programmable digital sigmoid function circuit for neuro-computing

Implementation of programmable digital sigmoid function circuit for neuro-computing

A programmable digital sigmoid function generator circuit has been designed and implemented. The circuit is based on second order approximation of the nonlinear sigmoid function and utilizing its characteristics. It offers an easy way of changing the steepness of the function in the transition state. In addition, the squaring function operation was implemented using combinational logic. This offers a compact realization of the system as a parallel bit architecture. The maximum output error is .047 and the average error is within 0.01. Simulations show that the circuit operates satisfactorily up to 50 MHz. The digital VLSI circuit has been fabricated using MOSIS/AMI N-well 1.2 /spl mu/m CMOS double metal process. It occupies an area of 900/spl times/450 /spl mu/m/sup 2/.

H.S. Abdel-Aty-Zohdy | M. Al-Nsour

[1] Stamatis Vassiliadis,et al. Sigmoid Generators for Neural Computing Using Piecewise Approximations , 1996, IEEE Trans. Computers.

[2] C. Alippi,et al. Simple approximation of sigmoidal functions: realistic design of digital neural networks capable of learning , 1991, 1991., IEEE International Sympoisum on Circuits and Systems.

[3] D. J. Myers,et al. Efficient implementation of piecewise linear activation function for digital VLSI neural networks , 1989 .

[4] H.S. Abdel-Aty-Zohdy,et al. VLSI design and implementation of an improved squaring circuit by combinational logic , 1997, Conference Record of the Thirty-First Asilomar Conference on Signals, Systems and Computers (Cat. No.97CB36136).

[5] A. Tsoi,et al. Implementation issues of sigmoid function and its derivative for VLSI digital neural networks , 1992 .

[6] Andrew J. Viterbi,et al. Error bounds for convolutional codes and an asymptotically optimum decoding algorithm , 1967, IEEE Trans. Inf. Theory.