PROJECT TITLE :

A Hardware-Efficient Sigmoid Function With Adjustable Precision for a Neural Network System

ABSTRACT:

A hardware-economical sigmoid operate calculator with adjustable precision for neural network and deep-learning applications is proposed in this brief. By adopting the bit-plane format of the input and output values, the computational latency of the processing time will be dynamically reduced per the user configuration. To reduce the hardware price, the coefficients used to calculate the sigmoid value can be shared for multiple calculators while not any structural hazard. Still, the restricted constraint is applied in the coefficients' coaching stage to further simplify the computation within the calculation stage with a negligible quality loss. A test module is designed for the proposal and operated at three hundred MHz to realize 75 million sigmoid calculations per second. Implemented in ninety-nm CMOS technology, the core of the calculator costs 1663 gates, and a 1-kb globally shared memory is used to store the coefficients.