FPGA implementation and comparative analysis of sigmoid calculators processing the full argument range in view of symmetry

This paper implements on FPGA and analyses two sigmoid calculation circuits: the first circuit uses the function’s symmetry, and the second one processes the full argument range. Each circuit is implemented with a bit width of 7 to 10 bits. It is shown that, compared to the second circuit, the first one consumes a smaller amount of resources. But the first circuit uses auxiliary blocks containing long carry chains. sigmoid function; neural network; FPGA; bit-level mapping method; symmetry; critical path.