Abstract: STAM algorithm is an efficient approach to implement non-linear function approximation with hardware, in which lookup tables and addition arc involved to approximate non-linear function accurately, and size of table could be reduced remarkably due to symmetry of function. Nevertheless the first coefficient still uses relatively big table. An improved method based on differential was brought up aiming at relieve the problem, and with the algorithm hyperbolic tangent sigmoid function was implemented on FPGA. Experimental results show that 17 0 0 to 30 0 0 of memory was saved after improvement.

Key words: STAM algorithm, Non-linear function approximation, Hyperbolic tangent, Sigmoid function, FPGA