Title :
Generation of a neuron transfer function and its derivatives
Author_Institution :
Electron. Inst., Tech. Univ. of Denmark, Lyngby, Denmark
Abstract :
A new simple circuit is presented that can be used to realise both a neuron transfer function and its approximate derivative, making it possible to use the same hardware in both learning and recall phase of the ANN. The mode of operation is digitally controlled. A CMOS differential pair has been used for realising the neutron transfer function, but a bipolar differential pair could be used equally well.
Keywords :
CMOS integrated circuits; bipolar integrated circuits; learning (artificial intelligence); neural chips; CMOS differential pair; bipolar differential pair; digitally controlled; learning phase; neuron transfer function; recall phase;
Journal_Title :
Electronics Letters
DOI :
10.1049/el:19931243