Physical and HDL designs for a library of neuron components for use in intelligent systems
V. Srinivasan, Joothiram Jayamani Athreya, M. Mailavaram, Carla Purdy · 2005
A library of neural network components suitable for hardware implementation has been created to enable development of entire networks. Designs for the unit step, linear threshold, sigmoid and Gaussian activation function circuits have been developed in the Verilog-AMS hardware description language (HDL) and performances have been compared with SPICE simulations. This work demonstrates how an HDL can be used to conveniently design and simulate complex systems to build a library of codes for networks, which will ultimately be efficient and cost-effective for automatic synthesis