Cascading Neural Network Blocks of Transistor Level Transceiver Models

This paper presents a feed-forward neural network (FNN) framework devised for cascade-able transceiver behavior modeling. Voltage waveforms after each cascading block are predicted by the proposed method and directly fed to the next block as input. Unlike conventional IBIS-AMI models, FNN models do not interact with any SPICE-like transient solver, and hence are not limited by slow convergence or convolution. Comparison of accuracy and time efficiency are drawn and compared with a commercial simulation to demonstrate the effectiveness of employing FNN in transceiver macromodeling.