Sub-micron Parameter Scaling for Analog Design Using Neural Networks

In present work we aimed to develop a NN based approach to translate the design parameter from a submicron technology to a long channel one. The proposed approach is able to find the superseded design parameters in 1.2 mum technology using the input information, which are design parameters of Gain, Phase Margin, Unity Gain Bandwidth and Power in TCMS 0.18 mum. The training data are obtained by various simulations in the HSPICE design environment with TSMC 0.18 mum process nominal parameters. The neural network structure is developed and trained in the C++ based program. To observe the utility of proposed neural network model it is tested through at least 50 samples. Experimental results show validity of our approach in less than 0.5 db error for gain, less than 1 degree phase error for phase margin, 0.1 db for unity gain bandwidth and 0.05 db for power.