Compact Model Parameter Extraction Using Modular Q Learning for Nano-Scale Transistors

Authors:

Abstract
A reinforcement learning algorithm, using modular Q - Learning based policy, for the extraction of parameters in compact models is proposed in this work. The unique effects of each of the nine parameters on the $\mathrm{I}_{\mathrm{d}}-\mathrm{V}_{\mathrm{g}}$ and $\mathrm{I}_{\mathrm{d}}-\mathrm{V}_{\mathrm{d}}$ curves are used to extract the parameter values. The results show that the average and maximum relative errors lie between 0.1-1%. The $\mathrm{I}_{\mathrm{d}}-\mathrm{V}_{\mathrm{g}}$ (log and linear scale), $\mathrm{I}_{\mathrm{d}}-\mathrm{V}_{\mathrm{d}}$ (log and linear scale), the first derivative of current (gm), and second derivative of current obtained with the extracted parameters using the proposed algorithm show very good matching with the ideal characteristics.

This publication has 2 references indexed in Scilit: