Explorando Predição da Caracterização Elétrica com Machine Learning
Gabriel Lima Jacinto, Lucas Yuki Imamura, Mateus Grellert, Cristina Meinhardt · 2023
ABSTRACTWith the advancement of integrated circuit manufacturing technology,more and more aspects must be considered during the electricalcharacterization of circuits in order to solve challenges such as processvariability effect. This increases the characterization time dueto traditional techniques based on exhaustive electrical simulations.The adoption of machine learning techniques already helps digitaldesign at many levels of abstraction. Thus, the main objective ofthis research is to evaluate machine learning regression algorithmsas an alternative to exhaustive electrical simulation in the cell characterizationproject. In this step, multiple linear regression, supportvector regression, decision trees and random forest algorithms wereconsidered. This work presents the results of NAND2 and NOTgates using bulk CMOS technology. Specifically, the energy valuesand the propagation times of this circuit will be predicted separately.A comparative analysis, together with the inference time,is made for each dependent variable between the models, in orderto understand which is the best regression model for the task. Thealgorithm with the lowest cost function and shortest inference timeproved to be the decision tree for all predicted variables in bothgates.