An Artificial Neural Network Assisted Optimization System for Analog Design Space Exploration

Analog and Mixed-Signal IC Security via Sizing Camouflaging

“…There exist several commercial CAD tools for automatic analog circuit sizing, for example the Optimizer in Eldo tool by Mentor Graphics, A Siemens Business, the WiCkeD tool by MunEDA, and the ID-Xplore by Intento Design. There are also several tools proposed in the literature (for example, see [31]- [37]).…”

Transfer learning with bayesian optimization-aided sampling for efficient AMS circuit modeling

“…Nevertheless, the P2M function illustrated in [13] is almost linear because the sample region is not sufficiently wide. For more complicated situations with more parameters and larger sample region, as shown in [17], artificial neural networks (ANNs) dominantly outperform SVR.…”

“…ANNs are proved to be very suitable for representing the complex behavior of AMS circuits because of their large model capacity and the recent development of computational capacity. In [17], the author used a shallow ANN and a deep neural network (DNN) to model the P2M function of an opamp. In comparison with SVR [13], ANN can provide a more precise metrics prediction of a complicated analog circuit.…”

“…In comparison with SVR [13], ANN can provide a more precise metrics prediction of a complicated analog circuit. Furthermore, as demonstrated by [17], the well-trained ANN can be used to efficiently optimize the circuit that has been modeled. However, the intrinsic drawbacks of optimizing a circuit based on (NN) -i.e., the inevitably large number of training samples as well as long simulation and training time-are still unresolved.…”

“…With the recent development of advanced machinelearning algorithms and computational capacity, there is an emerging opportunity to accurately characterize the complicated function between the input design parameters and the performance metrics of an AMS circuit by, for example, representing the parameter-to-metric (P2M) function using machine-learned regression models, such as neural networks (NN) [6,17] or support vector regression (SVR) [7,13]. Thus, because of its relatively lower computation cost, the objective is to use the regression model to expedite the parameter search process over wide target specifications.…”