Transistor level simulations are generally used for testing the performance of electronic chips. The models used in these simulations are complex and computationally expensive. To counter this, behavioral models are used which capture the circuit response on a high level. Neural networks can be instrumental in modeling the circuit behavior accurately. Models like Transformer networks and Random Forest Regressor model will be implemented since these prove to be capable of learning or fitting time series problems, similar to our use case. The model will then be used to replace the circuit block for testing simulations, which are done by an Infineon in-house simulator. These simulations are based on solving the required differential equations iteratively based on configurations and stimuli provided to the circuit according to a set pipeline. We will also further enhance the use case by trying to predict the behavior of a slightly modified circuit without retraining or by transfer learning. This could in turn help quantify the generalizing capability of the developed models. A potential extension for this approach would be to utilize the domain and the circuit knowledge to break down the problem into smaller sub-problems, allowing for more control and exibility over the modeling of the circuit as a whole. With this thesis, we have successfully applied the approach to an LDO circuit by predicting the output waveforms to an acceptable accuracy, thereby bolstering the idea and applicability of using a machine learning model for modeling the behavior of electronic circuits.     «

Transistor level simulations are generally used for testing the performance of electronic chips. The models used in these simulations are complex and computationally expensive. To counter this, behavioral models are used which capture the circuit response on a high level. Neural networks can be instrumental in modeling the circuit behavior accurately. Models like Transformer networks and Random Forest Regressor model will be implemented since these prove to be capable of learning or fitting time...     »