MACHINE LEARNING BASED ANALOG CIRCUIT DESIGN AUTOMATION

Abstract

This study proposes a machine learning-based approach for the automated design of a common-source (CS) amplifier with a diode-connected load [19]. The objective is to optimise the circuit's performance characteristics, such as gain and power consumption while considering the constraints imposed by the diode-connected load utilitsing the NN model. A dataset comprising various input parameters and corresponding circuit performance metrics is collected from a set of simulated CS amplifiers with diode-connected loads to achieve this. Deep learning algorithms, specifically Multi-layer perceptron models, are then trained on this dataset to establish the relationships between the input parameters and the desired performance metrics. The trained models are subsequently utilised for automated circuit design. Given the desired specifications and constraints, the machine learning algorithm predicts the optimal values for the circuit parameters, including transistor dimensions, biasing, and load characteristics. This approach reduces the design iteration time and the reliance on manual tuning, enabling faster and more efficient circuit design. The effectiveness of the proposed method is evaluated through extensive simulations and comparisons with different number of datasets in order to compare performance of estimation with respect to dataset. The results demonstrate that the machine learning based automated design approach achieves improved RMSE and MSE of aspect ratio estimation Moreover, the automated design process exhibits robustness and scalability across different design specifications. In conclusion, this research introduces a novel approach to automate the design process of CS amplifiers with diode-connected loads using machine learning techniques. The proposed method offers significant advantages in terms of efficiency, performance, and adaptability. It holds great potential for accelerating the development of analog circuits and fostering advancements in the field of circuit design.