Combining Machine Learning and Optimization Techniques for the High-Level Design of Σ∆Ms

This paper presents a toolbox designed for optimizing and automating the design process of Analog-to-Digital Converters (ADCs). The tool combines Artificial Neural Networks (ANNs) and optimization algorithms together with behavioral simulations to perform high-level sizing, turning system-level specifications into component-level requirements. The ANNs are trained to choose the most suitable ADC architecture for a given set of specifications and to find the optimal design parameters that meet those specifications. The predicted designs are further refined through time-domain behavioral simulations, using different optimization engines such as Simulated Annealing (SA), to maximize the figure of merit. The toolbox, demonstrated for Sigma-Delta Modulators (Σ∆Ms), includes a MATLAB-based Graphical User Interface (GUI) that assists users through the entire process, from specifications to optimization and validation¹