Machine Learning Meets Analog Circuit Design: Intelligent Automation of IC Design

Abstract

Automating AMS circuit design procedure has always been challenging as it tightly ties to human expertise and intuition to make a relationship between various parameters and performances. Given the main circuit topology, in order to satisfy the desired specifications, the circuit parameters should be chosen optimally. So, this circuit optimization problem, i.e. determining the circuit parameter values to meet the required specifications, can be solved using mathematical optimization methods. As a result, a Computer-Aided Design (CAD) tool would be able to automate this circuit sizing optimization procedure leveraging techniques such as gradient-based, convex optimization, and evolutionary algorithms. This thesis proposes a series of works that uncover unique methods for automating the design of AMS circuits. First, we propose a fully automated Single-Board Computer (SBC) generator tool, FASCINET. FASCINET uses a Neural Network (NN) model to design customized peripheral circuits for SBCs. The tool creates a large Commercial Off-the-Shelf Database (COTS DB) of existing components, efficiently searches through them, and selects optimal components for both main and peripheral components based on the user’s requirements. Creating such a broad COTS DB requires processing abundant datasheets. In order to automate this process, we describe a novel NN-based approach for automatically categorizing datasheets and propose an extraction technique for parsing relevant functional information. Our evaluations show that FASCINET is able to design SBCs that are identical to the manually-designed ones except for minor differences. Second, we go over our table extraction tool, Tablext, which can be used on datasheets or any other documents. Third, we present AnGeL, a fully-automated analog circuit generator framework. AnGeL performs all the schematic circuit design steps from deciding the circuit topology to determining the circuit parameters i.e. sizing. Furthermore, we present a method to reduce the size of AnGeL’s database. For this purpose, we use NNs to determine the behavior of complicated circuit topologies by combining the more simple ones. By generating such unlabeled data, the time for providing the training set is significantly reduced compared to the conventional approaches. Our results show that for multiple circuit topologies, in comparison to the state-of-the-art works while maintaining the same accuracy, the required labeled data is reduced by 4.7x - 1090x. Also, the runtime of AnGeL is 2.9x - 75x faster. Fourth, we propose FuNToM, a functional modeling method for Radio Frequency (RF) circuits. FuNToM leverages the two-port analysis method along with NNs for modeling multiple topologies using a single main dataset and multiple small datasets. This significantly reduces the required number of training data. Our results show that for multiple RF circuits, in comparison to the state-of-the-art works while maintaining the same accuracy, the required training data is reduced by 2.8x - 10.9x. In addition, FuNToM needs 176.8x - 188.6x less time for collecting the training set in post-layout modeling. Finally, we conclude our work and give meaningful insights about the current challenges and open issues.