Absynth: A Comprehensive Approach for Full Front to Back Analog Design Automation

The benefits of any form of automation are well known, however unlike most other forms of automation, analog design automation (ADA) is not yet accepted by most designers as part of the work-flow. The goal of this work is to explore the needs and the targets to make ADA viable in industry and research. There are various approaches targeting different sets of designers of different experience levels. In this work, we present ABSYNTH, an ADA tool which can be used with or without expert knowledge depending on the interest and proficiency of the user. It also makes use of advanced modeling techniques like Support Vector Regression, and other neural network techniques for learning the circuit models during the initial runs and contribute during repeated runs of the circuit. Four circuits have been presented where automation of different steps have been carried out. three of these circuits have been used in a manufactured chip, that will soon be tested. A generic hybrid layout generator based on a constrained template which can be optimized while maintaining the matched structure has been presented. This improves the quality of the layout by producing results comparable or better than those predicted by simulators without parasitic extraction.