Our paper, Automated Transient Input Stimuli Generation for Analog Circuits, was accepted in IEEE Transaction on CAD. This was a joint work with my advisor Shobha Vasudevan.
A pre-print of the paper can be downloaded from here or the publication page. The source code of the tools developed for and used in this project is available on project’s Github page. The article is available for early-access from IEEE

In this paper, we introduced multi-objective rapidly-exploring random trees. We use MO-RRTs to generate transient input stimuli to validate hardware circuits. Depending on the application, MORRTs can be tuned to be improve goal-orientedness or coverage or both. Goal-oriented stimuli generation is useful for exposing design or manufacturing bug whereas coverage is useful to improve the coverage of the state space. For example, the figure above shows how we used MORRT to generate stimuli for the Josephson circuit. In comparison to the standard Monte Carlo simulation, MORRT is able to efficiently expose erroneous behaviors of the circuit.

We present an automated directed random input stimulus generation algorithm with high coverage for nonlinear analog circuits. Our methodology is able to generate input stimuli to meet two kinds of objectives: (i) to reach user-defined goal regions and (ii) increased coverage of state space. The principal benefit of our approach is that it can provide directed input stimulus generation, as opposed to the randomly generated input stimulus by Monte Carlo-based methods. The methodology introduces Multiple Objective Rapidly-exploring Random Trees (MORRTs), which add a bias and a feedback loop to the standard RRT algorithm. The biasing is provided by a statistical inference algorithm. Simultaneous biasing towards goal regions and coverage is possible in MORRT to a user-defined extent. Our methodology generates several input stimuli that are concentrated in the goals or relevant operating regions, while providing high coverage of the state space. We demonstrate the efficiency and scalability of our approach on high-dimensional analog case studies.