1: \begin{abstract}

2: 	This paper introduces new perspectives on analog design space search.

3: 	To minimize the time-to-market, this endeavor better cast as constraint satisfaction problem than global optimization defined in prior arts.

4: 	We incorporate model-based agents, contrasted with model-free learning, to implement a trust-region strategy. As such, simple feed-forward networks can be trained with supervised learning, where the convergence is relatively trivial.

5: 	Experiment results demonstrate orders of magnitude improvement on search iterations.

6: 	Additionally, the unprecedented consideration of PVT conditions are accommodated.

7: 	On circuits with TSMC 5/6nm process, our method achieve performance surpassing human designers.

8: 	Furthermore, this framework is in production in industrial settings.

9: 

10: \end{abstract}

11: