Efficiently finding the ‘best’ solution with multi-objectives from multiple topologies in topology library of analog circuit

Abstract

This paper presents a new method using multi-objective optimization algorithm to automatically find the best solution from a topology library of analog circuits. Firstly this method abstracts the Pareto-front of each topology in the library by SPICE simulation. Then, the Pareto-front of the topology library is abstracted from the Pareto-fronts of topologies in the library followed by the theorem we proved. The best solution which is defined as the nearest point to specification on the Pareto-front of the topology library is then calculated by the equations derived from collinearity theorem. After the local searching using Nelder-Mead method maps the calculated best solution back to design variable space, the non-dominated best solution is obtained.Comparing to the optimization methods using single-objective optimization algorithms, this work can efficiently find the best non-dominated solution from multiple topologies for different specifications without additional time-consuming optimizing iterations.The experiments demonstrate that this method is feasible and practical in actual analog designs especially for uncertain or different multidimensional specifications.

Performance Driven Design Space Exploration for Multi-Objective Optimization Using Stability in Competition

“…Typically, DSE encompasses design space modeling and performance evaluation, system level tradeoffs to most effectively capture the targeted application in silicon real estate [11,13,5,7,36,12]. Although effective, DSE is often cursed by high dimensions of the exponentially exploding design space for practical size applications with an enormous amount of architectural configurations possible for each modular implementation.…”

“…DSE is thus one of the most imperative and beneficial phases of embedded project development, leveraging maximal tradeoffs and performance gain benefits, especially when applied at the early stages of system development. Space walk tools are designed to automatically generate possible design configurations variants for the given high level application task [36,34,19]. The DSE explorer takes the input from the Space Walk tools and decides on the direction and the best possible way for further exploration.…”

“…Design Space Exploration (DSE) techniques for multi-objective optimization problems have been an active area of research, which is receiving even more importance for its power aware high level synthesis applications. In recent times extraction and evaluation of the pareto optimal [19,33,7,17,36,12] design points have emerged as the most promising and time efficient way to overcome the design space complexities. In [7], the authors have developed Pareto Front Arithmetic (PFA) techniques to explore very large design space in a time efficient manner.…”

“…Additionally, the presented work develops an analytically exact method to utilize the most suitable range of the ordered design space variables making the search more accurate in terms of solution resolution, faster, as well as more easily traceable on to the design space configurations. Another noteworthy approach based on pareto-optimal design configuration is developed by authors in [36], addressing the multi-dimensional output-input space mapping technique to extract the pareto-optimal design points for analog circuit topologies. Although the research showed promising results in selection of optimal design variants in performance space it fails to trace back the performance space parameters to the design space variables, thus making the mapping exhaustive.…”

Toward high performance solution retrieval in multiobjective clustering

“…Other approaches are focused on ad-hoc methods. Those identify the desirable solution according to the specific domain of the problem [41] but they are not useful when the domain is not well-known and, unfortunately, this is the case of most real-world problems.…”