A multi-population evolution strategy and its application in low area/power FSM synthesis
A multi-population evolution strategy and its application in low area/power FSM synthesis
复制标题
多群体演化策略及其在低面积/功率FSM综合中的应用
DOI：
10.1007/s11047-017-9659-5
复制
发表时间：
2017
影响因子：
2.1
通讯作者：
Qinyu Wang
中科院分区：
文献类型：
--
作者：
Yanyun Tao;Lijun Zhang;Qinyu Wang
寻找低面积/功率状态分配是有限状态机综合中的 NP 难题。为了解决这个问题，本研究提出了一种多群体进化策略，记为MPES。 MPES 通过使用内部 ES 和外部 ES 来完成任务。在inner-ES中，亚群单独进化并负责不同区域的局部搜索。采用交替(μ+λ)策略和(μ,λ)策略从排序群体中选择亲本个体进行突变。三个突变算子“替换”、“2-交换”和“转移”对亲代个体进行操作以产生后代。分别为面积和功率评估定义了不同的适应度函数。外层 ES 充当外壳来优化内层 ES 的子群，以获得越来越好的解决方案。在outer-ES中，进化亚群的参数由outer-population的个体来表示。 Outer-ES 对外部群体进行选择和变异，以改变内部 ES 中不断演化的子群体的参数，从而生成更好的解决方案。两名助理操作员，竞争者和新生儿，共同努力消除贫困亚群并创造新的亚群。通过使用两级ES，MPES能够获得多个好的解。我们在基准测试上对 MPES 进行了广泛的测试，并从各个方面将其与以前的状态分配方法进行了比较。实验结果表明，与之前的出版物相比，MPES 显着降低了面积和功耗成本。
Finding a low area/power state assignment is a NP-hard problem in finite-state machines synthesis. In order to solve this problem, this study proposes a multi-population evolution strategy, denoted as MPES. MPES accomplishes the task by using inner-ES and outer-ES. In inner-ES, subpopulations evolve separately and are responsible for local search in different regions. Alternating (μ+λ) strategy and (μ,λ) strategy are employed to select parental individuals from the ranked population for mutation. Three mutation operators, ‘replacement’, ‘2-exchange’ and ‘shifting’, perform on the parental individuals to generate offspring. Different fitness functions are defined for area and power evaluation, respectively. Outer-ES acts as a shell to optimize the subpopulations of inner-ES for better and better solutions. In outer-ES, the parameters of evolving subpopulations are represented by individuals of outer-population. Outer-ES performs selection and mutation on the outer-population to change the parameters of evolving subpopulations in inner-ES for generating better solutions. Two assistant operators, competition and newborn, work together for poor subpopulations elimination and creating new subpopulations. By using two-level ES, MPES is able to obtain multiple good solutions. We test the MPES extensively on benchmarks, and compare it with previous state assignment methods from various aspects. The experimental results show MPES achieved a significant cost reduction of area and power dissipation over the previous publications.