CAJ | 학술논문

针对基于2D Mesh结构的片上网络功耗优化问题，提出了一种类电磁优化的片上网络低功耗映射算法.该算法采用实数编码机制，将类电磁算法应用于求解离散问题.使用轮盘赌的选择机制进行种群初始化，提高初始化粒子的质量，从而提高算法效率；利用调整序的方法进行局部搜索，提高粒子在局部范围内的精细搜索能力；设计电荷计算公式求解合力，用阈值滤掉作用力甚微的粒子，提高搜索最优解的效率.实验表明：改进类电磁的映射算法与现有的遗传算法、蚁群算法相比，平均节能达20.35%和12.58%，有效地降低了片上网络通信能耗，并且能耗分布更加均匀，算法效率更高.
침대기우2D Mesh결구적편상망락공모우화문제，제출료일충류전자우화적편상망락저공모영사산법.해산법채용실수편마궤제，장류전자산법응용우구해리산문제.사용륜반도적선택궤제진행충군초시화，제고초시화입자적질량，종이제고산법효솔；이용조정서적방법진행국부수색，제고입자재국부범위내적정세수색능력；설계전하계산공식구해합력，용역치려도작용력심미적입자，제고수색최우해적효솔.실험표명：개진류전자적영사산법여현유적유전산법、의군산법상비，평균절능체20.35%화12.58%，유효지강저료편상망락통신능모，병차능모분포경가균균，산법효솔경고.
A low-energy mapping algorithm based on the modified electromagnetism-like mechanism is proposed.Real number coder is used to realize the transformation of electromagnetism-like algorithm from continuous to discrete space.To improve the initializing particle quality and the algorithm efficiency, roulette wheel selection is used for population initialization.Adj ustment sequence is designed to improve the efficiency of searching local fine particles.The new charge formula is proposed to get the force,and some particles are filtered out to improve the efficiency.Experimental results shows that the proposed algorithm is more efficient in energy optimization.Compared with the existing genetic algorithm and ant colony algorithm,the electromagnetism-like mapping algorithm has a more evenly distributed energy consumption, and the algorithm can also save 20.35% and 12.58% energy on average.Thus the energy consumption of the NoC is effectively reduced.