基于Faster R-CNN的光刻热点检测

Lithographic Hotspot Detection Based on Faster R-CNN

光刻热点检测是集成电路可制造性设计的一项重要环节。已有研究将卷积神经网络应用于光刻热点的检测,但在卷积运算的重复性、检测结果准确度等方面存在较多问题。为了解决上述问题,提出一种基于Faster R-CNN并结合在线难例挖掘和软性非极大值抑制的光刻热点检测算法。采用ICCAD 2012Contest的版图基准作为验证载体。实验结果表明,该算法能有效提高检测的精度和效率,平均检测耗时为0.6h/mm2,召回率为96.1%,精确率可达40.3%。

Lithographic hotspot detection takes an important role in design for manufacturability (DFM)of integrated circuits.Although convolutional neural network (CNN)had been employed in hotspot detection by early researchers,many problems such as the redundancy of convolution operations and inaccuracy of detection results still exist.A lithographic hotspot detection algorithm based on Faster region-based convolutional neural networks (Faster R-CNN),in combination with schemes of online hard example mining (OHEM)and soft non-maximum suppression (Soft-NMS),was proposed to tackle the problems mentioned above.Using ICCAD 2012Contest layout benchmarks as the verification vehicle,the experimental results showed that the proposed algorithm could effectively improve detection accuracy and efficiency,with an average detection runtime of 0.6h/mm2,a recall of 96.1%and a precision of 40.3%.