Demand for products of dynamic random-access memory (DRAM) has dramatically increased since 2019. To satisfy the soaring demand, many companies have increased their supply for DRAM products. The DRAM CoSi2 resistance significantly affects the quality of a DRAM product. The case company under this study suffered from deviations of the CoSi2 resistances of its DRAM products from a target value of 11 ohms. The deviation of the CoSi2 resistance has resulted in a low yield rate in manufacturing the DRAM products. In this paper, we propose a new method consisting of Taguchi’s method, a neural network (NN) and a genetic algorithm (GA) to reduce the deviation of the average CoSi2 resistance from a target value. The experimental result showed that the proposed method helped the case company to successfully reduce the deviation of its average CoSi2 resistance from the target value of 11 ohms, from 1.440 ohms to 0.302 ohms. As a result, the yield rate has been significantly improved, and no defective DRAM products have been returned from its customers after applying the proposed method.

中文翻译：

---

应用田口方法、人工神经网络和遗传算法降低DRAM产品的CoSi2电阻偏差

自2019年以来，动态随机存取存储器（DRAM）产品的需求急剧增加，为了满足不断增长的需求，许多公司增加了对DRAM产品的供应。DRAM CoSi2 电阻显着影响 DRAM 产品的质量。本研究中的案例公司的 DRAM 产品的 CoSi2 电阻偏离了 11 欧姆的目标值。CoSi2电阻的偏差导致DRAM产品的良品率偏低。在本文中，我们提出了一种由田口方法、神经网络 (NN) 和遗传算法 (GA) 组成的新方法，以减少平均 CoSi2 电阻与目标值的偏差。实验结果表明，所提出的方法帮助案例公司成功地将其平均 CoSi2 电阻与目标值 11 欧姆的偏差从 1.440 欧姆降低到 0.302 欧姆。结果，良率得到显着提高，并且在应用所提出的方法后，客户没有退回有缺陷的DRAM产品。