PREDICTING THE YIELD OF INTEGRATED CIRCUITS BY TISM WITH NEUTRAL NETWORK

通过基于全解释结构模型的神经网络预测芯片成品率

陈思言

11849217

硕士

集成电路工程领域工程

骆宗伟

2020-05-30

2020-07-20

哈尔滨工业大学

深圳

The importance of integrated circuits has again attracted attention recently due to the trade friction between US and China, and between Japan and Korea recent year. In recent years, for example, ZTE has been unable to cope with US sanctions. Huawei has also been banned from importing US semiconductor products on the grounds of endangering national security since May 26, 2020. As the integrated circuits processing performance and technology level of integrated circuits become increasingly polarized, the importance of IC output is becoming more prominent. The field of integrated circuits has gradually become a pawn in the game between great powers. One of the most important indicators of IC output is chip yield. The yield determines the manufacturing cost and process limits of the chip. In the past, many studies on the IC chip yield have remained on specific processes and specific types of integrated circuits. For example, some studies start with photolithography in the first step of IC processing, while others start with encapsulation in the last step of IC processing. This research adopts method called deep learning networks to explain the entire process of integrated circuit processing and further to calculate and predict the IC chip yield innovatively. We use Totally Interpretive Structural Modeling and two neural networks with strong explainability to combine the method we called TISM+DLNN. And we get good result by using the data we collected. More importantly, we propose a highly explainable method based on neural network and TISM for solving multi-factor complex problems. It provides ideas for dealing with such problems in the future.

近些年由于美国和中国之间以及日本和韩国之间的贸易摩擦，使得集成电路的重要性最近再次引起关注。例如近些年中兴就因为美国的制裁一蹶不起。而华为也从2020年5月26日起被美国以危害国家安全为由禁止进口美国半导体制品。由此可见随着集成电路的处理性能和集成电路的技术水平分化程度越来越大，集成电路的产量的重要性变得越来越突出。集成电路领域也逐渐成为大国之间博弈的棋子。集成电路产量的最重要指标之一是芯片成品率。成品率决定着芯片制造成本和工艺极限。过去，对集成电路芯片成品率的许多研究仍保留在特定的工艺和特定类型的集成电路上。比如说有些研究从集成电路前段加工的第一步光刻入手，也有的研究是从集成电路加工的最后一步封装入手。本研究创新性地采用称为深度学习网络来解释集成电路生产的全过程，并进一步计算和预测集成电路芯片的产量。我们使用了全解释结构模型加上两种可解释性很强的神经网络构成了我们称之为全解释结构模型+深度学习神经网络的方法。并且通过这个方法在我们收集到的数据中进行运算，得到了不错的结果。更重要的是，我们提出了一种高度可解释的基于神经网络和全解释结构模型的方法，用于解决多因素复杂问题。它为将来解决此类问题提供了思路。

integrated circuit yield prediction Totally Interpretive Structural Modeling machine learning explainability

集成电路 产量预测 全解释结构模型 机器学习 可解释性

英语

联合培养

南方科技大学

Chen SY. PREDICTING THE YIELD OF INTEGRATED CIRCUITS BY TISM WITH NEUTRAL NETWORK[D]. 深圳. 哈尔滨工业大学,2020.