【作者】 郭晓光；

【导师】 郭东明；

【作者基本信息】 大连理工大学 ， 机械制造及其自动化， 2008， 博士

【摘要】 超精密磨削技术是先进制造技术领域的前沿课题,是未来发展我国微电子产业的关键技术。在超精密磨削过程中,特别是进行纳米级加工时,材料以离散的数个原子或原子层的方式去除,因此加工过程中的能量分配、已加工表面的形成、材料的去除等都与常规加工存在巨大差别。因而对加工过程采用建立在传统连续介质力学基础上的切削理论来解释显然是不合适的,并且难于应用仪器对微观现象进行观察和测量。而分子动力学仿真是在理论上研究超精密加工过程的一种非常有效的方法,它提供了一条从系统的微观细节探索实验过程宏观特性的捷径,其具有可实验性、安全性、可进行超前研究、减少实验量等特点,目前在物理、化学、生物、医药以及材料等多个研究领域得到了广泛应用,并逐渐渗透到机械加工领域。对分子动力学基本理论和方法进行了深入研究,将固体物理学中的Debye模型引入到单晶硅原子动能和温度之间的转换过程中,建立了适合于单晶硅纳米级磨削过程的分子动力学仿真模型。研究了分子动力学并行算法,提出了基于区域分解法的区域二次划分的分子动力学并行算法、原子亲属表的概念和基于“永久序号”的消息传递策略,这些算法和策略的应用大大简化了编程,减小了程序出错的概率,同时也有效节省了通信开销和仿真时间。借助于联想深腾1800高性能服务器编制了分子动力学并行化仿真程序,进行了计算机实验,结果表明:并行化仿真程序相对串行仿真程序,模拟规模由几千个原子提高到数十万个原子,计算时间至少缩短了十倍,并且该程序具有很好的加速比、并行效率和扩展性。对单晶硅纳米级磨削过程进行计算机仿真,从磨削过程中瞬间原子位置、磨削力、原子间势能、损伤层深度等角度分析了纳米级磨削加工的机理,发现硅原子间势能的变化是导致单晶硅亚表面损伤的重要原因,在原子量级条件下单晶硅亚表面损伤层深度主要指沿磨削深度方向原子发生不规则排列的原子层的最大厚度。分子动力学仿真结果还发现:单晶硅纳米级磨削过程产生的亚表面损伤的主要形式是非晶结构,无明显的位错产生;磨粒原子与硅原子之间有粘附现象发生,这是由于纳米尺度磨粒的表面效应产生的。应用分子动力学方法,从理论上研究了磨粒钝圆半径、磨削深度和磨削速度对单晶硅纳米级磨削机理和工件亚表面损伤的影响,研究表明:磨粒钝圆半径、磨削深度和磨削速度对单晶硅纳米级磨削的机理几乎没有影响,只是在磨削力、能量和损伤层深度等方面有些差异。在磨削深度和磨削速度相同情况下,随着磨粒钝圆半径的增加,单晶硅亚表面损伤层变厚,这时与宏观实际情况一致。对于相同磨粒钝圆半径和磨削速度条件下的仿真结果表明:随着磨削深度的增大,单晶硅表面和亚表面质量恶化。在磨削深度和磨粒钝圆半径相同的情况下,在20～200m/s范围内,磨削速度对单晶硅亚表面损伤影响很小,说明分子动力学仿真对磨削速度的变化不敏感,因此可以适当提高仿真速度,从而缩短仿真时间和扩大仿真规模。为了进一步扩大分子动力学仿真的规模,开发了数百万粒子规模单晶硅超精密磨削的分子动力学并行化仿真软件。提出了数据“挖空”过滤算法,结合球形绘图法实现了大规模粒子仿真结果的可视化。最后应用该软件对单晶硅超精密磨削过程进行了仿真计算,将仿真结果与单颗磨粒超精密磨削加工实验和单晶硅片纳米级刻划加工实验结果进行了对比分析,结果表明:理论仿真结果与实验结果在单晶硅磨削后沟槽深度、材料堆积高度、表面形貌以及主切削力方面都比较接近,证明分子动力学仿真结果是有效可靠的,可以应用到纳米级机械加工的机理研究中。更多还原

【Abstract】 Ultra-precision grinding technology is at the forefront of modern manufacture, and plays an important role in developing our country’s future IC industry. In ultra-precision, especially nanometric grinding process, chip removal takes place in a limited region containing only a few atoms or atomic layers. So some phenomena including energy dissipation, machined surface formation and chip removal, and so on, differ from those of general grinding process. It is extremely difficult to observe and measure various microscopic physical phenomena occurring in nanometric machining through experiments, nor can the conventional theory based on "continuum mechanics" explain these phenomena. And many facts have proven that molecular dynamics (MD) approach is a very effective tool for prediction and analysis of ultra-precision machining in theory, which provides a shortcut from micro phenomena to macro characteristics. MD method has many advantages, such as experimental test, security, advance study, reduction of experiments, etc. Since then, MD simulation has been applied to a wide range of fields, including physics, chemistry, biology, medicine and material, to name a few. It also has been introduced to machining in 1990’s.Through profound research of basic MD theory and method, Debye model is introduced from solid-state physics for conversion between kinetic energy and temperature of the silicon atom, the grinding model of monocrystalline silicon are established. Based on the detail investigation of MD parallel algorithms, a new MD parallel algorithm in which the spatial domain is divided twice is developed according to the spatial decomposition, an atom transferring strategy based on "invariable sequence number" is designed and the concept of "atom relative list" is proposed by studying the listing method of atom neighbor list. The application of all these stratygies greatly simplies the programming, reduces the probability of error in programming and also saves the communication overhead and the simulation time. The lenovo shenteng 1800 server works out the newly developed MD parallel program and the results show that the simulation scale is expanded from several thousand atoms of serial program to one hundred thousand level compared with the MD serial programe. Accordingly the computing time can be decreased to 1/10 of serial computing time at most. And the parallel program also has good accelerator, parallel efficency and augmentability.Besides, the grinding processes are simulated with the help of MD approach. Nanometric grinding mechanism is analyzed from the viewpoint of instantaneous distribution of atoms, grinding force, potential energy between silicon atoms and depth of damage layers. It is found that subsurface damage of the monocrystal silicon is mainly concerned with the variation of potential energy between silicon atoms. On atomic scales, the depth of subsurface damage layer of monocrystal silicon is defined as the maximal thickness of the atomic layers with random array in the subsurface of monocrystal silicon in the direction of grinding depth. Furthermore, under the conditions of present simulations, it is discovered that the subsurface damage is mainly composed of the amorphous layers, no obvious dislocations are found and the sorption between silicon atoms and diamond atoms is occurred due to the surface effect of the single grit.Moreover, the effects of cutting edge radius, cut depth and grinding speed on grinding mechanism and subsurface damage are studied in response to simulation results of different grinding conditions from the viewpoint of theory. It is shown that cutting edge radius, cut depth and grinding speed have little effect on the mechanism of the nanometric grinding and there are some differences in the value of the grinding force, potential energy between silion atoms and the depth of subsurface damage layers. From the results of MD simulations, it is shown that when the cutting edge radius increases in the nanometric grinding process with the same cut-depth and grinding speed and the depth of damage layers will become larger, which accord with those of the conventional grinding process. Then, when cut depth rises, both the depth of damage layers will increase. When the grinding speed ranges between 20m/s and 200m/s, the depth of damage layers doesn’t change much with the increase of the grinding speed under the same cutting edge radius and cut depth conditions. That means MD simulation is not sensitive to the change of the grinding speed, thus advancing the grinding speed properly can shorten the simulation time and enlarge the simulation scale. Finally, to enlarge the scale of the MD simulation, the MD parallel software is developed, which can simulate the monocrystal silicon grinding process with several hundred million atoms. A new hollowing-out data filtering algorithm is deveopped to realize the visualization of massive particles combining with the mapping method based on the spherical shape. The simulation of monocrystal silicon ultra-precision grinding process is carried out with the aid of the newly developed MD parallel software. Through comparison the MD simulation results and the results of the single grit ultra-precision grinding experiments and the nanometric scratching experiments by means of atomic force microscope (AFM), it is shown that the theoretical results are very similar to thoses of the experiments from the viewpoint of the depth of the groove, the hight of chip piled up on two sides of the abrasive grain, silicon surface topography after grinding and the main cutting force. That means MD simulation is very effective, reliable and successful to fulfill the investigation on nanometric machining mechanism.更多还原