Research On Optimization Of Key Parameters Of Silicon Multi-wire Cutting Based On PSO-BP Neural Network

In recent years,the integrated circuit industry has developed rapidly,and the market's demand for semiconductor products such as memory,logic devices,analog devices,and power devices has increased dramatically.As the base material of integrated circuits,the improvement of the manufacturing technology of silicon wafers is crucial.Multi-wire cutting is one of the important processes in the manufacture of silicon wafers.The cutting quality directly affects the subsequent processes and the quality of the silicon wafers.The traditional multi-wire cutting adopts free mortar type cutting.This cutting method has low efficiency and high cost.It is difficult to separate the waste liquid after cutting,which is easy to cause environmental pollution.The diamond wire cutting method can significantly improve production efficiency and reduce costs,but most diamond wire cutting equipment comes from imports,which is expensive,and domestic equipment has poor stability,which makes it difficult to meet production and processing requirements.In addition,compared with solar-grade silicon wafers,IC-grade silicon wafers have higher quality requirements,and the quality control of silicon wafers for diamond wire cutting is higher.Domestic IC-grade silicon wafer manufacturers have not promoted the use of diamond wire cutting equipment.Therefore,there is an urgent need to design a new type of high-precision diamond multi-wire cutting equipment to meet the domestic IC-level silicon wafer manufacturing needs.First of all,this article introduces the overall structure,sub-structures and key parameter design models of a new type of diamond multi-wire cutting equipment,introduces the design of mechanical structure,electrical control,data acquisition and monitoring system,and conducts silicon wafer cutting tests.And compared with the cutting quality of mortar multi-line cutting,the results show that the average value of TTV,Warp,Bow of silicon wafers cut by diamond wire equipment is better than that of traditional mortar cutting,but the data is unstable and the overall yield is less than mortar cutting.Secondly,in order to improve the stability of equipment cutting,BP neural network is adopted to predict the quality of silicon wafer cutting by modeling and analyzing the process,process parameters and quality parameters of multi-wire cutting.Because the training of BP neural network weights uses the gradient descent method,it is easy to fall into the local optimum,so the PSO algorithm is introduced to train and simulate the BP network weights.By comparing and analyzing the BP network and the PSO-BP network,the PSO-BP network greatly reduces the possibility of the algorithm falling into a local minimum,improves the prediction accuracy,and accelerates the convergence speed.Finally,the key parameters that affect the quality of diamond multi-wire cutting are studied,combined with the PSO-BP quality prediction model,and the orthogonal test method is used to obtain the degree of impact of each key parameter on cutting,and the optimal parameter combination is selected The diamond wire speed is 1300 m / min,the feed speed is 1.1mm / min,the silicon rod length is 150 mm,and the steel wire tension is 13N).Through the construction of PSO-BP prediction model and the optimization of key parameters,it can improve production efficiency,ensure product quality,and provide a reference solution for the silicon wafer manufacturing industry.