Improving the bond quality of copper wire bonds using a friction model approach

Abstract

In order to increase mechanical strength, heat dissipation and ampacity and to decrease failure through fatigue fracture, wedge copper wire bonding is being introduced as a standard interconnection method for mass production [1].To achieve the same process stability when using copper wire instead of aluminum wire a profound understanding of the bonding process is needed. Due to the higher hardness of copper compared to aluminum wire it is more difficult to approach the surfaces of wire and substrate to a level where van der Waals forces are able to arise between atoms. Also, enough friction energy referred to the total contact area has to be generated to activate the surfaces.Therefore, a friction model is used to simulate the joining process. This model calculates the resulting energy of partial areas in the contact surface and provides information about the adhesion process of each area. The focus here is on the arising of micro joints in the contact area depending on the location in the contact and time. To validate the model, different touchdown forces are used to vary the initial contact areas of wire and substrate. Additionally, a piezoelectric triaxial force sensor is built up to identify the known phases of pre-deforming, cleaning, adhering and diffusing for the real bonding process to map with the model.Test substrates as DBC and copper plate are used to show the different formations of a wedge bond connection due to hardness and reaction propensity. The experiments were done by using 500 µm copper wire and a standard V-groove tool. Model conception for wire bondingCurrent model concepts divide the bonding process into four phases: (I) Touchdown/pre-deformation [2], (II) Sticking friction (III) Slipping friction/surface activation and (IV) diffusion phase [3,4].Before the tool starts vibrating, the wire is pressed with a defined force and velocity on the substrate. Hereby, the wire is pressed into the V-groove of the tool and an initial contact area between wire and substrate is created. The touchdown velocity has an influence on the size of this contact area. A high touchdown velocity (e.g. 25mm/s) leads to an overshoot of normal force (up to 30%) and therefore to a higher predeformation of the wire. Also the hardness of the wire and substrate after the touchdown depends on the touchdown force and velocity. This hardness change is described in [5]. Both contact bodies increase their hardness in and next to the contact area by up to 40% for typical touchdown forces.After powering the sonotrode, the tool tip starts to vibrate and excites the wire. Depending on the bonding parameter (here: normal bonding force and ultrasonic voltage amplitude) the short sticking friction phase starts and is followed by the slipping friction phase. Due to this, the surfaces grind their oxide layers and start to strain the contact near parts of the bodies. This results in an activation of surfaces.While activating the contact bodies, the ultrasonic softening effect begins to soften the materials and reduces the yield points....

Experimental and Numerical Simulation Study of Pre-deformed Heavy Copper Wire Wedge Bonds

“…Nevertheless, the contact status indicates a sticking area in the center of the touchdown interface. The region of maximum pressure and the status in the contact are confirmed by the microscopic analysis in [2]. There, it was found that the center region of separated bond connections was not fully bonded if low ultrasonic voltages and high bonding forces were used.…”

“…Recent studies, e.g. [2], confirmed the strong impact of the pre-deformation process. There it was found that highly pre-deformed bonds adhere faster than less pre-deformed bonds while bonding on a soft DBC surface.…”

Intelligent Production of Wire Bonds using Multi-Objective Optimization – Insights, Opportunities and Challenges

“…By this it the simulation time could be reduced by more than a decade without losing significant accuracy. The ultrasonic transducer is modelled as a linear system as introduced in [10]. Fig.…”