Abstract: Based on the research of the thermo-ultrasonic flip-chip bonding process and the current development status of thermo-ultrasonic flip-chip bonding equipment at home and abroad, a thermo-ultrasonic flip-chip bonding machine for chips was developed. Keywords: Thermo-ultrasonic flip-chip bonding, mechanical vision, motion control, HexSight Since the 1990s, driven by the application fields, microelectronic packaging technology has developed rapidly. Currently, the main primary packaging technologies are wire bonding and flip-chip bonding. Wire bonding technology uses metal wires to connect the electrode leads on the integrated circuit chip to the external leads of the integrated circuit substrate. By reducing the lead diameter and lead spacing, the packaging density can be increased. However, small-diameter leads have poorer strength and rigidity, which makes lead bending operations difficult and reduces the reliability of wire bonding. With the continuous increase in the number of I/Os in chip packaging and the increasingly higher requirements for internal connection reliability, chip bonding processes show a trend of development from wire bonding to flip-chip bonding. Flip-chip bonding is a packaging process based on arrayed solder balls. Currently, the main flip-chip bonding processes include thermo-ultrasonic bonding, reflow soldering, thermocompression bonding, and epoxy conductive adhesive bonding. Reflow soldering offers high reliability and a large number of bumps, but the use of Sn/Pb solder is extremely detrimental to environmental and human health. Epoxy conductive adhesive bonding is simple and can be performed at low temperatures, but it suffers from low reliability and high parasitic resistance. Thermocompression bonding is pollution-free and highly efficient, but it has drawbacks such as low reliability and stringent bonding conditions. Thermo-ultrasonic flip-chip bonding achieves direct interconnection between the chip's I/O ports and the substrate under the combined action of ultrasonic energy, pressure, and heat. Thermo-ultrasonic flip-chip bonding offers advantages such as high packaging reliability, high connection efficiency, simple process, low cost, and strong adaptability. The lower bonding temperature reduces the possibility of Au-Al intermetallic compounds forming between the bumps and pads. Furthermore, it is a lead-free, green soldering process and is considered a promising new process and technology to meet the requirements of next-generation chip packaging. SY Kang et al. from the University of Colorado have successfully flip-chipped GaAs with 64 gold bumps onto a silicon substrate using thermo-ultrasonic bonding, achieving a bonding strength of 0.23 N/bump. This chip has been applied in multichannel memory systems and optoelectronic components. Toshiba Corporation of Japan and ASM Corporation of Singapore are also researching and adopting thermo-ultrasonic bonding technology. Domestic research on this topic is still in its early stages, with only a few research institutes and joint ventures besides Central South University conducting such research. 1 Thermo-ultrasonic Flip Bonding Principle 1.1 Chip Thermo-ultrasonic Flip Bonding Process Flow The chip and substrate used in the thermo-ultrasonic flip bonding process are shown in Figure 1. This bonding process flow can be divided into the following four steps: (1) Chip pickup. Before the bonding process begins, the chip and substrate are placed on a designated worktable. If the chip and substrate are not within the visual range of the vision system, the translation stage is driven to search for the chip and substrate according to a specified algorithm. After the chip position is obtained, the chip position is adjusted according to the substrate placement, assuming that the chip will not deflect after being picked up by the substrate, to achieve alignment (alignment is achieved through coordinate transformation, but there is still a fixed position difference in physical reality, which is recorded by the program). The vacuum adsorption system is driven, the nozzle position is lowered, and the chip is adsorbed by vacuum suction to complete chip picking. At the same time, the substrate heating system is started to heat the substrate to about 150°C. (2) Alignment of chip and substrate. After the chip adsorption is completed, the translation stage is driven to move by a fixed position difference to complete the physical alignment of the chip and substrate. However, it was found in the experiment that During the chip picking process, a shift will occur, so the vision system must be restarted before the chip moves to the top of the substrate for alignment. The chip is then viewed from below after being adsorbed to obtain the actual position of the chip after adsorption. Based on the actual shift and angle of the chip adsorbed by the nozzle, the electrode position is readjusted to achieve alignment between the chip bumps and the substrate pads (alignment is achieved through coordinate transformation, but there is still a fixed physical difference. The mechanical motion mechanism achieves physical alignment after the program is recorded). (3) Apply bonding force. After the chip is aligned with the substrate, the chip moves to the top of the substrate (physical alignment). The vacuum suction is maintained, and the nozzle slowly descends until the chip bumps and the substrate pads are a small distance apart. Then, the bonding pressure control system is driven to slowly apply bonding pressure to the chip. Ideally, the chip surface is parallel to the substrate, and the bonding force is perpendicular to the substrate direction. (4) Apply ultrasonic waves. When the bonding force reaches the predetermined pressure, the bump contacts the substrate and is flattened and deformed to a certain extent; at this time, the ultrasonic generator is started, and the transducer applies ultrasonic energy to the chip in the direction parallel to the substrate through the suction nozzle, so that the interface between the bump and the substrate is rubbed, removing the oxide and contamination layer on the surface of the bump, the temperature rises sharply, the bump is deformed, and the atoms of the bump and the substrate pads permeate each other to achieve the effect of mutual connection. (5) Bonding head reset. After the bonding work is completed, the vacuum suction of the suction nozzle is released, the chip is separated from the suction nozzle, the suction nozzle is lifted, and one bonding cycle is completed. The steps of thermo-ultrasonic flip chip and substrate connection are shown in Figure 2. 1.2 Composition of the bonding machine According to the above process flow, the flip bonding machine designed in this paper consists of the following parts: (1) Chip adsorption stage. The chip is fixed on the platform by vacuum adsorption, and the platform adjusts the chip's X and Y plane positions and W axis rotation angle under the drive of the vision system. (2) Bonding work stage (i.e., substrate adsorption stage). In addition to having the function of a chip adsorption stage, it can also heat the substrate and is equipped with a temperature sensor to monitor the temperature in real time during the bonding process. (3) Bonding head. It has a vacuum adsorption function and is connected to the ultrasonic transducer bonding pressure controller at the rear end to pick up the chip on the chip adsorption stage. When the chip is aligned with the substrate, bonding pressure and ultrasonic energy are applied. (4) Mechanical vision system. It collects images of the chip and substrate positions to provide position parameters for the mechanical movement of the whole machine and guides the movement of the mechanical mechanism. The relationship between each part is shown in Figure 3. 2 Hardware composition of the visual positioning system of the thermal ultrasonic flip bonding machine 2.1 Design of mechanical vision unit The vision system is a relatively independent module. It is connected to the computer through the PCI local bus. The image data is transmitted to the computer control system through the bus to control the movement of the X and Y translation stages and the W and R rotary stages of the bonding machine. The vision system consists of the following parts: (1) Optical system. The optical system is mainly a fixed focal length objective lens, which focuses the light from the observed target onto the CCD sensor photosensitive device. Since the CCD sensor is very sensitive to the object distance. An adjustment screw is installed behind the CCD camera. The best imaging effect is achieved by rotating the adjustment screw. The focal length cannot be adjusted during the operation of the bonding machine. The focusing is performed on the surface of the chip. As for substrate alignment, even if the image is out of focus, the focal length cannot be changed. However, the thickness of the substrate and the chip is similar. This kind of image defocus does not have a serious impact on the image. Moreover, this system has a certain tolerance for image defocus. (2) CCD camera. The CCD camera is the front-end component for image acquisition. It is composed of a linear array or rectangular array of photodiodes. It outputs the voltage pulse of each diode in a certain order, converting the image light signal into an electrical signal. The output voltage pulse sequence can be directly input into a standard display in RS-170 format or into the computer memory for numerical processing. 2.2 Motion unit design The bonding machine micro-motion system adopts a piezoelectric ceramic driven precision positioning stage. This system is attached to the Z-axis and forms a position and force control device with the Z-axis translation stage. After the macro-motion part completes the rapid and accurate positioning, the low inertia and high precision characteristics of the micro-control console are used to slowly apply bonding pressure to the chip. This allows for precise control of flip-chip bonding pressure, achieving good bonding strength, protecting the chip from damage, and facilitating the analysis of the flip-chip bonding process. 3. Development of Visual Positioning Software for Thermo-Ultrasonic Flip-Chip Bonding Machine 3.1 Image Processing Software After image data enters the computer via the data acquisition card, it must undergo image preprocessing to extract image information and complete the positioning work. HexSight provides a rich set of processing procedures. The development of this software is based on the ActiveX components provided by HexSight, and the addition method is the same as other standard components. The component properties are modified, and application processing procedures are added. Here, only the image acquisition (HSAcqulsitionDevice) and positioning (HSLocator) processing procedures are used. The image acquisition processing procedure acquires images into a public database and simultaneously compensates for pixel distortion, lens distortion, and projection distortion. The positioning processing procedure extracts and searches the geometric features of the object's contour, using state-of-the-art contour detection technology to identify objects and patterns. This technology can still achieve good results even when the image is cluttered, the light source brightness fluctuates, or objects overlap. 3.1.1 Establishing a Search Model The effectiveness of the model directly affects the positioning accuracy, response speed, and stability of the system. A good model can provide accurate and effective object contour features. To improve the robustness of the system, the model should meet the following characteristics: (1) The model image should be acquired under the ideal conditions of the lighting system and optical system. (2) The background of the model image should be uniform. (3) The image should reflect the common geometric contour features of the object to be identified as much as possible. (4) The graphic features (size, shape) used in the model must be stable elements in subsequent images. The model establishment results in this application are shown in Figure 5. 3.1.2 Visual Positioning Main Program This application uses two cameras to correspond to two processing systems working at different times. The program drives different cameras to acquire images by receiving events from the motion unit, loads the corresponding model library, identifies and positions the images, gives the X and Y coordinate differences between the chip and the reference position and the rotation angles of the W and R axes, guides the bonding head to pick up the chip, obtains the position of the substrate, performs coordinate transformation to complete the alignment of the chip and the substrate. The motion control software of the flip bonding machine is developed around the vision software. After the image data acquired by the vision system is processed by the software, the specific positions of the chip and the substrate are output. Then, this data is transmitted to the motion part to drive the external servo control card. 3.2 Motion control software In this system, the basic motion control mainly includes two aspects: (1) Position motion control. It mainly includes position motion control in the XY plane, position motion control in the vertical direction of the Z axis, and rotation angle control of the R and W axes. In this system, the position feedback signal is provided by the grating ruler, which belongs to closed-loop control. The host computer performs path planning and transmits the motion command and position data to the servo controller. The servo controller performs interpolation and acceleration/deceleration control to generate the path. The position control principle is shown in Figure 6. (2) Micro-motion control. After the macro stage moves to the specified position, the micro-motion control is started to apply bonding force to the chip. The preset bonding pressure is set by the user in the host computer, and the host computer transmits the data to the lower computer. At the same time, the control is also handed over to the lower computer, which controls the piezoelectric ceramic drive power supply. The micro stage is controlled to give a small displacement. The preset bonding pressure is achieved. If the predetermined pressure cannot be reached after the micro-controller reaches its maximum stroke, the lower-level machine feeds back a signal to the main controller, which then drives the macro-motion stage to give an appropriate displacement. The micro-motion control principle is shown in Figure 7. This paper introduces the flip-chip bonding process of the thermo-ultrasonic flip-chip bonding machine, the system composition of visual positioning, and the implementation of visual positioning software. A mechanical vision positioning system was developed based on HexSight vision software. Actual operation shows that when the main contour curves such as chip outline and bumps are used as feature attributes, the positioning time for chips at any angle is only about 20ms. This vision software, combined with the closed-loop servo motion control system, has successfully completed the alignment of 8 bump chips from picking them up to aligning the chips with the substrate. However, there is a certain gap between the overall system efficiency and accuracy and foreign chip packaging equipment. The following improvement scheme is proposed: (1) The motion control system should be further improved by optimizing the mechanism motion scheme and control algorithm. (2) For the vision system, since two cameras are used, the alignment of the chip and the substrate is achieved by coordinate transformation in the program, but not physically. In practical applications, the positioning of the vision system depends to a large extent on the accuracy of coordinate transformation. Therefore, to improve the positioning accuracy of the vision system, high-precision parameter calibration of the two cameras is necessary.