1 Introduction
In semiconductor back-end packaging, multiple processes are typically involved. Wafers undergo thinning, dicing, bonding, interconnection, and lead trimming to form finished products suitable for later applications. Yield is the most critical requirement for the entire semiconductor industry. The entire back-end packaging process is shown in Figure 1. With the rapid development of markets such as mobile phones and wearable devices, chip sizes are becoming smaller, pin counts are increasing, and advanced packaging technologies such as flip-chip FC, wafer-level packaging (WLP), and system-in-package (SiP) are being increasingly introduced. These developments have led to increasingly higher requirements for the precision and stability of upstream and downstream equipment throughout the entire process.
Figure 1. Chip packaging process flow and equipment
After the wafer undergoes the initial thinning process, it enters the dicing stage, and the dicing machine is the key piece of equipment. The basic process of the dicing machine is to fix the wafer on a blue film (Mylar) to prevent it from scattering after dicing, and then feed it to the worktable. The entire wafer is precisely divided into individual chips along the X and Y directions by the grinding wheel blades. The machine is then visually inspected and adjusted online. After dicing, the chips are unloaded by the unloading mechanism for use in subsequent processes, as shown in Figure 2.
Figure 2 Wafer slice
Currently, dicing machines still traditionally rely on imports, and the systems are often quite old—a significant characteristic of the semiconductor industry, where system switching is not easily undertaken to avoid potential risks. A domestic microelectronics equipment company began developing this dicing machine early on, and through relentless efforts, it achieved a breakthrough in this field and initiated the process of domestic production.
2. Requirements for Algorithms and Data Processing Capabilities
The wafer fabrication process is extremely complex. After going through countless hardships, the wafers arrive here for packaging. If defects occur here and lead to scrap, it means a huge waste of costs. Therefore, dicing must be precise and stable, and yield is the core requirement. To ensure this, high-speed synchronization between visual inspection and motion control must be achieved so that deviations can be corrected at any time.
This dicing process includes loading the sheet, automatic measurement (height, thickness, etc.), machine vision positioning, calculating the blade path, generating a trajectory file, then controlling the cutting blade to continuously cut according to the trajectory file, and finally unloading the sheet.
Typically, for higher precision, the cutting grinding wheel operates under the control of a motor supported by air bearings, reaching a speed of 60,000 RPM. This makes the dicing itself extremely demanding in terms of detection accuracy and response. Furthermore, the visual inspection data must be transmitted and controlled in real time, and stored for quality improvement analysis.
In summary, real-time performance and data processing capabilities are typical high requirements for dicing machines and semiconductor equipment.
3. Traditional solutions and their problems
Due to the complexity of algorithm design and the requirements of graphics and image processing, the semiconductor industry traditionally uses Windows-based industrial PCs for programming and task processing, typically expanding I/O and other devices via PCI cards. A common architecture is shown in Figure 3. However, Windows systems frequently experience software crashes and freezes, which can interrupt the dicing process and damage some wafers—creating significant challenges, as yield and efficiency are crucial to the semiconductor industry's survival.
Besides this reason, the PCI I/O expansion method is not as easy to expand as the distributed I/O architecture commonly used in PLCs, since the number of available PCI slots is limited.
Figure 3. Traditional control system using PC + PCI card
However, machines require more detection and control, which makes the original architecture inadequate to meet the demands. Therefore, change is an inevitable choice for users—but change also means risk, so they need to evaluate partners very carefully.
4. Flexible cooperation
At this time, Yide Tongchuang, a partner from B&R, contacted the client and recommended a complete solution. However, the semiconductor industry must be cautious. After reaching a consensus, they decided to proceed in two steps: first, solve the problem of machine stability and scalability, which is fundamental and related to basic quality issues.
Initially, Yide Tongchuang believed that the entire existing system could be directly replaced using APC or Panel PC series. However, considering the lengthy development time spent on the original software and its PCI card architecture—a software-hardware coupling relationship that presented potential migration difficulties, primarily due to the need to process diced trajectory files and handle graphics tasks—the biggest problem was solved by using a PLC for trajectory file access and processing. Even if the host PC crashes, it won't affect the stable operation of the PLC program. Furthermore, this PLC can replace the original PCI cards to control frequency converters, servos, and I/O.
Yide Tongchuang recommends B&R's X20CP1382, a compact PLC known as the "popular PLC." As shown in Figure 4, it is only the size of a palm, but it is equipped with POWERLINK real-time Ethernet, as well as a standard Ethernet, CAN, RS232, and USB2.0 interface. It also comes with 32 I/O points (including one PT1000 temperature channel and two analog channels), making it relatively economical.
A scribing trajectory file is generated on the existing PC. This file is saved to a specified shared directory. A read command is sent to the X20CP 1382 via communication. The PLC then reads the trajectory file through the shared file, saves it in memory, accesses the file in memory through the program structure pointer, parses it, and sends it to motion control and I/O to execute the task.
Secondly, on the other hand, the task interaction between image processing and PLC is also a problem. It is necessary to be able to see the slice status in time in order to adjust the precision of the coordinate control servo axis of the machine tool. For two different systems, it is quite troublesome to handle such a thing. However, it is not a difficult problem for the engineers of Yide Tongchuang. They wrote a TCP/IP interaction program that can achieve adjustments on the order of 10ms, thus solving this dynamic interaction problem.
5. Higher performance Hypervisor technology
Following a successful initial collaboration, users began to trust the products and technologies from Yide Tongchuang and B&R. They believed that adopting the new PC architecture would directly achieve a more integrated and stable architecture—and that program portability would not be a problem. PCs based on Hypervisor technology can find seamless solutions for different tasks, as shown in Figure 5. Today, chips like the Intel Core i series or Apollo Lake series, employing multi-core technology and excelling in graphics and data processing, run general-purpose operating systems like Windows/Linux to handle complex graphics and algorithm tasks. Other cores are used to run RTOS like Automation Runtime to ensure real-time task processing capabilities.
This effectively solves the various problems of the original commercial PCs. The industrial-grade PC features a fanless design, greater stability, and vibration resistance. Both existing Windows tasks and previous X20CP1382 PLC programs can be easily ported to the new industrial PC, ensuring overall stability and reliability. At the same time, storage capacity, data processing capabilities, and real-time performance have all reached a new level.
The semiconductor industry needs a stable hardware and software architecture to support its long-term innovative designs. Therefore, platform architectures like B&R's are well-suited to meet the industry's needs.
Figure 4 X20CP1382, a "celebrity" PLC
Figure 5. B&R Hypervisor Technology
500 Million DeChuang, Professional Partner
As a partner of B&R, Yide Tongchuang possesses experts who have a deep understanding of customer needs. They can tailor solutions to various task requirements based on B&R's control systems. Regarding hardware integration, Yide Tongchuang offers third-party driver technologies based on B&R's complete solutions and POWERLINK and CANopen connections, tailored to the performance and functional requirements of each task. Furthermore, they have a highly professional engineering team providing software support.
Since partnering with B&R in 2017, Yide Tongchuang has established deep collaborations with customers in the packaging, pharmaceutical, electronics, and semiconductor sectors. By maintaining B&R's solution capabilities and fully leveraging the potential of B&R systems, they have earned the high trust of their clients, and the company's business has continued to grow.
As Chen Wei, CTO of Yide Tongchuang, said, "B&R Systems' greatest strength lies in its flexibility, which is crucial for partners like us who primarily provide industry solutions, because we are always facing changing demands. Semiconductor devices have very high algorithm and data processing capabilities, and B&R's Automation Studio platform and Hypervisor controller are particularly well-suited to the application needs of this industry."
