The growth and competitiveness of the mobile phone industry have driven significant investment and innovation across various sectors, including imaging, software, and metallurgy. Among these, the semiconductor market has been particularly affected, experiencing a persistent increase in demand for higher performance in increasingly compact designs over the past several decades.
In 2023, Apple introduced a new iPhone model equipped with the cutting-edge A17 Bionic chips, manufactured using TSMC’s advanced 3 nm process. Reports suggested that Apple was securing the entire production capacity of TSMC for these chips.
The A17 Bionic chips are touted as being smaller, faster, cooler, and more power-efficient than their 5 nm predecessors. According to Apple, each chip boasts 19 billion transistors, with some as small as just 12 silicon atoms wide.
Similar pressures are being felt in printed circuit board (PCB) manufacturing as well. Apple is reportedly planning to use resin-coated copper (RCC) foil for its new PCB materials, aiming to produce even thinner boards.
However, this poses significant challenges for manufacturers, as RCC foil is highly delicate and prone to damage from heat and pressure during the lamination process, according to IEEE researchers.
Innovation alone is not enough; profitability also plays a crucial role. Reports from The Information highlight how Apple has secured favorable terms to reduce costs.
In exchange for placing large orders, TSMC is reportedly responsible for covering the costs of defective processors. While companies like TSMC are advancing their competitive edge by developing smaller node processes to reduce chip size and power consumption, ensuring quality amidst these innovations remains a challenging task.
Automated Optical Inspection for PCB Quality Inspection
For many companies, quality control is a major roadblock in the PCB manufacturing chain. This includes reliability testing and reworking defective PCBs. Boosting the speed and efficacy of quality control may lead to a significant increase in production yield and throughput, thus lowering manufacturing costs and waste.
Most PCB manufacturers rely on automatic optical inspection (AOI) to detect defects in printed circuit boards. AOI is particularly effective at identifying issues with soldering, connections, pads, and traces, ensuring high-quality and reliable circuit boards.
AOI is also invaluable for early detection of issues during assembly, such as shorts, open circuits, thinning solder joints, and scratches on traces. Scratches, in particular, can be detrimental, altering a board's electrical properties and leading to complete failure of the final product.
One key advantage of AOI is its placement at the end of the PCB production line, prior to lamination and etching, which allows for earlier detection of defects compared to other methods.
AOI systems use high-resolution imaging, capturing details down to a few microns, and compare these images against a 'perfect' model board, also known as the "golden board," or an image database of both acceptable and defective samples
Example PCB Defects. Huang, W., Wei, P., Zhang, M. & Liu, H. Hripcb: A challenging dataset for PCB defects detection and classification. J. Eng. https://doi.org/10.1049/joe.2019.1183
As well as performing tests on the PCB in the process of assembly, the AOI method can also monitor the manufacturing process. Pick-and-place machines can address identified defects instantaneously, rectifying errors in assembly such as misplacement and misalignment of components.
Moving Past the Conventional toward Artificially Intelligent Imaging
As the demand for smaller, higher-performance components grows, the complexities and nuances of material defects have increased to the point where traditional manual inspection or rule-based imaging methods are often inadequate.
For example, one semiconductor OEM was required to identify multiple subtle defects on PCB components, such as breakage, abrasion, contamination, fragments, and air bubbles. In this case, traditional rule-based image processing was unable to provide the required level of accuracy, and thus the manufacturer saw an increase in faulty parts that went unidentified in their existing process, increasing costs. A new solution was necessary.
To resolve these issues, the OEM made steps to investigate the potential of machine learning to address their accuracy needs for detecting faults on PCBs and their parts. They decided to use Teledyne’s Sapera AI inspection software suite.
The Sapera AI software made it possible for them to build on their rule-based algorithms with AI functions within their AOI machine. The Sapera AI software was an ideal solution for the OEM, making it possible for them to utilize much of their existing system while offering more precise detection of the subtle faults that other methods left unnoticed, such as breakage, abrasion, contamination, and fragments.
With Sapera AI, it was possible for the OEM to attain 98 % accuracy in continual classification with 12-14 millisecond speed for 200 images and 100 % accuracy with 453 good+ 11 bad images. They were also able to reach 99.62 % accuracy with 259 images and 20 millisecond speed for object detection when searching for multiple faults on a part image simultaneously.
Transistors on a PCB can have many small variations that may or may not affect performance. Image Credit: Teledyne DALSA
This is just one example of the significant improvements that have occurred in machine learning in recent years.
Just a few years ago, AI systems typically had to be trained from scratch and required hundreds or even thousands of image samples. Today’s deep learning software, however, is generally pre-trained, meaning that users might only need to add tens of extra samples to adapt the system for their specific use case.
The end result for the OEM was a production line that could accurately identify subtle faults on printed circuit boards without labor intensive manual inspection. The AI functions proved to be a reliable and consistent replacement for traditionally used rule-based image processing which had formerly lacked reliability in identifying subtle faults.
Overall, the OEM saw significant improvements in accuracy and speed of detecting defects on PCBs with the help of Teledyne’s Sapera AI software. This allowed them to generate less fallout while producing higher quality products according to their requirements.
Getting Ready for 2030
Today, the industry is still recovering from the global semiconductor shortage that started back in 2021.
Although analysts like McKinsey say that almost 70 percent of growth in semiconductors over the next ten years will be led by three industries: automotive, computation and data storage, and wireless communications, these industries are still catching up from missed product launches, delayed rollouts, steeper prices, and higher expectations. There is a lot of pressure in the space.
Machine learning and AI-informed software systems can quickly increase speed and accuracy in their largest challenge: quality control. With the new technology, companies can make quality control a competitive advantage, increasing speed and reducing costs while securing trust in an extremely tight-knit industry.
Improving products could be just the start. Semiconductor firms, already at the forefront of data generation and analysis, stand to gain even more from the insights provided by machine learning and AI. These technologies have the potential to enhance various aspects of their operations, including predictive maintenance, yield optimization, research and development investments, and even market strategies and product refinement.
By leveraging more data and advanced learning techniques, these companies can achieve significant benefits across their operations.
This information has been sourced, reviewed and adapted from materials provided by Teledyne DALSA.
For more information on this source, please visit Teledyne DALSA.