Hello, my name is Akio Osawa. Let me quickly introduce myself. I joined Advantest in 1989 and was assigned to the SE department, which stands for System Engineering. Since then, I have mainly worked as an associate test system engineer, promoting testers, providing test solutions to customers' test issues, and providing technical support. I currently oversee a global department called System Solutions, which is responsible for systems and application engineering for all testers, including memory as well as SoC. I will explain about the business environment surrounding SoC testers. Advantest plays a role in supporting the quality of all semiconductors through testing. From smartphones and other devices to data center infrastructure, semiconductors are now used in all kinds of applications. The technological advancements in semiconductors, such as the addition of various functions and improvements in energy efficiency to enable long-term use of devices without recharging, have enhanced the convenience of our lives today. In recent years, AI technologies that efficiently collect, learn, infer, and generate large amounts of data with high efficiency have been emerging. With AI becoming a pervasive force in our daily lives, it is essential to enhance the processing power of semiconductors to create an environment that can process dramatically increasing amounts of data at higher speeds. Currently, new semiconductor packaging technologies are rapidly advancing the heterogeneous integration and performance improvement of high-performance semiconductors such as HPC and HBM. As a system engineer during my tenure at Advantest, I have long been working closely with customers who develop and evaluate various semiconductors. The role of an SE is to contribute to the customer's time to market, time to quality, time to volume through technical support. Specifically, we support the optimization of the test environment for each device by providing support for the development of test programs and working with the customers to identify defective areas of the semiconductor. I have listened to their technology challenges and explored enhancements of test solution requirements. I would then provide feedback to our R&D team to develop next-generation test solutions. I have experienced such cycle numerous times. Today, we will focus on HPC. Later on in this presentation, Iino will explain in detail about the characteristics of the technological advancement associated with these semiconductors and test needs they are creating. In our presentation of the third midterm management plan announced in June of this year, we stated that the semiconductor industry is shifting into an era of complexity. The challenges in the semiconductor industry can be summarized as shown on the left side of this slide, where the six circles representing the challenges are all interconnected. This means that each of these challenges is significant and interrelated in terms of complexity. Furthermore, complexity has a multiplier effect. For devices responsible for advanced computing, such as for AI, the interdependence between these challenges become even stronger. In the semiconductor process, the test process is essential for solving interrelated problems. In the era of complexity, the more the interdependence between issues increases, the more it is necessary to strengthen, advance, and integrate test solutions. In other words, test intensity increases. The diagram on the right shows the value which advanced test solutions provide. With the proliferation of AI, the semiconductor value chain is tackling the challenge of integrating advanced computing functions such as learning, inference, and generation into a single device. In this process, many interdependence challenges have become apparent already. Going forward, we anticipate that this will become even bigger industry challenges. In response to this trend, we are confident that the technology, assets, and know-how we have cultivated will increase in value even further. Let's take a more detailed look into the challenges which the semiconductor industry faces. Let me outline the detail of the challenges in the six circles. Scaling wall. Process, fabrication, technology, miniaturization. Power increasing power consumption. Memory wall. Increasing memory complexity. I/O wall. Communication with faster, more advanced protocol. Device complexity. Device structure complexity. Energy efficiency. Power consumption reduction. Among these, the items that are highlighted in red are prominent challenges in these several years. As I explained earlier, these challenges are becoming even more advanced while being mutually interdependent and interdisciplinary. In today's presentation, we will lay out the characteristics and test needs of HPCs and AI devices. In doing so, we will explain how these challenges are leading to the increase in test volumes for HPC AI devices. I'll hand the presentation over to Iino-san. Hello, I am Takashi Iino. From here, in my presentation titled, "Characteristics and Testing Needs of HPC/AI Device Test," I will explain the testing needs of HPC AI devices, including the technical background. Let me introduce myself briefly. I joined Advantest in 2001, and have been supporting our domestic and international customers with SoC device test technology ever since. Currently, I belong to the Center of Expertise under the marketing organization for the SoC tester V93000, and I'm supporting customers in solving their problems from a technological perspective through development of test technologies and in the high-performance digital domain. In the following slides, I will introduce the characteristics and technological trends of HPC/AI devices. The large block diagram on the left represents the processors operating in data center or AI servers. It is mainly composed of processor dies, I/O dies, and high-bandwidth memory. Common features of HPC/AI devices include chiplet design, multiple processor dies, multiple I/O dies, and multiple HBM. As a recent change, cryptocurrency technology emerged around 2014. In response, ASIC mining and GPU technology were developed rapidly, and parallel computing capabilities gained traction. At the end of 2022, OpenAI's GPT-3 and ChatGPT were introduced, bringing generative AI technology to the public. While further advancements and expansions of advanced AI technologies are anticipated, achieving this will require computational capabilities far exceeding current levels. The required processing power is increasing year by year. To keep up with this, technology trends such as new process technologies, larger, faster, and more complex packaging are emerging. As indicated by the chart showing average transistor count by product category in the upper right side on this slide, the number of transistors in server GPU could potentially increase fivefold over the next five years compared to the current levels. There are many challenges in test technology for volume production of HPC/AI devices. Among these, the most significant challenges are related to test volume and test time. Let me walk through the test challenges caused by the advancement of HPC/AI devices. This slide shows five major technology trends and their associated test challenges, as well as their impacts on test processes. These can be categorized into longer test time and/or increase in test processes. First, the process node miniaturization. For products that adopt cutting-edge process nodes, the number of integrated transistors increases dramatically, requiring a vast number of test patterns to ensure their quality compared to the past. Next, the new process technology. The evolution of transistor architecture continues to support miniaturization. FinFET technology has become widespread, and recently, Gate-All-Around is being introduced. Furthermore, a new architecture called Complementary FET is being proposed for sub-2nm processes. The miniaturization and innovative transistor architecture are likely to trigger new failure modes that did not exist before, necessitating additional test patterns to detect them. Third, the multi-die integration. To maximize yield, the quality and individual characteristics of each die are examined before packaging. In some cases, dies with similar performance and characteristics are considered for inclusion in the same package. After packaging, it is necessary to reconfirm that all these dies function correctly. The increase in the number of transistors and dies also leads to thermal control challenges during volume production. In order to carry out tests efficiently, a large number of transistors have to be run in parallel unit time. As a result, a large amount of current continuously flows through the device, and this manifests as heat generation. Heat energy changes in proportion to the square of the current, as expressed by P=RI^2. It is necessary to consider and optimize the trade-off between heat generation and test time in order to prevent incidents such as burn-in and overheating. High quality assurance. In data centers, tens of thousands of devices operate in clusters. If even one device fails or experiences performance degradation in the field, it can significantly impact overall performance. It is essential to ship devices that have been tested under stress conditions with high coverage to ensure very high quality. In order to efficiently test whether the hundreds of millions of circuits integrated into a device are performing correctly, a test method called scan test has been used for SoC devices. The features of the scan test will be explained in the following slide. Let me explain about the changes in scan test volume. As illustrated in the chart on the left, there has been significant changes in volume and test time for large-scale device test, especially between the 2010s and the recent 2020s. The increase in scan test volume has been a major driver of this change. In the 2010s, in addition to DC test and BIST test, scan test was also performed, but the proportion of scan test was not that significant. Entering the 2020s, however, the proportion of scan test has significantly increased, and this trend continues until today. While the increasing complexity of devices has also caused an increase in other test items such as DC test, their growth rate is not as substantial as that of scan test. EDA companies have developed pattern compression technologies to improve the transistor test rate per unit time. Although new efforts to further enhance efficiency are actively ongoing, the reality is that they still can't keep up with the increase in the number of integrated transistors. scan test remains the most efficient method for testing exponentially increasing transistors, and its volume is exploding. In order to reduce the increase in test time, customers, EDA companies, and ATE vendors are also conducting various research and development activities. EDA companies have proposed solutions to improve scan test efficiency, such as networked scan circuits and scan over HSIO technology that uses high-speed I/O interfaces such as PCIe and USB as scan input and output pins to which Advantest is also involved in joint technology development. Depending on the result, the growth rate of scan test may vary in the future. Now, let me explain how SoC testers measure devices in the context of HPC/AI device test and their characteristics. The left side of this slide outlines the process for HPC/AI devices' logic testing. Signals are input into the device, and after various processes are performed within the device, the output signals are received by the tester. These signals are then compared to the preset comparison pattern to determine whether the device performs as intended. Generally, the input patterns and comparison patterns used in test are automatically generated by EDA tools, which are used during a semiconductor design phase based on the device specification and test specifications. On the right side of this slide, as we have described, the scan test which I have referred to earlier. A scan test is a standard test item that emerged based on a concept called Design for Testability. First proposed around 1990, scan test has continued to evolve, and is now a major test item for SoC devices. A key feature is the insertion of a test circuit called 'scan circuit' into the device circuit. With a scan circuit, conditions of specific areas to be tested can be set arbitrarily, and the outcome after running a test can be checked. Using this mechanism, EDA tools generate test patterns that incorporate target transistor areas and simulate failure models. If the number of transistors were to double and if the performance efficiency of scan patterns also doubled, it would be possible to maintain the same test time as before. However, as of now, the efficiency of pattern generation algorithms have plateaued, and there are limits to the parallelization of scan circuits. The lengthening of test patterns due to the increase in the number of transistors continue. This is a major factor leading to longer test times. I explained that the test volume has significantly increased in the 2020s compared to the 2010s. In this slide and the next slide, I will delve a bit deeper into what has changed. Let me review tests of the 2010s. The slide title, Screening of Defective Devices, refers to the primary purpose of test, which remains to this day. Compared to the present day, test in the 2010s was simpler. The chart at the bottom of this slide represents typical test processes and main test contents. In the wafer test, which is carried out in the front end, a scan test is first performed. After the device is packaged, a final test is carried out. Here, a scan test of similar nature is conducted. In reality, the test items that were performed in a wafer test and package test are different. However, as a result of selecting and carrying out the most optimal test items, there was no significant difference in the test volume. Furthermore, for certain devices, system level test was inserted to further enhance test quality. To clearly visualize the total test volume of the test process, the main test items are stacked on the far right. How has the test evolved in the 2020s? The test contents has grown in both complexity and test time. First, for wafer test, the volume of scan test has increased due to migration to new process nodes and the increase in the number of integrated transistors. The trend is similar for final test. However, the volume of scan tests increases even more to test the multiple dies assembled in a package. Furthermore, an additional process called die level test is being considered to more accurately determine the individual characteristics of each die. Although similar test items have been performed during the wafer test in the past and are still being performed today, there is an ongoing active discussion about inserting additional process to test the die individually after dicing due to several reasons, including heat generation issues. For devices such as automotive that conventionally face high reliability requirements, tests such as burn-in, which apply high thermal stress and mission mode test have been performed. Recently for similar reasons, stress application has also been implemented for HPC/AI devices and mission mode test has become a requirement for the most part. Lastly, we have stack test items compared with the 2010s, as in the previous slide. Please note that the height of each box does not accurately represent test volume. However, you may still get an idea of the significant increase in test volume compared to the 2010s. Up to this point, we have broken down the factors contributing to the increase in test volume for HPC/AI devices and explained the reasons behind them. This slide serves as a summary. The performance improvements in HPC/AI devices are bringing numerous technological and test challenges. These challenges are intricately linked and interdependent, as shown in the diagram on the left. Moreover, the number of transistors integrated within a single package continues to increase with miniaturization, new process technologies, and multi-die assembly. To efficiently test the vast number of transistors, scan test is increasingly utilized, and as shown in the chart on the right, their volume continues to grow significantly. As part of the AI technology roadmap, the goal over the next decade is to achieve artificial general intelligence, AGI, capable of handling a wide range of tasks like a human. To this end, HPC/AI devices will continue to evolve in terms of data processing capabilities, communication speeds, and energy efficiency, leading to even more test challenges than we face today. In response, we will continue to listen to the latest technical challenges faced by our customers and explore test technologies. We will feed this back to our R&D to create next-generation test solutions. Finally, let me touch on our SoC tester business. This slide shows the 25 year revolution of our flagship SoC tester, V93000. Starting in 1999, during the era of personal computers, V93000 met the test demands of PCs and Internet-related devices with the Pin Scale platform. The next generation, Smart Scale, increased its presence in the SoC market, driven by mobility/smartphones. In 2020s, we introduced the latest platform, EXA Scale, which is now supporting the HPC/AI era. Currently, we hold nearly 60% share of the SoC testing market and recognize our position as a leader in the field of high-performance computing. There are three core competencies behind the company's number one market share. The first competence is the strong relationship with HPC customers, nurtured through 20 years of communication. By keeping abreast of customer technology trends and requirements in a timely manner, we have continued to provide speed, accuracy, density, and performance that are one step ahead of the rest. The second competence is our test platform that combines capable scalability and compatibility. When transitioning to next generation devices, our customers can upgrade the necessary measurement modules to enhance the value of their existing platforms and continue to utilize them. Furthermore, our platform offers compatibility that allows for the smooth transition of existing assets, such as test programs, to the new platform. Finally, our global engineering support capability is our third competence. Our engineers are optimally positioned against the globally disaggregated semiconductor supply chain. We provide a high level of engineering support required by our customers working together on a global basis. These have enabled us to share our long-term roadmap with our customers and maintain a virtual cycle in which we continue to develop and supply the test solutions required for the new generation of devices in a timely manner. While embodying the core competence of our SoC tester business, we have been supporting the world's leading customers advance their semiconductors. We are excited to continue to work closely with our customers facing increasingly complexity, and to expand our test solutions. This concludes my presentation. Thank you very much.