As part of the academic program TUMKolleg, students are given the chance to undergo an internship abroad, preferably in an English-speaking country. However, given the current circumstances and uncertainties surrounding the COVID-19 pandemic, many decided to do a domestic internship at an institute or company with main operations in the sciences. In fact, I myself had the wonderful opportunity to work at the Infineon Campeon in Munich Neubiberg (at the so-called “Campeon”) from July 19th to August 13th, 2021. The “Camp” of Campeon referring to campus and the “eon” part referring to Infineon, this site to the south of Munich acts as the headquarters for Infineon’s global operations mainly in research & development.
Consisting of 15 large buildings, 13 of which are occupied by Infineon, the Campeon has a superb connection to both the S3 train line (station Fasanenpark) and the A8 highway. Infineon Technologies AG is a spin-off of the former semiconductor departments of Siemens AG from 1999. As a market leader in automotive and power semiconductors, Infineon is one of the ten
largest semiconductor manufacturers worldwide. Consequently, Infineon plays a key role in our modern digital world: Semiconductors, more commonly referred to as integrated circuit/computer chips, are essential for the operation of all modern electronic devices, making them the foundation of modern technology – Billions of connected devices would not function without them. Apart from its role in technology, Infineon’s verdant Campeon with all-around green park landscape and lakes is open to all visitors and employees for a delightful walk or a comfortable jog – After lunch at the so-called Casino, from ongoing projects to the newest advances in technology, these walks were the home of countless intriguing insights into all kinds of topics.
Even Infineon’s goal of being sustainable is reflected in its thoughtful design of the campus. By pumping water into the ceilings of the buildings at night, the water can be cooled in summer and warmed in winter efficiently.
Microchips are produced in special semiconductor fabrication plants, also referred to as “fabs”. The countless steps taken in such “fabs” to produce such devices require utmost precision and sanitation. However, the design process of integrated circuit chips starts in the hands of system engineers, who sketch out the concept, specification, and architecture of the intended microchip in collaboration with analog designers. After further layout generation, the detailed chip prototype is sent to the “fabs”. From refining the basic material silicon to processing the wafer made out of sliced pure silicon ingots using photolithography to sketch out a template for the circuit patterns to finally installing the circuit components, even the physical production of such a chip takes up to 15 weeks (design & fabrication process heavily nutshelled). However, one has to keep in mind that this is merely the design & manufacturing of a single prototype – In order to develop a market-ready chip, the requirements and product goals must be determined and fulfilled. Such requirements and product scopes are usually dependant on factors such as, but not limited to, market and customer demand, feasibility, fabrication costs, and innovation. Therefore, to achieve a minimum viable product (MVP) the microchip has to be re-designed and -fabricated in many cases.
Quelle: InfineonI, contrary to the company’s primary industry, was mainly working on software and artificial intelligence-related projects – Artificial Intelligence (AI) has the power to transform all industries, therefore, machine learning as a substantial part of AI is also experimented with and used at Infineon Technologies. From company internal solutions to the entire design and production process, the Applied Machine Learning (AML) team aims to enable, accelerate, and optimize processes company-wide. I myself am deeply intrigued by the possibilities of AI and hence decided to work on the AML team despite Infineon mainly being a semiconductor design and manufacturing firm. I worked with my mentor from the United Kingdom on the AML team of the SCS department for the whole duration of my internship. Unfortunately, I wasn’t able to meet the majority of my team in person due to the current circumstances – Infineon still strongly encourages home-office, fortunately. Nevertheless, I was able to frequently collaborate with colleagues from other teams both in person – In fact, employees rushing through the hallway to other teams or departments is no uncommon sight. From capacity-limited meeting rooms to virtual Webex meetings, Infineon’s environment and the concept of software development in such an environment promote interdepartmental cooperation over several industries.
Quelle: Shizhe HeOn my first day of work, I got a thorough tour of the previously introduced Infineon campus and the current work of the apartment, including the team, from my referee. After picking up my employee ID with which I got a 50 percent discount at lunch right after, I got to meet my mentor from the AML team in person.
He introduced me to the project I’ll be working on as well as the projected goals. Over the course of my internship, I was given the chance to improve and add several features of an Infineon internal solution/application with the purpose of assisting the usage of an industrial solution. After assembling a micro-chip, it still has to be tested on the respective requirements via pre-determined verification domains/methods. Product requirements such as the ability to run at a specific electrical current under a certain temperature limit can be validated both before (pre-silicon) and after (post-silicon) producing a physical prototype of the microchip. This industrial solution (whose name I am not allowed to mention) is a centralized, traceable solution to manage and assign requirements to validation domains for requirements engineers. However, certain improvements in terms of efficiency and usability can be made to the existing application.
So WHAT did I specifically work on?- Firstly, I solved the changes and drawbacks of the existingapplication, including but not limited to facilitating the accommodation of large datasets due to certain limits of the previous approach, improving user experience, and improving the existing code-base in terms of scalability (refactoring to a repository design pattern). Furthermore, I was also able to develop new features, especially to incorporate certain features of the original application. An example would be implementing/replicating the hierarchical structure of the requirements according to the nested structure displayed in the industrial application. Unfortunately, I am not able to visually illustrate this due to data confidentiality reasons. Finally, we aimed to automate this tedious, extensive (hour- to day-long) manual process of setting relations between requirements and verification domains using a deep learning algorithm, which would otherwise require a requirements engineer. Previous work, including a baseline machine learning model and sample datasets, has proven this natural language processing (NLP) problem to be solvable. During my last week at Infineon, I was able to apply my existing knowledge with insightful input from my mentor to analyzing the pre-set requirement-verification domain datasets from existing real-world semiconductor products. In these datasets, requirements and the respective descriptions of varying lengths have been matched with one or more verification domains. From quantifying certain keywords for the verification domains and finding clusters to applying LDA topic modeling, the results of my data analysis were the foundation of the later machine learning model development. As previously mentioned, the application already had a functional, though unreliable, baseline machine learning algorithm integrated. However, the reliability of the algorithm is the utmost priority for an industrial application. My favorite part of the internship was improving upon this model, which I was able to achieve using a state-of-the-art pre-trained transformer – BERT by Google. Transformers – no, not the human-like robots in the sci-fi franchise – deep learning models characterized by self-attention and its encoder-decoder structure, have proven to be very successful in a wide variety of NLP tasks, especially for machine translation, biological sequence analysis, and natural language generation. Despite not being able to travel abroad due to public health regulations during the pandemic, I must say that the fruitful discussions, the culturally diverse team, and the challenging projects have made this internship at Infineon an extremely insightful and wonderful experience.
Appendix:
On my first day of work, an embarrassing moment occurred on my way to the Campeon: During my 1-hour transit to Infineon, I got off the S3 at Fasanengarten instead of Fasanenpark. So
always stay aware of your position and surroundings, especially when traveling abroad or during extended rides. Luckily, I still arrived in time thanks to the time buffer I incorporated