Building smarter semiconductor manufacturing with AI
In this interview, Ivan Nikitski, Photonics Technology Expert from EPIC – European Photonics Industry Consortium, talks to Aftkhar Aslam, CEO at yieldWerx, a US developer of data analytics and AI software for the semiconductor manufacturing industry.
Aftkhar Aslam: In 1994, after studying computer science and pure maths, I joined Texas Instruments in Dallas, Texas, where I stayed for 15 years eventually becoming Director of Design/R&D Engineering Applications. Because TI is an integrated device manufacturer (IDM), it was a great learning experience as I got to know about the entire end-to-end process- from design to tape-out, through manufacturing, assembly, and testing.
In 2009, I joined Accenture and did management consulting, implementing digital transformations for companies like Intel, Qualcomm, and Lam Research. During this period, a colleague and I came to realise that there was a business opportunity. We could take the best of what we had learned at TI and apply it to help other companies overcome challenges. Accordingly, in 2015, we started yieldWerx with me as CEO/CTO.
Ivan Nikitski: How has yieldWerx developed?
Aftkhar Aslam: We set up the company as an enterprise platform to enable IDMs, foundries, EDA vendors, and equipment manufacturers to analyse data in order to identify yield, process and design issues as well as to be better able to monitor processes, reliability, quality, sustainability, and overall portfolio health.
We started with primary modules for engineers to use for data analysis. Since then, we’ve developed our own IPs to help customers improve yield, particularly in life sciences and automotive. We’re now applying those same reliability and quality modules to the quantum computing and photonics industries, because reliability will become increasingly important there.
Another major milestone has been the introduction of AI, which we are adopting cautiously because any product or quality escape can have catastrophic consequences in our industry. The next major milestone will be the use of agentic AI (autonomous AI systems to achieve specific goals), to tackle many of the industry’s challenges and make our solutions even more comprehensive.
As regards customers, in the beginning we secured the US company Monolithic Power Systems and another anchor account who are still with us. Since then, we’ve developed an extensive network of partners who are trained on our platform and able to provide local support. This has allowed us to serve customers in the US, Europe and China, Japan, Korea, Taiwan, Singapore, Malaysia, and the rapidly growing market in India. In this period, our workforce has grown to more than 60 full-time staff based across the US, Europe, and Asia.
Ivan Nikitski: What are the main challenges in finding new customers?
Aftkhar Aslam: Our software is very modular to make it configurable and flexible and also beneficial from a pricing perspective because customers aren’t paying for functionality they don’t need. For example, a customer may have one product destined for the automotive industry, where the highest levels of reliability and quality are required. They may have other products for agricultural equipment or consumer markets, where those requirements are different.
Having said this, one of the biggest challenges has been helping customers understand the long-term vision. Many companies have homegrown solutions and believe they can solve problems better themselves. We try to demonstrate that we’ve worked with customers across many industries, platforms, and vertical markets, and that our solutions have been tried, tested, and trusted often providing advantages over internally developed systems.
Another challenge is convincing customers to adopt a brownfield implementation, where we integrate the best of our platform with the best of theirs.
There’s also the fundamental question of trust. Customers ask why they should trust us and what makes us different. This is where our experience from Texas Instruments becomes valuable as we speak the same language and can demonstrate how we’ve solved similar problems for other companies. Nevertheless, platform decisions are complex, lengthy, and often involve significant bureaucracy. And navigating that process is one of our biggest challenges.
Ivan Nikitski: What are the main challenges in working with AI?
Aftkhar Aslam: For the traditional semiconductor industry (analogue, signal processing, RF) AI is mature enough to be applied effectively. But when we move into areas like photonics, quantum computing, or microLEDs, there are several challenges.
The first is data quality. Labelling data correctly is very important. You need to know whether data is good, bad, or what we call “maverick” data—data whose quality is uncertain. In many cases, customers don’t even have a methodology for labelling their data, which makes it difficult to achieve reliable results.
The second challenge is building and training the models and then proving to customers that the AI solution works for their specific products. Every product is unique, from die size to the test data collected and even the naming conventions used for test parameters.
The third challenge is security. Customers naturally want to know whether the data they’re providing is secure. They also want assurance that the knowledge gained from their data won’t be used to benefit another customer, particularly a competitor. That’s something we take very seriously.
The final challenge is managing expectations. Customers sometimes expect AI to perform miracles. In reality, AI is a journey. You observe, learn, train, refine, and gradually build trust in the system. Eventually, AI can take on more responsibility in processing incoming data and supporting decision-making.
Ivan Nikitski: What are the main challenges in accessing new markets?
Aftkhar Aslam: As regards photonics and quantum computing, one of the biggest challenges is data protection as there are many companies competing with one another.
Another challenge is the lack of collaboration, standards, and established methodologies within the photonics industry. This means that when it comes to testing photonic devices, questions remain such as what data should be collected, what standards should be used, and how electrical and optical data can be combined while maintaining traceability from design through wafer fabrication to a single die. This problem is going to be exacerbated as moving toward multi-stack dies and optical modules will create even greater complexity.
Ivan Nikitski: How do you see the future of yieldWerx?
Aftkhar Aslam: I’m very optimistic about the future. The semiconductor industry has always been cyclical, but what we’re seeing now is significant long-term growth, driven in large part by AI. It’s not only software that’s growing. The hardware ecosystem is expanding as well, and AI will touch many different aspects of people’s lives.
We see quantum computing, photonics, lasers, and microLED technologies as a tremendous opportunity for us because these sectors will require increasingly sophisticated platforms like yieldWerx to improve yield, optimize designs, and refine manufacturing processes.
Looking ahead, quantum computing and photonic technologies will be instrumental not only in accelerating AI but also in making data centers more efficient, reducing power consumption, increasing data transfer speeds, and supporting sustainability. Over time, these technologies will also expand into industries such as automotive and life sciences.
For yieldWerx, the challenge is keeping pace with the increasing volume of data, the growing complexity of products, and the rapid evolution of technology. And we will need to continue accelerating our own platform development to meet those needs.
Ivan Nikitski: What’s your advice for the next generation of entrepreneurs?
Aftkhar Aslam: Even if you’re an excellent data scientist and you’re able to leverage AI effectively, you still need to understand the basic underlying engineering principles of semiconductor technology: what a semiconductor is and how products are designed, manufactured, assembled, and tested.
As technology nodes continue to shrink, manufacturing equipment is becoming increasingly sophisticated, and the amount of data being generated is enormous. But without knowledge of basic engineering principles, it’s very difficult to build solutions that truly serve the industry.
