Who Is Most Likely To Link Financial And Manufacturing Data?

Experts at the Table: Semiconductor Engineering sat down to discuss which companies have the most to gain from linking financial data with manufacturing data analytics platforms with Dieter Rathei, CEO of DR Yield; Jon Holt, senior director of product management at PDF Solutions, Alex Burlak, vice president of advanced analytics and test at proteanTecs; and Dirk de Vries, technical program manager and senior architect at Synopsys. What follows are excerpts of that conversation. To view part one of this discussion, click here. Part two is here.

(L-R): ProteanTecs’ Burlak, DR Yield’s Rathei, Synopsys’ de Vries, PDF Solutions’ Holt

SE: Is considering financial data in manufacturing decisions more advantageous for wafer fabs or OSATs?

Holt: The OSATs probably don’t care about the financial data, but the fabless companies care. The OSATs typically are leasing tester time or tool time. You get the tester, and we test for this many hours on this number of parts. But the fabless manufacturers care because they’re making decisions regarding which products to ship to the OSATs.

Burlak: Both sides can leverage financial data to optimize their operations and be in a win-win position. There are a lot of optimizations that can be taken by the fabless company and OSATs using the data they have from the chip, testers, and assembly. Collaborating on this data and translating it as a part of financial data models benefits both sides.

Rathei: I agree that the OSATs probably don’t care at this time about these financial matters. I made this statement about OSATs earlier in the context of the next generation of 3D integration, because at some point OSATs will play a significant role in supporting these technologies. Then we’ll face the problem of making these kinds of advanced decisions in order to optimize the performance of these platforms.

Holt: That is valid point because it’s in the news. TSMC is partnering with SPIL on a next generation OSAT. Intel also is planning very similar activities. So it is becoming more of a shared ownership in the future. But it’s in progress.

Rathei: I was always surprised that data exchange between foundries and fabless companies was, at least in the past, not a big topic in the contract negotiations. I hope that as more decision-makers understand how important this kind of data access is for optimizing the manufacturing flow they will start specifying this in the contracts.

SE: Within your specific companies are manufacturing decisions made in a financially aware manner?

Holt: There were existing solutions out there from Boomi, MuleSoft, and others called enterprise application integrations that link factory data with financial data. We have combined that with a manufacturing data analytics platform. Currently there are two types of data that are readily compatible. AI or business logic rules can be applied to that data to share information. That data remains on-premise, within the customer ownership. They can utilize both the financial data and the manufacturing data for decision-making or for analytics. It’s a Kubernetes-based platform, which has advanced analytics capabilities that performs decision-making without necessarily transferring the IP or data from the platform. We’re seeing big applications that use financial data to manage and secure the supply chain or to perform cost setting on the parts. And there’s a fourth one that comes through the financial side, the business-to-business side, and that is field returns. Field return data comes through the financial system in the form of RMAs, and that information is very important for tracking down quality issues on the manufacturing floor.

Rathei: We started in 2005 with big data. About six years later I heard the term “big data” for the first time. Then we started in 2017 with the development of AI applications, which the industry is beginning to use now. Today we are developing these financial decision-making algorithms, and the interfaces to the financial data. As a solution provider, we can develop these features. Whether they will be adopted by customers is their choice.

de Vries: The way that we’ve approached this is to specify and provide APIs to product data management systems so there’s a smooth exchange. Fabless companies track target and actual test time, test costs, etc. We provide APIs to have a seamless interaction between the analytics platform and the customer PDMs. Then, internally, we apply a combination of black box and normalization to do the analysis, to visualize results, and to report on the findings. That is in place. It is mostly through the fabless company and not through the OSATs. We get that data in an indirect way. Certain customers have a keen interest in these applications. Other customers are a little bit less advanced. But these types of analyses are beginning to take off.

Burlak: From the financial data perspective we’re focusing on three major elements — the cost of parts, optimization of quality, and the time-to-market. To help our customers with these three elements we have the infrastructure that uses two capabilities. One is the capability to extract parametric data that provides visibility into different aspects from each chip. For this data we have a platform that is capable of taking the data and training machine learning models or different types of advanced analytics to connect between these data versus the desired KPI – whether that’s RMAs, DPPM, yield, binning, material selection, or other. The second capability is to take these models and implement them into the ATE and system-level test flow. This is an important component, because eventually, even if you create a model, you want to make a decision on a per-part basis on the manufacturing line. This model needs to be implemented back at the test program for inline predictions and analysis. This is how we close the loop.

SE: Which companies do you expect to be the early adopters?

Holt: Let me give a quick answer. We’ve seen a huge amount of interest in this. But those folks that already are implementing flows are typically customers that are in the process of digital transformation. They’re going to a large data lake and they’re updating their ERP systems. Most of the leading-edge companies are investing billions of dollars to do this now.

Rathei: We are in a very competitive environment, and everybody has to optimize all the data flows in order to survive.

de Vries: Where I see the most interest is in the complex SoCs — for example, processors with a lot of cache, etc. But there are also complex systems with optical components like face recognition. The biggest gains will be in those complex MCMs because the packaging cost is relatively high.

SE: What are your parting thoughts?

Holt: This is a semiconductor discussion, but we’re seeing this same demand from battery companies at the system level. And I have some requests in agriculture. It’s not just high tech combining financial and manufacturing data.

de Vries: You can see that within the discussion that we had today. We have four independent trajectories and we’re see the same things. It’s not a very contentious topic. So we’re really moving forward jointly to support our customers and make these kinds of analytics-supported decisions possible.

Rathei: I agree. I also noticed that we were very much aligned on the big picture. There are some differences in opinions and observations on the details, but the industry has to proceed this way to be more efficient.

Burlak: At the end of the day, everything we spoke about applies to decision-making and optimization in-field for chips that have been deployed and are running the application. For example, if you’re now bringing up a new generation, there is a lifetime amount of maybe three, four, five years that has accumulated from previous generations. There are all kinds of decisions on optimization that can be done and tied to the financial data that can optimize these types of transitions between generations and how you are managing the deployment of new systems or extending the lifetime of the existing systems. These topics are beyond the manufacturing process.

Read parts one and two of the discussion
Secure Handling Of Financial Data In Manufacturing
Data sharing becomes more challenging when AI and multi-die assemblies are involved.
Cutting IC Manufacturing Costs By Combining Data
Mixing financial data with manufacturing analytics can boost efficiency, but there are still pockets of resistance.