Supporting the renaissance of 200mm fabs
Earlier this year, Intel revived its construction of the 450mm factory in Chandler, Arizona. While that grabbed a lot of media attention, equally of significance is a quite renaissance, in the demand for 200mm fabrication happening across the world. Interest in IoT and automotive sector has particularly contributed a surge in demand for MEMS, power, analog and discrete semiconductors, much of which can be manufactured on 200mm wafers.
According to SEMI, by 2019, installed capacity of 200mm will reach close to 5.4 million wafer starts per month. This will be almost as high as capacity in 2006, adding a net capacity of more than 600k wpm in 5 years. Of the 19 new fabs that started construction in 2016-17, 4 are 200mm.
Despite the healthy growth in this installed capacity, the industry is experiencing an acute shortage of 200mm fab capacity. And, should the IoT and EV and various other changes in automotive markets take off as expected, the surge in demand could make the capacity shortage even more acute over the next few years.
Many of the existing 200mm semiconductor fabs operating in US and Europe were expected to close soon a few years back. However, now there is a much stronger rationale to extend their lives, and continue to operate them for longer, given the expected demand growth.
To continue to run these fabs profitably, there needs to be a renewed focus on improving fab cost efficiencies, and increasing yields. For this, there are a number of levers that could be deployed. To do so, it is important to understand the economics and operating models of these fabs. They are very different from the 300mm fabs, which are highly automated. 200mm fabs have much higher labor intensity per square-unit of silicon. They are typically used with smaller production runs. They use old equipment, much of which is difficult to maintain, and cost of replacement parts has surged in the last few years.
There are a number of emerging levers that be used to improve the efficiency and yields, and especially have the potential to address these operating specifics of the 200mm fabs. Four of these are particularly important:
- Improving downtime prediction and station utilization: Old fab equipment understandably tends to have more downtime. Unfortunately, 200mm equipment parts are in scarce supply, and are getting increasingly expensive. Latest machine learning techniques can reduce downtime and improve maintenance planning significantly by better predicting out of control parameters of the equipment and process. These go well beyond control charts that apply rules such as Nelson rules. Furthermore, they can predict using large number of parameters and the interplay between those parameters. Alarms for timely intervention can also be dynamically triggered based on cost and impact functions.
- Improving factory traceability and creating a real-time information bus: Production costs can be reduced by improving feedback loops with upstream activities. Manufacturing production orders can be prioritized based on demand criticality, for e.g., by allowing putting some orders on hold to let the important orders progress. The most important capability to build here is a granular lot traceability. It also provides more flexibility through allowing lot splits and merges. Furthermore, a building a central data bus can allow real-time reaction to any events across the factory, allowing faster response to potential downtimes and wastages.
- Improving low-hanging operations automation: 200mm fabs have a heavier load of direct and indirect labor costs. These fabs were built in 1990s, when semiconductor manufacturers hadn’t yet developed the automated tools to handle hundreds of steps required in die manufacturing, and manual activities were needed, especially since the processes varied significantly. Although automation in these fabs improved with time, there are many fabs where surprisingly large amount of low-hanging automation opportunities still exist. Techniques, such as extensive use of RFID chips to all movable parts, can improve automation significantly. Updating IT systems e.g. for maintenance management can further reduce operating costs significantly.
- Left shift of quality predictions: Yield improvement is the most critical goal of all semiconductor operations. Yield has become even more important in 200mm factories given the capacity constraints. , probe testing, post-assembly testing and customer defect tracking, provide the feedback too late. Having better quality interventions earlier can make a significant difference in yield improvements. There are two parts to the shift-left of quality interventions. First, supervised or unsupervised modelling to help detect failures throughout the manufacturing process, as the wafer goes through hundreds of steps. Second, linking failures during probe testing to the exact cause of failure can help eliminate the root cause. Given the volumes of data that is generated during the manufacturing process, these techniques are not easy to apply. The first part however, is much easier compared to the second, and small investments in the direction can make a big difference.
Semiconductor companies with 200mm fabs have a reason to be excited. They are certainly in demand. And opportunities have emerged to improve their efficiency and yield further, so that they can continue to meaningfully contribute to the companies’ bottom-line for possibly even decade more than was earlier thought!
More from Pankaj Chugh
Convergence is happening all around, and rapidly. The distinction between the physical and…
R&D constitutes 15-20% of the sales in semiconductor companies. Top 10 semiconductor companies…
Latest Blogs
How will extended reality transform CX? Extended Reality (XR) is a term that brings together…
In today's digital era, ransomware attacks and other cyber threats are more prevalent than…
In the evolving landscape of technology, the rise of quantum computing stands out as a frontier…
In contemporary corporate landscapes, the pursuit of human resources (HR) transformation remains…