By Paula Doe, SEMI
The changing market for ICs means the end of business as usual for the greater semiconductor supply chain. Smarter use of data analytics looks like a key strategy to get new products more quickly into high yield production at improved margins.
Emerging IoT market drives change in manufacturing
The emerging IoT market for pervasive intelligence everywhere may be a volume driver for the industry, but it will also put tremendous pressure on prices that drive change in manufacturing. Pressure to keep ASPs of multichip connected devices below $1 to $5 for many IoT low-to-mid end applications, will drive more integration of the value chain, and more varied elements on the die. “The value chain must evolve to be more effective and efficient to meet the price and cost pressures for such IoT products and applications,” suggests Rajeev Rajan, VP of IoT, GLOBALFOUNDRIES, who will speak on the issue in a day-long forum on the future of smart manufacturing in the semiconductor supply chain at SEMICON West 2016 on July 14.
“It also means tighter and more complete integration of features on the die that enable differentiating capabilities at the semiconductor level, and also fewer, smaller devices that reduce the overall Bill of Materials (BOM), and result in more die per wafer.” He notes that at 22nm GLOBALFOUNDRIES is looking to enable an integrated connectivity solution instead of a separate die or external chip. Additional requirements for IoT are considerations for integrating security at the lower semiconductor/hardware layers, along with the typical higher layer middleware and software layers.
This drive for integration will also mean demand for new advanced packaging solutions that deliver smaller, thinner, and simpler form factors. The cost pressure also means than the next nodes will have to offer tangible power/performance/area/cost (PPAC) value, without being too disruptive a transition from the current reference flow. “Getting to volume yields faster will involve getting yield numbers earlier in the process, with increasing proof-points and planning iterations up front with customers, at times tied to specific use-cases and IoT market sub-segments,” he notes.
Rapid development of affordable data tools from other industries may help
Luckily, the wide deployment of affordable sensors and data analysis tools in other industries in other industries is developing solutions that may help the IC sector as well. “A key trend is the “democratization” – enabling users to do very meaningful learning on data, using statistical techniques, without requiring a Ph.D. in statistics or mathematics,” notes Bill Jacobs, director, Advanced Analytics Product Management, Microsoft Corporation, another speaker in the program. “Rapid growth of statistics-oriented languages like R across industries is making it easier for manufacturers and equipment suppliers to capture, visualize and learn from data, and then build those learnings into dashboards for rapid deployment, or build them directly into automated applications and in some cases, machines themselves.”
Intel has reported using commercially available systems such as Cloudera, Aquafold, and Revolution Analytics (now part of Microsoft) to combine, store, analyze and display results from a wide variety of structured and unstructured manufacturing data. The system has been put to work to determine ball grid placement accuracy from machine learning from automatic comparison of thousands of images to select the any that deviate from the known-good pattern, far more efficiently than human inspectors, and also to analyze tester parametrics to predict 90% of potential failures of the test interface unit before they happen.
“The IC industry may be ahead in the masses of data it gathers, but other industries are driving the methodology for easy management of the data,” he contends. “There’s a lot that can be leveraged from other industries to improve product quality, supply chain operations, and line up-time in the semiconductor industry.”
Demands for faster development of more complex devices require new approaches
As the cost of developing faster, smaller, lower power components gets ever higher, the dual sourcing strategies of automotive and other big IC users puts even more pressure on device makers to get the product right the first time. “There’s no longer time to learn with iterations to gradually improve the yield over time, now we need to figure out how to do this faster, as well as how to counter higher R&D costs on lower margins,” notes Sia Langrudi, Siemens VP Worldwide Strategy and Business Development, who will also speak in the program.
The first steps are to recognize the poor visibility and traceability from design to manufacturing, and to put organizational discipline into place to remove barriers between silos. Then a company needs good baseline data, to be able to see improvement when it happens. “It’s rather like being an alcoholic, the first step is to recognize you have a problem,” says Langrudi. “People tell me they already have a quality management system, but they don’t. They have lots of different information systems, and unless they are capturing the information all in one place, the opportunity to use it is not there.”
Other speakers discussing these issues in the Smart Manufacturing Forum at SEMICON West July 14 include Amkor SVP Package Products Robert Lanzone, Applied Materials VP New Markets & Services Chris Moran, Intel VP IoT/GM Industrial Anthony Neal Graves, NextNine US Sales Manager Don Harroll, Optimal+ VP WW Marketing David Park, Qualcomm SVP Engineering Michael Campbell, Rudolph Technologies VP/GM Software Thomas Sonderman, and Samsung Sr Director, Engineering Development, Austin, Ben Eynon.
All the arguments here are too obvious and are not different from a regular manufacturing of any device!!!