Travis Stevens, on August 09, 2021, 11:06 AM
A New Way to Optimize Semiconductor Yield: Make the Machines Hum Again
Each adult in the developed world today requires at least a hundred or so semiconductor chips to run their various devices, another 1000 for every vehicle they own. The semiconductor industry has been desperately trying to keep up with consumer demand – churning out over a trillion chips per year. And yet, there is now a severe chip shortage, and it’s getting worse. Global supply chains are stretched to the limit, and this is especially true in the semiconductor industry.
Multi-million dollar machines are required to manufacture these microscopic chips but there is a huge shortage of these machines worldwide, so the ethos for manufacturers today is “use what is there”. That is, squeeze 5% to 10% more yield out of their existing machines. Make the machines hum! No easy task in a 70-year-old industry that has been chasing maximum yields since its inception, but as is often the case, technology opens a window of opportunity.
But before looking forward, it is useful to look back and see how the machines were made to hum in the days of old. All the way back to the analog era of the 1930s, when factories depended on the master technicians who perfected their ability to maintain their machines over the course of 20–30 years. So much so that many of these technicians could simply walk up to a machine, place their hand on it, feel the hum and know exactly what was wrong with it, and then, with a slight tap of a hammer in the exact right spot, the machine would hum again, and produce at optimum capacity.
That was then, this is now
Those days are over. Clean rooms in semiconductor fabs, which require technicians to wear body suits and gloves, mean that the human hand can no longer touch the machine. Additionally, the technician is often located in a different city than the machines themselves, and so he or she can only monitor their machines using a Dashboard on their computer screen. The great advance of globalization means that a technician sitting in Munich can monitor their machines running in Singapore, but the downside is of course that the hum will never be felt by a human hand again.
So, without the human hand on the machines, how can yields be maximized by making the machines hum? The answer: a digital hand on the machines that can be just as powerful, or perhaps even more so. How is this possible? Look to the logs of course…
Almost every machine in a semiconductor fab generates endless slews of information in their multiple gigabyte logs. Every millisecond is now tracked, and while 10 years ago this information was simply too overwhelming to process and analyze, this has now changed due to the exponential growth of processing power, and the advent of technology for handling these massive logs – primary among which is ELK.
Introducing ELK – A digital foundation for driving continuous improvements
The E in ELK stands for Elastic, and it delivers ultra-fast searches across the data in the logs. The L stands for Logstash and it is the entry point for working with logs. And K is for Kibana – the graphical front-end GUI that enables the creation of dashboards and charts – which can be used to create almost any imaginable visualization of the mass of data from the logs.
Altogether, ELK can deliver every detail related to each machine in a fab and depict this on a computer screen. Perhaps not the subtlety of the sense of touch offered by the human hand – but maybe even better – a multitude of hands reaching out to a multitude of machines, making not only each machine hum but the entire fab. 120% capacity might just be possible after all.
What about the hammer to fix a problem? Also possible! With all of the data collected and delivered through the ELK stack, a multitude of additional applications are possible, including the identification of root causes of problems, which enables quick remediation of quality issues, and a return to optimization with minimal if any downtime.
Transforming data into actionable optimizations
And this is not a theoretical concept. SYSTEMA has implemented such systems, such as a project that was initiated by a large global manufacturer who wanted to track lot movements over time across their fab floor, with the purpose of helping with the identification of root causes of problems caused by lot quality defects, which you can read about in this case study.
So using the ELK stack as a foundation, the mass of information stored in your machine logs can be transformed into optimization and investigation applications. The hand on the machine and the accurate swing of a hammer to fix it will never return but lest we yearn for days of yore, technology fills the void with something less tactile, but perhaps even more powerful.