Big data in semiconductors—how to collect, detect, and act

Big data, defined as an all-encompassing term for any collection of data sets so large or complex that it becomes difficult to process using traditional data processing applications, continues to be a major topic in virtually every business segment. While big data solutions have gained a foothold in retail, sales, and financial organizations over the past 10 years, there are other industries, like semiconductor manufacturing, that stand to reap significant gains from extreme analytic solutions that can mine these enormous sources of data.

Why does big data benefit the semiconductor manufacturing industry?

Collecting and analyzing all the data sets across a semiconductor company’s global supply chain enables employees from operations, engineering, and executive management to improve a broad number of metrics to achieve the company’s manufacturing goals: to improve yield, quality, and productivity. Improvements in these areas lead to the semiconductor company’s ultimate goals of increased profitability and market share.

The key to successfully incorporating a big data solution into a company’s semiconductor test and manufacturing environment is to understand the true value and benefits of collecting, detecting, and acting on the timely data that is generated about every chip. Each chip has its own test DNA, and through big data solutions, that DNA can be analyzed to determine its true value to the semiconductor manufacturer and ultimately its customers.

Collection

For many semiconductor companies, testing can occur at many different geographic locations depending on the size and scale of their manufacturing operations. It is not unusual for wafer manufacturing to occur at a foundry in one country, wafer sort test in a different country, and final test or system-level test in yet a third country—leading to data fragmentation across the manufacturing landscape. Many semiconductor companies are resigned to the fact that this is what they are forced to deal with—highly fragmented data that is difficult to collect, validate, and analyze. However, there are ways to overcome this apparent deficit, and to have a single database that contains all of the data generated by their manufacturing supply chain regardless of the physical location of where the data is generated or the ATE equipment used to test the devices.

Test data (typically STDF) is generated on every tester in a semiconductor supply chain whether it is located at a foundry or OSAT. Traditionally, this data is stored and only reviewed at periodic intervals or after a serious manufacturing problem has been detected, and likely long after it started. Unfortunately, in the time it takes to go back and analyze the STDF data, far too much time has passed to correct the issue that was originally identified. The only remaining option by that point is to try to keep that problem from recurring and affecting yield, quality, or productivity in the future. As a result, many companies have just learned to tolerate the ongoing problem of losing millions of chips to yield issues that could have been saved if the source of the manufacturing problem was found sooner.

Furthermore, the IT effort of carrying out data collection and storage is becoming more challenging because of the sheer quantity of data being generated on a daily basis, combined with the extended lengths of time that the data needs to be made readily available for detailed analysis. Many semiconductor companies are now regularly storing parametric data for many years after final test to enable the analysis of product RMAs to pinpoint the source of unexpected field failures. Big data solutions allow companies to access and store enormous quantities of data (tens of terabytes up to petabytes) in support of RMA analysis, and save countless hours of engineering effort spent tracking the data for future RMA prevention.

Detection

Rapid insight into a chip’s test DNA enables any semiconductor company, big or small, to make better operational decisions, which can lead to impressive ROI. Big data solutions provide a real-time view of test data to address a wide range of manufacturing issues that can increase device yield or prevent problematic parts from going to market. The ability to collect STDF data directly from every tester in a global supply chain is the foundation to enable companies to make actionable business decisions within minutes of test completion– dramatically improving yield, quality and productivity.

Another benefit of big data solutions is the democratization of big data usage across a company’s multiple organizations. In the case of manufacturing data, different groups within a company can find value unique to them via the ability to access and analyze this data in a timely fashion. For operations managers, near real-time analysis of bin data is very helpful in uncovering certain signatures of devices based on geographical wafer location, helping to decrease test escapes. Test engineers can take full advantage of parametric data, analyzing tens of thousands of test measurements within seconds to detect trends that are impacting product yield. And for product planners, big data solutions enable them to have an up-to-the-minute view of how much product is in their supply chain, driving confidence in their ability to meet the product demand for their key market segments.

Action

Access to all the data in real time can help solve many problems at hand, but only if the ability to mine the data is equally as fast. The availability of automated data rules and algorithms enables engineers to easily and rapidly determine the root of a problem in test, which frees them up to focus on solving problems instead of looking for problems. By leveraging extreme analytics capabilities, semiconductor companies can realize the full benefit of big data solutions. Companies who invest in these solutions see immediate returns in test time reduction, yield recovery, and escape prevention.

Test time reduction

Collecting and evaluating data in real time alerts users when equipment problems arise. Thousands of algorithms feed test tools to detect abnormalities and alert operations to potential issues. The data helps determine where there are equipment problems and where things are running smoothly. This is a key frequent use of data that saves manpower and provides better accuracy in assessing equipment and other errors that may occur.

Recovery

Retesting a highly recoverable bin may recover 80% of the chips that failed during test. Retesting a poor bin, however, may only recover 1%. Operations teams can analyze the data and intelligently determine whether to retest one bin (data pointing to it being highly recoverable) vs. another bin (data pointing to it being less recoverable). This saves companies hundreds of thousands of dollars in tester time while achieving maximum yield recovery.

Reallocation of chips

Many semiconductor companies test chips for speed and tier them based on the results. For example, a first-tier chip may be 2.6 GHz while a second-tier chip may be in the 2.3 GHz range. One problem that is often seen is that during wafer sort these first-tier chips continue to run at top-speed, but as these chips are packaged and moved to final test, the chips may show some “drift” in performance, running at a slower, tier-two rate. Data from a chip’s test DNA can determine if the final test was an anomaly or if something else is affecting the chip in some way. Based on this information, the vendor can automatically reallocate and re-bin suspect chips to preserve quality and performance in the devices they deliver to their end-customers.

Escape Prevention

Understanding the DNA of each chip allows engineers and operations to comb through data and determine when a “good chip” may not really be “good.” One example is the identification of a good die in a bad neighborhood. While a particular die may test as “good,” being able to leverage geographical data that shows that this “good” die comes from a region of the wafer that has much lower yield can enable a company to do more testing to be sure that the chip is really “good” or even decide that regardless of the test result, that die will not be shipped to the end customer out of quality concerns. This type of big data analytics can dramatically improve overall quality and lead to higher customer satisfaction.

Why big data matters

Whatever a semiconductor company’s pain point—yield, quality, or productivity—big data solutions can help address the problem and provide substantial ROI. Today, many of the world’s largest semiconductor companies, both IDM and fabless, are leveraging the power of big data analytics to help them collect, detect, and act on their manufacturing data to ensure substantial ROIs in yield and productivity while improving overall quality – ultimately increasing profit margins and market share for their respective companies.