This text summarizes the content material of a paper developed and offered by Advantest at ETS 2022.
Written by Sonny Benwari and Matthias Sauer
In keeping with the market analysis agency Gartner, Inc. When evaluating the completion charge of information science tasks, in addition to the underside line worth they generate for his or her firms, solely 15 to twenty p.c of those tasks are accomplished. Furthermore, amongst those that handle to succeed in completion, lower than 10 p.c generate worth, in accordance with feedback offered by company CEOs. Backside line: Lower than 2 p.c of information science tasks are ever accomplished and ship worth. How can the waste of company investments and efforts be mitigated? A method is to make use of superior machine studying (ML) strategies.
Nonetheless, implementing ML in on-line manufacturing testing presents its personal set of challenges. ML-based functions problem conventional take a look at flows and infrastructures, as they require:
- massive quantities of information, usually by means of a number of entries which are geographically unfold throughout a number of continents and positioned in several company entities relying on the place within the worth chain;
- safe, scalable, and built-in computing infrastructure based mostly on open requirements; And
- An infrastructure for executing dynamic testing.
A few of these traits really battle with conventional take a look at setups, resulting in non-standard take a look at flows and creating further work that impacts time to market and return on funding—specifically slowing the success and adoption of machine studying functions.
Repeatability and repeatability are important for testing procedures, particularly for the automotive and different markets that depend on standardization and a excessive diploma of conformance. This requires making a extra structured desirous about ML and its affect on testing. Determine 1 reveals an industry-ready machine studying lifecycle designed to convey information science onto the testing ground.
Determine 1: The 4 main ACS-supported machine studying lifecycle steps are proven right here.
Empowering the ML ecosystem
Advantest Cloud Options (ACS) is a extremely safe, scalable information platform that allows an ecosystem of open options that helps clients handle probably the most urgent challenges of the period of clever manufacturing. The open facet of the ecosystem is crucial, because it permits any firm alongside the availability chain to not solely use it however so as to add worth, create partnerships, deploy their very own options, and so forth. ACS offers the crucial infrastructure piece, in addition to a variety of improvement choices by means of the ACS Options Retailer, whereas Advantest offers different software program services that clients can buy once they want complementary providers to reinforce or improve current deployments. Let’s take a fast take a look at every of the 4 principal steps of the machine studying lifecycle.
Discover and perceive the issue
The extra early adopters search to establish issues within the manufacturing course of, the extra information they want. Early detection of unhealthy molds is an important element of this effort, as prediction of doable failures after packaging may also help scale back prices considerably and enhance high quality within the packaging course of. To realize this requires massive quantities of top quality information. Determine 2 reveals the normal take a look at circulate with out the advantage of ML strategies on the left. On the best, our ACS know-how evaluates and correlates information collected from earlier entries to precisely predict issues, enabling the person to avoid them by fixing issues on the root trigger, thus stopping unhealthy deaths from reaching closing take a look at entries. This not solely reduces the price of testing, but additionally the price of supplies and processing wanted because the chips journey by means of the three-month manufacturing course of throughout continents and firms.
Determine 2: With Advantest ACS applied sciences, clients can delete unhealthy molds early within the testing cycle to cut back packaging prices and enhance high quality.
Mannequin engineering
Mannequin engineering is a crucial step for implementing the evaluated enterprise necessities and turning them right into a data-driven ML software, both utilizing a customized implementation or utilizing a predefined resolution from the ACS Answer Retailer.
To judge the “ACS Yield Optimization” reference software described above, we compiled a real-world dataset containing greater than 200 related comfort outcomes for every template from the probe take a look at and a number of failure bins from the ultimate take a look at.
By working ACS-based information analyzes on aggregated, unscavenged information, utilizing a single device-in-test (DUT) identifier for every entry, the instrument makes use of deep learning-based variable choice to establish the variables with the best affect on return. It then generates new test-probe thresholds based mostly on the distribution of this information, eradicating false passes and confirming the return enchancment. Within the above case, the end result was a 5% enchancment in yield, from 88% to 93%, which interprets to six-figure annual financial savings in US {dollars}.
Determine 3: ACS Yield Optimization makes use of deep studying to investigate and optimize a variable, leading to larger returns and vital value financial savings.
Publication and implementation
This particularly refers back to the integration of ACS’ high-performance safe testbed with traceable deployments, as proven in Determine 1. Determine 4 reveals our core ACS Edge product, which incorporates ACS Edge Server and ACS Container Hub.
ACS Edge is a high-performance, extremely safe computing and analytics resolution that allows ultra-fast AI algorithm resolution making with millisecond latency throughout take a look at execution. It communicates with the person’s take a look at gear over a non-public, high-speed encrypted hyperlink and makes use of the superior container hub to run the person’s protected functions whereas defending and preserving the person’s information and analytics.
Determine 4: Advantest ACS Edge and its accessible extensions assist clients simply and securely combine ACS into their take a look at circulate, enabling them to understand the total advantages of its ML-enabled capabilities.
Monitoring and validation
To ensure that fashions to go from laboratory to quantity manufacturing, they should be monitored for any sudden behaviors attributable to adjustments in product design or testing setting. Optimization effectiveness should be verified utilizing real-world eventualities to ensure that the information science challenge to finish and contribute to the general worth of the corporate.
Semiconductor manufacturing is enormously affected by course of variations from die to die, wafer to wafer, or batch to batch, particularly in smaller course of nodes with tighter geometries that present much less room for deviation. Thus, there’s an inherent threat of degradation of the “silent” mannequin when the properties of the acquired course of change, which can have an effect on the standard of the mannequin (yield, take a look at time, system high quality, take a look at escapes, and many others.) ACS makes use of a steady studying loop with steady monitoring, which considerably reduces These dangers are so nice that fashions keep their integrity.
ACS Answer Retailer
One other important a part of the ACS ecosystem that helps create a standardized and repeatable workflow is the ACS Answer Retailer, which offers quick access to real-time ACS information infrastructure options and software program functions. This on-line platform permits clients to find, buy and securely deploy all accessible ACS options from Advantest and a variety of analytics ecosystem companions throughout the semiconductor lifecycle course of. As well as, the ACS Answer Retailer allows app builders from these companion firms to publish, promote, distribute, and handle their Advantest-certified apps.
This final facet of ACS’s providing is significant to sustaining an open ecosystem, because it makes all ACS choices simply accessible to clients, in addition to giving them and our companions the flexibility to develop and deploy their very own functions. This enables new capabilities and finest practices to be shared in order that the capabilities of ACS applied sciences could be optimally leveraged throughout firms all through the semiconductor ecosystem.
We proceed to develop and evolve our Advantest cloud options to satisfy evolving buyer necessities. By inserting ACS of their take a look at environments, clients can guarantee they’re armed and prepared for the way forward for semiconductor testing.
Matthias Sauer is an utilized analysis challenge supervisor at Advantest Europe.