The semiconductor industry has always been at the forefront of technology and automation. With complex manufacturing processes and wafer thin margins, the industry also experiences constant pressure to keep improving overall process yields, quality, efficiency and thereby profits. Due to the high level of process automation in the industry, data availability has never been a challenge. What has proved rather difficult is the utilization of available data, in real time, to assert control and improve performance.
The industry now faces an unprecedented challenge, as data availability peaks due to the inculcation of IoT. Sensors and devices are deployed across the fab and the already-vast amounts of data being generated from process equipment, third party applications, in-house legacy tools, manual/spreadsheet data, material logistics data, supply chain partner data and existing MES applications all contribute to data overload. Data abundance can be a massive challenge unless the manufacturing application deployed is able to capture and cultivate it to derive value.
For a traditional MES, data collection alone from the myriad of sources for each and every transaction happening on and beyond the shop floor is a massive challenge. Leveraging existing data through the applications of AI/ML and advanced analytics to generate near-real time action is virtually impossible.
Semiconductor industry use cases for a data platform
A truly Industry 4.0-enabled semiconductor fab needs more than a MES. It needs an IoT data platform, which encompasses the core MES functionality, but has the capability to unleash the potential of IoT through enhanced data storage and manipulation capabilities.
A modern data platform creates a sort of ‘process envelope’ where integration with automation and enterprise-level applications enables end-to-end connectivity. Cloud and container-based infrastructure bring enhanced data capture and processing. AI/ML and advanced analytics allow for deployment of APC mechanisms in real-time; and standard MES and Quality control functionalities deployed ensure that a single, overarching application orchestrates, controls and self-ameliorates the process.
Let’s examine industry use cases to better understand why an IoT data platform is not just a requirement, but a necessity for semiconductor manufacturing.
Equipment Data Capture
The challenges when it comes to equipment data in semiconductor manufacturing are multi-layered. First is the sheer volume of data being generated. There might be hundreds, if not thousands, of possible data collection points for transactions happening during process execution. Even data from standard interfaces may yield thousands of data points. A classic example in high-end manufacturing is Interface A (SEMI 300 standard), which is beyond what any normal/legacy MES deployed can handle. Traditionally, process engineers would pre-define a few parameters to be measured and subsequently be used for analysis. This approach invariably delivers sub-par results as the full data picture is never captured, contextualized and correlated to uncover possible defects and issues.
A second challenge is the way in which data is structured. While data from SECS/GEM and other standard interfaces might be structured, unstructured data from manual collections, spreadsheets, third party sources and low level machines need standard formatting in order to be processed. Third, with thousands of data points generated across the process every second from equipment, the in-process controls and oversight applications issuing alarm data and other status indications may range in the thousands of points per equipment. The need is to go beyond data collection in real-time, but correlate the data to provide context which drives action on the shop floor.
Solution – An IoT data platform is equipped to capture every single transaction occurring on the shop floor, which includes data from all generation sources. Data is structured to allow processing in real-time and prompt action. The IoT platform eliminates the need for any third party application as it integrates with the automation and equipment, allowing data collection and complex analysis, be it overlay analysis or stacked mapping, in real-time, for effective yield and quality management.
Processes which have extremely small tolerances and require frequent adjustment of parameters to stay in the acceptable process window are subject to R2R control. This is typically achieved through a third party application or in-house legacy tool in most semiconductor fabs. Applying R2R control requires data collection in real-time and deployment of sophisticated algorithms which define output parameters based on input values to enable requisite control in an automated manner. The dependence on third party applications and the need to have a controller at minimum for every piece of equipment means traditional methods have a potential to slow down the entire operation. It could also cause the potential loss of vital data in case the controllers aren’t integrated to the process execution application.
Solution – When deployed through an IoT data platform, R2R control can be established as cutting-edge APC functionality; all controls are defined, configured and deployed through a single platform. The platform uses advanced analytics to create controllers, which self-optimize due to ML, meaning with each new transaction the controllers iteratively perform better than the previous transaction. The platform enables controller lifecycle management within its functionality for versioning, logging, debugging, simulations and approvals to happen on the same platform. When deployed through the data platform, R2R control enables better and more proactive recipe management, as data is collected in real-time and controller outputs created in near real-time. This is just not possible with traditional process tools from a speed and scale perspective.
Fault Detection and Classification
FDC monitors sensor data coming from equipment in a continuous stream, analyzes it and applies user-defined limits to detect process anomalies. The primary challenge in FDC is the collection of relevant data points. For each production step executed, there might be hundreds or even thousands of data instances for process parameters which can be measured and recorded. Hundreds or thousands of data points may be compounded across the process for each wafer and chamber tool. Even if the vast amounts of data generated are successfully captured, the next challenge becomes the analysis of data and classification of failures. The main hurdle is correlating the data in a manner which allows the correct classification of a given defect. It must trigger a corrective action which may range from putting material or equipment on hold to changing sampling patters and updating maintenance requests. This needs to happen in real-time as the process executes. This immediacy is where third party or legacy applications threaten to slow down the process and thereby impact/decrease the performance and yield.
Solution – As with R2R control, the IoT data platform collects all process data, cleans it, sorts it, and augments it with context to provide the most accurate fault profile. The platform deploys advanced analytics and ML to compare data recorded with established standard data to create precise actions and alarms. A classic example would be an oven profile for a tempering process, where there are several stages from warm up, baselines and fluctuations in temperature. The data is compared with a ‘golden curve’ resulting in a good/bad indicator and/or a failure classification. The IoT platform allows streaming analytics of time series data, creating KPIs and qualitative indicators which enable actions from detections and classifications to be made.
Sampling standards dictate rigid rules on the frequency of sampling and testing in a predefined manner. While this practice has been effective in the past, Industry 4.0-level semiconductor operations must move towards dynamic sampling, which creates a self-auditing and self-improving testing process, leading to exception-based rather than standard sampling protocols.
Solution– The IoT data platform enables dynamic sampling to become a reality for semiconductor manufacturing. By correlating historical testing data and real-time sampling data, it can predict the frequency and need of testing based on the deployment of advanced analytical tools, which pave the path towards exception- based testing.
Yield Management and Wafer Map Analysis
In many semiconductor fabs, yield management or defect density engineering is performed by third party applications, which point to good/bad (or pass/fail) calls being made and prompts decisions on whether or not to raise an NC. Unless data pertaining to defects is fed back to the SPC module, detailed analysis and formation of trends which could lead to insights remains amiss. If the MES isn’t integrated with the third party application or if the data needs to be fed manually or by uploading a file, the classification of defects, correlation between layers, calculation of killer defects and cross-reference of material and process information leading to a root-cause analysis becomes a major challenge. Similarly, for Wafer Map Analysis, it is imperative to consider and correlate data pertaining to material logistics, tracking, inline measurement data, parametric and functional test data.
Solution– An IoT data platform enables yield management and wafer map analysis, as it is the application which encompasses the entire production process; all requisite data is available within the platform itself. With the right data platform, defect-driven yield analysis becomes a reality. The platform allows algorithms to correlate defects to their causes, permitting RCA (root cause analysis) to precisely point out the origin of a given defect type. Only through an IoT data platform can yield management and wafer map analysis be truly optimized.
Complex processes, millions of shop floor transactions, thousands of possible measurable process parameters, dependence on legacy tools and third party point applications. All of this with extremely high pressure to improve production lead times, increase quality and maintain margins. It beckons for a step change in semiconductor manufacturing operations management.
This is the right time to act on digital transformation for the semiconductor industry. Supply chain strain, component shortages and unfettered demand have placed pressures on semi like never before. A step change is needed in manufacturing, and without the accompanying costs of building a new fab, upgrading and modernizing existing operations.
One means to progress: adopt an overarching, singular, modular, IoT-enabled MES data platform. The use cases underscore the necessity and criticality of having an IoT platform as the starting point of your digital transformation.