Benefits of Advanced Process Control in Semiconductor Manufacturing
What is Advanced Process Control (APC) ?
Today's semiconductors have become an integral part of our way of life, almost every appliance has a plethora of chips making our lives easier. This rapid increase in demand for chips coupled with the increasing complexity is pushing high volume, state of the art, semiconductor manufacturing facilities to adopt advanced manufacturing technologies. The methods of process control such as qualification, wafer send ahead, extended tool preventive maintenance, and SPC alone are no longer sufficient to achieve the tight tolerance budgets of today but, are also detrimental to the cost effectiveness of today’s manufacturing. In order to achieve both the tight tolerance budgets and cost-effective manufacturing, methods of Advanced Process Control (APC) must be applied.
APC is used to control a process affected by systematic variations. The components of this control methodology are most commonly broken down into two main control mechanisms, Run-to-Run (R2R), and Fault Detection Classification (FDC). Because both mechanisms address systematic variation, tool and process models can be applied to automate process control optimization.
R2R control of a process on a tool uses data from outgoing and incoming wafers in combination with a model of the process on that tool, to adjust process parameters. FDC on the other hand, is the prediction of tool failure where application of an appropriate model to classify failure root cause.
Application of these control methodologies is not to say that traditional SPC techniques no longer play a role in today’s manufacturing, on the contrary, because SPC is intended to handle normal variation, it can be more appropriately applied with the systematic variants accounted for by APC. For example, the drift of a process over time does not trigger SPC violation because the given drift is addressed through R2R compensation.
APC has been applied to varying degrees in the semiconductor industry, however the most prosperous manufacturing facilities are those that have the highest level of understanding how to integrate APC into their production flow. The goal is to reach as real time as possible for both R2R as well as FDC.
What is Run-to-Run (R2R) Process Control?
R2R control is a method where process parameter adjustments are made via feedback (FB), feed forward (FF), or FF/FB combination models. (See figure 1a, b, and c). Process parameter setting updates are derived from on-going analysis of historical process and metrology data. (See algorithm Example 1 below).
1.1 Feedback Model
The feedback R2R model is the simplest form for implementation into the production environment. The reason for this is that the process modeling is straight forward, i.e., the number of variables is small and the time between process and metrology is relatively short.
For several years in lithography this type of model has been utilized for lot-to-lot Dose, and Alignment offset settings. Below are generic examples of both CD and OL FB algorithms.
1.2 Feed Forward Model
The feedforward (FF) approach entails incorporating key pre-process measurement data in the process control model. The simplest application of this type of model is for lithography Overlay (OL) control. In this case, the measured alignment for a given lot is fed forward to the next alignment process for that lot. FF for Dose compensation is also possible based on film thickness measurements, however this type of compensation is more appropriately handled with a combined FF/FB model.
Example 2 Lithography Overlay (OL) Algorithm:
New Parameter = current tool setting + measured offset from previous process
1.3 Feedforward/Feedback Combination Model
This combination model considers pre-process results as well as drift of the current process. With this hybrid approach a higher degree of systematic disturbances can be addressed. A comprehensive characterization of the process on a large data set can be used for building the algorithms for this type of hybrid model.
For example, a material film thickness can be shown through simulation to have a significant and systematic effect to printed CD, however, this must also take into consideration other variables, both pre and current process, influencing CD.
To gain the confidence to code a R2R algorithm, the variations of pre-process and process variables must be tested. To do this, the process engineer may start out with the appropriate DOE’s, compare results to simulations and verify against available in-line product data or yield result.
To give one example of a simple approach to this model, consider a typical lithography FB model in operation.
An exposure dose and CD may be trending normally and a pre-process event (e.g., nitride) may shift to one end of its tolerance. If the parameter shift is then observed as a spike or shift in CD or exposure dose, then a passive FF/FB system could alarm a user of a correlation. The advantage of this approach versus relying solely on the Dose FB to bring CD on target is that it may take several lots with the same pre-process conditions to effectively re-center CD or the pre-process parameter may then also re-center and further FB adjustment will be needed.
The result is two out of control (OOC) events for printed CD, convergence to target CD is slow and high rework can be expected. The passive FF/FB system would serve to avoid this situation by causing a process flag to trigger a send ahead for pre-process operation point change.
1.4 Parameter Set Points, Alarm, Reset, Timers, and Data Integrity and filtering Considerations
For effective R2R, basic functionality must be in place to monitor (i.e., SPC) recipe set points to safeguard against adjustment/correction limit violation. That is, the point a given tool or process can not physically correct beyond…e.g., “the knob will not turn past 10!” As such conditions are approached in a R2R environment, it is necessary to alarm or flag the event and trigger a system reset.
Alarm basic functionality, will flag the user when thresholds, which may affect the control capability of the R2R system, are breached. With FB controllers, the time variable between process, measurement, and response to next process event is critical. Therefore it is desirable to enable alarm and reset functionality for time delay events.
Also, to assure effective R2R, all sensor/measurement data must be filtered and validated for, data integrity. This means that erroneous data points must not be allowed to feed into control algorithms. To achieve this, the R2R system must have the intelligence to apply statistical flier models, e.g., Q limits test built into its basic functionality.
Benefits of implementing APC Systems like R2R ?
Improves process capability (Cpk) through optimizing recipe parameters lot-to-lot, wafer-to-wafer, die-to-die or batch-to-batch.
Reduces and prevent out-of-spec (OOS) & scrap events by compensating for both incoming and inherent systematic variation.
Lower Cost & Cycle-Time
Improves the throughput by reducing SAHD wafer and Pilot lots, eliminating the need for manual adjustment allowing more wafers to be processed.
A Return on Investment (ROI) can be estimated by calculating a yearly Revenue Opportunity Loss (ROL) as a function of; number of rework lots per event, average events per year, wafer per lot, number of layers per part, chips per wafer, average wafer yield, and current market price per chip. See the following equation below.
In the equation, LM represents Lost Moves in wafers, Ycpw is yieldable chips per wafer, Ppc is the market price per chip, and Rn is the average number of significant process-adjustment (rework) events per year.
To illustrate this conservatively, consider the following sample calculation:
In summary, the rising demand of chips in everyday life coupled with performance driven device complexity, will continue to push semiconductor manufacturing facilities to adopt Advanced Process Control systems that must allow for both Feed-Back and Feed-Forward Run-to-Run control. With such a systems in place, process adjustment compensation is made more consistently, and potentially harmful processing faults can be avoided without introducing avoidable rework or scrap.
In the upcoming blogs more detailed examples will be presented to illustrate real-word practical applications for both SPC and R2R with a focus on semiconductor processes (Lithography, Etch, CMP, Plating....).
The introduction of current Data Science and Machine Learning Applications will also be reviewed where both Supervised and Unsupervised learning models provide benefit to better tune control threads and identify unaccounted for features and variables impacting outgoing process indicators.
Gould, Christopher, "Complementary Feed-Forward and Feedback Method for Improved Critical Dimension control", SPIE Vol. 5378 (SPIE, Bellingham, WA, 2004)
Sentient’s “Software-as-a-Solution” offers end-to-end Knowledge, Guidance and Direction tailored to your unique needs, through Consulting, Training and Project Support. Let us help you Build Smarter and discover Value.
Abeer has over a decade of experience in semiconductor processing and advanced manufacturing. He is co-inventor on U.S. patents in the field of semiconductor manufacturing and has extensive experience implementing advanced manufacturing systems for high volume and complexity products. Prior to his role at Sentient, Abeer has worked as an equipment technician, process engineer, advanced process control systems engineer, and manufacturing systems architect. He has a degree in Microelectronic Engineering.