Abstract

This discussion illustrates real-time, automated process control for model-based recipe tuning in a Semiconductor manufacturing application. Advanced Process Control (APC) applications such as Run-to-Run (R2R) recipe tuning will be discussed in concert with classical Statistical Process Control (SPC).

For confident, transparent tracking of such active process control applications, SPC is applied to classify ‘special cause’ events in the Recommended recipe settings (aka Manipulation Variable), Applied settings, Metrology or model based Optimized settings.

It is important to consider, Advanced Process Control (APC) as an extension of the classic SPC methods that have been deployed in manufacturing for nearly a century, since Walter A. Shewhart first introduced it at Bell Laboratories in the early 1920s.

Introduction

Shallow Trench Isolation (STI) is a very common process, and a relatively straight forward application for APC. The basic stack-up thickness and depth control also applies to many MEMS processes. The Figure 1 depicts a typical cross-sectional flow for the process. 

Although, there are many points of process control considerations, like deposited material thickness control, for this discussion, we begin with a pattern developed in photo resist (step 2. – Fig.1).

The feature of interest or Control Variable (CV) here is the outgoing Depth in Silicon, as measured after a reactive ion etch (RIE) process. The practical application of traditional SPC and an active, real-time Run-to-Run controller deployed using Sentient will be described in terms of the Process Engineering setup and capabilities for optimal process control with transparent monitoring and reporting.

Basic Theory and Application

In this example, we consider the incoming measured film stack thicknesses (Step 3. - Fig1.) as input Model Parameters (MP), the Etch Process Recipe Settings as Manipulation Variables (MV) and the Targeted output Depth in-Silicon as the Control Variable (CV) as the key variables for monitoring and optimal process adjustment.

For direct process feedback, we consider a Process Model for the CV, Depth_InSi based on process characterization results for Etch Depth versus Etch Time from experimentation (Fig.2.). The final applied model will include pre-process thickness (Fig.3 Etch Time Controller Schematic).

The characterized linear process Feedback model is configured into the Sentient.cloud system, together with the upstream metrology and material context. These variables are illustrated in Figure 4 below.

Statistical Process Control for Run-to-Run Application

Now, we consider what is to be monitored, and what metrics and conditions will trigger process intervention or investigation.

In this case, we assume the pre-process film thicknesses have their own process controls applied and we start with the first parameter for monitoring as "Recommended Etch Time" that will be applied to the Process Recipe used and then the outgoing CV itself (i.e., Depth In-Si) and the calculated Optimized value determined for Etch Time, based on the current measured sample.

Automated Run-to-Run (R2R) process Recipe Tuning should not be based solely on process characterization and process assumptions. To avoid process excursion from a jump in the controller updates for process state, statistical monitoring, and reaction to SPC violation is key to maintain process modelling accuracy.

The key SPC metrics for Run-to-Run supported SPC include, Moving Range of the (1)Recommended Recipe Setting, (2)Sample Optimized Setting and (3) the Control Variable Metrology result.

The Run-to-Run controller is designed to continuously tune the Feedback signal gain (i.e., weighing of historical Optimized values) based on the delta between a lot’s Recommended Setting and the lot’s Optimized Setting, determined after measurement. Appropriate metrology data filters should be applied to remove measurement errors, that can impact the controller state and subsequent recommended settings.

Key Statistical Process Control (SPC) Charts

1. Optimized Setting (e.g, Etch Time) – Moving Range

The ‘Optimized Setting’ is the calculated optimal process setting based on a single sample being measured. That is, based on the setting used and the resulting measurement of a single sample. Since process drift is expected, SPC is not applied to the optimized setting values, but rather the Moving Range of these values calculated from the measurement of the control variable after processing.

Run-to-Run update filtering of the calculated ‘Optimized’ values guards against not only measurement error, but also unaccounted factors in the applied Run-to-Run model. This is often in addition to filtering applied to the measurement itself. For instance, measurement filtering may include within sample variation checks. The Optimized calculation would then be based already on filtered metrology. If the metrology filter is too aggressive or incorrect for a given sample, the Optimized setting filter can protect the controller update to be impacted by the influence of this sample.

The application of SPC on the Optimized value provides the process engineer insight to samples that may have a valid assignable cause of a statistically significant change in the optimal setting. The assignable causes for this may include a true measurement error, or even a sample dependent cause like a significant material stack change. Investigation of these statistical violations can lead to improved process models, or important actions to minimize similar upstream process excursions.

2. Recommended Setting – Moving Range

Like the application of SPC for the ‘Optimized’ setting, the Recommended setting are expected to have non-Gaussian, systematic drifts. Unlike Optimized setting SPC, Recommended setting SPC allows notification prior to process start. This allows confirmation by operations to avoid any potential process excursion. Engineering review of the updated recommended setting may result in a controller reset, historical data confirmation and re-running the process recommendation model.

SPC applied at this step is an important safeguard, even with automated safeguards like Recommended setting clipping. Clipping is a setting that allows process engineering to keep adjustment settings within a plausible range, e.g., Process Etch Time from 0.5min to 5 min are allowed. This is especially useful for new adopters of Run-to-Run to keep updates within a “plausible” range.

3. Measured Control Variable – Xbar, Range, Sigma, & MR

SPC of measured values applied correctly is a powerful tool to maximize feedback model accuracy and precision. The post compensated metrology signal is expected to be randomly distributed, with any systematic process influence removed.  In this situation, it is very important that review any results that are flagged as a statistically special cause event.

In most cases, the metrology signal is already filtered and as a result an SPC violation may not influence the controller update. Metrology should be carefully considered to maximize controller state updates without introducing model disturbance.

A comprehensive R2R process control solution will include a range of metrology filters as well as historical weighting functions. An example of an effective metrology filters includes within sample variation checks and metrology tool to tool matching algorithms. Feedback weighting function examples include Exponentially Weighted Moving Average (EWMA) or similar historical Percent Weighted Average (PWA) methods. With solid metrology filters in place, the historically weighted feedback will best follow the actual process drift that is compensated. Figure 7 illustrates two historical “optimized” value feedback update algorithms.