 |
Triant Technologies Inc. sales@triant.com 1-604-697-5090 |
Universal Process Modeling
UPM Fundamentals
Outputs
Inputs
Benefits of UPM
Universal Process Modeling (UPM) is a unique modeling algorithm originally developed in 1990 by Triant's Chief Scientist, Dr. Jack Mott. As a modeling technology, UPM has the ability to predict how a complex system should behave, and compare that prediction to how the system is actually behaving. Unlike many other modeling techniques, UPM is very robust mathematically, and requires much less effort in the actual building of a model. In fact, by using ModelWare®'s intelligent software agent, a UPM model can be created automatically for all recipes running on a particular piece of semiconductor processing equipment. UPM, with its attributes and advantages, is one of the key factors that influenced Motorola to establish a Joint Development Agreement (JDA) with Triant in 1994, and Applied Materials to establish a strategic alliance in 1999.
UPM Fundamentals - Top of Page
When modeling a system, 2 main alternatives exist:
UPM is based on empirical modeling, as the physics and chemistry of most semiconductor processing equipment is far too complex to model using first principles.
From a practical point of view, UPM can be described as a "black box" with a number of inputs and outputs, as shown in the diagram below.
A simplistic view is that sensor data is collected in real-time (up to 64 sensors sampled at, say, once per second) and fed to UPM in real-time. Using other pre-defined inputs (discussed below), UPM proceeds to calculate predicted sensor data and validation data, also in real-time. The results are available before the next sample enters UPM.
This means that each time the sensors are sampled, UPM calculates a prediction for each sensor, and provides a statistical measure (in standard deviations) of the error in the sensor data as well as in the overall equipment operation. Alarms may be triggered if the deviation is significant enough.
Outputs - Top of Page
Here is a brief explanation of UPM outputs:
| Predicted Data: |
Calculation of what the sensor value should be.
A prediction can be calculated for each sensor of the input data. The prediction data is in the same units as the sensor data. | Validation data: |
Statistical measures of the differential between actual sensor data and predicted data. In that sense, the validation data are a measure of "health". These measurements are normalized in units of standard deviations. This way, the "health" of each individual sensor can be measured (Sensor Health), as well as the "health" of the overall system (System Health). Alarms can be triggered when a "health" measure goes beyond a pre-defined number of standard deviations. |
Inputs - Top of Page
As mentioned previously, 2 important inputs to UPM need to be defined before any modeling occurs, namely the Reference Data and the Alarm & Configuration Information.
| Reference Data: |
A library of several hundreds or even thousands of examples of "good" sensor data. As a group, these examples describe the various "good" modes of operation of the system. When calculating a prediction, UPM chooses a subset of "good" data samples from the Reference Data. | Alarm & Configuration Information: | A set of files and a Graphical User Interface used to define an alarm strategy. An advanced user can fine-tune a number of modeling parameters by using the alarm and configuration settings. |
The process of defining those 2 inputs is often referred to as building a model. A model may be built manually by an engineer (a few hours), or automatically generated with only some fine-tuning required.
Benefits of UPM - Top of Page
The key benefit in using UPM is in its ability to monitor complex signals that are not easily predicted. For example, in a simplified case of a pump establishing vacuum conditions in a chamber, one may easily monitor the chamber pressure without requiring a tool like UPM. However, should a minute leak appear, the pump will compensate for the leak, maintain vacuum condition, and there will not be any way of detecting the appearance of the leak until it becomes catastrophic. However, if UPM is used to monitor the chamber pressure and valve angle, along with other system parameters, UPM could detect an unusual valve position resulting from the minute leak, despite the oscillations that may be normal in the valve angle signal. In a real-case example, up to 40-50 signals may need to be monitored simultaneously in order to extract all the necessary information. No other technology has been able to perform such complex multivariate analysis in a production environment, and with so little input required from an end user.
A second key advantage of UPM is that it has been shown to work with "real-life" data, in a production environment. In such an environment, the data sampled from the various sensors may be defective (faulty sensor, noisy environment, etc.) or even missing altogether. Such a harsh environment requires a fault tolerant algorithm like UPM in order to produce meaningful results.
Finally, being example-based, UPM is relatively easy to set up and use, especially given Triant's introduction of the first intelligent software agent to automatically create or even update UPM models.
|
© 1996-2005 Triant Technologies Inc. All Rights Reserved. |
 |