US20030236759A1 - Neural network for determining the endpoint in a process - Google Patents
Neural network for determining the endpoint in a process Download PDFInfo
- Publication number
- US20030236759A1 US20030236759A1 US10/176,065 US17606502A US2003236759A1 US 20030236759 A1 US20030236759 A1 US 20030236759A1 US 17606502 A US17606502 A US 17606502A US 2003236759 A1 US2003236759 A1 US 2003236759A1
- Authority
- US
- United States
- Prior art keywords
- curves
- neural network
- training
- group
- endpoint
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Abandoned
Links
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/0265—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion
- G05B13/027—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion using neural networks only
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0224—Process history based detection method, e.g. whereby history implies the availability of large amounts of data
- G05B23/024—Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks
-
- H—ELECTRICITY
- H01—ELECTRIC ELEMENTS
- H01L—SEMICONDUCTOR DEVICES NOT COVERED BY CLASS H10
- H01L22/00—Testing or measuring during manufacture or treatment; Reliability measurements, i.e. testing of parts without further processing to modify the parts as such; Structural arrangements therefor
- H01L22/20—Sequence of activities consisting of a plurality of measurements, corrections, marking or sorting steps
- H01L22/26—Acting in response to an ongoing measurement without interruption of processing, e.g. endpoint detection, in-situ thickness measurement
Landscapes
- Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Automation & Control Theory (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Software Systems (AREA)
- Medical Informatics (AREA)
- Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Manufacturing & Machinery (AREA)
- Computer Hardware Design (AREA)
- Microelectronics & Electronic Packaging (AREA)
- Power Engineering (AREA)
- Drying Of Semiconductors (AREA)
Abstract
There is provided a system and method for pattern recognition of an endpoint curve for a dry etch process. The system trains a neural network with a group of training curves corresponding to the dry etch process, wherein the training curves contain normal and abnormal features. The system receives an endpoint curve at the neural network representing a dry etch process and detects an abnormal feature in the endpoint curve.
Description
- 1. Field of the Invention
- This invention relates to endpoint detection in a semiconductor process utilizing a neural network.
- 2. Background of the Invention
- Today in semiconductor processing, different structures, such as trenches and vias in semiconductor devices are formed by etching processes. FIGS.1A-1C illustrate the steps in a known process for etching a semiconductor device. As shown in FIG. 1A, a
semiconductor device 100 comprises asilicon substrate 102, apolysilicon layer 104, and aninsulating layer 106. On insulatinglayer 106 is deposited aphotoresist layer 108.Photoresist layer 108 is patterned into a predetermined structure. The predetermined structure may be, for example, a via hole or trench.Photoresist layer 108 serves as a mask for the etching process. - Then, as shown in FIG. 1B,
photoresist layer 108 functions as a mask during the etch process. In the etching process,insulating layer 106 is etched untilpolysilicon layer 104 is reached. Oncepolysilicon layer 104 is reached, the etch process is terminated. As shown in FIG. 1C, after the etching process is completed,photoresist layer 108 is removed. The etching process used may be, for example, a wet etching. Alternately, a dry etching, such as ion bombardment or plasma etching, may be preformed. - In semiconductor device manufacturing processes, dry etching is one technique for forming micropatterns. Dry etching is a method of generating plasma in a vacuum using a reactive gas and removing an etching target by using ions, neutral radicals, atoms, and molecules in the plasma. Etching is performed on an etching target until an underlying layer is reached. If etching is continued after the etched target is completely removed, the underlying material may be excessively etched, or the etching shape may be undesirably changed. Therefore, to obtain a desired structure, it is advantageous to detect the endpoint in the etching process accurately.
- FIGS. 2A and 2B illustrate etching a semiconductor device by a dry etching process. As shown in FIG. 2A, a
semiconductor device 200 comprises asilicon substrate 202, aninsulating layer 204, and aphotoresist layer 206. Photoresist layer (PR) 206 is patterned into a predetermined structure, such as a trench or via hole. - As shown in FIG. 2B,
plasma gas reactant 208 is introduced tosemiconductor device 200. Plasma reactant 208 reacts withlayer 204, thereby etchinglayer 204. Asplasma reactant 208 reacts withlayer 204, a by-product is produced, the significance of which will be described hereinafter. Oncelayer 204 is etched so thatsubstrate 202 is exposed,plasma reactant 208 is removed, thereby ending the etching process. - The point at which the etching process is stopped is called the endpoint. There are several methods in determining the endpoint in an etching process. Optical emission spectroscopy is one method for endpoint detection. Optical emission spectroscopy is easy to implement and offers high sensitivity. Optical emission spectroscopy relies on the change in the emission intensity of characteristic optical radiation from either a reactant or by-product in a plasma. Light is emitted by excited atoms or molecules when electrons relax from a higher energy state to a lower one. Atoms and molecules emit a series of spectral lines that are unique to each individual reactant or by-product. FIG. 3 is a graph illustrating the emission spectrum of exemplary by-products in a conventional etching process. As seen in FIG. 3, each different reaction product emits radiation at a different wavelength. The emission intensity is a function of the relative concentration of a species in the plasma. By observing the intensity of a particular by-product, the endpoint of the etch process may be determined.
- A known
system 400 utilized for endpoint detection is shown in FIG. 4.System 400 comprises areaction chamber 410 which contains asemiconductor device 412 to be processed and awindow 414.System 400 operates by recording with arecorder 440, such as a meter or oscilloscope, the emission spectrum, otherwise known as an endpoint curve, during the etch process in the presence and absence of the material that is to be etched. Emission radiation from the reaction passes throughwindow 414 through alens 420 and is received by adetector 430.Detector 430 is equipped with a filter that lets light of specific wavelengths pass through for detection. To detect the endpoint, the emission intensity of a particular reactant or by-product is monitored at a fixed wavelength and recorded on arecorder 440. When the end point is reached, the emission intensity changes. The change in emission intensity at the endpoint depends on individual reactant or by-product being monitored. The intensity due to reactant increases, while the intensity due to etch by-products decreases as the endpoint is reached. - FIG. 5 is a graph illustrating the emission spectrum intensity for a by-product in an etch process. As shown in the graph, as the etch process begins (502), the intensity of the emissions for the by-product from the reaction process increase as a function of the elapsed time (504). Then, as the etch process nears the endpoint (506), the emission intensity for the by-product decreases. Thus, the endpoint (506) can be determined by monitoring the emission of a reaction product for a decrease in intensity. FIG. 6 is a graph illustrating an experimental endpoint curve for a exemplary by-product. The endpoint is determined from observing the curve to locate normal features that denote an endpoint. Normal features are features which regularly occur in an endpoint curve.
- Optical emissions spectroscopy also has some drawbacks.
- Oftentimes, it is difficult to determine the exact endpoint of an etching process. Such is the case when an endpoint curve contains abnormal features that may be hard to detect by conventional optical spectroscopy. Many times an over-etching is required. However, in the case of overetching, the underlying materials may be damaged.
- In certain aspects consistent with the present invention, there is provided a system and method for determining an endpoint of an etching process by utilizing a neural network. By learning the features of a group of endpoint curves containing normal and abnormal features for an etch process, the neural network may determine the endpoint of the etch process through pattern recognition.
- In one aspect consistent with the present invention, there is provided a system and method for pattern recognition of an endpoint curve for a dry etch process. The system trains a neural network with a group of training curves corresponding to the dry etch process, wherein the training curves contain normal and abnormal features. The system receives an endpoint curve at the neural network representing a dry etch process and detects an abnormal feature in the endpoint curve.
- Additional features and advantages consistent with the present invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The advantages consistent with the present invention will be realized and attained by means of the elements and combinations particularly pointed out in the appended claims.
- It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed.
- The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate certain aspects consistent with the present invention and, together with the description, serve to explain the principles of the invention.
- FIGS.1A-1C are diagrams illustrating known etching processes for a semiconductor device;
- FIGS.2A-2B are diagrams illustrating known dry etching processes for a semiconductor device;
- FIG. 3 is a graph illustrating the wavelengths of emissions for exemplary by-products for a dry etch process;
- FIG. 4 is a diagram illustrating a known system for detecting the endpoint of a dry etch process;
- FIG. 5 is a graph illustrating the intensity of a particular exemplary by-product in a dry etch process;
- FIG. 6 is a graph illustrating an exemplary recorded endpoint curve;
- FIG. 7 is a diagram illustrating a node in a neural network;
- FIG. 8 is a diagram illustrating a neural network having two nodes;
- FIG. 9 is a neural network implementing a back propagation learning rule for endpoint detection in an etch process according to an aspect consistent with the present invention;
- FIG. 10 is a method of training a neural network according to an aspect consistent with the present invention;
- FIGS. 11 and 12 are graphs illustrating endpoint curves with stable and unstable front features, respectively;
- FIG. 13 is a flowchart illustrating a method for training a neural network;
- FIGS.14A-14C are graphs illustrating exemplary recorded endpoint curves with normal features;
- FIGS.15A-15C are graphs illustrating exemplary recorded endpoint curves with abnormal features;
- FIGS. 16A and 16B are graphs illustrating exemplary experimental endpoint curves;
- FIG. 17 is a diagram illustrating a system for endpoint detection for an etch process according to an aspect consistent with the present invention; and
- FIG. 18 is a flow diagram showing the steps in a method for endpoint detection according to the present invention.
- Reference will now be made in detail to aspects and features consistent with the present invention, examples of which are illustrated in the accompanying drawings. Wherever possible, the same reference numbers will be used throughout the drawings to refer to the same or like parts.
- Certain aspects of the present invention overcome disadvantages of known endpoint detection systems and methods by utilizing a neural network. By learning the features of a group of endpoint curves containing normal and abnormal features for an etch process, the neural network may determine the endpoint of the etch process through pattern recognition.
- A neural network is an interconnected assembly of simple processing elements, called nodes, whose functionality is loosely based on the human brain, in particular, the neuron. The processing ability of the network is stored in inter-node connection strengths, called weights, obtained by learning from a set of training patterns. Learning in the network is achieved by adjusting the weights based on a learning rule and training patterns to cause the overall network to output desired results.
- The basic unit of a neural network is a node. FIG. 7 is an example of a
neural network node 700.Neural network node 700 functions by receiving an input vector X composed of elements x1, x2, . . . xn. Input vector X is multiplied by a weight vector W composed of elements w1, w2, . . . wn. That is, each vector element is multiplied by a corresponding weight vector element. The resultant product is inputted into a linear threshold gate (LTG) 710.LTG 710 sums the product of X and W. The sum is operated on by a transfer function f(s). An output value y is output fromLTG 710 after the sum is operated on by the transfer function f(s). For example, the transfer function f(s) may be a step function. In this case, if the sum is greater than the threshold of the step function, abinary 1 is output as y. If the sum is less than the threshold of the step function, abinary 0 is output as y. - A neural network is composed of multiple nodes arranged in layers. FIG. 8 is a diagram of a
neural network 800.Neural network 800 is composed of two layers, afirst layer 810 and asecond layer 820.First layer 810 comprises k number of nodes.Second layer 820 comprises a single node. The nodes infirst layer 810 andsecond layer 820 may be, for example, an LTG, such asLTG 710 illustrated in FIG. 7.Neural network 800 functions by receiving an input vector X composed of elements x1, x2, and xn atfirst layer 810.First layer 810 processes input vector X and outputs the results tosecond layer 820. The nodes offirst layer 810 may process input vector X by the method described for the node shown in FIG. 7. The results outputted fromfirst layer 810 are inputted intosecond layer 820.Second layer 820 processes the results and outputs a result y. The node ofsecond layer 820 may process the result fromfirst layer 810 using the method described for the node shown in FIG. 7. - One advantage of neural networks is the ability to learn. The learning process of neural networks is facilitated by updating the weights associated with each node of a layer. The weights for each node in a layer may be updated based on a particular learning rule. One type of learning rule is a back-propagation (BP) learning rule. The BP learning rule dictates that changes in the weights associated with the learning process are based on errors between actual outputs and desired outputs. In the network, an error associated with the output of a node is determined and then back propagated to the weights of that node. A number of training sets are inputted into the network in order to train the network. The neural network processes each training set and updates the weights of the nodes based on the error between the output generated by the neural network and the desired output. BP learning rule neural networks may be used for pattern matching and classification.
- FIG. 9 is an example of a
neural network 900 implementing a BP learning rule according to certain aspects related to the present invention.Network 900 serves as a pattern classification network to recognize normal and abnormal features in etch endpoint curves.Network 900 is trained by submitting a series of sample endpoint curves which contain characteristic features. -
Neural network 900 compriseslayers Layer 910 is an input layer and contains an I number of nodes.Layer 910 receives vector X[x1 . . . xi,xn].Layer 920 is a hidden layer and contains a K number of nodes.Layer 920 is connected to inputlayer 910.Layer 930 is an output layer and contains a J number of nodes.Layer 930 is connected to layer 920 and serves to output the results of the neural network processing. The nodes of both layers may be LTG, as illustrated in FIG. 7. The number of nodes in each layer will be determined by the number of size of the input vector, the number of training endpoint curves, and the number of possible output. - In order to train
network 900 according to the BP learning rule, a series of training endpoint curves with characteristic features are fed tonetwork 900.Network 900 processes the endpoint curves, in order to learn, by updating the weights. The learning process is described with reference to FIGS. 9 and 10 and equations 1-5. FIG. 10 is a diagram illustrating alearning process 1000 among the ith node oflayer 910, the kth node oflayer 920, and the jth node oflayer 930. A transfer function for the nodes inlayers - where α is a gain factor and s is a sum of the weighted inputs. One skilled in the art would realize that the transfer function is not limited to a sigmoid function, but may be another transfer function, for example, hyperbolic tangent given by ƒ(s)=tan h(βs), where β is a gain factor and s is a sum of the weighted inputs.
-
- where Yj is the output for the jth node of
layer 930 and Tj is the desired output for the jth node oflayer 930. -
-
-
-
-
-
- where Tk is an “estimated target” for the kth node of
layer 920 and ρh is a learning rate. One skilled in the art would realized that the above equation may be adapted for other transfer functions, f(s). -
Network 900 is trained by inputting sample endpoint curves and training the network to recognize the endpoint curves by the characteristic features. FIGS. 11 and 12 are graphs illustrating sample endpoint curves with different characteristic features. FIG. 11 illustrates anendpoint curve 1100 which has astable front feature 1102. A stable front feature is a feature in the endpoint curve in which the differential point or endpoint of the etch may be easily determined. In the case of stablefront feature 1102,feature 1102 is defined by a leveling of the intensity. FIG. 12 illustrates anendpoint curve 1200 which has anunstable front feature 1202. An unstable front feature is a feature in the endpoint curve in which the differential point or endpoint of the etch may not be easily determined. In the case ofunstable front feature 1202,feature 1202 is defined only by a slight leveling of the intensity. -
Network 900 is trained by setting the desired output Tj to classify the desired features and then inputtingendpoint curves 1000 as vector X repeatedly until the output of the network approaches the desired output. FIG. 13 is a flow chart illustrating the method fortraining network 900. - The method begins by setting up
network 900 to use a back propagation learning rule such as the rule illustrated in FIG. 10 (stage 1300). -
- where Vm are the transformed values for the curves, Vin are the collect values for the curves undergoing transformation, Vmax is the select maximum value for the curves, and Vmin is the selected minimum value for the curves. In the case of FIGS. 11 and 12, the curves have been normalized such that the curves range between 0 and 1.
- Classification involves selecting the desired output Tj for each curve in the group of curves. In the case of FIGS. 11 and 12, the curves may be classified as stable front and unstable front and a desired output Tj selected for each. One skilled in the art would realize that the group of curves may contain additional curves other than those illustrated in FIGS. 11 and 12.
- Next, the group curves that have been transformed for classification are divided into a first set of curves and a second set of curves (stage1330). The first set of curves is used to train network 900 (stage 1340) by inputting the curves as vector X into
network 900. Training ofnetwork 900 continues until a desired learning is achieved. Innetwork 900, training stops when the root mean square errors between desired output Tj and outputs Yj are reduced to a predetermined error. The predetermined error is selected based on required accuracy of network. For high accuracy, the predetermined error is set to a small value. - When training comes to an end, weights of
network 900 will be set, through learning, to values that will allow classification of curves.Network 900 may be tested by inputting the first set of curves which is used for training network 900 (1350). Since the first set of curves was used to trainnetwork 900,network 900 should be able to classify the curves of the first set with 100% accuracy. If the accuracy is not 100%, an error in the classification of the learning samples is present. For example, identical types of curves may be given different classifications and different desired outputs Tj, thus producing inaccuracy. - After the first set of curves is verified with
network 900, the second set of curves are inputted in thenetwork 900 for classification and detection of the endpoints (stage 1360). If the recognition percentage is high for the second set,network 900 weights are reliable. Thus,network 900 is set to process curves in an etching process for endpoint detection. - If there is a need to add a new classification to network900, a new group of training curves may be collected representing the new classification (stage 1370).
Network 900 is then retrained using the new group of curves and the first group of curves. - Once
network 900 has been trained (stage 1380) with the endpoint curves, an experimental endpoint curve may be input as vector X intonetwork 900. Becausenetwork 900 has been trained,network 900 can recognize the endpoint curves from the characteristic features. One skilled in the art would realize thatnetwork 900 may be input with parts of the endpoint curve instead of the entire curve. Thus, an experimental endpoint may be processed in part and in real-time. For example, a selected portion representing a portion of the endpoint curve may be input as vector X. Thus, as the endpoint curve is received in real time, the portions which have been received may be input intonetwork 900 to determine if a characteristic feature is present. - FIGS.14A-14C are graphs illustrating exemplary training endpoint curves with normal features used to train a neural network, such as
network 900. FIGS. 15A-15C are graphs illustrating training endpoint curves withabnormal features network 900. Abnormal features are features in an endpoint curve which may not regularly occur in an etch endpoint curve and which may be difficult to identify in known etch systems. As in the case of the endpoint curves illustrated in FIGS. 11 and 12,network 900 is trained by setting the desired output Tj to classify the desired features and then usingendpoint curves network 900 according to the method illustrated in FIG. 13. The method is also repeated forendpoint curves network 900 has been trained with the endpoint curves, an experimental endpoint curve may be input as vector X intonetwork 900.Network 900 would receive an experimental endpoint curve. The experimental endpoint curve is processed by the layers ofnetwork 900. Becausenetwork 900 has been trained with exemplary training endpoint curves with normal features and abnormal features,network 900 can recognize the endpoint curves from the characteristic features. - FIGS. 16A and 16B are exemplary experimental curves for endpoint determination after a neural network, such as
network 900, has been trained. FIGS. 16A and 16B illustrate curves in which changes in time and output voltages of the etch endpoint curves are not identical or proportional. For example, in FIG. 16A, assuming that the etch rate is the same forcurves curve 1600 the etching endpoint occurs at time 91 (x axis), while incurve 1610 the etching endpoint occurs at time 112 (x axis). In FIG. 16B, for each of thecurves - FIGS. 16A and 16B show that the changes in time (axis x) and output voltages (axis y) of the etch endpoint curves often do not occur at the same time. Therefore, a point-to-point matching method (i.e. visual inspection of the curve) may not be reliable in matching the curves. From a visual inspection, the profile of the curves may be classified at best as similar. However, for each of the curves, the voltages of the corresponding sites are not identical. Thus, utilizing a back propagation neural network to recognize patterns is the best method.
- For such a classification, each of the
experimental curves network 900.Network 900 will determine match features of the experimental curves with features contained in the example training endpoint curves used in the training of thenetwork 900. Once a feature has been matched bynetwork 900, the matched pattern is output. One skilled in the art would realize thatnetwork 900 may be input with parts ofendpoint curves - FIG. 17 is a diagram of a
system 1700 for performing endpoint detection according to an aspect consistent with the present invention.System 1700 comprises areaction chamber 1710 which contains asemiconductor device 1712 to be processed and awindow 1714.System 1700 operates by recording with arecorder 1740, for example, a meter, oscilloscope, or chart recorder, the emission spectrum during the etch process in the presence and absence of the material that is to be etched. Emission radiation from the reaction passes throughwindow 1714 andlens 1720 and is received by adetector 1730.Detector 1730 is equipped with a filter that lets the light of specific wavelengths pass through to be detected. For example, if SiF is to be detected in the etching of a silicon or polysilicon layer, the filter would allow light of 777 nm to pass. To detect the endpoint, the emission intensity of a particular reaction product is monitored at a fixed wavelength and recorded on arecorder 1740. An analog to digital converter (ADC) 1750 is connected torecorder 1740.ADC 1650 converts the analog signal ofrecorder 1740 to a digital signal and the digital signal is transferred to acomputer unit 1760. -
Computer unit 1760 may contain standard components for inputting, outputting, processing, manipulating, and storing data. For example, the computer unit may comprise a central processing unit (CPU), random access memory (RAM), video card, sound card, magnetic storage devices, optical storage devices, input/output (I/O) terminals, monitor, and a network interface card (NIC) (all not shown).Computer unit 1760 may optionally be connected to a printer (not shown) through the I/O terminals. Examples of the I/O terminals to which the printer may be connected are parallel, serial, universal serial bus, and IEEE 1394. -
ADC 1750 may also be connected to the I/O terminals. Alternately,ADC 1750 may be a component incomputer unit 1760. For example,ADC 1750 may be located on an expansion card which is internally contained incomputer unit 1760. - Also,
computer unit 1760 may be connected to the other computing units through a network (not shown). For example, the network may be a local area network (LAN) or wide area network (WAN), such as the Internet, or wireless network. -
Computer unit 1760 is configured to perform endpoint determination using a neural network, such as the one illustrated in FIGS. 9 and 10. The neural network may be configured in the form of a process created using various programming languages or software suites. For example,neural network 900 determination may be a program written in the Java™ programming language, hypertext markup language, or C programming language. In this example,neural network 900 would be configured as a set of instructions stored on the storage unit of computer unit 1660. The CPU unit ofcomputer unit 1760 would utilize the instruction forneural network 900 stored in the storage unit. Also, during the endpoint determination, the endpoint curve may be displayed on the monitor or printed on the printer ofcomputer unit 1760. - One skilled in the art would realize that
neural network 900 may be implemented in different forms oncomputer unit 1760. For example,neural network 900 may be implemented in hardware incomputer unit 1760. In such an example,network 900 would be configured as a group of logic and processor units contained internally incomputer unit 1760 or externally located and connected tocomputer unit 1760 via the I/O terminals. - FIG. 18 is a flowchart illustrating a method for detecting an endpoint of a
curve utilizing system 1700 illustrated in FIG. 17 according to an aspect consistent with the present invention. - The method begins by setting up a neural network at computer unit1760 (stage 1800). The neural network is set up by creating the design of the network and determining a learning rule for the network. For example, the neural network can be designed, such as
network 900 illustrated in FIG. 9, using the learning rule illustrated in FIG. 10. - Next, a group of endpoint training curves are collected (stage1810). These curves may be collected, for example, by
detector 1730. These curves may also be prerecorded and stored atcomputer unit 1760 or generated bycomputer unit 1760. - Then, the group of endpoint training curves are converted from analog data to digital data (stage1820). This may be achieved by, for example,
ADC 1750. Once the data is converted, the endpoint training curves are used to train the neural network (stage 1830). Training is achieved by using a using a network training method, such as the one illustrated in FIG. 13. - Next, a group of experimental curves are collected (stage1840). These curves may be collected, for example, by
detector 1730. Then, the group of endpoint training curves are converted from analog data to digital data (stage 1850). This may be achieved by, for example,ADC 1750. - Once the data is converted, the experimental curves are input into the trained neural network (stage1860). Based on the training from
stage 1830, the neural network determines the endpoint of the experimental curves (stage 1870). Since the network has been trained to recognize different features for multiple curves, the endpoint may be determined with high accuracy. - Other aspects consistent with the present invention will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. For example, other neural network learning rules which operate as pattern classifiers may be used instead of the BP learning rule. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the invention being indicated by the following claims.
Claims (27)
1. A method for pattern recognition of an endpoint curve for a process comprising:
training a neural network with a group of training curves corresponding to the process, wherein the training curves contain normal and abnormal features;
receiving an endpoint curve at the neural network representing a process; and
detecting an abnormal feature in the endpoint curve.
2. The method as set forth in claim 1 , wherein the neural network uses a back-propagation learning rule.
3. The method as set forth in claim 1 , further comprising collecting the group of training curves.
4. The method as set forth in claim 1 , further comprising identifying a normal feature from the group of training curves.
5. The method as set forth in claim 1 , further comprising collecting an analog signal of the endpoint curve.
6. The method as set forth in claim 5 , further comprising transforming the analog signal to a digital signal.
7. A system for pattern recognition of an endpoint curve for a process comprising:
means for training a neural network to learn a feature corresponding to the process;
means for receiving the endpoint curve; and
means for detecting an abnormal feature in the endpoint curve.
8. The system as set forth in claim 7 , further comprising a group of training curves corresponding to the process.
9. The system as set forth in claim 8 , further comprising means for collecting the group of training curves.
10. The system as set forth in claim 7 , further comprising means for collecting an analog signal of the endpoint curve.
11. The system as set forth in claim 10 , further comprising means for transforming the analog signal to a digital signal.
12. The system as set forth in claim 7 , further comprising means for comparing the endpoint curve with the feature.
13. The system as set forth in claim 7 , wherein the means for training the neural network uses a back-propagation learning rule.
14. A computer readable medium including instructions for performing a method, when executed by a processor, for pattern recognition of an endpoint curve for a process comprising:
training a neural network with a group of training curves corresponding to the process, wherein the training curves contain normal and abnormal features;
receiving an endpoint curve at the neural network representing a process; and
detecting an abnormal feature in the endpoint curve.
15. The computer readable medium as set forth in claim 14 , wherein the neural network uses a back-propagation learning rule.
16. The computer readable medium as set forth in claim 14 , further comprising collecting the group of training curves.
17. The computer readable medium as set forth in claim 14 , further comprising identifying a normal feature from the group of training curves.
18. The computer readable medium as set forth in claim 14 , further comprising collecting an analog signal of the endpoint curve.
19. The computer readable medium as set forth in claim 18 , further comprising transforming the analog signal to a digital signal.
20. A method for pattern recognition of an endpoint curve for a process comprising:
setting up a neural network;
collecting a group of analog training curves;
converting the group of analog training curves to a group of digital training curves;
training the neural network with the group of digital training curves;
collecting a group of analog experimental curves;
converting the group of analog training curves to a group of digital training curves;
inputting the group of digital experimental curves into the neural network; and
determining by the neural network the endpoint of the group of digital training curves.
21. The method as set forth in claim 20 , wherein the group of analog training curves includes at least one of a normal or abnormal curve.
22. The method as set forth in claim 20 , wherein the group of analog experimental training curves includes at least one of a normal or abnormal curve.
23. The method as set forth in claim 20 , wherein the neural network uses a back-propagation learning rule.
24. The method as set forth in claim 20 , wherein training the neural network comprises:
classifying the group of digital training curves;
dividing the group of digital training curves into a first and second set of digital training curves;
training the neural network with the first set of digital training curves;
testing the neural network;
inputting the second set of digital training curves into the neural network; and
determining whether the neural network is trained.
25. The method as set forth in claim 24 , wherein training the neural network further comprises:
adding a new classification to the group of digital training curves.
26. A system for pattern recognition of an endpoint curve for a process comprising:
a process chamber for etching a semiconductor substrate;
a detector for detecting radiation from the process chamber during the etching of the semiconductor substrate and producing a radiation signal;
a recorder coupled to the detector for recording the radiation signal;
an analog to digital converter coupled to the recorder for converting the radiation signal; and
a pattern recognition apparatus coupled to the analog to digital converter for detecting the endpoint curve of the converted radiation signal.
27. The system as set forth in claim 26 , wherein the pattern recognition apparatus comprises a neural network using a back-propagation learning rule.
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US10/176,065 US20030236759A1 (en) | 2002-06-21 | 2002-06-21 | Neural network for determining the endpoint in a process |
TW092116734A TWI243324B (en) | 2002-06-21 | 2003-06-20 | Method and system for pattern recognition of an endpoint curve for a process and computer readable medium including instructions for performing the method |
CN03137451.4A CN1275198C (en) | 2002-06-21 | 2003-06-20 | Graphic identification method and system for process endpoint curve and computer readable media having instructions executing the same method |
US11/646,205 US7403930B2 (en) | 2002-06-21 | 2006-12-27 | Neural network for determining the endpoint in a process |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US10/176,065 US20030236759A1 (en) | 2002-06-21 | 2002-06-21 | Neural network for determining the endpoint in a process |
Related Child Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US11/646,205 Continuation US7403930B2 (en) | 2002-06-21 | 2006-12-27 | Neural network for determining the endpoint in a process |
Publications (1)
Publication Number | Publication Date |
---|---|
US20030236759A1 true US20030236759A1 (en) | 2003-12-25 |
Family
ID=29734051
Family Applications (2)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US10/176,065 Abandoned US20030236759A1 (en) | 2002-06-21 | 2002-06-21 | Neural network for determining the endpoint in a process |
US11/646,205 Expired - Lifetime US7403930B2 (en) | 2002-06-21 | 2006-12-27 | Neural network for determining the endpoint in a process |
Family Applications After (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US11/646,205 Expired - Lifetime US7403930B2 (en) | 2002-06-21 | 2006-12-27 | Neural network for determining the endpoint in a process |
Country Status (3)
Country | Link |
---|---|
US (2) | US20030236759A1 (en) |
CN (1) | CN1275198C (en) |
TW (1) | TWI243324B (en) |
Families Citing this family (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100138575A1 (en) * | 2008-12-01 | 2010-06-03 | Micron Technology, Inc. | Devices, systems, and methods to synchronize simultaneous dma parallel processing of a single data stream by multiple devices |
KR101069602B1 (en) | 2009-09-29 | 2011-10-05 | 한국과학기술원 | Computer system combining neuro-fuzzy system and parallel processor and objects recognizing system using thereof |
US9287097B2 (en) * | 2011-11-30 | 2016-03-15 | Sony Corporation | Predicting ultraviolet ray damage with visible wavelength spectroscopy during a semiconductor manufacturing process |
US10453653B2 (en) * | 2016-09-02 | 2019-10-22 | Tokyo Electron Limited | Endpoint detection algorithm for atomic layer etching (ALE) |
TWI807987B (en) * | 2016-11-30 | 2023-07-01 | 美商應用材料股份有限公司 | Spectrographic monitoring using a neural network |
GB201620820D0 (en) * | 2016-12-07 | 2017-01-18 | Univ Oxford Innovation Ltd | Characterisation of dynamical statistical systems |
US11056098B1 (en) * | 2018-11-28 | 2021-07-06 | Amazon Technologies, Inc. | Silent phonemes for tracking end of speech |
CN110479636B (en) * | 2019-07-19 | 2021-11-09 | 深圳市微蓝智能科技有限公司 | Method and device for automatically sorting tobacco leaves based on neural network |
CN110323242A (en) * | 2019-08-15 | 2019-10-11 | 德淮半导体有限公司 | The forming method of imaging sensor |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US4317698A (en) * | 1980-11-13 | 1982-03-02 | Applied Process Technology, Inc. | End point detection in etching wafers and the like |
US5467883A (en) * | 1992-12-14 | 1995-11-21 | At&T Corp. | Active neural network control of wafer attributes in a plasma etch process |
US5653894A (en) * | 1992-12-14 | 1997-08-05 | Lucent Technologies Inc. | Active neural network determination of endpoint in a plasma etch process |
US6641746B2 (en) * | 2001-09-28 | 2003-11-04 | Agere Systems, Inc. | Control of semiconductor processing |
Family Cites Families (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
AU2485788A (en) * | 1987-07-28 | 1989-03-01 | Maxdem, Inc. | Electrically settable resistance device |
JPH03236275A (en) * | 1990-02-14 | 1991-10-22 | Olympus Optical Co Ltd | Memory element and memory device |
JP2715722B2 (en) * | 1991-08-23 | 1998-02-18 | 日本電気株式会社 | Semiconductor synapse circuit, method of manufacturing the same, semiconductor neuron device, and semiconductor-superconductor composite neuron device |
US5311360A (en) * | 1992-04-28 | 1994-05-10 | The Board Of Trustees Of The Leland Stanford, Junior University | Method and apparatus for modulating a light beam |
US5475341A (en) * | 1992-06-01 | 1995-12-12 | Yale University | Sub-nanoscale electronic systems and devices |
US6320200B1 (en) * | 1992-06-01 | 2001-11-20 | Yale University | Sub-nanoscale electronic devices and processes |
US5393994A (en) * | 1993-02-08 | 1995-02-28 | Matsushita Electric Industrial Co., Ltd. | Optical semiconductor device for neural network |
US5737496A (en) * | 1993-11-17 | 1998-04-07 | Lucent Technologies Inc. | Active neural network control of wafer attributes in a plasma etch process |
US5961923A (en) * | 1995-04-25 | 1999-10-05 | Irori | Matrices with memories and uses thereof |
US6340588B1 (en) * | 1995-04-25 | 2002-01-22 | Discovery Partners International, Inc. | Matrices with memories |
US6329139B1 (en) * | 1995-04-25 | 2001-12-11 | Discovery Partners International | Automated sorting system for matrices with memory |
US5859964A (en) * | 1996-10-25 | 1999-01-12 | Advanced Micro Devices, Inc. | System and method for performing real time data acquisition, process modeling and fault detection of wafer fabrication processes |
US6175831B1 (en) * | 1997-01-17 | 2001-01-16 | Six Degrees, Inc. | Method and apparatus for constructing a networking database and system |
US6750881B1 (en) * | 1997-02-24 | 2004-06-15 | America Online, Inc. | User definable on-line co-user lists |
US6269369B1 (en) * | 1997-11-02 | 2001-07-31 | Amazon.Com Holdings, Inc. | Networked personal contact manager |
NL1009376C1 (en) * | 1998-06-11 | 1998-07-06 | Boardwalk Ag | Data system for providing relationship patterns between people. |
WO2000016209A1 (en) * | 1998-09-15 | 2000-03-23 | Local2Me.Com, Inc. | Dynamic matchingtm of users for group communication |
US6889213B1 (en) * | 2000-02-18 | 2005-05-03 | Hewlett-Packard Development Company, L.P. | E-service to manage contact information with privacy levels |
US7047202B2 (en) * | 2000-07-13 | 2006-05-16 | Amit Jaipuria | Method and apparatus for optimizing networking potential using a secured system for an online community |
US7099862B2 (en) * | 2001-08-02 | 2006-08-29 | International Business Machines Corporation | Programmatic discovery of common contacts |
US6889216B2 (en) * | 2002-03-12 | 2005-05-03 | Knowm Tech, Llc | Physical neural network design incorporating nanotechnology |
US6844582B2 (en) * | 2002-05-10 | 2005-01-18 | Matsushita Electric Industrial Co., Ltd. | Semiconductor device and learning method thereof |
-
2002
- 2002-06-21 US US10/176,065 patent/US20030236759A1/en not_active Abandoned
-
2003
- 2003-06-20 CN CN03137451.4A patent/CN1275198C/en not_active Expired - Fee Related
- 2003-06-20 TW TW092116734A patent/TWI243324B/en not_active IP Right Cessation
-
2006
- 2006-12-27 US US11/646,205 patent/US7403930B2/en not_active Expired - Lifetime
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US4317698A (en) * | 1980-11-13 | 1982-03-02 | Applied Process Technology, Inc. | End point detection in etching wafers and the like |
US5467883A (en) * | 1992-12-14 | 1995-11-21 | At&T Corp. | Active neural network control of wafer attributes in a plasma etch process |
US5653894A (en) * | 1992-12-14 | 1997-08-05 | Lucent Technologies Inc. | Active neural network determination of endpoint in a plasma etch process |
US6641746B2 (en) * | 2001-09-28 | 2003-11-04 | Agere Systems, Inc. | Control of semiconductor processing |
Also Published As
Publication number | Publication date |
---|---|
CN1482578A (en) | 2004-03-17 |
US7403930B2 (en) | 2008-07-22 |
TWI243324B (en) | 2005-11-11 |
US20070214098A1 (en) | 2007-09-13 |
CN1275198C (en) | 2006-09-13 |
TW200400447A (en) | 2004-01-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US7403930B2 (en) | Neural network for determining the endpoint in a process | |
CN108351600B (en) | Analog image is generated from design information | |
EP3465174B1 (en) | Generating simulated images from input images for semiconductor applications | |
EP3465552B1 (en) | Systems and methods incorporating a neural network and a forward physical model for semiconductor applications | |
CN111837225B (en) | Defect detection, classification and process window control using scanning electron microscope metrology | |
JP3179997B2 (en) | Method and system for analyzing plasma data | |
US7523076B2 (en) | Selecting a profile model for use in optical metrology using a machine learning system | |
EP1891492B1 (en) | Process change detection through the use of evolutionary algorithms | |
CN100381799C (en) | Method and appts. for endpoint detection using partial least squares | |
US7069257B2 (en) | Pattern recognition method for reducing classification errors | |
KR20110084302A (en) | Etching apparatus, analysis apparatus, etching treatment method, and etching treatment program | |
CN112596016A (en) | Transformer fault diagnosis method based on integration of multiple one-dimensional convolutional neural networks | |
CN114648528A (en) | Semiconductor detection method and device and computer readable storage medium | |
KR101117928B1 (en) | Plasma process diagnosis system, method and apparatus of detecting an end point in the same | |
CN113610747A (en) | Automatically selecting an algorithm module to examine a sample | |
US6675154B2 (en) | Method and system for the quantum mechanical representation and processing of fuzzy information | |
KR20220001603A (en) | Plasma Etching Process Monitoring Device Using Machine Learning Model For Multi-channel Optical Emission Spectroscopy Analysis And Plasma Etching Process Monotoring Method Using The Same | |
US10734207B2 (en) | Plasma processing apparatus and analysis method for analyzing plasma processing data | |
KR101895707B1 (en) | Method of determining an etching end point of plasma process | |
JP3117355B2 (en) | End point detection method for plasma processing | |
US20220066410A1 (en) | Sequenced Approach For Determining Wafer Path Quality | |
Hong et al. | Neural network modeling of reactive ion etching using principal component analysis of optical emission spectroscopy data | |
Hong et al. | Neural-network-based time series modeling of optical emission spectroscopy data for fault detection in reactive ion etching | |
KR20200088012A (en) | Method and appratus for predicting fault pattern using multi-classifier based on feature selection method in semiconductor manufacturing process | |
US20230401694A1 (en) | Active learning-based defect location identification |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
AS | Assignment |
Owner name: MACRONIX INTERNATIONAL CO., LTD., TAIWAN Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:HO, TSUNG-HSUAN;REEL/FRAME:013276/0051 Effective date: 20020827 |
|
STCB | Information on status: application discontinuation |
Free format text: ABANDONED -- FAILURE TO RESPOND TO AN OFFICE ACTION |