SG176567A1 - Arrangement for identifying uncontrolled events at the process module level and methods thereof - Google Patents

Abstract

A method for detecting an in-situ fast transient event within a processing chamber during substrate processing is provided. The method includes a set of sensors comparing a data set to a set of criteria (in-situ fast transient events) to determine if the first data set includes a potential in-situ fast transient event. If the first data set includes the potential in-situ fast transient event, the method also includes saving an electrical signature that occurs in a time period during which the potential in-situ fast transient event occurs. The method further includes comparing the electrical signature against a set of stored arc signatures. If a match is determined, the method yet also includes classifying the electrical signature as a first in-situ fast transient event and determining a severity level for the first in-situ fast transient event based on a predefined set of threshold ranges.

Description

ARRANGEMENT FOR IDENTIFYING UNCONTROLLED EVENTS AT THE

PROCESS MODULE LEVEL AND METHODS THEREOF

BACKGROUND OF THE INVENTION

[0001] Advances in plasma processing have provided for growth in the semiconductor industry, To be competitive, a manufacturing company needs (0 be able to process the substrates into quality semiconductor devices. Tight control of the process parameters is generally needed to achieve satisfactory resets during substrate processing. When the processing parameters (e.g, RF power, pressure, bias voltage, ion flux, plasma density, and the likes} fall outside of a pre-defined window, undesirable processing results {e.g , poor etch profile, low selectivity, damage to the substrate, damage to the processing chamber, and the hikes) may resadt. Accordingly, the ability to identify conditions when the processing parameters are outside the pre-defined windows is important im the manefacture of semiconductor devices.

[8002] During substrate processing, certain uncontrolled events may happen that may damage the substrate andfor canse damage to the processing chamber components. To identify the uncontrolled events, data may be collected during substrate processing.

Monitoring devices, such as sensors, nay be emploved to collect data about the various process parameters {such as bias voltage, reflected power, pressure, and the likes) during substrate processing. As discussed herein, sensor refers to a device that may be enploved to detect conditions andior signals of a plasma processing component, For case of discussion, the form “component” will be used to refer to an atomic or a multi-part assembly ina processing chamber.

[0003] The type and amount of data that are being collected by the sensors have increased in recent years. By analyzing the data collected by the sensors in relation to the process module data and the process context data (chamber event data), parameters that are outside of the pre- defined window may be identified. Accordingly, corrective actions {such as recipe adjustment) may be provided to stop the uncontrolled event(s), thereby preventing further damage from occurring to the substrate andior the processing chantbher components.

BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWINGS

[06164] The present invention is iHustrated by way of example, and not by way of limitation, in the figures of the accompanying drawings and in which like reference numerals refer to similar elements and in which:

[8003] Fig | shows a prior art overall logic view of an interconnecting tool environment with a host-lovel analysis server.

[6006] Fig. 2 shows a simple block dragrant of an interconnecting tool environment with a chuster tool level solution for correlating data between the sensors and the process model controllers.

[0007] Fig. 3 shows, in an embodiment of the invention, a simple logic overview of a process-level troubleshooting architecture.

[0008] Fig. 4 shows, in an embodiment of the vention, a simple functional diagram of a process mode level analysis server

[0009] Fig. 5 shows, a simple diagram of a pHcro-areing event.

[80168] Figs. 6A and 6B show, in embodiments of the invention, simple block diagrams of a

Processing environment.

[6011] Fig. 7 shows, in an embodiment of the invention, a simple Bowchart iHastrating a method for detecting a real-time fast transient event within a production environment in which the fast sampling transient detection algorithm is not part of an analyzing module.

DETAILED DESCRIPTION OF EMBODIMENTS

[0012] The present wvention will now be described mn detail with reference to a fow embodiments thereof as Hlustrated in the accompanying drawings. In the followmg description, mamerous specific details are set forth in order to provide a thorough understanding of the present invention. It will be apparent, however, to one skilled in the ast, that the present mvention may be practiced without some or all of these specific details. In other instances, well known process steps andfor structures have not been desertbed fn detail inn order to not unnecessarily obscore the present invention.

[6013] Various embodiments are described hereinbelow, including methods and techniques.

It should be kept mn nund that the invention might also cover articles of manufacture that cludes a computer readable medium on which computer-readable Instructions for carrying out entbodiments of the mventive technique are stored. The compater readable medium may include, for example, semiconductor, magnetic, opto-magnetic, optical, or other forms of computer readable medium for stoning computer readable code. Further, the mvention may also cover apparatuses for practicing embodiments of the brvention. Such apparatus may include cireuits, dedicated and/or progravomable, to carry out tasks pertaining to embodiments of the invention. Examples of such apparatus mclude a general-purpose computer andfor a dedicated computing device when appropriately programmed and may inchide a combination of a computer/computing device and dedicatedfprogrammable circuits adapted for the various tasks pertaining to embodiments of the mvention.

[0014] As aforementioned, to gain a competitive edge, manufacturers have to be able to effectively and efficiently woubleshoot problems that may arise daring substrate processing.

Troubleshooting generally involves analyzing the plethora of data collected during processing. To facilitate discussion, Fig. 1 shows a prior art overall logic view of an mterconnecting tool environment with a host-level analysis server.

[0015] Consider the situation wherein, for example, a manufacturing company may have one or more cluster tools (such as eich tools, cleaming tools, steip tools, and the hikes). Each cluster tool may have a plurality of processing modules, wherein cach processing module is configared for one or more specific processes. Each cluster tool may be controlled by a chuster tool controller (CTC), sach as CTC 104, CTC 106, and CTC 108. Each cluster tool controtler may nteract with one of more process module controller (PMC), such as PMCs

THO, 112, 114, and 116. For case of discussion, examples will be provided in relation to PMC

Pi

[6016] In order to identify conditions that may require intervention, sensors nay be crploved to collect data (sensed data) about processing parameters during substrate processing. In an example, during substrate processing a plurality of sensors {such as sensors 118, 120, 122, 124, 126, 128, 130, 132, 134, 136, 138, and 140) may interact with the process modile controllers to collect data about one or more processing parameters. The ype of sensors that nay be avadable may depend upon the type of data that may be collected. For example, sensor 118 may be configured to colfect voltage data. In another example, sensor 120 may be configured to collect pressure data. Generally, the sensors that may be employed to collect data from a process module may be of different brands, makes, and/or models. As a result, a sensor may have litle or no interaction with another sensor.

[0017] Usually, a sensor is configured to collect measurement data about one or more specific pacameters. Since most sensors are not configured to perform processing, cach sensor may be coupled fo a computing module (such as a computer, user interface, and the hikes}. The computing module is usually configured to process the analog data and to convert the raw analog data into a digital format.

[6018] in an example, sensor 11& collects voltage data from PMC 1H) via sensor cable 144.

The analog voltage data received by sensor 118 is processed by a computing module 18h

The data collected by the sensors are sent to a host-level analysis server (such as data box 142). Before sending the data onward to data box 142 over the network connection, the data is first converted from an analog format to a digital format by the computing models. nan example, computing module [ED converts the analog data collected by sensor 118 mito a digital format before sending the data over a network path 146 to data box 142,

[8019] Data box 142 may be a centralized analysis server that 1s configured to collect, process, and analyze data from a ploradity of sources, inclading the sensors and the process modufes, Usually, one data box may be available fo process the data collected during substrate processing by all of the cluster weols of a single manufacturing company.

[0620] The actual amount of data that may be transmitted to data box 142 may be sigmificantly less than the amount collected by the sensors, Usually, a sensor may collect a massive amount of data. In an example, a sensor may collect data at rates of up to megabyte per second. However, only a fraction of the data collected by the sensors is seat to data box 142. [$621] One reason for not transmitting the entire data streams collected by the seasors to data box 142 is due to the network bandwidth limitation when using cost-effective, commercially available communication protocols. The network pipeline to data box 142 may not be able to handle large volume of data from a plurality of sources {such as sensors HEE, 120, 122, 124, 1206, 128, 130, 132, 134, 136, 138, and 140) being sent to a single receiver {such as data box {42}. In other words, the network path between the sensor mvangements (sensor and computing module) and data box 142 may experience major traffic congestion as data box 142 tries to receive the massive amount of data coming fron all of the sensor arrangements.

As can be appreciated from the foregomng, if data box 142 15 unable to handle the mconing wraffic, the data packets being sent may be dropped and may have to be resent, thereby putting any additional burden onto the already heavily congested network pipeline.

[6022] In addition, data box 142 may not be able to handle a high volome of incoming data from multiple sources while at the same time performing other important fimetions, such as processing and analyzing data. As aforementioned, data box 142 is not only configured to receive the incoming data packets but data box 142 is also configired to process and analyze all of the incoming data streams, for example. Since data box 142 is the analysis server for the different data streams being collected, data box 142 needs sufficient processing capability to perform analysis on the plethora of data streams.

[6823] Since data box 142 has limited processing resource, only a fraction of the data collected front cach sensor is sont to data box 142, Tn an example, of the thousands of data ems that may be collected by a single sensor, only 104-15 data tems at 1-5 hertz may be forwarded to data box 142. In one example, only a summary of the data collected by sensor {18 may be sent to data box 142.

[6024] in addition to receiving data from the plurality of sensors, data box 142 may also be receiving data from the process module controlfers. In an example, process modele data and process context data {chamber-event data) may be collected by each process module controller and forwarded to data box 142. For ease of discussion, process module data and process context data may also be referred to as process module and chamber data. For example, process module data and process context data may be collected by PMC 110 and be sent to CTC 104 via a path HE. CTC 104 is not only managing the data from PMC 116 bat may also be handling the data from the other processing module controllers within the chuster tool (such as PMC 112, PMC 14, and PMC 110).

[8025] The data collected by the cluster tool controller is then transniitted to a fab host 102 via a semiconductor equipment conpnunication standard generic equipnaent module {SECS/GEM) interface. In an example, CTC 104 transmits data collected from PMCs 114, 112, 114, and/or 116 to fab host 102 through SECS/GEM 156 via a path 150. Fab host 12 may not only be receiving data from CTC 104, but also may be receiving data from other cluster tool controllers, such as CTCs 106 and 108, for example. The data collected by fab host 102 is then forwarded to data box 142 via a path 138. Due to the sheer volume of data bemy collected, not all data being sent to fab host 102 1s forwarded to data box 142. In many stances, only a semmary of the data may be transmitted to data box 142.

[0026] Data box 142 may process, analyze andfor correlate the data collected by the sensors and the process module controlers. If an anomaly is wdentified, data box 120 may then determine the source of the problem, such as a parameter that is not in conformance with a recipe step being performed in PMC THO, for example. Once the source of the problem has been identified, data box 142 may send an interdiction in the format of an Ethernet message to fab host 102. Upon receiving the message, fab host 102 may forward the message through

SECS/GEM 156 10 CTC 104. The chaster tool controller may then relay the message to the infended process module controller, which is PMC HO m thus example.

[0027] Unfortunately, the interdiction is usually sot provided in real-time. Instead, the interdiction is wsaally recetved by the intended process modale after the affected substrate has been processed of even after the entire substrate lot has exited the process module.

Accordingly, not only have the substrate/substrate tot been damaged, but one or more processing chamber components may have also been negatively impacted, thereby increasing waste and increasing ownership cost.

[6G28] One reason for the delay 1s dug to the sheer volume of data being received from a pluthora of sources. Even if data box 142 may be configured with a fast processor and have sufficient memory fo handle the large vohume of data streams, data box 142 may still need time to process, correlate andfor analyze al of the data being collected.

[8029] Another reason for the delay in receiving the interdiction by the process module is dae the incomplete data streams that are being received by data box 142. Since data box 142 is receiving data from a plethora of sources, the actual data that 1s being sent to data box 142 1s significantly less than the data being collected. Tn an example, instead of sending the gigahertz data stream that is being collected by sensor 118, only a fraction {about 1-3 hertz) of the data is actually being sent. As a result, even though data box 142 is receiving a high volume of data from all of its sources, the data that is bem received 1s usually incomplete.

Thus, determining an wicontrotied event may take time given that data box 142 may not have access to the complete data set from all sources.

[0030] In addition, the paths by which the data are being sent to data box 142 may vary. In an example, data are sent directly from a sensor arrangement (that is sensor and iis computing module) after the analog data has been converted indo digital data. In contrast, the data collected by the process module is ansmitted over a longer network path (twough at least cluster tool controller and fab host). Accordingly, data box 142 1s unable to complete its analysis until all related data streams have been received.

[8031] Not only is the network path between a process modude and data box 142 longer but the data streams sent through this path are usually faced with at least two bottlenecks, The first bottfeneck 1s at the cluster tool controller. Since the data collected by the process modales within a cluster tool is being sent to a single cluster too! controller, the first bottleneck occurs since the data streams from the various process modules have to be processed through a single cluster tool controlfer. Given the sheer vohune of data that can be transoitted from each process module, the network path to the cluster tool controler usually cxperiences heavy traffic congestion.

[0632] Once the data has been received by the cluster tool controller, the dats is transautied to fab host 102. The second bottleneck may occur at fab host 102. Given that fab host 102 nay be receiving data from vanoas cluster tool controllers, traffic into fab host 102 may alse be experiencing congestion due to the high volume of data being received,

[6033] Since data box 142 needs the data from the different sources in order fo determine an uncontrolled event, the traffic condition between a process module and data hox 142 prevents timely delivery of the data streams to data box 142. As a result, precious time 1s ost before f data box 142 has gathered all the necessary data to perform analysis. Furthermore, once an interdiction is prepared, the interdiction has to raved through the same lengthy path back to the affected process module before the interdiction can be applied to perform corrective action.

[8034] Another factor contributing to the delay is the challenge of corvelating data from the various data sources, Since the data streams being received by data box 142 1s usually summary of the data collected from cach sensor andior process modules, correlating the data may be a challenging task since the data streams available may be of different time mtervals.

In an example, the selected data streams ransautied to data box 142 from sensor 118 may be at a one second interval while the data streams from PMC 110 may be at a two second mierval, As a result, correlating data streams may require thine before an uncontrolled event may be definitively determined. [#035] An additional challenge for correlating the data is die to the different paths by which the data are being sent to data box 142. As the data is being transmitted through different computers, servers, and the hikes, the data may be exposed to computer drift, network latency, nctwork loading and the kes. As a result, data box 142 may have difficelty correlating the data from the various sources. Given that a tight correlation is required to quackly identify uncontrolled events, more analysis may be required to be performed before an uncontrolled gventt may be accurately Wdentified.

[8036] Another disadvantage of the solution provided in Fig, 1 is the cost of ownership, In addition to the cost of maintaining a cluster tool system, additional cost is associated with the sensor arrangements. Since cach sensor may be of different brands/makes/ models, cach sensor arrangement useally inchsdes a sensor and a computing module. Physical space is asaally required to house cach of the sensor arrangements. Accordingly, the cost of housing the sensor arrangement can become pricey, especially in areas in which real estate prices can be high. (8637) To reduce the actual time delay between the actual cccarrence of the sncontrofled event within the process module and the receipt of the interdiction by the process module, a claster-level analysis server is provided. Fig. 2 shows a simple block diagram of an interconnecting ol environment with a cluster-tool-level solution for correlating data between the sensors and the process model controllers. [G38] Simudar to Fig. 1, a cluster tool may include a plurality of process modules {such as

PMCs 210, 212, 214, and 216). To collect data for analysis, each process module may he coupled to a plurality of sensors {such as sensors 218, 220, 222, 224, 226, 228, 230, 232, 234,

236, 238, and 240). Each sensor may interact with its corresponding proccss module controtler via a sensor cable (such as sensor cable 244} to collect processing parameter data.

The data collected by the sensor may be in analog format. A computing module {such as computing module 2185) may process and convert the data into a digital format before forwarding the data via a path 246 to a cluster-level analysis server {such as remote controller 242),

[0039] Similar to Fig. 1, cach process module controller may also transout data (such as process module data and process context data} to a cluster tool controller {such as CTC 204 and 206). In an example, data collected by PMC 210 may be transmitted to CTC 204 vin a path 248. Besides receiving data from PMC 210, CTC 204 may also be receiving data from the other processing modele controllers (such as PMC 212, 214, and 216). The data received by the cluster tool controller 1s then foraarded via a path 250 to a fab host 202.

[6040] Between fab host 202 and CTC 204, a serial tap may be connected to network path 254 to duplicate the data being forwarded to fab host 202. In an example, a serial tap 208 may intercept the data bemyg forwarded by CTC 204 to fab host 202. The data is duplicated and a copy of the data stream is sent to remote controller 242 via a path 254. If the fab host is connected to more than one cluster tool controller, than for cach claster tool controller, a dedicated remote controller 1s associated with the cluster tool controller. In an example, the data being sent fron CTC 206 to fab host 202 via a path 232 is intercepted by another serial tap (256). The data is duplicated and sent via a path 25% to a remote controller (260) that is different than the remote controller (242) associated with CTC 204

[0041] Hence, mstead of a single data box to handle all the data from the various cluster tools, multiple remote controllers may be available to handle the data from the various claster tools. In other words, cach cluster tool 1s associated with its own remote controller, Stace cach remote controller is handling data from a fower number of data sources (such as the process module controllers and the sensors associated with a single cluster tool), each remote controler is able to handle a higher volume of data from each source. In an example, nstead of 30-110 data items being sent, about 40kB — 100kB data tems at 10 hertz may now be received by each remote controller.

[6042] Data received from the sensors and the process module controllers are analyzed by the remote confrolier. If a problem is identified, the remote controler may send an mierdiction to the cluster tool controller. In an example, remote controller 242 identifies a problem within PMC 210. An interdiction is sent via paths 254 and 250 through serial tap 208 10 CTC

204. Upon receiving the interdiction, CTC 204 forwards the interdiction to the intended process module comtrolier, which is PMC 210 in this example.

[6043] Since the remote controller is only responsible for handling data from one cluster tool mstead of a plurality of clusters wols (as being done by data box 142), more data may be analyzed and better correlation may exist between the different data sets. As a result, the remote controfler may perform better and faster analysis, thereby providing more timely mtervention to correct an uncontrolled event within a processing module. In an example, mstead of receiving an mterdiction to prevent an identified uncontrolled event from happening in the next substrate lot {such as the micrdiction provided by data box 142), the witerdiction sent by remote controller 242, for example, may enable the process engineers to salvage at least pant of the substrate lot that is scheduled tw be processed.

[8044] Although the remote controler sohution is a better solution than the data box solution, the remote controller solution still depends upon summary data to perform its analysis. Asa result, problems that may be occurring during substrate processing may remain unidentified.

Further, the path between the process module and the remote controller 1s still not a direct path. As a result, computer drift, nctwork latency, and/or network loading may cause tine discrepancy that may make it difficult for the remote controller to correlate the data from the sensors with the data from the process modules. [#045] Thas, even though the remote controller solution has increased the tmeliness of the mnterdiction, the remote controtler solution is still inadequate. At best, the isterdiction may be able to prevent a problem experienced by the affected substrate from occarring during the processing of the next substrate. In a fiercely competitive market where cost needs to be minimized, waste due fo damaged substrate and/or downtime due to damaged processing chantber components may translate into market loss. Accordingly, a real-time solution for whentifying uncontrolled event is desired.

[0046] In accordance with embodiments of the present invention, a process-level troubleshooting architcetire (PLTA) is provided in which troubleshooting is performed at the process module level, Embodiment of the invention includes a process-level troubleshooting architecture that provides for real-time analysis with real-time interdiction. Embodiment of the vention further includes arrangements for load balancing and fault tolerance between

SENSOrS,

[0047] in an embodiment of the invention, the process-level troubleshooting architecture is a network system 1 which an analysis server is comnumicatng with a single processing module and its corresponding sensors. In an embodiment, the information being exchanged g in the network 1s bidirectional. In an example, the analysis server may be continually receiving process data from the processing module and sensors. Conversely, the sensors may be receiving data from the processing module and the processing module may be receiving mstructions from the analysis server.

[8048] Consider the sitsation wherein, for example, a substrate 1s being processed. During substrate process, a plurality of data may be collected. In an example, data about pressure is collected every 100 pulhiseconds. If the processing takes one hour, 36,000 data tems have been collected for the pressure parameter. However, a plurality of other process data {e.g., voltage bias, temperature, otc), besides pressure data, may alse be collected. Thus, a considerable amount of data is being collected by the time the substrate process has completed.

[8049] In the prior art, the data ave transmitted to an analysis server that may be configured to service data collected from a plurality of processing modules (such as remote controller 242 of Fig. 2} if not from a plurality of cluster tools (such gs data box 142). Since the data streams are coming from a plurality of sources, time 1s required to analyze andfor correlate the data, Further, singe the analysis seover of the prior art may not be able to process and analvze all of the data collected, only a fraction of data collected from cach sowree is transmitted to the analysis server. As a result, the complex task of coordinating, processing, correlating, and/or analyzing the data streams requires ime that may not always be readily avatlable.

[058] In one aspect of the invention, the biventors herein realized that a nore accurate and quicker analysis may be performed if more granular data is available for analysis. In order to analyze more data from a single source, the analysis server has to be analyzing data from fewer sources. In an embodiment, an arrangement is provided for processing andfor analyzing data af a process module level. In other words, a process-module-level analysis server 1s provided for performing analysis for each process module and its corresponding

SENSOLS.

[0051] In an ombodiment, the process-module-level analysis server includes a shared memory backbone that may mchude one or more processors. Each processor may be configured to interact with one or more sensors. In an example, data collected by sensor may be processed by processor T while data collected by sensor 2 is processed by processor 2.

[6052] Unlike the prior art, the processors may share iis processing power with one another to perform load balancing and fault tolerance. In the prior art, a computing modale is configured to handle the data colleeted by a seaser. Since cach computing module is an individual waif and usually does not interact with one another, oad balancing is usually not performed. Unlike the prior art, the set of processors within the process-module-fevel analysis server may perform load balancing. In an example, if processor 1 1s experiencing data overload while processor 216 receiving litte or no data, processor 2 may be recruited to assist processor 1 in processing the data from sensor 1.

[0053] Furthermore, wn the prior ant, if a computing module 1s malfunctiomng, other computing modules is unable to take over the processing performed by the malfunctioning computing modale since the computing modules tend to be of different brands/makesfmodels.

Unlike the prior art, workload may be redistributed between the processors as needed. For example, if processor 2 is unable to perform its function, the workload may be redistributed to other processor until processor 2 is fixed. As can be appreciated from the foregoing, the processors eliminate the need for mdividual computing modales, thereby also reducing the physical space required to house the computing modules,

[0054] In an cobodiment of the Invention, the processors may be divided nto two types of processors: primary processor amd secondary processor. Both primary aad secondary processors are configured to handle data from sensors. In an example, if secondary processor {1s associated with sensor 1 then secondary processor | usually only process data coming from sensor §. Likewise, if secondary processor 2 Is associated with sensors 2 and 3, then secondary processor 2 usually only process data coming from those two sensors (2 and 3) [#55] In an cmbodiment, the shaved memory backbone may include one or more prinwary processors. The set of primary processors may be configured not only fo handle data from the sensors but may also be configured to handle data conning from the processing modude.

In addition, the set of primary processors is configured to correlate the data between the various sources (such as the sensors and processing module) and perform analysis. If an mterdiction is needed, the set of primary processors is configured to send the mterdiction to the process modude controller.

[0056] The features and advantages of the present vention may be batter understood with reference to the figures and discussions that follow.

[6057] Fig. 3 shows, in an embodiment of the mvention, a simple fogic overview of a process-tevel troubleshooting architectire, Although a manufacturing company may have more than one cluster tool, 3 single cluster tool is used as an illustration of onc erabodiment of the invention. Although a cluster tool may have a varying number of processing modules, the example Hhustrated m Fig. 3 include a single cluster tool with four processing modudes.

[6658] The data collected by cach processing modude is collected by ts corresponding processing module controllers (PMC 306, PMC 308, PMC 310, and PMC 312) and transmitted to a fab host 302 via a cluster tool controller (CTC) 304. The data that may be transmitted by the PMCs may be the same type of data {process module data and process context data} that has been previously sent in the prior at. Unlike the prior art, the data being transmitted to fab host 302 is not relied upon by the processing modules to perform troubleshooting. Instead, the data may be archived and be made available for future analysis,

[0059] In an cobodiment, a process-module-level analysis server (APECS 314) 1s provided tos perform the analysis needed for troubleshooting. Consider the situation wherein a substrate 1s being etched in PMC 308. During substrate processing, sensors 316, 318, and 320 are collecting data from PMC 308. hy an example, sensor 316 is configured to collect voltage bias data from PMC 30. Analog data colfected from PMU 308 15 sent via sensor cable 328 to sensor 316. Likewise, sensors 318 and 320 may be collected data via sensor cables 330 and 332, respectively, The data collected by the sensors may then be ransoutted via one of the paths 322, 324, and 326 to APECS 314 for processing and/or analysis.

[6060] Unlike the prior art, data collected by the sensors do sot have fo be preprocessed {such as sununarized, for example) before being transmitted fo the analysis server (APECS 34). In an embodiment, instead of having a computing module to process the data, cach sensor may include a simple data converter that may be employed to convert the analog data nto digital data before forwarding the data to APECS 314. Alternatively, a data converter, such as a field-programmable gate array (FPGA) tray be built into APECS 314. nan embodiment. In an example, cach processor may include a data converter algorithm for converting the data into a digital format as part of ifs processing. As can be appreciated from the foregoing, by climinating the need for a computing module, fess physical space is required to house the cluster tool and its hardware. As a result, the cost of ownership may be reduced. [#061] Since APECS 314 is dedicated to processing data only from one processing module and its corresponding sensors, APECS 314 1s able to handle a higher volume of data conning from a single source. In other words, instead of having to pare down the volume of data ansiuited from each sensor, APECS 314 is configured to handle most, if not all, of the data collected by each sensor, In an example, instead of just 10-15 data items being sent for analysis, now two thousands plus data ems from cach sensor may be avaiable for analysis by APECS 314. As a result, the data stream that is available for APECS 314 to process and analyze 1s a more complete data set.

[6662] In an embodiment, APECS 314 1s also configured to handle the data coming from the processing module. Unlike the prior art in which the data stream is sent through a lengthy data path through various servers (e.g, cluster tool controller, fab host, ete.) before being received by the analysis server {such as the data box or the remote controller), the data coltected by the process module is sent divectly to APECS 314 without having to go through other servers, In an example, process module data may be sent from PMC 308 to APECS 314 via a path 334 If an uncontrolled event Is identified, an interdiction may be sent directly to PMC 308 via a path 336 without having to go through other servers first,

[6663] Further details about the process module level analysis server are provided mn Fig. 4.

Fig. 4 shows, m an embodiment of the invention, a simple functional diagram of a process» modude-fevel analysis server. A process-modele-level analysis server {such as APECS 401) may be assigned to each process module. APECS 40) 1s a bi-directional server and is configured for processing coming data and for sending interdictions when weontrolied events gre wentified, {0664] Data sources may flow from two mun sources, data collected by sensors and data cotlected by a process modale. In an embodiment, APECS 400 is configured to receive incoming data from a plurality of sensors (sensors 410, 412, 414, 416, 4206, 422, 424, and 4263. (iven that some cluster tool owners may have already invested a considerable amount of money into the traditional sensor arrangement (sensor with a computing modale), APECS 4030 15 configured to accept data from both the traditional sensor arrangements and the modified sensors (sensor that does not require a computing module).

[0065] In an embodiment, APECS 400 may inchsde an interface, such as Ethernet switch 418, for interacting with traditional sensor arrangements {such as sensors 410, 412, 414, and 416). In an example, data collected by sensor 410 is first converted from an analog format mito a digital format by computing module 410b before the digital data is transmitted to

APECS 400 (via paths 430, 432, 434, or 436). Ethemet switch 418 1s configured to interact with the traditional sensor arrangements to accept the data streams. The data streams are then passed (via paths 446, $48, 450, or 452) to onc of the processors (402, 404, 406, and 408) within APECS 400 for processing.

[6066] Instead of utilizing a traditional sensor arrangement for measuring process parameters, a modified sensor {one without a computing module) may be employed. Since the data collected does not have to be summarized, a computing nodule 1s no longer required for processing. Instead, a modified sensor may include a data converter {not shown), such as an expensive FPGA, for converting data from an analog format to a digital format, in an embodiment. Alternatively, instead of installing a data convertor within the sensors, a data converter {not shown) may be installed within APECS 400. Regardless if the data converter ts stalled externally or intoreally to APECS 400, the elimination of the computing module provides a cost saving in the ownership of the cluster tool. In an example, the cost to purchase, house, and maintain the computing module is substantially elinunated.

[8067] In an embodiment of the invention, APECS 400 inchsde a set of processors (402, 404, 406, and 40%) for handling the incoming data. The set of processors may be physical processing umts, virtual processors, or a combination thereof. Each processor is responsible for handhng the data streams from the sources associated with the processor, In an example, data streams flowing m from sensor 422 via a path 440 are handled by processor 404. In another example, data streams collected by sensor 424 are transnsitted to processor 406 via a path 442 for processing. [$068] The number of processors and 11s relationship with the sensors may depend upon a user's configuration. In an example, even though Fig. 4 only shows a one-to-one relationship between the processors and the sensors, other relationships may exist. In an example, a processor may be configured to handle data from more than one source. In another example, mote than one processor may be configured to handle data streams from one sensor.

[6869] Each of the processors shares a shared memory backbone 428, in an enbodiment. As a result, load balancing may be performed when one or more processors are overloaded. In an example, if the data streams flowing in from seasor 426 via a path 444 is overwhelming processar 408 processing capability, other processors pay be recruited to help reduce the load on processor 408.

[6070] Besides oad balancing, a shared memory backbone also provides an environment for fault tolerance. In other words, if one of the processor is not working properly, the processing previously supported by the malfunctioning processor is redistributed to the other processors. In an example, if processor 406 is not functioning properly and is unable to process the data streams coming from sensor 424, processor 404 may be directed to handle the data streams from sensor 424. Accordingly, the ability to redistribute the workload enables the improperly functioning processor to be replaced without incurring downtime for the entire server.

[6871] in an embodiment, two types of processors pray exist within APECS 404. The first type of processors is a secondary processor {such as processor 404, 406, or 408). Each secondary processor 1s configured to process the data streams received from its corresponding sensors). Additionally, cach processor 1s configured to analyze the data and to identify any potential problem that may exist with the corresponding sensar{s), in an embodiment. [6872} The second type of processor is known as a primary processor (4023. Although Fig. 4 only shows one primary processor, the number of primary processors may depend upon user's configuration. In an embodiment, a primary processor nay be configured to handle data streams from one of more sensors. In an example, data streams collected by sensor 420 arc sent via a path 438 to primary processor 402 for processing,

[0673] Another source of data for a primary processor is a process modhide. In other words, the process module data and the process context data collected by a process moduls 1s processed by the primary processor. In an example, data collected by a process module is sent through a process controd bus via a path 454 to APECS 404. The data first traverses through Ethernet switch 418 before flowing via path 446 to primary processor 402. [#074] In addition to processing data, the primary processor 1s also configured to analyze data from multiple sources. In an example, data correlation between data streams from sensors 422 and 424 is performed by primary processor 402. In another example, data correlation between data streams from one or more sensors with data streams from a process module is alse performed by primacy processor 402.

[6875] Since the data paths for each of the data scurces are now of about similar length, correlating the data is significantly less challenging than that experienced in the prior art. In an example, since the data flow from the process modiide to APECS 400 without having to go through other servers (such as a cluster fool controller andior a fab host), the data streams from the process modude does not experience changes due to computer andor network conditions {such as computer drift, network latency, network foading and the kes) that may have occurred when the data streams have to be transmitted through other servers (such as a cluster tool controller, a fab host, and the likes) as described in Fig. 1 and Fig. 2. In addition, the wait time for receiving all of the relevant data streams require to perform correlation and analysis is now significantly reduced. Thus, correlating data from different sourees is significantly simplified when external conditions {such as computer drift, network latency, network loading and the hikes) have been substantially eliminated.

[6076] Besides the data path, quicker and more accurate analysis may be performed since a higher volume of data with more granularities from a single source provides more data points for performing correlation. in the prior art, correlation between data sources is usually difficult because the data that is available for analysis is usually incomplete since the prior ant analysis server is unable to handle a high volame of data from a plethora of data sources.

Unlike the prior art, the number of data sources is significantly reduced since cach analysis serves is now only responsible for analyzing data from a limited number of sources (the process module and the sensors associated with the process module). Since the number of data sources has been significantly reduced, the analysis server has the capacity to handle a higher volume of data from a single soorce. Given that more granular detasls are provided, better correlation may be achieved between the data streams of the various sources,

[0077] If a problem (such as an uncontrolled event) is identified, primary processor is configured to send an mierdiction to the process module. In an embodiment, a divect digital output line 430 is emploved to send an interdiction from APECS 4060 to the process mode.

With a direct digital output Ine between the two devices, the interdiction does not have to be first converted into an Ethernet message before the interdiction can be transmitted.

Accordingly, the time required to properly format the interdiction and then convert i back is substantially chminated. Thus, APECS 4M) is able to provide real-time interdictions or near- real time wterdictions to the process module to handle the uncontrolled event.

[0078] In an embodiment, a primary processor may also be configured to mteract with other devices via a path 438. In an example, if a cluster tool controller sends a request to APECS 4040, the request ray be sent via path 438 and be handled by primary processor 402. In another oxample, notification to the fab host may be seit via path 458 and the cluster tool controller. [8679) As can be appreciated from one or more embodiments of the present invention, a mocess~-level troubleshooting architecture is provided. By localizing the analysis server at the process modele level, data grambarity is provided for analysis resubing in a quicker and more accurate analysis. With a similar data path for the varions data sources, better correlation exists between the various data streams. With quicker and more gecurate analysis, troubleshooting may be performed on more timely basis with the interdiction provided in a amely manner to provide corrective action that may be employed to not only prevent the next substrate from being damaged bat alse to provide corrective action to fix the uncontrolied event impacting the affected substrate, thereby saving the affected substrate from being damaged. Thus, fewer numbers of substrates are wasted and damages to the processing chamber components may be substantially reduced.

[6080] in another aspect of the vention, the mventors herein realized that with a process- level troubleshooting architecture capable of performing timely, quick and accurate analysis, real-ime in-situ detection of fast transient events {such as micro-arcing events, dechucking events, spiking events, ete.) may be identified and managed. As discussed herein, a fast lo transient event refors to an event (sech as a micro-areing event, dechucking ovo, spiking event, ofc.) that may happen quickly and usually for a short duration duning substrate processing. Due to the speed and the short length of time cach event may last, the task of wdentifving a fast transient event has usually been performed offline, if at all possible, after an entire sabstrate lot has been processed.

[0081] In an example, one or more substrates nay be inspected using an optical metrology tool, for example. Unfortunately, the inspection does not provide for real-time detection.

Instead, by the time a nucro-arcing event, for example, has been wdentified as ocourrmg on the substrate, the substrate has not only been damaged but the rest of the substrate lot may have also been damaged. Additionally, damages to the hardware components within the processing chamber may have also occurred.

[8082] hirecent years, fast transient sensors have been developed enabling fast transient cleetrical signatures {which is a result of fast transient events) fo be captured. However, most fast transient sensors do not have the ability to classify the electrical signatures. In other words, the fast transient sensors may be capable of collecting the data; but, the fast transient sensors usually do not have the capability to classify the data into meaningful clectrical signatures that mav be employed to identify potential harmful events. [6083 Consider the situation wherein, for example, during an etch process, electrical charge may build op causing pcro-arcing to ocowr. As discussed herein, micro-areing refers to an gvent that cocurs when power is quickly dissipated and the dissipation causes damages to the patter on the substrate {such as destruction of the layer, destruction to the pattern. melted layers, ote). By employing a VI probe, data about suoro-areing may be collected. However, most fast transient sensors, such as VI probes, lack the itelligence to interpret the data and identify when a fast transient event, such as a nucro-areing event, has happened.

[0084] Instead, the data collected by the fast transient sensor may have to be analyzed by a third party, such as a human user or by a software program. In an example, a human user may have to analyze the plethora of data and make a determination {based on his expertise} if a fast transient event has occurred during substrate processing. The task of analyzing the data may take hours if not weeks. Even if the data analysis is performed by a software program, analyzing million of data samples may require time. By the time the problem is identified, damages [0 onc or more substrate lots and/or to the hardware coraponents of the processing chamber may have already occurred, [BOBS] Detecting fast transient events, such as micro-arcing events, can be a difficult task since a micro-arciyg event 1s usually not a predictable phenomenon. In other words, micro-

arcing, for example, does not always occur on every substrate. In one aspect of the mvention, the mventors herein realized that even though the timing of 2 micro-arcing cvent is unpredictable, the electrical signature of a micro-arcing event is not. In other words, each micro-arcing event may be represented by a unique signature.

[8086] Fig. 3 shows, a simple diagram of a micro-arcing event {curve 5023. Ag can be seen from curve 502, when an on-wafer micro-arcing event occurs, the voltage and current signals experience a steep drop (304) simultaneously. Then the voltage and current signals undergo a reverse decay as the voltage and current signals gradually rise to a plateau (506) that may be at a different level than the pomt at which both signals dropped.

[6087] in accordance with embodiments of the vention, methods and arrangements are provided for handling a fast transient event, such as a nicro-areing event, within a processing chamber of a plas processing system. Embodiments of the invention include methods for detecting a fast transient event (e.g, micro-arcing). Embodiments of the invention also include methods for classifying a fast transient electrical signature by performing a signature comparison with known fast transient signatures {such as arc signatures). Embodiments of the ivvention further inclade methods for classifiving the severity of the fast transien! event.

Embodiments of the invention vet also include methods for managing the fast transient event to munimize damages during real-time production environment.

[088] In this document, various implementations may be discussed using nHcre-arcing as an example. This invention, however, is not inated to micro-arcing and may include any fast transient event that may occur during substrate processing. Instead, the discussions are raeant as examples and the invention is not Himited by the examples presented. [B08] In an embodiment of the invention, methods and arrangements ave provided for detecting a potential micro-arcing event. As aforementioned, fast-transient sensors (such as

V1 probes} that are capable of performing a high sampling rate (e.g. collecting nuilions or billions of data points in a second) may be employed to collect data daring substrate processing. In an embodiment, a fast sampling transient detection algorithm may be ronning while the V1 probe, for example, collects data during substrate processing. nan embodiment, the fast sampling transient detection algorithm may include criteria for defining a potential fast transient electrical signal. In an example, to identify a potential on-wafer nrcro-arcing ovent, the fast sampling transient detection algorithm may be searching for an event in which both the voltage and the current signals stmultancously drop. In another example, to identify a potential chamber micro-arcing event, the fast sampling transient detection algorithm may be emploved to search for an event in which both the voltage and the current signals are both spiking.

[6094] in an embodiment, the fast sampling transient algorithm is performed by a sensor controller {such as a VI probe controller), a computing module that is coupled wo the sensor {e.g VI probe) and is configured fo provide an interface to the sensor {e.g., VI probe) and to receive data from the sensor (eg, Vi probe). In another embodiment, the fast sampling transient algorithm is performed by a computing module that ss interacting with the sensor controller {e.g., V1 probe controller). In yet another embodiment, the fast sampling wansient algonthm 1s performed by an analyzing module that is interacting directly with the seasor {e.g., VI probe).

[8091] If a potential micro-arcing event is identified by either the sensor (e.g. VI probe) ora computing modude that is interacting with the sensor {e.g., VI probe), then in an embodiment, the waveform of the voltage and current signals {e.g , electrical signatures) that ocowr at around the occurrence of the event may be saved and forwarded to an analyzing module, such as a process-module-lovel analysis server (eg, APECS 314), for analysis. In other words, by performing the detection at the sensor level, only data about potential fast ransient electrical signatures {such as micro-arcing) are forwarded omard to an analyzing modale for further analysis. Thus, instead of sending all the data to the analyzing module for analysis, filtering may be performed to reduce the amount of data traffic being sent along a data path, thereby reducing bandwidth requirement and reducing the processor capability of the analyzing module.

[0092] However, if a potential micro-arcing event is identified by an analyzing modude that ss interacting directly with the sensor (e.g, VI probe), then in an embodiment, data filtering is not required. Instead, the analyzing modide {such as APECS 314}, which is part of the process-tevel troubleshooting architecture may have a fast processor that is capable of handling a large volume of data. Given the enique inventive process-level troubleshooting architecture, common data traffic congestion that may ocowr im other type of analysis architecture may be substantially eliminated. As a result, the analyzing module 1s capable of analyzing millions of data samples quickly and efficiently.

[6093] In an embodiment of the invention, classification of a potential fast transient electrical signatwre may be performed. In an example, once the wavelons of the potential fast transient event is received by the analyzmg module, the analyzing wodule may compare the potential fast transient electrical signature against a set of fast transient signatures {such as a set of arc signatres). In an embodiment, different knoven waveforms that may be axamples of a fast transient event, such as micro-arcing, may be stored within a library.

[6094] if the potential fast transient electrical signature matches one of the set of fast transient signatures saved in the hbravy, the severity of the fast transient event may then be deternyined, in an embodiment. In an example, the fast transient event may be an event that may have little or no impact on the substrate being processed. Thus, the event may be classified as an event with a low severity level. In another example, the fast transient event may be an event that may have damaged the current substrate being processed. Thus, the fast fransient event may be classified with a high severity level

[0095] By identifying the severity of the fast transient event, a determination can be made on

Bow best to handle the fast wansient event. In an embodiment of the invention, predefined course of actions may be provided depending upon the severity of the fast transient event. In an example, a fast transient event with a low severity level may trigger a warning while a fast transient event with a high severity level may result in the etch process, for example, being ternunated.

[6096] To facHitate discussion, Fig. 64 shows, in an embodiment of the invention, a simple block diagram of a processing environment. A processing system 600 may inclede a processing chamber 602 i which a substrate 604 is being processed. Durning substrate processing, gas (not shown) may interact with power {provided through a set of RF generators 606 via a set of match boxes 608) to create plasma for etching the substrate.

[8697] During substrate processing, if an electrical charge bud up cocwrs causing a fast wransient event to occur, the data may be collected by a VI probe 610 and identified by a fast sampling transient detection algontun module 616. Fast samphng transient detection algorithm module 616 may clade criteria for defining a fast transient event, in an embodiment. In an embodiment, the fast sampling transient detection algonthm module may be configured to run during substrate processing.

[8098] In an embodiment, the data collected may be forwarded to a V1 probe controler 612 along a set of paths 614. VI probe controller 612 is configured at least for managing Vi probe 610. In an embodiment, VI probe controller 612 may also include fast sampling transient detection algorithm module 616.

[6099] in another embodiment, fast sampling transient detection algorithm module 616 may be an independond computing module that may communicate with VI probe controller 812.

In other words, the data collected by VI probe 610 may be sent via VI probe controller 612 10 fast sampling transient detection algorithm modale 616. By making fast sampling transient detection algorithm module 616 an independent modele, VI probe controller 612 does not have to be modified if VI probe controller 612 is not capable of handling additional

Processing. [O10] in another embodiment, instead of sending the data to VI probe controller 612, the data nay be sent directly from VI probe 611 via a path 650 to an analyzing modale 618 {as shown in Fig. 6B), which may house fast sampling transient detection algorithm modide 616. By transnutting the data divectly wo analyzing module 618, data collected by the VI probe 610 do not have to be preprocessed. In addition, a computing module {such as Vi probe controller 612) may be chiminated to reduce real estate overhead. Instead, analyzing modole 618 may be employed to identify a potential fast transient electrical signature.

[00101] Once a potential fast transient electrical signature has been detected based on the predefined criteria, the potential fast transient electrical signature may be classified by analyzing module 618, such as a process-module-tevel analysis server (eg, APECS 314) In an embodiment, analyzing module 618 may perform signature comparison by comparing the potential fast transient electrical signature against a set of fast transient signatures stored within a library, such as a set of are signatures. If a match is identified, a fast transient event is considered to have occured.

[60102] in an embodiment, analyzing module 618 is configured to deternune the severity of the fast transient event. Those skilled in the act are aware that fast transient events may have different severity (e.g. intensity) levels, Accordingly, an algorithm is provided determining the severity of cach fast transient event. In an embodied, the severity level/threshold range may be predefined and may be user-configurable. As an example, a drop greater than 4dB in the current or voltage signal and a duration {defined as from the drop to the recovery) {onger than 15 microseconds nay be deented as appropriate thresholds for detection of damage on the wafer,

[68103] Once the severity level for a fast transient event has been classified, a course of action may be applied. in an embodiment. the course of actions may be predetermined and may be associated with the severity levelfhreshold range. hian embodiment, the course of action may be user-configurable. In an example, a fast transient electrical signature (such ag micro-arcing) with a small voltage and current drop may be considered as harmless and may require only a notification to be sent to the operator. In another example, a fast transient clectrical signature with a large voltage and current drop may be considered as an event with a high severity tevel and a termination of the substrate process may be triggered.

[00104] Fig. 7 shows, in an embodiment of the invention, a simple flow chart

Hlustrating a method for detecting a real-time fast transient event within a production environment in which the fast sampling transient detection algorithm 1s not part of an analyzing module.

[80105] At a first step 702, substrate processing commences. Consider the situation wherein, for example, substrate 604 is being processed within processing chamber 602.

[060106] At a next step 704, substrate processing within the processing chamber is being monitored. At a step 704a, fast transient sensors, such as VI probes, may be monitoring clectrical parameters {e.g voltage and current signals at different phases, fundamentals and harmonies). At about the same time, at a step 704b, a fast sampling transient detection algoridun may be executed,

[00107] At a next step 706, a determination is made about the existence of a potential fas{ transient event. In other words, the fast sanpling transient detection algorithm may mclude criteria for defining a potential fast transient event, such as nuicro~-areing, for example. 1f the data collected by the VI probe does not meet the criteria defined by the fast sampling transient detection algorithm, then no potential fast transiont event has occurred and the VI probe continues monitoring the substrate process (step 704).

[60108] However, if a potential fast transient event 1s identified, then at a next step 0%, the voltage and current waveform at around the occurrence of the potential Fast transient event may be saved,

[30109] Af anext step 710, the saved waveform is transmitted to an analyzing module.

In an embodiment, only the data related to the ocaurrence of the potential fast transient event may be saved and vansmitted. By only sending the potential fast transient electrical signature, resource drain may be nunimized. In addition, since preprocessing has been performed by the sensor controller (such as the VI probe controller). the analyzing module may not need to clude a fast processor to analyze the data and quickly classify and deterpiine a course of action for the potential fast transient event,

[060116] At a next step 712, signature comparison is performed by the analyzing modole. In an embodiment, the analyzing module may compare the potential fast transient electrical signature against a set of fast transient signatwres. In an embodiment, the set of fast transient signatures may be stored within a library, In an embodiment, the library mav also inchade non-fast transient signatures to enable correlation to be performed.

[00111] At a next step 714, a determination is made on the classification of the potential fast transient electrical signawwre. {f the signatee comparison results in no match being identified, then the potential fast transient electrical signature is not classified as a fast transient electrical signature of mterest (step 716). In an embodiment, the potential fast transient electrical signature may be discarded, a another erubodiment, the potential fast transient electrical signature may be added to the Bbrary as a now fast transient electrical signatare (step 718).

[60112] However, if the signature comparison results in 3 fast transient electrical signature being wdentified, then at a next step 724, the seventy of the fast transient event is determined. In an example, the severity may range from low tw high. In an embodiment, the severity may be based on a predefined set of threshold ranges. In an embodiment, the fast transient electrical signature may be added to the library {step 718). Step 718 1s an optional step and is not requived in detecting real-time fast transient events, [B8113] At a next step 722, a course of action is determined. Ounce the severity level has heen determined, a course of action may be excouted. In an embodiment, the course of action may be predefined. In an example, a fast transient electrical signature with a low severity level may trigger a notification to the operator. In another example, a fast transient clectrical signature with a medium severity level may trigger an alamo In yet another example, a fast transient electrical signatare with a high severity {evel may trigger a termination of the substrate process. As can be appreciated from the foregomg, the severity levels and he course of actions associated with the severity levels may be wser-configurable. {60114} Fig. 7 shows, but one embodiment for implementing a method for detecting a real-time fast transient event within a production environment. In another example, the method may also be applied to detect a real-time fast vansient event in which the fast samphing transient detection algorithm 1s part of an analyzing module, in an embodiment. In this type of environment, the excoution of the fast sampling transient defection algorithm may be performed by an analyzing module {such as APECS 314) instead of a Vi probe controller.

In an embodiment, the analyzing module 15 a fast processing computing module that 1s capable of handling a high volume of data. In an embodiment, the analvzing module is directly coupled with the sensor. Thus, data is collected by the sensor and transmitied directly to the analyzing module,

[06115] As can be appreciated from the foregoing, arrangements and methods are provided for detecting an w-situ real-time fast transient event. In the prior art, detection of a fast transient ovent is usually performed after substrate processing has been comnpleted fora substrate lot. Further, complex metrology tools may he required to determine the existence of a fast transient ever. Since the existence of a fast transient event is enpredictable, cach substrate within a substrate lot may have to be measured in order to determine the potential damage that may have ocowrred. [Ga 16] in contrast to the prior art, embodiments of the mvention provide for the detection of fast transient events during substrate processing in real-time, thereby nuinimizing damages to the rest of the substrate lot andfor the processing chamber. In addition, unlike the prior art, the detection process is an automated process that require little or no heman miterference. Instead, once the user-configurable conditons/eritersa‘thresholds have been defined, the system is configured to detect a fast transient event automatically. [661171 Given that fast transient events (such as micro-arcing ovenls) may be wlentified in real-time within a production environment, the latency between the actual occurrence and the course of action taken to manage the occurrence may be reduced. In the prior art, the latency may take hours or even weeks. However, with the methods andfor arrangements described herein, the latency may be reduced to mere vulfi-seconds, thereby reducing the overall cost of ownership.

[06118] While this invention has been described in terms of several preferred embodiments, there are alicrations, pornuiations, and equivalenis, which fall within the scope of this mvention. Although various examples are provided here, it is intended that these examples be Hlustrative and not limiting with respect to the mveation,

[06119] Also, the title and sunmmary are provided herein for convenience and should not be used to construe the scope of the claims herein. Further, the abstract is written ina fughly abbreviated form and is provided herein for convenience and thus should sot be employed to construe or limit the overall invention, which is expressed in the claims. ihe term “set” is employed herein, such term is intended to have its commonly understood mathenatical meaning to cover zero, ong, of more than one member. It should also be noted that there are many alternative ways of implementing the methods and apparatuses of the present mvention. It is therefore mended that the folowing appended claims be interpreted as including all such alterations, permutations, and equivalents as fall within the true spirit and scope of the present invention.

Claims (1)

1. CLAIMS What is claimed is:

   i. A method for detecting an in-situ fast transient event within a processing chamber of a plasma processing system during substrate processing, said method comprising: analyzing a first data set collected by a set of sensors, wherein said analyzing includes comparing said first data set to a set of criteria to determine if said first data set includes a potential m-situ fast transient event, wherein said set of criteria defines a set of in-situ fast transient events; if said first data set includes said potential in-situ fast transient event, saving an clectrical signature that occurs in a time period during which said potential in-situ fast transient event oocurs, comparing said electrical signature against a set of stored are signatares; if a match ts determined, classifying said electrical signature as a first in-sito fast jransiont event; and determining a severity level for said first in-situ fast transient event based ona predefined set of threshold ranges.

   2. The method of elaim 1 wherein sald analyzing of said first data set includes performing a fast sampling transient algorithm, 3 The method of claim 2 wherein said fast sampling transient algorithm is executed by a sensor controller.

   4. The method of claim 2 wherein said fast sampling transient algorithm is executed by a computing module, wherein said computing module is configured at least to be coupled to one of a sensor and a sensor controller. 3 The method of claim 2 wherein said fast samphing transient algorithm is executed by an analyzing module that 1s configured to interact directly with a sensor of said set of sensors.

   6. The method of claim § wherein said analyzing module is a process-module-level analysis server that is configared to perform analysis for cach process modvde and a set of sensors associated with said cach process module.

   7. The method of claim 1 firther including determining a course of action based on said severity level of said first in-situ fast transient event. 8, The method of claim 1 wherein said first in-situ fast transiont cvent 1s 3 NUCIO-areing event.

   9. The method of claim 1 wherein said first data set 1s being collected by fast-transient sensors that are capable of performing a high sampling rate.

   10. The method of elatim | wherein said electrical signature 1s added to a library as a non- fast transient event signature if said electrical signature does not match one of said set of stored arc signatures. 11 An arrangement for detecting an in-sita fast transient event within a processing chamber of a plasni processing system, wherein sad processing chamber inclades a plurality of sensors configures for collecting data during substrate processing, said arrangement COMPTISIG: a fast sampling transient algorithm module configured for comparing said data against a set of criteria and extracting an electrical signature from said data, wherein said set of criteria defines a set of predefined m-situ fast transient events; and an analyzing module, wherein said analyzing module comumanicates directly with said fast sampling transient algorithm module, wherein said analyzing module is configured for performing at least receiving said electrical signature, comparing said electrical signature against a set of stored are signatures, classifying said electrical signature as a fast transicnt event if a match ocours, and deternuning a severity level for smd fast transient event based on a predefined set of threshold ranges.

   12. The arrangement of claim 11 further incloding a library, wherein said hibrary is configured for storing saud set of stored are signatures

   13. The arrangement of claim 12 wherein sad Hibvary is configored for storing non-fast ransient signatures. 14, The arrangement of claim 1 wherein analyzing modide is configured for sending said course of action directly to a process module condrofler when said fast transient cvent is wentitied during said substrate processing.

   15. The arrangement of clam 11 wherein said analyzing module is further configured for deternuning a course of action based on said severity level of said fast transient event.

   16. The arrangement of claim 11 wherein smd fast transient event is a nucro-arcing event.

   17. The arrangement of claim 11 wherein said fast sampling transient algorithm nodule is controlled by an analvezing module that is configured to miteraet directly with said plurality of SENSOrS,

   18, The arrangoment of claim 11 wherein said analyzing module is a process-modale- tevel analysis server that is configured to perform analysis for cach process modude and a set of sensors associated with said cach process module.

   19. The arvangement of claim 11 wherein said fast sampling transient algorithm module is controlled by a sensor controller.

   20. The arrangement of clam 11 wherein said fast sampling transient algorithm module is controlled by a computing module, wherein said computing module is configured at least to be coupled to one of a sensor and a sensor controller.