In-Situ Metrology: the Path toReal-Time Advanced Process Control

G.W. RubloffUniversity of Maryland, 2145 A. V. Williams Building, College Park, MD 20742-3285, USA

rubloff@isr. umd. edu, www. isr. umd. edu/gwrubloff/Abstract. While real-time and in-situ process sensors have been effectively applied to fault detection, process

control through course correction has been mainly focused on in-line metrologies to drive run-to-run feedback andfeedforward control We have developed in-situ metrologies based on mass spectrometry, acoustic sensing, andFTIR techniques which enable real-time thickness metrology and control in CVD processes at a level of about 1%accuracy. These developments open the door to real-time sensors as the basis for both fault management andcourse correction, i.e., for real-time advanced process control We have also employed in-situ metrology todevelop robust control schemes for CVD precursor delivery from solid sources, and we are exploring a newspatially programmable reactor design paradigm for which real-time, in-situ sensing, metrology, and control ofacross-wafer uniformity is fundamental

As indicated in Table 1, the goals of CC and FMare quite different: CC is aimed at keeping productquality on track, while FM is intended to monitor andmaintain equipment so that product is not wasted andequipment repair is timely. The other characteristicwhich differentiates APC approaches is a distinctionbetween run-to-run and real-time control: run-to-run

ADVANCED PROCESS CONTROLMetrology is a mainstay of advancement and

practice in semiconductor technology development andmanufacturing. [1,2] Advanced process control (APC)[3, 4] is aimed at exploiting metrology to increaseequipment productivity and lower manufacturingcosts. Early efforts toward APC were in evidence fromthe late 1980's, including TI's MMST program [5-8],Sematech's J-88 program [9] , and the inclusion of ametrology and process control roadmap as asupplement to the first NTRS in 1994. [10-12] Today,much progress has been made in developing andimplementing APC in the industry, and APC hasbecome a mainline component of the ITRS(Metrology section) [13] and the subject ofprofessional meetings [14, 15], consortia [16], andpublications [17, 18].

In both conceptual and operational senses, APCis perhaps best regarded in terms of two components,course correction and fault management.

Course correction (CC) is aimed at adjustingparameters in a process or process sequence such thattargets metrics are maintained in the presence ofprocess drift and short-term variability. In principle,CC may be accomplished as run-to-run (or wafer-to-wafer) feedback and/or feedforward control, or as real-time control within a unit process.

Fault management (FM) is intended to detectequipment and process problems as soon as possible(hopefully in real time), and then initiating correctiveaction, ranging from immediate shut-down and repairof equipment to a more extensive exercise in faultclassification and prognosis that leads to reschedulingof tool maintenance.

control can provide benefits from in-line or off-linemeasurements, while real-time control can deliver adifferent set of benefits if real-time/in-situ metrology isavailable. [19]

ControlstrategyGoal

Run-to-runmetrology

control

apply to

corrects forReal-time

metrologycontrol

apply tocorrects for

CourseCorrection

maintainproduct quality

mainline

FaultManagement

monitor &maintain

equipmentemerging

in-line, off-linefeedback &feedforward

unit process &process sequence

acceleratedmaintenanceunit process

long-term driftemerging mainline

real-time/in-situreal-time

correction orend-point

equipment stop,accelerated

maintenanceunit process

long-term drift & short-termvariability

TABLE 1. Goals and modes for run-to-run and real-timeadvanced process control.

CP683, Characterization and Metrology for VLSI Technology: 2003 International Conference,edited by D. G. Seiler, A. C. Diebold, T. J. Shaffner, R. McDonald, S. Zollner, R. P. Khosla, and E. M. Secula

© 2003 American Institute of Physics 0-7354-0152-7/03/$20.00583

An early, major success for APC was IBM'sapplication [20] of in-situ residual gas analysis (massspectroscopy) to detect faults in process equipment,terminate the process, and alert the operator and/orengineer, thereby avoiding further yield loss. Since anumber of important equipment failure modes are wellunderstood and readily distinguishable this way (e.g.,gas source impurities from inadequate purging upongas bottle change), this real-time fault managementmode has provided substantial return and has beenbroadly implemented in manufacturing, as indicated by"mainline" in Table 1. More sophisticated faultdetection algorithms have been reported. [21]

Expanding fault management to more complexfault mechanisms, equipment health prognosis, andoptimized maintenance scheduling is morechallenging: in the general case, multiple sensorsignals, system modeling, and probabilistic riskassessment may be required to classify faults, specifylikely origins, and predict when they will demandrepair. On the other hand, this more general FM goalcan benefit from in-line and off-line as well as real-time/in-situ metrologies, since they deliver trendinformation helpful to both classification andprognosis. A variety of interfacing and software toolsare currently in use for data collection, integration andalgorithm application [22] to develop these emergingstrategies to the point that they can define equipmentrepair actions needed and appropriate schedules forthem.

The primary thrust of course correction APC hasbeen in run-to-run control. [23-26] In this case in-lineor off-line measurements are used to guide processrecipe changes for succeeding wafer steps - eithersubsequent wafers in the same tool (feedback) orsubsequent steps for the same wafer (feedforward).Such approaches can detect and compensate for long-term drift of process/equipment behavior, where long-term signifies variation evident over multiple wafers,but not detectable within a single wafer run. Indeed,the Sematech J-88 program was aimed specifically atrun-to-run control, and since then run-to-run controlbecome mainline, with substantial success andadoption in manufacturing for feedback, feedforward,and multi-step applications. [27-29] This is due insignificant part to the availability of in-line and off-linemetrology technology, which is a prerequisite fordevelopment and manufacturing even without explicitAPC intentions. Platforms for APC implementation -emphasizing run-to-run control - have been developedto meet industry needs. [30]

In contrast, real-time course correction has beenlimited, primarily by the fact that real-time/in-situmetrologies with sufficient quantitative accuracy havenot been available. It is important to note that real-time/in-situ sensors for fault detection do not demandthe quantitative accuracy that is needed for CCapplications. A primary advantage of real-time CCwould be that unit process control could compensate

for short-term random variability within a process step,as well as for long-term process drift, thus achieving ahigher degree of overall course correction APC. Inaddition, sensors suitable to drive real-time metrologymay be capable of serving both real-time coursecorrection and fault detection goals simultaneously.

REAL-TIME APCNow that important elements of APC have seen

wide acceptance, one can envision a next-generation,i.e., real-time APC. In this scenario, quantitative real-time metrologies would drive real-time coursecorrection to achieve enhanced controllability,effective for both short-term random variability andlong-term drift, and these metrologies would becompatible with (if not identical to) those employed forreal-time fault detection. With both real-time coursecorrection and fault detection controlling controlresponses at the unit process level (rather likeregulatory controllers), the existing benefits offeedback and feedforward run-to-run control could beretained at a more supervisory level of control, andfuture advances in fault classification and prognosiswould ultimately add to this picture.

Over the past years our research has pursued thedevelopment of chemical analysis techniques as real-time, in-situ process metrologies, as well as thedemonstration of these in control applications, withprimary emphasis on mass spectrometry, acousticsensing, and infrared (FTIR) sensors applied tochemical vapor deposition (CVD) processes andrate/thickness control. Currently, these methods - withthree different sensors - have demonstrated better than1% film thickness metrology and 1.5-2% real-timethickness control, and they have broad application(e.g., etching, heterostructures, etc.). Opportunities forsensing and controlling spatial uniformity - thoughlonger-term - are also in view. And similarapproaches to real-time sensor-based control have beenshown, e.g., in the control of MOCVD source deliveryfor difficult materials (solids with low vapor pressure).This paper provides examples of such progress towardreal-time APC in the hope that the next steps oftechnology transfer to manufacturing can be taken.

MASS SPEC METROLOGYInitial investigations using in-situ mass

spectrometry (mass spec) clearly revealed time-dependent behavior of reactive gases in CVD [31] andplasma [32, 33] processes. For example, RTCVDpoly Si and oxide results showed generation of reactionproducts and depletion of reactants to an extentdependent on reaction rate, and they indicated thepossibility of using these for deposition rate metrology[34] . More recently, we have evaluated the metrologyaccuracy obtainable using downstream mass spec forseveral W CVD processes, both H2 reduction [35, 36]and SiH4 reduction [37], as carried out in a UlvacERA-1000 W CVD cluster tool. With reasonablereactant conversion rates (-20%) in the SiH4 reduction

584

process, thickness metrology of 1-1.5% accuracy wasdemonstrated and exploited to achieve real-time endpoint control consistent with this metrology. [37]

The experimental setup is indicated in Fig. 1,showing differentially pumped mass spectrometry(Inficon Transpector™) arranged to sample reactantand product species in the CVD reactor. As indicatedby the time-dependent HF product signal in Fig. 2 forthe H2 reduction reaction, the sensor system activelyprocess gases to determine dynamics through theprocess cycle. These in turn are consistent withdynamic simulations of equipment, process, and sensorsystems. [38]

The SiH4 reduction process for W CVD,3 SiH4 + 2 WF6 => 6 H2 (g) + 2 W (s) + 3 SiF4 (g),produces both H2 and SiF4 as reaction products. For alow pressure (0,1 torr) process, a thickness metrologyof 1-1.5% accuracy was obtained using the H2 productgeneration signal. [37] By tracking the integratedproduct signal and terminating the process when thetarget value was reached, run-to-run variability wasreduced from ~4% to -1.5%. Figure 3a shows theimprovement for the mass spec product signal uponimposing real-time end point control, while Fig, 3bshows the corresponding improvement in control of thedeposited W film thickness, as determined post-process

Ulvac ERA-1000 SelectiveW CVD Chamber #2

H2/SIH4

WF6 M-

Lamp Heater

i i i i i i i i i i i i i i i i i i u n n i m i i i i i i i iShowerhead

Wafer

High-ConductancePressure Control Valve

WF6 By-PassTo Mech

Pumping Stack

Orifices Closed IonSource

200AMU QMSManifold Electronics Box

Orifice

FIGURE 1. Mass spec sampling of W CVD process.

/r' A

r Reactor Total Pressure

i • i * iWF6&H2

A /i • i ' i

i N2i i

|

r*-

1x10*-

1000 1200 1400 1600 1800 2000 2200 2400 2600Time (sec)

FIGURE 2. HF product signal through process cycle usingH2 reduction process for W CVD, i.e., 3 H2 + WF6 => 6HF(g) + W(s).

DifferentialPumping Diaphragm

Pump

by weight measurements. In these experiments, "nocontrol (experiment)" corresponded to a sequence ofwafers where no end point control was imposed; "nocontrol (estimate)" reflects an estimate of what thevalues would have been if end point control had notbeen imposed, judging from the signal levels observed.

At higher pressures (10 torr) the quality of thethickness metrology is even better, which is importantbecause W CVD processes in manufacturing operate at100-500 torr. [36] Using the HF product signal fromthe H2 reduction reaction, a linear regression fit to theweight (thickness) vs. integrated HF product signalyielded a standard deviation of 0.67% uncertainty, asindicated in Fig. 4. These results indicate that massspectrometry is a viable sensor-based metrology forreal-time APC.

585

no control (estimate) with control

o5.0

4.0

3.0 -

1.0 -

0.0

O -target H2 value

no control '(experiment)

1 2 3 4 5 6 7 8 9Wafer number

FIGURE 3a. Real-time end point control of integrated H2

reaction product signal for SiH4 reduction W CVD process.FIGURE 3b. Real-time end point control of film weight

no control (estimate)

O)

O)

0.04 -

0.03 -

0.02 -

0.01 -

0.00

with>control

no control (experiment)

1 2 3 4 5 6 7 8 9Wafer number

corresponding to Fig. 3 for SiH4 reduction W CVD process.

,=, 260-

Eil Linear Regression

Average Uncertainly +/- 0.56%Standard Deviation 0.67%

0 10* 2.1X10"5 2.2X10"8 2.3X10"5 2.4x10* 2.5x10* 2.6x10* 2.7x10* 2.8x10*

Integrated HP Signal (A-s)

FIGURE 4. Mass spectrometry based thickness metrologyfor H2 reduction W CVD process at 10 torr pressure.

MASS SPEC FOR ULTRATHINLAYERS

There is an increasing need for thickness ofultrathin layers (in the 10 nm range), driven by specificapplications (e.g., barrier and seed layers, gatedielectrics). Mass spectrometry based metrologyshows promise for such applications. Figure 5 showsmass spec signals for the initial stages of H2/WF6 CVDreaction (H2 reduction) on the Si surface. In this case,the H2 reduction reaction is considerably slower thandirect reaction of the WF6 reactant with the Si surface(the Si reduction reaction), which generates a SiF4volatile product. [3 9] The Si reduction reactioncontinues until the initial Si surface is completelycovered by W, at about 30 nm thickness.

6.0X1012

SiF4 product fromSi reduction reaction

"hot" waferHF signal -***®Z&^*K**&*-K

2.0time (min)

FIGURE 5. Mass spectrometry product signal for Sireduction reaction in forming 30 nm W on Si surface.

The sharp SiF4 peak in Fig. 5 demonstrates thatreal-time mass spec has sensitivity in the 10 nm range,suggesting a thickness metrology for such ultrathinlayer applications of order 5% precision or better. Thissuggests at least two key areas of application. First, itmay mean that mass spec metrology will be useful forultrathin barrier, seed, or gate dielectric applications.Second, it shows that the approach can certainly revealfaults, such as inadequately cleaned contacts where theshape and intensity of the resulting peak changesconsiderably if the CVD chemistry is reasonablyselective (e.g., reduced deposition rate on oxide cf. Si).

ACOUSTIC METROLOGYAcoustic sensors enable the determination of gas

composition from the measured velocity of sound inthe gas. Given the charactistic product generation andreactant depletion which occur in processes like CVD,this implies that acoustic sensors should also becapable of reaction rate and thickness metrology. Infact, real-time in-situ metrology for depositionthickness has been demonstrated at a comparable levelof accuracy using acoustic sensors. [40, 41]

586

valve

4^

11

Gas

Vit

lit

gimp

to

FIGURE 6. Acoustic sampling of W CVD process.

0.27-

0.26-

0.25-

^2Jf 0.24-O)

5 0.23-

£ 0.22-

0.21-

0.20-80C

__-*— — «"̂ *^__^t— - — • "\_____•̂ ~"

Run numbers

-

-

)00 85000 90000 95000

- 17000 .

- 16000

- 15000 '

- 14000

(frequency integrated signal, Hz-s) in Fig. 7 predictsthe deposited W film weight to about 0.5%,comparable to the mass spec result.

FTIR METROLOGYFinally, optical sensing, carried out using Fourier

transform infrared spectroscopy (FTIR) has been usedto monitor reactant depletion and product generation inthese CVD processes. [42] In this case the depletion ofWF6 is more sensitive than the generation of HFreaction product. As shown in Fig. 8, the integratedFTIR absorption peak of WF6 provides an in-situthickness metrology accurate to about 0.5%, i.e.,comparable to the mass spec and acoustic sensorresults.

oin 680

660

640

620

600

580

R=-0.99672

40tfC40CfC

B406°C

410°C

41CfC

Frequency integrated signal (Hz.s)

0.14 0.15 0.16 0.17 0.18 0.19 0.20 0.21Mass of deposited tungsten (g)

FIGURE 8. FTIR sensor based thickness metrology for H2reduction W CVD process at 10 torr pressure.

FIGURE 7. Acoustic sensor based thickness metrology forH2 reduction W CVD process at 10 torr pressure.

Acoustic sensing requires pressures sufficientlyhigh to sustain acoustic waves in the medium, typically50-100 torr, from which the measured sound velocityreflects the composition of the gas mixture. Theexperimental arrangement to achieve this is depicted inFig. 6, where in this case the 10 torr pressure for the H2reduction W CVD process required pressuretransduction to higher pressure in the sensor, which inturn was achieved using gas compression by adiaphragm pump. The sensor was an InficonComposer.™

The metrology accuracy of the acoustic sensingtechnique for the 10 torr H2 reduction W CVD process(as reflected in Fig. 4) is shown in Fig. 7. [40, 41] Inthis case, 10 wafers processed at 490°C showed awafer-to-wafer drift, as evident in the Fig. 7. Thoughthe algorithm for extracting a metrology signal fromthe acoustic sensor data is rather more complex thanthat for the mass spec data, the metrology metric

WAFER STATE METROLOGY FORREAL-TIME APC

The three sensors described above are, strictlyspeaking, process state sensors, yet by suitablymonitoring the dynamic state of the process andrecognizing that the process conditions reflect waferstate conditions, all three generate a wafer statethickness metrology of order 1% accuracy or better,seemingly sufficient to deliver value in manufacturing.In addition, at least one of them - mass spectrometry -is already in widespread use for real-time faultdetection, indicating that it could provide dual use toreal-time APC. Indeed, we have exercised massspectrometry for fault detection purposes in concertwith its role in thickness metrology for real-timecourse correction.

While these techniques have achieved themetrology accuracy originally sought, they must beused carefully. For example, they reflect the totalmaterial deposited, though some of this could be onother internal surfaces of the reactor (e.g., wafer chuck,chamber walls). In general, they require pressure

587

transduction systems to transform the sampled gasfrom the process pressure used to the pressure regimeneeded by the sensor, e.g., lower pressure for massspectrometry and higher pressure for acoustic sensing.And of course they do not indicate across-waferuniformity, an equally important metric formanufacturing productivity, but probably less sensitiveto sources of process variability.

Nevertheless, these results are very promising inproviding a base for expansion of APC to real-timeAPC. The problems of implementing sophisticatedsensors like mass spectrometry have already beensurmounted for real-time fault detection, and importantissues of sensor conditioning and reliability have beendealt with.

In some sense, there remain two key issues fortransfer of these real-time metrologies tomanufacturing. First, in the context of pervasive run-to-run course correction applications and existing real-time fault detection, how will real-time coursecorrection be integrated into the larger APC context?This may be a significant architecture issue, though nota fundamental obstacle given that real-timequantitative metrologies like these would primarilydrive real-time feedback control at the unit processlevel, thereby functioning more like regulatorycontrollers within a larger supervisory control systemwhich executes feedforward as well as feedbackcontrol involving multiple process steps and tools.

Second, what will be the accuracy of thesemetrologies when implemented in a manufacturingtool? Of course this can only be judged by experiment,and it should be emphasized that this must be the nextstep. While we have processed and tested 10-20 wafersper day in our experiments, it is doubtful that ouracademic research environment (or any other) can addmuch more in evaluating these techniques formanufacturing. Technology transfer to a developmentor manufacturing environment is the essential nextstep, where considerably larger wafer lots can beprocessed through real manufacturing tools.

Nevertheless, we can make some predictionsbased on our experience. A major hurdle in achievingthe metrology accuracy reported here has been thepresence of wall reactions (e.g., WF6 adsorption andreaction with subsequent H2 to form HF), and theconcomitant need to condition the reactor system(chamber, gas lines, etc.) in order to stabilize thesefactors, and when this is done the results are veryfavorable. In manufacturing tools, where wafers areprocessed at high rate over long periods of time, thisconditioning is substantially guaranteed. Furthermore,when variations from continuous operation ofmanufacturing tools are made (e.g., preventivemaintenance), the tools undergo conditioningprocedures (e.g., monitor wafers for recalibration,recognition of first-wafer effects, etc.). Because theregularity of the equipment environment is muchhigher in manufacturing, we can only predict more

accurate metrology for real-time APC than we haveachieved in the research cited here.

CVD is a pervasive semiconductor manufacturingprocess and thickness control is important, so thisexample may serve as a useful entry point fordevelopment and integration of such sensors into alarger real-time APC system. However, semiconductortechnology demands a considerably broader set ofprocesses as well as different kinds of metrology.Therefore, one should consider the prognosis forexpansion of a real-time APC capability.

Other chemical processes are certainly amenableto the same or similar techniques. Besides otherthermal CVD processes, these methods can be appliedto plasma-enhanced CVD and plasma etching, asalready demonstrated. [32, 43] But the general notionof sensing chemical reaction signatures for chemicalprocesses is much broader. For example, we havebegun investigations of the curing process used tocreate nanoporous low-K dielectrics by spin-casting,using time-of-flight secondary ion mass spectrometry(ToF-SIMS) [44]. The properties of the materialdepend sensitively on the details of how the sacrificialporogen species are volatilized (they are theplaceholders for the nanopores that will remain in thelow-K matrix). Initial results clearly show the kineticsof porogen depletion with curing (annealing), and wehave begun experiments to identify the volatileproducts directly through thermal desorption massspectrometry.

UNIFORMITY CONTROLWhile most in-situ sensors target the average

behavior across the wafer, or a single point on thewafer, across-wafer uniformity is a key metric formanufacturability and an intense focus in processdevelopment. Clearly, it would be highly desireable toextend real-time process and wafer state metrology toencompass uniformity measurements for real-timeAPC.

Some sensor approaches are promising in thisregard. Spatially resolved optical emissionspectroscopy and multichannel laser interferometryhave been used to achieve process and wafer stateuniformity measurements in plasma etching processes[45], and the latter has led to development of real-timeinterferometric imaging instruments. [46]Ellipsometry has been employed using in-situ etchrates measured at a single point on the wafer togetherwith offline multi-point measurements and responsesurface modeling for etch uniformity control. [47]Transient behavior at end point sensed by real-timemass spectroscopy has distinguished signatures ofuniformity. [43]

We have been exploring a new concept inequipment design aimed at uniformity control, namelyspatially programmable reactor design. The goal isreactor designs which ultimately enable spatialuniformity to be modified through arrays of sensorsand actuators, so that the performance characteristics

588

of the reactor are software-programmable. Such aparadigm would provide for spatial arrays of processand wafer state sensors, or multiplexing of spatialsampling to such sensors, along with arrays ofactuators and control systems, to accomplish spatialcontrol of the process in run-to-run and real-timecontrol modes.

Programmable reactor designs would enable threebroad applications: (1) programmable uniformity,assuring that manufacturing requirements foruniformity can be achieved at any process design pointwhich is sought for materials performance; (2)programmable nonuniformity, in which intentionalnonuniformity is introduced to expedite experimentalprocess optimization, essentially a "one-wafer design-of-experiments"; and (3) combinatorial materialsdiscovery, in which gradients of stoichiometry areintentionally introduced to accelerate the developmentof complex new materials.

The initial testbed for this concept is thedevelopment of a spatially programmable CVDshowerhead, applied initially to W CVD under NSFITR sponsorship. [48] The current prototype is a three-zone showerhead [49] to test the approach, evaluatenovel design concepts, and validate models [50].Future embodiments will involve higher levels ofintegration and MEMS devices for sensors andactuators.

EQUIPMENT SUBSYSTEMS CONTROLMajor manufacturing tools are complex equipment

systems, comprising not only the reactor environmentin which processes modify the wafer, but alsoequipment subsystems for reactant delivery, exhaustmanagement, wafer transport, and other functionalities.Delivery and control of reactive gas species has been anotorious problem, e.g., relating to mass flowcontroller reliability.

More recently, the management of delivery andcontrol of liquid and solid sources has become moreimportant due to the profusion of new and complexmaterials needed to support the ITRS roadmap.Delivery of liquid metallo-organic CVD (MOCVDprecursors has been accomplished by standard bubblertechniques, in which an inert carrier gas is bubbledthrough the liquid source to take up its saturation vaporpressure of the source material for delivery to thereactor. More recently, liquid delivery systems havebecome common, in which the liquid is delivered tothe reactor and vaporized as it enters. Whilereproducible liquid delivery remains a challenge,perhaps the most difficult is the use of solid sourceswith low vapor pressure, where the precursor deliveryrate is destabilized by variability e.g., in temperature orin the active area of the powder source as the precursoris consumed. Indeed, because of such difficultiessuppliers have adopted the alternative of dissolvingsuch sources in organic solvents and then using liquiddelivery systems, but this raises new issues ofcontamination and chemical complexity.

We have applied acoustic sensing for real-timecontrol of such problematic precursors, particularly oneof the most challenging - solid Cp2Mg, with a very lowvapor pressure (0.04 torr at 25°C). This materials isimportant in optoelectronics materials growth. Thereal-time APC setup is shown schematically in Fig. 9.The inert H2 carrier gas draws precursor from the solidsource, which in turn is mixed with a H2 diluentstream, after which the composition of the mixture issensed by the acoustic sensor.

Carrier gas flow set point

50 g. Cp2Mgsolid source

FIGURE 9. System for real-time control of solid sourcedelivery using acoustic sensor.

_ 0.020-,

0.016

:§ 0.012:

"coot 0.008

0.004Q_O 0.000

oo

0 40 80 120 160 200Time (min)

60

40

FIGURE 10. System for real-time control of solid sourcedelivery using acoustic sensor.

Results in Fig. 10 show the performance of thisreal-time control system. With a low Cp2Mgconcentration (0.01 mol-%) as the target, the sourcetemperature was varied from 40 to 32°C, causing a50% decrease in vapor pressure, to simulate a majordisruption of the precursor delivery rate. With theacoustic sensor providing a real-time compositionmetrology and driving a controller for both H2 streams,precursor delivery rates can be controlled to within 1%of their targets.

This source delivery control strategy should bewidely applicable, not only to the most difficult ofprecursors like solid Cp2Mg, but to a broad range ofliquid and solid precursors as well. Such sensors canaccomplish both real-time course correction and faultdetection. And the ability to control delivery of exoticchemical precursors is particularly timely in this era of

589

new multicomponent materials for semiconductortechnology.

CONCLUSIONSAdvanced process control, only a vision a decade

ago, has now become a mainstream thrust forsemiconductor manufacturing. Its two areas ofprimary acceptance and success are in run-to-runcontrol and real-time fault detection. Run-to-runcontrol relies on continuing advances in in-line andoff-line metrology tools, whose technologyadvancement is already required to meet the futurenodes on the ITRS roadmap. In contrast, real-timefault detection demands real-time/in-situ sensors;however, high quantitative accuracy is not required,since the sensors are intended only to quickly identifyknown faults with serious consequences.

Improvements in quantitative accuracy of real-time, in-situ sensors opens the door to a fullcomplement of real-time APC. High accuracy sensorswhich resolve chemical or spectroscopic informationpromise to resolve subtle changes in process and toenable course correction in real time. Amidst thevarious available in-situ sensor technologies, we havedemonstrated that chemical sensors of process state canreveal wafer state consequences associated withreactions occurring in the process module, leading towafer state metrologies. For mass spectroscopy,acoustic sensing, and FTIR we have demonstratedthickness metrology with accuracy of 1% or better,certainly suitable for manufacturing requirements, andin the case of mass spectroscopy we have demonstratedits application in real-time end point control. Theprocess spectrum available with these sensors isconsiderably broader than the thermal CVD exampleaddressed here, and other in-situ approaches (e.g.,optical [51, 52]) add to the available sensor portfoliofor real-time APC.

With research to date indicating that in-situmetrology can support real-time course correction, it istime to move these efforts into a manufacturingenvironment. Our experience to date suggests thattheir performance will be even better because of thehigh volume conditioning which manufacturingequipment experiences.

New directions for research emphasis arenecessary and evident for in-situ metrology and real-time APC. One is certainly in development ofspatially resolved sensors which can determine across-wafer uniformity, or possibly in new concepts forspatially programmable reactor design. Another is inemploying in-situ metrologies for real-time APC inequipment subsystems, where manufacturingvariability and fault generation is often found. To thisend, the application of acoustic sensing to a broadrange of precursor delivery systems seems an excellentexample of focusing attention on the larger equipmentsystem, rather than solely around the wafer.

ACKNOWLEDGEMENTSThe author is extremely grateful to his

collaborators and students for their contributions andcooperation over some years of work on in-situmetrology and real-time APC, especially to LaurentHenn-Lecordier, John Kidder, Theodosia Gougousi,Yiheng Xu, Soon Cho, Charles Tilford, and in theearlier phases Laura Tedder and Guangquan Lu. He isalso appreciative of many discussions on sensors,metrology, and APC with B. Ellefson, L. Frees, C.Gogol, A. Wajid, S. Butler, G. Barna, A. Diebold, B.van Eck, and J. Hosch. We are grateful for financialsupport from NIST, NSF, SRC, Texas Instruments, andInficon.

REFERENCES[I] Melliar-Smith, M. and Diebold, A. C., "Metrology

Needs for the Semiconductor Industry Over the NextDecade," in Characterization and Metrology for VLSITechnology: 1998 International Conference, vol.CP449, Walters, E. J., Ed.: The American Institute ofPhysics, 1998, pp. 3-20.

[2] Diebold, A. C., Solid State Technology, 45, 48, Jan2002

[3] Butler, S. W.,J. Vac. Sci. Technol. B, vol. 13, pp. 1917-1923, 1995.

[4] Butler, S. W., Hosch, J., Diebold, A. C., and Van Eck,B.,2,pp. 315-321, 1997.

[5] Barna, G. G., Loewenstein, L. M., Robbins, R., Obrien,S., Lane, A., White, D. D., Hanratty, M., Hosch, J.,Shinn, G. B., Taylor, K., and Brankner, K.Jeee TSemiconduct M, vol. 7, pp. 149-158, 1994.

[6] Barna, G. G., Loewenstein, L. M., Brankner, K. J.,Butler, S. W., Mozumder, P. K., Stefani, J. A., Henck,S. A., Chapados, P., Buck, D.? Maung, S., Saxena, S.,and Unruh, A.,J. Vac. Sci. Technol. B, vol. 12, pp.2860-2867, 1994.

[7] Maung, S., Banerjee, S., Draheim, D., Henck, S., andButler, S. W.Jeee T Semiconduct M, vol. 7, pp. 184-192, 1994.

[8] Moslehi, M. M., Davis, C., and Bowling, A., pp. 44-63,1992.

[9] Sofge, D. A., "Virtual Sensor Based Fault Detection andClassification," 1995.

[10] Diebold, A. C., "Metrology Roadmap: A Supplement tothe National Technology Roadmap forSemiconductors," Sematech, Austin, TX TechnologyTransfer #94102578A-TR, 1995.

[II] Diebold, A. C., presented at 1995 IEEE/SEMIAdvanced Semiconductor Manufacturing Conference,Gaithersburg, MD, 1995.

[12] Diebold, A. C., presented at SemiconductorCharacterization: Present Status and Future Needs,Gaithersburg, MD, 1995.

[13] Chism, W., Diebold, A., Canterbury, J., and Richter,C.,Sol St Phen, vol. 76-77, pp. 177-180, 2001.

[14] Advanced Equipment Control / Advanced ProcessControl (AEC/APC) conference,http: //www. sematech. or g/publ i c/ne ws/conferences/aecapc/index.shtml

[15] AVS Manufacturing Science and Technology Group,www.avs.org

[16] Integrated Measurement Association,www. integratedmeasurement. com

590

[17] Butler, S. W., Solid State Technology, 42, 33-+, JAN[18] Butler, S. W.,J. Vac. Sci. Technol. B, vol. B13, pp.

1917-1923, 2000.[19] Helms, B., presented at AEC/APC Symposium XII,

2000.[20] Reath, M., presented at AEC/APC Symposium XII,

2000.[21] Lee, S. F., Boskin, E. D., Liu, H. C., Wen, E. H., and

Spanos, C. J.JEEE Trans. Semi. Mfg, vol. 8, pp. 17-25,1995.

[22] Sanders, M., presented at AEC/APC Symposium XII,2000.

[23] Stefani, J., Rao, S. P., Doke, N., Larsen, J., Comstock,S., and Paquette, B., presented at AEC/APC SymposiumXII, 2000.

[24] Sruber, J. D., Pagette, F., and Tang, S., presented atAEC/APC Symposium XII, 2000.

[25] Toprac, A. J., Campbell, W. J., and Purdy, M., presentedat AEC/APC Symposium XII, 2000.

[26] Stefani, J., Poarch, S., Saxena, S., and Mozumder, P.K.JEEE Trans. Semi. Mfg, vol. 9, pp. 366-383, 1996.

[27] Toprac, A. J., Sonderman, T., Peterson, A. O., andMiller, M., presented at AEC/APC Symposium XII,2000.

[28] Passow, M., presented at AEC/APC Symposium XII,2000.

[29] Atkins, N., presented at AEC/APC Symposium XII,2000.

[30] Sonderman, T. J., presented at AEC/APC SymposiumXII, 2000.

[31] Tedder, L. L., Rubloff, G. W., Shareef, I., Anderle, M.,Kim, D. H., and Parsons, G. N.,J. Vac. Sci. Technol. B,vol. 13, pp. 1924-7, 1995.

[32] Chowdhury, A. I., Read, W. W., Rubloff, G. W.,Tedder, L. L., and Parsons, G. N.,J. Vac. Sci. Technol.B,vol. 15, pp. 127-32, 1997.

[33] Srinivasan, E. and Parsons, G. N.,J Appl Phys, vol. 81,pp. 2847-2855, 1997.

[34] Tedder, L. L., Rubloff, G. W., Cohaghan, B. F., andParsons, G. N.,J Vac Sci Technol A, vol. 14, pp. 267-270, 1996.

[35] Sreenivasan, R., Gougousi, T., Xu, Y. H., Kidder, J.,Zafiriou, E., and Rubloff, G. W.,J. Vac. Sci. Technol. B,vol. 19, pp. 1931-1941,2001.

[36] Cho, S., Henn-Lecordier, L., Liu, Y., and Rubloff, G.W.,J. Vac. Sci. Technol. B, vol., (submitted).

[37] Xu, Y., Gougousi, T., Henn-Lecordier, L., Liu, Y., Cho,S., and Rubloff, G. W.,J. Vac. Sci. Technol. B, vol. 20,pp. 2351-2360, 2002.

[38] Lu, G. Q., Bora, M., Tedder, L. L., and Rubloff, G.W.,IEEE Trans. Semi. Mfg, vol. 11, pp. 63-74, 1998.

[39] Gougousi, T., Xu, Y., Kidder, J. N., Rubloff, G. W., andTilford, C. R.,18, pp. 1352-63, 2000.

[40] Henn-Lecordier, L., Kidder, J. N., Rubloff, G. W.,Gogol, C. A., and Wajid, A.,J. Vac. Sci. Technol. B,vol., (in press).

[41] Henn-Lecordier, L., Kidder, J. N., Rubloff, G. W.,Gogol, C. A., and Wajid, A.,J. Vac. Sci. Technol. B,vol. 19, pp. 621-6,2001.

[42] Singhal, A., Henn-Lecordier, L., Kidder, J. N., Gogol,C., and Kushneir, J. F., presented at AVS NationalSymposium, San Francisco, 2001.

[43] Chambers, J. J., Min, K., and Parsons, G. N.,J. Vac. Sci.Technol. B, vol. 16, pp. 2996-3002, 1998.

[44] Lazzeri, P., Vanzetti, L., lacob, E., Bersani, M.,Anderle, M., Park, J. J., Lin, Z , Briber, R. M., Rubloff,

G. W., and Miller, R. D., presented at 2003 Intl. Confon Characterization and Metrology for ULSITechnology, Austin, TX, 2003.

[45] Economou, D., Aydil, E. S., and Barna, G., Solid StateTechnology, 34, 107-111, APR

[46] Zhang, H. and Nikolaou, M., presented at AEC/APCSymposium XII, 2000.

[47] Stefani, J. and Butler, S. W.,J Electrochem Soc, vol.141, pp. 1387-1391, 1994.

[48] http://www.isr. umd. edu/Labs/C AC S E/researcn/progrxr/[49] Liu, Y., Choo, J., Henn-Lecordier, L., Rubloff, G. W.,

and Adomaitis, R. A., presented at AVS 49th NationalSymposium, Denver, CO, 2002.

[50] Choo, J.-O., Adomaitis, R. A., Rubloff, G. W., Henn-Lecordier, L., and Liu, Y., presented at AIChE AnnualMeeting, Indianapolis, IN, 2002.

[51] Huang, H.-T., Lee, J.-W., Stutzman, B. S., Klimecky,P., Garvin, C., Khargonekar, P. P., and Terry, F. L.,presented at AEC/APC Symposium XII, 2000.

[52] Hosch, J. W., Benson, K., Celii, F., and Harvey, K. C.,presented at AEC/APC Symposium XII, 2000.

591