A comparative study of dopant activation in boron, BF2, arsenic, and phosphorus implanted silicon

of 9

Please download to get full document.

View again

All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
PDF
9 pages
0 downs
2 views
Share
Description
A comparative study of dopant activation in boron, BF2, arsenic, and phosphorus implanted silicon
Tags
Transcript
  IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 49, NO. 7, JULY 2002 1183 A Comparative Study of Dopant Activation in Boron,BF    , Arsenic, and Phosphorus Implanted Silicon Ali Mokhberi, Peter B. Griffin, James D. Plummer  , Fellow, IEEE  , Eric Paton, Steve McCoy, andKiefer Elliott  , Senior Member, IEEE   Abstract— Ultra-low energy implants were used in combi-nation with rapid thermal anneals in the temperature range900 C–1050 C to study dopant activation in silicon. First,relatively long time anneals were performed in a conventionaltungsten-based RTA to investigate the activation mechanisms.The activation was monitored using Hall measurement, wherethe rate of electrical activation was considered by measuring thetime it takes to reach 50% activation. Using Arrhenius fits, anactivation energy was extracted, and it was found that while boronhas a mean activation energy for electrical activation of 4.7 eV inagreement with previous studies, arsenic and phosphorus havethermal activation energies of 3.6 eV and 4.1 eV, respectively. The4.7 eV activation energy for boron is believed to be related to apoint defect driven mechanism for electrical activation. Electricalactivation of arsenic and phosphorus, however, seems to be relatedto dopant diffusion. In the second set of experiments, an arc lampsystem was utilized to perform ultra-sharp spike anneals and toanalyze the effect of both ramp-up and ramp-down on boron andarsenic activation and diffusion. For both dopants, it was foundthat for a given temperature, there is an optimum ramp-rate thatproduces the desired dopant activation and junction depth.  Index Terms— Diffusion processes, doping, impurities, semicon-ductor process modeling. I. I NTRODUCTION T HE INTERNATIONAL Technology Roadmap for Semi-conductors [1] projects that for the 100-nm technologynode, the drain extension junction depth and sheet resistanceshould be in the range 20–33 nm and 200–625 / . Currently,the preferred method of introducing dopants into silicon is ultralowenergyionimplantation.However,therearetwomainprob-lems associated with this technique. First, the very high super-satuation of interstitials results in transient enhanced diffusion(TED). The second problem is dopant clustering or precipita-tion, which leads to solubility levels that severely limit the elec-trical activation that can be achieved. These challenges need tobeovercomeinordertosuccessfullycontinuethescalingtrends.In this paper, we present studies of boron, arsenic and phos-phorus electrical activation after implant and annealing. The Manuscript received January 21, 2002; revised March 19, 2002. This work was supported by the Semiconductor Research Corporation, Durham, NC. Thereview of this paper was arranged by Editor J. Vasi.A. Mokhberi, P. B. Griffin, and J. D. Plummer are with the Center for Inte-grated Systems, Stanford University, Stanford, CA 94305-4070 USA.E. Paton is with Advanced Micro Devices, Sunnyvale, CA 94088 USA.S.McCoyandK.ElliottarewithVortekIndustries,Vancouver,BCV5L1M5,Canada.Publisher Item Identifier S 0018-9383(02)05958-0. paper includes our experimental results, where we have mea-sured the activation energy of each dopant’s rate of electricalactivation and in the case of boron compared them to publishedresults. Furthermore, we have looked at the effect of ramp-upand ramp-down rates in spike anneals on dopant diffusion andactivation, where an arc lamp based system was used to achievean ultra-sharp spike.II. C HARACTERIZING  A CTIVATION  M ECHANISMS  A. Experimental Setup The implants considered for this experiment were boron,BF , arsenic, and phosphorus. All implants had a nominaldose of 1 10 cm and were made into bare silicon. Theimplant energies used are listed in Table I. The wafers werebroken into 1 1 cm pieces and capped with oxide using alow-pressure chemical vapor deposition at 350 C. The sampleswere then annealed at four different temperatures in the range900 C–1025 C for various times using an AG4108 rapidthermalannealer.Theramp-upusedwas100 C/sandattheendof the anneal, the lamps were shut down to allow the samplesto cool at a rate measured to be 70 C/s. The oxide was thenetched using dilute hydrofluoric acid, and dopant activationwas measured through Hall measurement. Finally, secondaryion mass spectrometry (SIMS) was performed on selectedsamples from the 1-keV boron set. These measurements weredone using an O primary beam and oxygen leak.  B. Method of Analysis The overall approach to analyzing the rate of dopant elec-trical activation was as follows. For each dopant, the active dosefrom Hall measurements was plotted versus time for each tem-perature, and the time to reach 50% electrical activation, called,wasextracted.Thisquantitywasthenplottedversusand the activation energy for electrical activation was extractedusing Arrhenius fits. However, before presenting the results,there are several issues that need to be addressed.One of the difficulties associated with Hall measurement isin the interpretation of the measured quantities, the Hall coef-ficient , and the conductivity . From these two quantities,one can calculate mobility and active dose, where based on asimple first-order approach, the relations are(1)or (2) 0018-9383/02$17.00 © 2002 IEEE  1184 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 49, NO. 7, JULY 2002 TABLE IL IST OF  I MPLANT  E NERGIES AND  M EASURED  H ALL  D OSES Fig. 1. Concept of effective time: Using (8), one can convert an anneal wheretemperature changes with time (profile 1) to an anneal at constant temperature.The conversion, however, depends on the process under consideration.Processes with different activation energies yield different effective times(profiles 2 and 3), where   decreases with increasing    . However, these relations neglect the statistical variations in thevelocities of free carriers. In general, this is taken into accountby introducing the Hall factor (or Hall ratio)(3)or (4)where depends on the relaxation time of the carriers, whichin turn depends on the various scattering mechanisms and canbe theoretically calculated from (see [2])(5)This calculation is, however, not an easy task, and in practice,becomes dependent on the material under study, carrier type,measurement temperature, and the magnitude of the magneticfield. In the case of boron, for example, published values rangefrom0.6to0.9formeasurementsmadeatroomtemperature[3],[4].Forhighboronconcentrations( 5 10 cm )measuredat room temperature, the Hall ratio is around 0.7.Ignoring the Hall ratio would mean that the measured Halldose could be up to 1.6 higher than the actual active dose. Inorder to minimize Hall measurement errors due to uncertaintiesinthevalueoftheHallfactor,weadoptedanempiricalapproachto measure the Hall factor. One sample of each implant wasannealedat1100 Cfor6htocompletelyactivatetheimplanteddopant,and themeasuredHalldosewasusedasanormalizationreference. Table I summarizes the measured electrical Hall doseof these samples. These measurements are within 15% of thedose measured by profile integration from SIMS, which servedas a second check on the implanted dose.The next issue in our analysis is the determination of an-neal times which must consider ramp-up and ramp-down times.Since dopantactivation is afunction of temperature, ramp timescannot be directly added to the anneal time and need to be av-eraged accordingly. For this purpose, we have applied the sametechnique used by Fiory  et al.  [5]. For any process that has asingle dominant activation energy , we can define a dimen-sionless time-dependant process variable(6)where is the normalization factor(7)is a reference temperature and is the tempera-ture–time profile. The effective process time at the referencetemperature can then be calculated by simply integratingover time(8)Equation (8) provides us with a convenient way to convert ananneal where the temperature is a function of time to an equiv-alent or effective anneal at the constant temperature . Forexample, if we choose to be the maximum temperature inthe anneal cycle, (8) will give us the equivalent anneal time atthe peak temperature. Note, however,that the effectivetime cal-culation depends on the activation energy of the process underconsideration through , and therefore, if two different pro-cesses have different activation energies, the effective times forthesetwoprocesseswillalsodiffer.Thisisshownschematicallyin Fig. 1.Inthecurrent context,weareinterested inacalculation oftheequivalent anneal time at the peak temperature, and thereforewill choose to be the peak temperature for each anneal.To be consistent with Fiory’s nomenclature, we will call thisthe  cycle-time . Furthermore, we are interested in the process of dopant electrical activation, the activation energy of which isinitially unknown, and is the quantity we want to extract. Inother words, this will be an iterative process, where we willassume a value for , process the data, extract , and repeatthe process until we obtain a stable value.When isextractedaccordingto(8),therampsaddapprox-imately an additional 1 s to the total anneal time. Obviously, 1 sis negligible when considering a 5-min anneal. However, it be-comes important as the time at the peak temperature decreases.Once the effective anneal time was extracted for each set of samples, the normalized Hall dose was plotted versus annealtime for the various anneal temperatures. An example of thisis shown in Fig. 2 for the 1-keV boron and arsenic samples.Using these plots, the anneal time required for 50% electricalactivation,referred toas was extractedat eachtemperaturebylogarithmicextrapolationandwasplottedagainst .Thisallowed us to extract the thermal activation energy for the rateof electrical activation. Notice that this is the value used in(8), and as mentioned earlier, the procedure requires iteration toarrive at the final value.  MOKHBERI  et al. : COMPARATIVE STUDY OF DOPANT ACTIVATION 1185 Fig. 2. Normalized Hall dose versus cycle time for 1-keV boron and 1-keV arsenic implants. As mentioned in the text, cycle time is the effective time at the peak temperature. Lines are logarithmic fits used to extrapolate    . Error bars are based on 10% uncertainty in Hall measurement.Fig. 3. Extraction of activation energy for the rate of activation for the four sets of implants considered. (a) Boron, (b) BF , (c) Arsenic, and (d) Phosphorus.Average    is extracted by fitting all the data with an Arrhenius line; solid lines have average    as slope. The main source of error is 10% uncertainty in Hallmeasurements, which translates to about 35% uncertainty in    . C. Results and Discussion Fig. 3(a)–(d) shows the plots of versus for eachof the four dopants. Our results for boron agree with previouslypublished results by Seidel [6] and Fiory [7] and show a meanactivation energy of 4.7 eV. It should be noted that Privitera et al.  [8] report activation energies in the range 2–3 eV, andMayur  et al.  [9] report a value of 2.0 eV. We believe the reasonfor the discrepancy is in the extraction method. Privitera  et al. assume that there is a saturation value for the activation and fitthe experimental data to the function(9)where is the active dose, and is a function of annealtemperature and represents the saturation value for the activedose.Inourexperiment,thehighesttemperatureannealsyielded100% activation, and although we did not perform long enoughanneals to study this behavior in the lower temperature regime,Seidel’s results [6] indicate that boron will reach 100% activa-tion at temperatures as low as 800 C if annealed for a longenough time. Thus, a model which assumes the activation levelsaturates at a temperature dependant value will predict a loweractivation energy.Mayur  et al.  extract by considering anneals at three dif-ferent temperatures, namely 950 C, 1050 C, and 1075 C.  1186 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 49, NO. 7, JULY 2002 However,thespecificannealconditionsuseddonotproducethesame electrical activation. Since the experimental matrix doesnot include enough time points to use extrapolation, the authorsextract bynormalizingtheelectricallyactivefractionateachanneal temperature to its cycle-time, and plotting this normal-izedvalueversus .Thisapproachassumesthatthetimede-pendence of electrical activation is linear, and as our data fromFig. 2 indicates, the dependance is clearly not linear, but loga-rithmic. In fact, if we use this particular approach on our datafrom Fig. 2(a) by choosing one anneal time from each temper-ature, depending on which anneal time is chosen the activationenergy could vary between 0.3 eV to 5.6 eV.From Fig. 3(b), we see that BF implants seem to have aboutthe same activation energy as boron, suggesting that the acti-vation process could be the same. However, comparison be-tween Fig. 3(a) and (b) indicates that anneal time required toreach a certain activation level, e.g., 50%, is about three timeslonger for BF implants than for boron implants. This is mostlikelyduetothepresenceoffluorine,whichwebelievebondstoboron atoms, causing less activation, but also a shallower junc-tion [10].Fig. 3(c) and (d) show that the activation energies for arsenicand phosphorus are about 1 eV lower than that of boron sug-gesting a different activation mechanism. The activation ener-gies of 3.6 eV and 4.1 eV for the rate of activation of As andP respectively are very close to the thermal activation energy of diffusion [11], suggesting that the processes of activation anddiffusion are the same. On the other hand, the 4.7 eV activa-tion energy for boron electrical activationisgreater thanits 3.46eV activation energy for diffusion [11], which indicates that theactivation process is linked to point defect diffusion which hasbeen reported to have an activation energy of 4.8 eV [12], [13].The problem of boron clustering has been studied exten-sively both experimentally and theoretically in the past fewyears [14]–[17], and although there are significant differencesin the details of the explanations proposed, there seems tobe a general agreement about the nature of the clustering.Experimental evidence suggests that boron clustering happensonly in the presence of a high supersaturation of interstitials[16]. Furthermore, it is believed that these clusters are of theform B I , where and are the number of boron andinterstitial atoms in the cluster respectively. Not only are thesedefects believed to be the source of interstitials driving boronTED [18], but they also remain in the system long after TED isover resulting in deactivated boron. It is speculated that there isa range of defects present in the silicon system although sometheoretical studies [15] suggest that after TED is over only afew defect sizes, such as B I, are stable.The final part of the process is defect dissolution and conse-quent boron activation, which is the focus of this paper. In somestudies [16], defect dissolution is presumed to occur by the ad-dition of a thermally generated interstitial to the BIC cluster,releasing an active boron atom. Alternatively, a thermally gen-erated vacancy might cause the break-up of a BIC cluster, re-leasing active boron. In either case, the rate of boron activa-tion would depend on the rate of native point defect formationandmigration(i.e.,ontheactivationenergyofsiliconself-diffu-sion). The approach of this experiment does not clarify whether TABLE IIC OMPARISON OF  F OUR  A NNEALS  T HAT  Y IELD THE  S AME  A MOUNT OF  A CTIVE B ORON . U SING  (8), A LL  A NNEALS  A RE  C OMPARED  U SING  910 C  AS THE R EFERENCE  T EMPERATURE .    I S  4.7 eV  AND  3.46 eV  FOR THE  E FFECTIVE A CTIVATION  T IME AND  E FFECTIVE  D IFFUSION  T IME , R ESPECTIVELY the boron-interstitial-cluster break-up is mediated by silicon in-terstitials, vacancies, or both, and further work is needed to in-vestigate this. We do note, however, that the sum of interstitialor vacancy formation and migration energies are approximatelyequal to 4.7 eV [13], so either point defect mechanism is pos-sible for boron activation.Fig. 3 also illustrates another interesting feature. For alldopants, the plots indicate that the activation energy of isindependent of implant energy (i.e., the data for each implantenergy has the same slope). This indicates that the activationmechanism is independent of implant energy at least in therange considered in this experiment. However, the same cannot be said about the activation time itself, where an energydependence is observed in some cases.The difference between the thermal activation energy of dopant electrical activation and dopant diffusion for boron hasan important implication from a technological point of view. Inthe case of arsenic or phosphorus, the two activation energiesbeing the same implies that other variables being constant, itis irrelevant what anneal time–temperature is used once theoverall product is constant. One will always be on thesame – trade-off curve. In other words, a spike anneal ata high temperature will produce the same – value as asoak anneal at a lower temperature if the product of bothanneals is constant. Thus, in building a device, the terminalcharacteristics would remain the same because the parasiticresistance does not change. On the other hand, the 1 eV dif-ference between boron activation and boron diffusion suggeststhat it is beneficial to use a higher-temperature shorter-timeanneal to maximize activation and minimize diffusion.We can analyze this quantitatively with the help of (8). Forexample consider the following two anneals: a) 910 C 220 sand b) 1010 C 5 s. Fig. 2(a) indicates that these two annealsresult in the same activation level for the 1-keV boron sample.If we include the ramp times, the  effective activation time  of both anneals at a reference temperature of 910 C is 221 s[ eV and C in (8)], and therefore it isnot surprising that both anneals produce the same amount of electrically active boron. Next, we can calculate the  effectivediffusion time  at 910 C, i.e., eV and Cin (8). The result is 221 s and 90 s for anneals a) and b)respectively. In other words, while from a dopant activationstandpoint, the two anneals are equivalent, from a diffusionstandpoint, anneal b) is equivalent to a 910 C 90 s anneal,and hence should produce a shallower junction than anneala), which is a 910 C 221 s anneal.  MOKHBERI  et al. : COMPARATIVE STUDY OF DOPANT ACTIVATION 1187 Fig. 4. (a) SIMS profiles and (b)    –    data for boron 1-keV (A)–(D) and arsenic 1-keV (E)–(F) samples. The anneal conditions are (A) 910 C 220 s, (B)945 C 60 s, (C) 960 C 30 s, (D) 1010 C 5 s, (E) 945 C 30 s, and (F) 1025 C 2 s. For each dopant, the anneal conditions yield about the same activation andhence same sheet resistance. The data clearly show the effect of 1 eV difference between the activation energies of electrical activation and diffusion for boron,where higher temperature, shorter time anneals results in improved junction depth. For arsenic, the activation energies of diffusion and electrical activation areabout the same, and hence, it is irrelevant what anneal temperature–time is used once the overall   product is constant. For comparison, arsenic data from [19] arealso included in (b), where the various spike or soak anneals were examined with a 1-keV 1 2   10 cm As implant. Notice that changing the anneal conditionsonly result in a movement along the same trade-off curve.TABLE IIIS UMMARY OF  S PIKE  A NNEALS  C ONSIDERED AND  C ORRESPONDING S HEET  R ESISTANCE  F ROM  F OUR -P OINT  P ROBE  M EASUREMENTSFOR  B ORON AND  A RSENIC The previous calculations, together with calculations on the945 C 60 s and 960 C 30 s samples (which have the sameactivation levels) are summarized in Table II. Secondary ionsmass spectroscopy performed on these four samples confirmsthese calculations. These results are illustrated in Fig. 4, whichalso includes arsenic data for comparison. As the plot indicates,increasing anneal temperature and decreasing anneal time forboron results in improved junction depth while the electricalactivation remains level. However, the arsenic data clearlyshow that changing the anneal time and temperature with theproduct kept constant does not change the – data.III. R AMP  R ATE  S TUDIES Althoughtoachieveultrashallowjunctions,anewapproachisneeded in the long term, currently the immediate scaling needsare met by optimizing the implant and anneal conditions [20].In particular, the thermal cycle is optimized by minimizing thesoak time at the peak temperature and maximizing ramp rates,which is referred to as a spike anneal [21]. It is believed that byminimizing the ramp time and time spent at peak temperature,dopant activation is achieved with minimum diffusion. Here weinvestigatetheeffectofbothramp-upandramp-downondopantdiffusion and activation. In particular, we are interested in ana-lyzing the benefits/disadvantages of increasing the ramp rates.For this purpose, we have used an arc lamp RTA developed byVortek Industries [22]. The advantage of the arc lamp systemcompared to a traditional tungsten based system is its fast re-sponse time, because the lamps can be switched off in a fewmicroseconds. In addition, a black chamber is used to minimizeradiation return to the sample and thus allow for faster cooling.Overall, the system provides close to ideal spike ramps, and isperfect for an investigation of the technological limits of dopantimplant and annealing.  A. Experimental Setup Using 1 1 in samples of B 0.5 keV and As 1-keV, spike an-neals with a peak temperature of 1050 C, and ramp rates in therange 150–400 C/s (ramp-up) and 75–135 C/s (ramp-down)wereperformed.TableIIIsummarizesthedifferentanneals,andFig. 5 shows the ramp characteristics of the arc lamp system forsample 2 (B 0.5 keV). As the plots indicate, thesystem providesa very sharp spike and stable ramp rates. Following the anneals,a four point probe measurement was used to obtain the sheetresistance, and junction depths were measured by performingsecondary ion mass spectroscopy. Boron was profiled using a500 eV O beam at normal incidence, while a 1-keV Cs beamat 60 degrees from normal incidence was used for profiling ar-senic.  B. Results and Discussion Sheet resistance data is summarized in Table III, whichindicates a rapid increase with increasing ramp rates. Fig. 6(a)shows the results of the SIMS measurements for bothdopants annealed at three different ramp-up/ramp-down rates:150–75 C/s, 250–90 C/s, and 400–135 C/s. As the plotsindicate, there is virtually no difference in the junction depthbetween the 250–90 C/s and the 400–135 C/s boron samples.On the other hand, the arsenic junction depth seems to improvewhen using the maximum ramp rates. – data for thesesamples is plotted in Fig. 6(b), which clearly illustrates that notonly is there a saturation of junction depth for boron beyond250–90 C/s, but there is also a significant loss in dopantactivation, manifested in increased sheet resistance. This isexpected since there is a time factor involved in electricalactivation, and the anneal time is reduced as the ramp rates areincreased.
Related Search
Advertisements
Similar documents
View more...
Advertisements
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks