Anda di halaman 1dari 13

A Simple Semiempirical Short-Channel MOSFET CurrentVoltage Model Continuous Across All Regions of Operation and Employing Only hysical

arameters
Ali !ha"ifiroo#$ Member, IEEE$ Osama M% &ayfeh$ Member, IEEE$ and 'imitri Antoniadis$ Fellow, IEEE

AbstractA simple semiempirical model ID (VGS , VD S ) for short-channel MOSFETs applicable in all regions of device operation is presented. The model is based on the so-called topof- the-barrier-transport model and !e refer to it as the virt"al so"rce #$S% model. The simplicit& of the model comes from the fact that onl& ten parameters are "sed. Of these parameters si' are directl& obtainable from standard device meas"rements( )% gate capacitance in strong inversion conditions #t&picall& at ma'im"m voltage VGS * Vdd %* +% s"bthreshold s!ing* ,% drain-ind"ced barrier lo!ering #-./0% coefficient* 1% c"rrent in !ea2 inversion #t&picall& Io at VGS * 3 $% and at high VD S * 4% total resistance at VD S * 3 $ and VGS * Vdd and 5% effective channel length. Three fitted ph&sical parameters are as follo!s( )% carrier lo!field effective mobilit&* +% parasitic so"rce6drain resistance ,% the sat"- ration region carrier velocit& at the so-called virt"al so"rce. 0astl& a constrained sat"ration-transition-region empirical parameter is also fitted. The modeled c"rrent vers"s voltage characteristics and their derivatives are contin"o"s from !ea2 to strong inversion and from the linear to sat"ration regimes of operation. 7emar2able agreement !ith p"blished state-of-theart planar short-channel strained devices is demonstrated "sing ph&sicall& meaningf"l val- "es of the fitted ph&sical parameters. Moreover the model allo!s for good ph&sical insight in device performance properties s"ch as e'traction of the $S$ !hich is a parameter of critical technologi- cal importance that allo!s for contin"ed MOSFET performance scaling. The simplicit& of the model and the fact that it onl& "ses ph&sicall& meaningf"l parameters provides an eas& !a& for technolog& benchmar2ing and performance pro8ection. Index Terms9MOS scaling inversion charge densit& MOSFET compact modeling virt"al so"rce velocit&.

online at http?AAieee0plore%ieee%org% 'igital O-.ect +dentifier 14%1148ATE'%3448%343B433

+% +&TRO ',CT+O& OSFET compact modeling has -een the su-.ect of a /ast amount of technical literature$ and models of /arious degrees of comple0ity and accuracy e0ist% Conceptually$ compact

Manuscript recei/ed Octo-er 12$ 34456 re/ised April 37$ 3448% Current /ersion pu-lished 9uly 33$ 3448% This :or" :as supported -y the Semiconductor Research Corporation Focus Center Research rogram under the Center for Materials$ Structures and 'e/ices% The re/ie: of this paper :as arranged -y Editor R% ;uang% A% !ha"ifiroo# :as :ith Microsystems Technology <a-oratory$ Massachusetts +nstitute of Technology$ Cam-ridge$ MA 43128 ,SA% ;e is no: :ith +=M Research at Al-any &anotech$ Al-any$ &> 13342 ,SA (email? "ha"i@us%i-m%com)% O% M% &ayfeh and '% Antoniadis are :ith the Microsystems Technology <a-oratories$ Massachusetts +nstitute of Technology$ Cam-ridge$ MA 43125 ,SA% Color /ersions of one or more of the figures in this paper are a/aila-le

models can -e considered to consist of a multile/el hierarchy% At the first le/el are the carrier charge and transport models$ in :hich some parameters are represented -y second-le/el models of geometry-dependent electrostatics and e/en -y third-le/el models that capture layout dependencies% Chile the ma.ority of compact models are concerned :ith computeraided-design applications and :ith accurate and detailed fittings to de/ice characteristics o/er a range of geometry and layout features$ the model presented in this paper -elongs strictly to the first le/el :ith the goal of pro/iding a simple and intuiti/e understanding of the underlying carrier transport in modern short-channel planar MOSFETs :ith the capa-ility$ for e0ample$ of e0tracting the /irtual source carrier /elocity% The Dne:E semiempirical model descri-es the short-channel MOSFET current /ersus /oltage characteristics and is /alid in all regions of opera- tion$ :ith continuity of -oth current and its deri/ati/es% To demonstrate the model$ this paper also presents some e0ample fittings to state-of-the-art MOSFETs% The model is /ery simple and reasona-ly accurate and is -ased on the physics of short- channel de/ice transport :ith only physical Fuantities that are either "no:n$ e%g%$ gate length$ or are easily o-taina-le from standard output and transfer characteristics% Chile much of the parameter reduction in this

model compared to standard com- pact models comes primarily from the fact that this is only a first le/el model$ still$ -y adopting a rather simple physical concept as its -asis$ e/en at this le/el$ the model has inherent simplicity and reFuires fe: fitting parameters% The "ey /alue of the model is that it allo:s e0traction of the so-called /irtual-source carrier /elocity$ :hich is a parameter of great technological importance G1H that cannot -e o-tained /ia direct measurement% Moreo/er$ the fact that the model is -ased only on a limited num-er of physically meaningful parameters allo:s for easy technology -enchmar"ing and performance pro.ection G1H% The term Dne:E here is used :ith some hesitation -ecause$ practically$ all ideas used in the model ha/e -een discussed in one form or another in the /ast topical literature$ -ut to the -est of our "no:ledge$ they ha/e ne/er -een com-ined in the :ay done in this paper% For reasons that :ill -ecome o-/ious in Section ++$ :e refer to this model as the /irtual source (VS) model% ++% VS M O'E< + & SAT , RAT + O & +n the Dcharge-sheet appro0imation$E the drain current normali#ed -y :idth (ID /W ) of a MOSFET can -e descri-ed -y

!;A!+F+ROOI et al.? S;ORT-C;A&&E< MOSFET MO'E< CO&T+&,O,S ACROSS A<< REJ+O&S OF O ERAT+O&

1K7L

:here Cinv is the effecti/e gate-to-channel capacitance per unit area in strong in/ersion$ t is the thermal /oltage (kB TGS /q )$ V is the internal gatesource /oltage$ i%e%$ corrected for the /oltage drop on the source resistance RS and is gi/en -y V GS * VGS ID RS $ n is the su-threshold coefficient$ :hich is related to the so-called Dsu-threshold s:ingE -y S * nt ln 14$ and VT * VT 0 V DS (2)

Fig% 1% Virtual source point !o in the channel of a modern shortchannel MOSFET% The carrier charge and /elocity$ used in (1) and through this :or"$ are defined at this point (at the pea" of the conduction -and profile)%

:here VT 0 is the strong-in/ersion threshold /oltage at VDS * 4$ V DS accounts for the /oltage drop on -oth RS and RD (drain resistance) as V DS * VDS ID (RS M RD )$ and is the '+=< coefficient in V /V % &ote that '+=< is the only term that introduces dependency of "#!o on VDS % The term follo:ing VT in (3) allo:s for the reFuirement of different /alues of threshold /oltage (-etter stated as Dreference /oltageE) in strong and :ea" in/ersion$ as discussed at length in the literature$ e%g%$ GLHG7H% +t :as found empirically in this :or" that a shift of VT -y 2%Lt is a /ery good appro0imation$ and hence$ % * 2%L in (3) is used in this :or"% The follo:ing Din/ersion transitionE function F& is a Fermi function that allo:s for a smooth transition -et:een the t:o /alues of reference /oltage and is centered at the point half:ay -et:een them? F& * 1 1 M e0p
V
GS

the product of the local charge areal density times the local carrier /elocity any:here in the channel% +t is particularly useful to :rite this e0pression at the location of the D/irtual source$E i%e%$ at the location of the top of the energy -arrier (! * !o ) -et:een the source and channel (see Fig% 1) -ecause the channel charge density there "#!o $ is easiest to model G3H$ G2H$ i%e%$ ID /W * "#!o $!o . (1)

For short-channel de/ices$ the /irtual source /elocity$ $!o is related to the so-called unidirectional thermal /elocity$ or -allistic /elocity $' $ as has -een discussed at length else:here G3H$ G2H% +t turns out$ as :ill -e sho:n later in this paper -y direct comparison :ith measured data in the literature$ that $!o is :ea"ly dependent on either VGS or VDS $ pro/ided that the de/ice is -iased in saturation (in either strong or :ea" in/ersion)% +n fact$ it has -een sho:n that in the saturation regime$ the -allistic /elocity and the in/ersion charge density at the /irtual source Ge0cept for drain-induced -arrier lo:ering ('+=<) dependenceH are independent of VDS G2H% +t is further demonstrated that e/en at the non-allistic regime in/ersion charge density at the top of the -arrier "#!o and carrier /elocity at this point $!o are almost independent of VDS if the de/ice operates in the saturation region% On the other hand$ :hile the -allistic /elocity increases monotonically :ith VGS G2H$ Monte Carlo simulations demonstrate that the /irtual source /e- locity is almost constant at high VGS G1H since carrier scattering also increases :ith the gate /oltage% &ote that instead of using the -allistic /elocity along :ith the -ac"scattering coefficient to descri-e the ID VGS characteristics in saturation$ as done$ for e0ample$ in G2H$ :e opt to use the a/erage /elocity of carriers at the /irtual source $!o to a/oid the complication caused -y e0act dependencies of the -ac"scattering coefficient on de/ice parameters and -ias$ and the am-iguity of the kB T layer% The /irtual-source charge density can -e appro0imated Fuite closely -y the ne: empirical function in (3)% This e0pression allo:s for a continuous e0pression for the in/ersion charge density at the /irtual source from :ea" to strong in/ersion%

(VT %t /3) %t

(B)

The form of the e0pression (i%e%$ :ithout the ne: % term in the e0ponential) :as first proposed -y Cright GBH as follo:s?

Fig% 3 compares the appro0imate solution of (3) :ith % * 2%L to the e0act solution from a 1-' oisson solution (e%g%$ G7H) under the sheet-charge appro0imation and assuming nondegenerate conditions for different /alues of the -ody factor? ( * 3q)Si * /C+ % As can -e seen in the figure$ the appro0imation in (3) produces an error less than 1LN o/er se/en orders of magnitude of channel-charge num-er-density up to 13 2 a ma0imum of 3%L 14 cm $ :hich is a sufficient range for most MOSFET applications% Moreo/er$ note that the error in the strong-in/ersion region is -elo: LN for more technologically 1 2 reasona-le /alues of -ody factor in the range of 4%11 V / % +t :as found that the errors remained nearly unchanged :ith temperatures o/er a :ide range$ e%g%$ form 34 C to 144 C% Of course$ the /irtual source charge density of a shortchannel de/ice can only -e deri/ed from a 3-' oisson solu (VT V (3) "#!o * Cinv nt 1 M e0p GS %t F& ) ln nt

tion% +n the 3-' case$ the su-threshold s:ing is dictated -y a com-ination of doping and geometrical electrostatics% ;o:e/er$ the o/erall shape of the charge /ersus VGS characteristics does not change$ and comparisons :ith 3-' de/ice simulations$ including degenerate conditions (not sho:n here)$ indicate that (3)(B) still yield a good appro0imation to the charge density :ith errors no larger than that sho:n in Fig% 3% All that is reFuired is that Cinv at ma0imum VGS $ n$ and the /alue of the charge "#!o at a particular /alue of VGS in :ea" in/ersion -e all matched to the theoretically calculated (or simulated) charge% &ote that matching these three parameters to data uniFuely determines the /alue of VT % Similarly$ it :as found -y compar- isons to theoretical calculations that included the in/ersion layer Fuantum-mechanical effect G5H and reasona-le amounts of poly depletion G5H that (3)(B) are still an e0cellent appro0imation$ pro/ided the a-o/e three parameters are matched%

1K7K

+EEE TRA&SACT+O&S O& E<ECTRO& 'EV+CES$ VO<% LK$ &O% 5$ A,J,ST 3448

-ut it :as generally found in this :or" that e0cellent fits to measured modern MOSFET data$ as :ill -e discussed later$ can -e o-tained :ith - * 1%5 for nFETs and - * 1%B for pFETs% +t is interesting to note that typical /alues of - :hen (K) has -een used to model carrier /elocity saturation are 3 for electrons and 1 for holes G14H$ G11H% Returning to (K)$ it is noted that the deri/ati/e of this eFuation :ith respect to V DS at VDS * 4 is 1/VDSAT $ irrespecti/e of the /alue of - $ and therefore$ :e ha/e 1 .ID W .V
DS VDS =0

"#!o (VDS * 4)$!o V W DR SA


T

1
C min

. (7)

;ere$ the appro0imate sign is used -ecause of the assumption that the VDS dependency of -oth $!o and "#!o is negligi-le% W RC min is the :idth-normali#ed resistance of the channel at VDS * 4 (in ohm centimeters)$ :here it has its minimum /alue for any gi/en /alue of VGS $ i%e%$ W RC min * /C " #!o 00 (5)

Fig% 3% (3)

(a) Relati/e error of the appro0imate solution for charge using

:here /C * /G 3/ov is the effecti/e channel length o-- tained from the gate length$ accounting for source and drain o/erlap (/ov )$ 0 is the channel carrier effecti/e mo-ility at VDS * 4 V$ :hich$ of course$ is a function of VGS -ut is assumed constant to the first order$ and "#!o 0 " #!o (VDS * 4) is defined for con/enience% ,sing (5) in (7) determines the /alue of VDSAT * VD SAT, in strong in/ersion$ :hich is independent of - and only dependent on physical parameters$ i%e%$ VDSAT, * $!o /C 0 . (8)

:ith % * 2%L as compared to 1-' oisson solution under the chargesheet appro0imation% For comparison$ the /alue of VT 0 :as fitted to match the /alue of VGS at a charge density of 148 cm12 in the e0act and appro0imate solutions% The su-threshold s:ing :as calculated from the classical e0pression n * 1 M (/3 3F % C+ :as matched to the theoretically calculated 2 /alue at VGS * 1%3 V$ :hich :as set to 11%8 0 FAcm in the theoretical 2 calculation :hile the -ody factor ( (in V / )$ :as /aried -y changing doping from 16 19 3 14 to 14 cm1 % (-) Model /ersus e0act charge solution for D:orst caseE appro0imation for ( * 4%42%

To properly account for saturation in :ea" in/ersion$ VDSAT in (K) in that region should -e eFual to t % Chile this is an appro0- imation -ecause F, is not e0actly eFual to 1 e0p(VDS /t ) as it should -e in :ea" in/ersion$ the resulting error is less than a-out 14N for a range of - /alues from 1%B to 3%L (not sho:n)% Finally$ to allo: a smooth transition -et:een the strong- and :ea"-in/ersion /alues of the saturation /oltage$ a generali#ed form of the saturation /oltage is introduced -y employing again the in/ersion transition function F& $ defined -y (B)$ as follo:s? VD SAT * VDSAT, (1 F& ) M t F& . ID /W * "#!o $!o F, V /VDSAT
1 /-

+++% &O&SAT,RAT+O& To account for the nonsaturation region$ the /elocity $!o in (1) is multiplied -y a Dsaturation functionE F, $ :hich increases smoothly from 4$ at V DS * 4$ to 1$ at V DS 2 VDSAT $ :here VD SAT is the saturation /oltage% EFuation (1) is then general- i#ed to (L)$ gi/en asDS follo:s$ :hich is /alid o/er all F, * regions of operation?

(14) (L)

(K)

+t can easily -e seen from (B) that F& in strong in/ersion tends to #ero$ :hile in :ea" in/ersion$ it tends to unity$ and thus$ VD SAT correspondingly /aries smoothly from VD SAT, 1 M (V DS /VDSAT )
-

This F, function is a"in to the carrier /elocity saturation function used else:here G8H$ G14H$ -ut here$ it ser/es to empirically reproduce the nonsaturation -eha/ior of the I V characteristics% At VDS * 4$ F, is 4$ :hile it -ecomes eFual to 1 for VDS VDSAT and$ hence$ allo:s the full /alue of $!o in (L)% +n principle$ - is a saturation-transition-region fitting parameter$

to t % Fig% 2 e0emplifies the -eha/ior of the model o/er the complete range of operation of a MOSFET% The smooth transition of the saturation /oltage from :ea" in/ersion to strong in/ersion is clearly e/ident% +t should -e emphasi#ed that (K) and (14) are heuristic% +t can -e easily sho:n from (1)$ (5)$ and (8) that VDSAT, is appro0imately eFual to the /alue of VDS :here the current through the resistor RC min is eFual to the saturation current$ i%e%$ it is determined -y the intercept of the 1/W RC min and "#!o 0 $!o loci in the ID /W /ersus VDS plane% Chile this is appealing$ it is not physically rigorous% +n addition$ the form of

!;A!+F+ROOI et al.? S;ORT-C;A&&E< MOSFET MO'E< CO&T+&,O,S ACROSS A<< REJ+O&S OF O ERAT+O&

1K77

tion in this appro0imate form is entirely closed form$ :hich is useful for some applications$ -ut in Section V$ only the e0act e0plicit solution gi/en -y (1)(14) is used% V% COM
A R+SO& O F T ;E MO'E< TO 'ATA A & ' '+SC,SS+O&

E0ample fits to de/ice data of a modern unia0ially strained KL- and 23-nm node technologies G1BH$ G1LH are sho:n in Figs% B and L% As can -e seen$ the model description of the data is /ery good o/er a -road range and of -ias /alues$ and o/er t:o different technologies$ i%e%$ :ith poly-SiO& and high-k metal- gate stac"s% Ta-le + sho:s the de/ice parameters that can -e considered as gi/en for this model$ i%e%$ Cinv $ Io (VGS Fig% 2% Model output characteristics as in Fig% B(a) in a semilog plot , VDS )$ S $ $ and /G % Since the data come from the literature$ to illustrate the continuity of the model from :ea" to strong in/ersion and a /alue for /ov * L nm had to -e assumedOthe same for from linear to saturation regions% -oth tech- nologies% This assumption does affect some:hat (K) is not fully physically .ustified$ e/en though it is -orro:ed the e0tracted /alue of 0$ as :ill -e discussed ne0t% The from the /elocity saturation e0pression G8H$ G14HOit is not e0tracted parameters are $!o $ W RS $ 0$ and - % As can -e .ustifia-le to assume that the lateral electric field is gi/en -y seen from the ta-le$ the e0tracted /alues are physically VDS //C under all conditions in strong in/ersion$ as actually reasona-le and come :ith the correct trend -et:een the t:o implied -y (K)% &e/ertheless$ as can -e seen in Section +V$ the technologies% The methodology for fitting can -e accomplished manually fitting of the VS model to real de/ice data is /ery good and :ith little iteration$ although standard optimi#ation methods comes :ith physically .ustifia-le parameters% can also -e used% First$ the capacitance Cinv is set to the strong-in/ersion /alue measured at VGS * Vdd $ :hich is +V% A ROP +MAT E EVA < ,AT + O & O F T ; E MO'E< typically reported% Second$ the su-threshold s:ing parameter n$ Io (typically at VGS * 4 V and VDS * Vdd )$ and the '+=< coefficient are o-tained directly from the measured transfer +t is clear from the fact that the internal /oltages V characteristics of the de/ice% &aturally$ in order to match the GS and VDS are reFuired in (L) and (K) that e0plicit solution of the calculated Io to the measured /alue$ an initial guess for $!o model eFuations reFuires iterations in order to o-tain self- is reFuired6 a good starting /alue is $!o * 7 cmAs$ and it 14 DS consistent /alues of VGS * VGS ID RS and V * VDS is then refined as descri-ed later% &ote that this process sets ID (RS M RD )% ;o:e/er$ iterations can -e a/oided the /alue of VT and esta-lishes the correct electrostatic -e:ith relati/ely little loss of accuracy -y simply replacing the ha/ior of the de/ice% Third$ from the ID /ersus VDS output internal /oltages in (3)(L) -y the applied e0ternal ones$ cur/e for VGS * Vdd $ the total resistance 3RS M RC min namely$ VGS and VDS $ :hile replacing the /irtual-source is it o-tained from the slope dID /dVDS |VDS =0 at VDS * 4 VO can typically -e assumed that RS * RD for symmetric /elocity in (L) -y the effecti/e /elocity $ $ :hich is gi/en -y? de/ices% At this point$ an assumption a-out /C and 0 is reFuired 1 F& in order to isolate and e0tract RS % A good typical /alue $ * F& M $!o . (11) for / ov is 4%1L the nominal short-channel /G $ and 0 1 M WRSC + (1 M 3 )$!o 2 2 344244 cm /V s for electrons and 0 1443L4 cm /V s As in the case of (14)$ the Fermi function F& $ :hich is gi/en in for holes in strained Si% Finally$ the /elocity $ is ad.usted !o (B)$ allo:s for a smooth transition of $ from a /alue of $!o in -y comparing model results in saturation :ith measured data :ea" in/ersion (F& 1)$ :here there is no degradation due to the presence of RS $ to the /alue that is appropriately degraded at VDS * Vdd and at /arious VGS /alues% The model is Fuite in strong in/ersion (F& 4)% The deri/ation of the second constrained$ and therefore$ relati/ely minor ad.ustments of the term in the right-hand side of (11) for the case of operation only in initial parameter /alues are typically reFuired or are$ indeed$ the strong-in/ersion region and$ therefore$ for F& * 4 has possi-le% Typically$ in nominal short-channel de/ices$ the par-een discussed in G13H and G12H% +n addition$ RS M RD asitic resistance dominates at the highest /alue of VGS $ and must -e added e0plicitly to the right-hand side of (5) so that therefore$ errors in the assumptions for / and 0 are not /ery ov the correct total resistance at VDS * 4 V can -e o-tained% critical for the e0traction of RS $ :hich$ in turn$ is critical for Then$ the strong- in/ersion saturation /oltage -ecomes o-taining an accurate /alue for $!o % Cith some e0perience and good engineering .udgment$ /ery reasona-le $/C physical pa- rameters can -e o-tained$ as can -e seen in

Figs% B and L VD SAT, * (RS M RD )W "#!o 0 $ M

(13)

Finally$ it is found that for this implicit solution$ the - factor in the saturation function F, Gsee (K)H should -e set to appro0imately 3%L for nFETs and 3%4 pFETs% The model implementa-

and Ta-le +6 it should -e noted that these particular MOSFETs e0hi-it the lo:est /alues of RS in the authorsQ e0perience$ -ut they are :ell :ithin the physical range for :ell-engineered de/ices% Furthermore$ although the /alues of carrier /elocity may appear high$ it should -e noted that the de/ices ha/e

1K75

+EEE TRA&SACT+O&S O& E<ECTRO& 'EV+CES$ VO<% LK$ &O% 5$ A,J,ST 3448

Fig% B% Comparison of model current (lines) :ith data (circles)% (a) and (-) For a KL-nm technology unia0ially strained nFET technology :ith poly-SiO& gate stac" G12H and gate length /G * 2L nm% (c) and (d) For a 23-nm technology unia0ially strained nFET :ith a metal-gate high-k stac" G1BH and gate length /G * 23 nm% 'e/ice parameters are gi/en in Ta-le +%

Fig% L% Comparison of model current (lines) :ith data (circles)% (a) and (-) For a KL-nm technology unia0ially strained pFET technology :ith poly-SiO& gate stac" G12H and gate length /+ * 2L nm% (c) and (d) For a 23-nm technology unia0ially strained nFET :ith a metal-gate high-k stac" G+ntel +e'M 45H and gate length /+ * 23 nm% 'e/ice parameters are gi/en in Ta-le +%

!;A!+F+ROOI et al.? S;ORT-C;A&&E< MOSFET MO'E< CO&T+&,O,S ACROSS A<< REJ+O&S OF O ERAT+O&

1K78

TA=<E + MOSFET 'EV+CE ARAMETERS

-een sho:n to -e non-allistic6 the thermal /elocity (-allistic limit) in these unia0ially strained de/ices has -een sho:n theoretically to -e significantly increased compared to rela0ed Si /alues G1H% The Fuality of agreement sho:n in Figs% B and L :as found to hold o/er numerous de/ices from the literature as :ell as e0perimental de/ices in the authorsQ la-oratory% +nterestingly$ the model produces good agreement e/en :ith non-MOSFET FETs$ such as L4-nm-gate-length +nJaAs ;EMTs G1KH using the same parameter - as for Si nFETs -ut :ith suita-ly increased electron mo-ility and /elocity$ as e0pected for +++V channel materials% As discussed in Section +$ the VS model in this paper :as de/eloped primarily in order to allo: for simple and largely uneFui/ocal characteri#ation of the historical e/olution of /irtual source carrier /elocity in Si-CMOS FETs operating in the saturation region and to permit e0trapolation of future reFuirements$ e%g%$ as done -y !ha"ifiroo# and Antoniadis G17H$ G15H% &e/ertheless$ it is rather remar"a-le that this minimalist model descri-es /ery :ell the I V -eha/ior of modern MOSFETs o/er their full range of operation and :ith only fe: physical parameters$ :hich$ moreo/er$ are held constant% Of course$ the model is not predicti/e in the sense that it reFuires the electrostatics (Cinv $ S $ $ and Io Oi%e%$ VT ) and /irtual source carrier /elocity and mo-ility as input parameters either from measurements or de/ice simulations$ or simply from educated guesses a-out future de/ice structures and materials% On the other hand$ if gate-length dependence of the electrostatics is a/aila-le$ the model can help in understanding the significance of the effects of the "ey MOSFET parameters and their /ariations o/er a range for gate lengths% As an e0ample$ Fig% K sho:s the VS model DpredictedE Io (VGS * 4 V, VDS * 1%3 V) /ersus Ion (VGS * 1%3 V, VDS * 1%3 V) for the nFET technology in Fig% B G1BH% The calculations :ere made -y /arying /G in the model from 24 to KL nm% For these calculations$ the /ariation of VTsat and VTlin (the typical constant-current threshold /oltages at VDS * 1%4 V and 4%4L V$ respecti/ely) /ersus /G in G1B$ Fig% 2H :as used to reproduce the '+=< parameter (/G ) and VT 0 /ersus /G dependence in the model% &ote that for VT 0 only the /ariation :ith /G :as used from the data -ecause the

Fig% K% Modeled /ersus measured Io (VGS * 4 V, VDS * 1%3 V) /ersus Ion (VGS * 1%3 V, VDS * 1%3 V) for the nFET technology in Fig% 2% 'ata from G12H are depicted -y points% For details of the model-calculated lines$ refer to the te0t%

definitions of VT 0 here$ and VTlin in that paper are different% As can -e seen in the figure$ e/en :ith no other /G parameter dependences in the model$ the calculated Ion /ersus Io (dashes) is in good agreement :ith the data% &e0t$ since some increase of S :ith decreasing length can -e e0pected e/en for these e/idently /ery :ell engineered de/ices$ a linear S (/G ) dependency on (/G ) :as assumed (empirically) as follo:s? S (/G ) * S (2L nm) M (4%1 VAdec)( (/G )( (2L nm))% The result is depicted in the dotted line% Finally$ a good match to the data is achie/ed (solid line) -y assuming a linear /ariation of $!o (/G ) as follo:s? $!o (/G ) * $!o (2L nm) M 7 (14 cmAs)( (/G ) (2L nm))% This form of dependency of /elocity for short-channel MOSFETs has -een sho:n in the literature G18H$ G34H$ and the coefficient here is Fuite reasona-le% Of course$ it cannot -e claimed here that the modeled /elocity dependency on '+=< is physically correct -ecause the S (/G ) dependency /ia '+=< :as assumed empiricallyOthe data are not a/aila-le in the literature% On the other hand$ S (/G ) is often a/aila-le from measurements$ and if so$ then the $!o (/G ) :ould ha/e real

physical significance$ for e0ample$ in Fuan- tifying the effecti/eness of strain engineering in a particular technology%

1K54

+EEE TRA&SACT+O&S O& E<ECTRO& 'EV+CES$ VO<% LK$ &O% 5$ A,J,ST 3448

V+% CO&C<,S+O& A minimalist MOSFET model that descri-es the I V -eha/- ior of short-channel MOSFETs o/er the complete region of op- eration has -een de/eloped% +n addition to physical parameters that are easily o-tained from measurements$ a minimum set of physically meaningful parameters$ all assumed constant o/er the region of operation$ is included$ and a methodology for e0tracting them from data has -een presented% The model is suita-le for analy#ing the effect of /ariation of these parameters and for technology road-mapping e0ercises% +n addition$ the model can form the -asis of a compact model for circuit simulation :ith the addition of models relating its parameters to physical de/ice dimensions andAor layout properties% AC!&OC<E'JME&T The authors :ould li"e to than" rof% >% Tsi/idis and rof% M% <undstrom for the useful discussions% R EFERE&CES
G1H A% !ha"ifiroo# and '% A% Antoniadis$ DTransistor performance scaling? The role of /irtual source /elocity and its mo-ility dependence$E in IEDM Te34. D#+%$ 344K$ pp% KK7K74% G3H M% S% <undstrom$ DElementary scattering theory of the Si MOSFET$E IEEE Ele3tron De$#3e /ett.$ /ol% 15$ no% 7$ pp% 2K12K2$ 9ul% 1887% G2H M% S% <undstrom and I% Ren$ DEssential physics of carrier transport in nanoscale MOSFETs$E IEEE Tran,. Ele3tron De$#3e,$ /ol% B8$ no% 1$ pp% 1221B1$ 9an% 3443% GBH J% T% Cright$ DThreshold modelling of MOSFETs for CA' of CMOSV<S+$E Ele3tron /ett.$ /ol% 31$ no% K$ pp% 33233B$ Mar% 185L% GLH >% Cheng$ M%-C% 9eng$ I% <iu$ 9% ;uang$ M% Chan$ !% Chen$ % !% !o$ and C% ;u$ DA physical and scala-le I V model in =S+M2/2 for analogA digital circuit simulation$E IEEE Tran,. Ele3tron De$#3e,$ /ol% BB$ no% 3$ pp% 377357$ Fe-% 1887% GKH &% Arora$ M5SFET Model, &or V/SI C#r36#t S#m6lat#on7 T4eor8 and 9ra3t#3e% &e: >or"? Springer-Verlag$ 1887% G7H >% Tsi/idis$ 5:erat#on and Model#n+ o& t4e M5S Tran,#,tor$ 3nd ed% &e: >or"? McJra:-;ill$ 1888% G5H >% Taur and T% ;% &ing$ F6ndamental, o& Modern V/SI De$#3e,% &e: >or"? Cam-ridge ,ni/% ress$ 1885% G8H '% M% Caughey and R% E% Thomas$ DCarrier mo-ilities in silicon empirically related to doping and field$E 9ro3. IEEE$ /ol% LL$ no% 13$ pp% 3183 3182$ 'ec% 18K7% G14H 9% A% Cooper and '% F% &elson$ D;igh-field drift /elocity of electrons at the Si-SiO2 interface as determined -y a time-of-flight techniFue$E ;. <::l. 948,.$ /ol% LB$ no% 2$ pp% 1BBL1BLK$ Mar% 1852% G11H R% C% Coen and R% S% Muller$ DVelocity of surface carriers in in/ersion lay- ers on silicon$E Sol#d State Ele3tron.$ /ol% 32$ no% 1$ pp% 2LB4$ 9an% 1854% G13H '% A% Antoniadis$ +% R-erg$ C% &% Chleirigh$ O% M% &ayfeh$ A% !ha"ifiroo#$ and 9% <% ;oyt$ DContinuous MOSFET performance increase :ith de/ice scaling? The role of strain and channel material inno/ation$E IBM ;. Re,. De$elo:.$ /ol% L4$ no% BAL$ pp% 2K227K$ 9ul% 344K% G12H A% !ha"ifiroo# and '% A% Antoniadis$ DThe future of high-performance CMOS? Trends and reFuirements$E in 9ro3. E6r. Sol#d State De$#3e Re,. Con&.$ 3445$ pp% 2427% G1BH S% Tyagi$ C% Auth$ % =ai$ J% Curello$ ;% 'eshpande$ S% Janna/aram$ O% Jolon#"a$ R% ;eussner$ R% 9ames$ C% !enyon$ S%-;% <ee$ &% <indert$ M% <iu$ R% &agisetty$ S% &atara.an$ C% ar"er$ 9% Se-astian$ =% Sell$ S% Si/a"umar$ A% St% Amour$ and !% Tone$ DAn ad/anced lo: po:er$ high performance$ strained channel KL nm technology$E in

IEDM Te34. D#+.$ 344L$ pp% 14741473% G1LH S% &atara.an$ M% Armstrong$ M% =ost$ R% =rain$ M% =ra#ier$ C%-;% Chang$ V% Chi"armane$ M% Childs$ ;% 'eshpande$ !% 'e/$ J% 'ing$ T% Jhani$ O% Jolon#"a$ C% ;an$ 9% ;e$ R% ;eussner$ R% 9ames$ +% 9in$ C% !enyon$ S% !lopcic$ S%-;% <ee$ M% <iu$ S% <odha$ =% McFadden$ A% Murthy$ <% &ei-erg$ 9% &eirync"$ % ac"an$ S% ae$ C% ar"er$ C% elto$ <% ipes$ 9% Se-astian$ 9% Seiple$ =% Sell$ S% Si/a"umar$ =% Song$ !% Tone$ T% Troeger$ C% Ce-er$ M% >ang$ A% >eoh$ and !% Ihang$ DA 23 nm logic technology featuring 3nd generation high-=M metal gate transistors$ enhanced chan-

G1KH G17H G15H G18H G34H

nel strain and 4%171 0m SRAM cell si#e in a 381 M- array$E in IEDM Te34. D#+.$ 3445$ pp% 12% '%-;% !im and 9% A% del Alamo$ D<ateral and /ertical scaling of +n0.7 Ja0.3 As ;EMT for post-Si-CMOS logic applications$E IEEE Tran,. Ele3tron De$#3e,$ /ol% LL$ no% 14$ pp% 3LBK3LL2$ Oct% 3445% A% !ha"ifiroo# and '% A% Antoniadis$ DMOSFET performance scalingO art +? ;istorical trends$E IEEE Tran,. Ele3tron De$#3e,$ /ol% LL$ no% K$ pp% 12811B44$ 9un% 3445% A% !ha"ifiroo# and '% A% Antoniadis$ DMOSFET performance scalingO art ++? Future directions$E IEEE Tran,. Ele3tron De$#3e,$ /ol% LL$ no% K$ pp% 1B411B45$ 9un% 3445% ;% ;u$ 9% =% 9aco-s$ <% T% Su$ and '% A% Antoniadis$ DA study of deepsu-micron MOSFET scaling -ased on e0periment and simulation$E IEEE Tran,. Ele3tron De$#3e,$ /ol% B3$ no% B$ pp% KK8K77$ Apr% 188L% A% <ochtefeld and '% A% Antoniadis$ DOn e0perimental determination of carrier /elocity in deeply scaled &MOS? ;o: close to the thermal limitSE IEEE Ele3tron De$#3e /ett.$ /ol% 33$ no% 3$ pp% 8L87$ Fe-% 3441%

Ali :ha2ifiroo; (MQ45) recei/ed the =%Sc% and M%Sc% degrees from the ,ni/ersity of Tehran$ Tehran$ +ran$ in 1887 and 1888$ respecti/ely$ and the h%'% degree from the Massachusetts +nstitute of Tech- nology (M+T)$ Cam-ridge$ in 3447$ all in electrical engineering% From 1887 to 3441$ he :as :ith the Thin Films Research <a-oratory$ ,ni/ersity of Tehran$ :here he :as engaged in lo:-temperature fa-rication of polysilicon thin-film transistors on fle0i-le su-- strates% From 3441 to 3445$ has :as :ith the Microsystems Technology <a-oratories$ M+T$ :here he studied carrier transport in deeply scaled MOSFETs and :or"ed on the fa-rication and characteri#ation of germanium-channel transistors% +n May 3445$ he .oined +=M Research at Al-any &anotech$ Al-any$ &>$ :here he is :or"ing on e0ploratory de/ice design for 33-nm node and -eyond% ;e has authored or coauthored more than L4 technical papers%

Osama M. <a&feh (MQ48) recei/ed the =%S% degree in electrical and computer engineering from the ,ni/ersity of +llinois at ,r-ana-Champaign in 3443 and the h%'% degree in electrical engineering and computer science from the Massachusetts +nstitute of Technology (M+T)$ Cam-ridge$ in 3445% ;is thesis in/estigated non/olatile memory de/ices using col- loidal silicon nanoparticles% ;e is currently a ostdoctoral Researcher :ith M+T$ :here his research focuses on no/el nanoelectronic de/ices for ad/anced computation and memory applications% ;e is the author or coauthor of 18 technical articles and has gi/en se/eral research presentations at academic$ go/ernmental$ and industrial locations% 'r% &ayfeh is a mem-er of the +EEE$ Materials Research Society and Sigma Pi% ;e :as the recipient of the +ntel Fello:ship in 3447 for his :or" on nanotechnology--ased de/ices%

-imitri Antoniadis (MQ78SMQ52FQ84) :as -orn in Athens$ Jreece% ;e recei/ed the =%S% degree in physics from the &ational ,ni/ersity of Athens$ Athens$ Jreece$ in 1874$ and the h%'% degree in electrical engineering from Stanford ,ni/ersity$ Stanford$ CA$ in 187K% ;e has -een :ith the Microsystems Technology <a-oratories$ Massachusetts +nstitute of Technology (M+T)$ Cam-ridge$ since 1875$ :here he is the Ray and Maria Stata Chair in Electrical Engineering% ;e is the 'irector of the multiuni/ersity Focus Research Center for Materials Structures and 'e/ices centered at M+T% ;e is the author or coauthor of more than 344 technical papers% ;is present research focuses on the physics and technology of e0treme su-micrometer Si$ silicon-on-insulator$ and SiASiJe MOSFETs% 'r% Antoniadis is a mem-er of the &ational Academy of Engineering and the recipient of se/eral professional a:ards%

Anda mungkin juga menyukai