Extended Abstracts of the lSth (1986 International) Conlerence on Solid State Devices and Materials, Tokyo, 1986, pp. 283-286

Forward-Bias Tunnelling Current Limits in Scaled Bipolar Devices

J. del AlamoNTT Labs. , Atsugi, Kanagawa, Japan 243-07

R. M. SwansonStanford University, Stanford, CA 94305, U.S.A.

Tunnelling assisted by midgap states in forward bias isobserv.ed in bipolar transistors containing heavily doped p-njunctions. A simple phenomenological model suitable toincorporation in device codes is developed. The model identifiesas key parameters the space-charge-region thickness at zeto biasand the reduced doping level at the SCR .edges. Both parameters canbe obtained from C-V characteristics. This tunnelling mechanism maylimit the maximum gain achievable from scaled bipolar devices.

Introduction

The scaling of silicon bipolartransistors demands a reduction in thebase thickness and an accompanyingincrease in the base doping 1eveI. Anumber of peculiar physical phenomena havebeen shown to arise from the high baseimpurity concentrations. Some of them arebeneficial, like bandgap narrowing in thebase which yields an i.mprovement in thegain [1], but most of them are harmful.Such are the reduction of breakdownvoltage and the appearance of Zenertunnetling at reverse bias I2l.

In this paper hre study a potentially moredeleterious mechanism: forward-biastunnelling in the base-emitter junction.If both sides of a p-n junction areheavily doped, a thin space-charge-region(SCR) results that may aIIow thetunnelling of electrons across it. If bothsides of the junction are degenerate, theEsaki effect is observed and a region ofnegative resistance appears in the forwardI-V characteristics. Without such highdoping levels, i.e., when one or neitherof the two sides are degenerate,tunnelling may also occur assisted bystates located inside the gap t3l.

Tunnelling currents of this nature havebeen recently observed at the periphery ofself-aligned poly-silicon bipolartransistors across the junction formed bythe emitter and the heavily-dopedextrinsic-base contact [4r5]. Adeguatespacer technology has been able to supressthis perimeter tunnelling. As devicescaling advances and the intrinsic basedoping level is raised, tunnelling isexpected to occur at the whole emitter-base area. This tunnellin!I ,current maypose a limitation to the maximum gain

B-5-2

Fig Cross section of bipolartransistor.

achievable from scaled bipolar devices.

Experimental

The transistors (Fi9. 1) used in thepresent work $rere designed to measure thehole tranrport parameters in heavily-dopedn-type silicon 161. The base wasfabricated by epitaxy and is thereforeuniformily doped with phosphorus to avalue that can be measured very precisely171. B was then implantp-d through an oxidemask to a dose of 9x10rr cm-2. Contact tothe base hras provided through rr* regionsoutdiffused from P-doped g1ass. MetaIcontact to the device \^ras realized withevaporated. A1.

The I-V and C-V characteristics of thetop p**-rrt diode were measured. in forwardbias at 297 K. Forward I-V curves ofselected devices are shown in Fig. 2, withthe base doping level used as 1abe1. At1ow voltages a strong increase in thecurrent occurs as the base dopinq level israised beyond about 5x1018

"*-3. This"excess current" is approximatellyexponential and can then be fitted with anequation of the form:

p++J + n++' n'-eDl

p- Substrate

++

sFz,t!(rGf,()

ND JoT vT rfSCR(o) N"N6/ (Na+Nd)

- - 1:t- I I - - - - - Jrt:tl t - - - tlyt - -- Jrt - _ - _ _ - J:l- I t _ - -8.8x1017 9.1xto-12 26 620

1.5x1018 1.lxto-l1 32 470

4.8x1018 1.3x10-lo 39 3Go

1.1x1019 2.5xto-9 45 340

1.4x1019 5.7xto-9 47 3oo

1.8x1019 1. ox1o-8 46 3to

2.4x1019 9.8x10-9 46 3to

4.0x1019 2.2xLo'7 46 3oo

3 .Oxlo17

5 .9x1017

9.3x1017

9.7xLOL7

1.4x1018

1.1x1018

1.2x1018

1.1x1018

o o.2 0.4 0,6 0.8VOLTAGE (V)

Fiq. 2 T-V characteristics of the p+*-n*Junction of selected devicps

Table I

The establishment of a rigorous theorythat accuratelly describes this tunnellingphenomenon presents considerablechallenges. Even in sirnple tunnellingprocesses, order of magnitude agreementbetween experiments and theory is at bestwhat can be expected t8l. With thisbackground, our intention in the presentwork is to demonstrate the tunnellingnature of the excess curtrents displayed inFig. 2 by developing a simplified theorythat contains the most relevant physics ina non-rigorous way. A model results thatis suitable for incorporation into devicecodes.

For the purpose of grasping thefundamental physics it is sufficient toconsider the tunnelling probabilitythrou-gh a.potential barrier of height Efand thjc:kneiss W [9]:

vI^=AJ^^un-r ut' . vr (1)

Since IT is postulated to be of atunnelling nature, Jot is denoted as thetunnelling saturation current and VT isthe tunnelling characteristic voltage. Ais the device area. JoT and VT vrereextracted from the measured data.

The space-charge-region width and thereduced doping Ievel N.Nd,/ (Na+Nd) at theSCR edge can be obtained from the C-Vcharacteristics for small forward bias. Naand Nd denote the net acceptor and donorconcentrations at the edges of the SCR.The results of all measurements arecollected in Table I.

T}:::YTunnelling assisted by midgap statesacross a forward-biased p-n junction waspostulated in order to explain the "excesscurrent" that appeared in Esaki diodes athigher voltages than required to uncouplethe conduction and valence bands t3l. Thisexcess current can exist even if one orneither of th.e two sides of the junctionare degenerate, as our experiments show.The basic process is shown in Fig.3. If alocalized state inside the forbidden gapis energetically aligned with an occupiedstate of the conduction band of the n-sideof the junctionr on electron can tunnelinto the localized state and Iaterrecombine with a valence band hole of thep region through a Shockley-Read-Hal 1process. This requires that the SCR besuf f iciently thin to achieve anappreciable tunnelling probability.

For a triangular or parabolic barrie,radditional numerical factors in theargument of the exponential appear [ 10 ] ,but identical dependences of T on W and'ffi are found.

There are several difficulties inextending this simple quantum mechanicaltheory to tunnelling assisted by midgapstate.s. Among them, the electron in thelocalized state cannot be described by afree-space wave function corrected by.the

Fig. 3 Illustration of basic tunnellingprocess assisted by midgap states.

roexPG+\@r) (2)

l. I x tore4'8 xlOrB1.5 x lOro

284

to

I

I

I

t\lE€

Fg,

=lrloztrlG,G3ozo

EDEsa@

(9z

lr,zzD

n

".'9xGl^l!(9{5o(DzFI

Jlrlzzt()HFot-lIllrlFo(rr(J

to8

6e

d0

01l

dl2bo

T.297K

Il-a

300 400SPACE fiARGE REGION

500a

WIDTH (AI

rcototoNo Nd /(NofNd) (tor?om'3)

Fig. 5 Vr2 versus reduced doping level.

exponentially increases with the forwardvoltage. The observation of such adependence in the experimental currentindicates that the midgap statedistribution is either exponential oruniform in energy. The first alternativeis expected if band tailing of theconduction and valence band edges occursdue to heavy doping t111. Tails areexpected to extend very deep into theforbidden gap in the SCR because of thelack of mobile carriers that screen thedopant impurities in neutral regions I121.

Eq. 5 implies that the tunnelling currentmay be approximatelly described by anequation like (1) with Jot dependingexponentially on the junCtion width atzero bias wscR(o) ( A;L'/' is a ratherinsensitive function of doping):

Jor= hroPI-hrw(o)l (6)

and VT depending linearly on the squareroot of the reduced doping leveI:

vr=kg (7)

The dependences contained in eqs. 6 and 7

are verified in our experiments. In fact,Fig. 4 displayes Jof versus WSCR(0) andindeed an exponential dependence is shownto appear owhen WSCR(0) is smalLer thanabout 400 A. Fig.5 plots the sguare of V,against the reduced doping leve1. The datafits a straight line that crosses theorigin, as expected from eq. 7. Thevalues of the constants that appe'ar ineqs. 6 and 7 can be extracted: k1=6.O2xLO7eJcn?,.^k2=1.L5x1-07

"*-1, and k3=4.28x10-11

v.c^3/2 .

The temperature dependence of the currents

Fig. 4 Jor versus WSCR(O)

effective mass of the host crystal, asassumed in the derivation of (2). Theef fective mass, **, of eq. 2 takestherefore an unclear meaning. Eq. 2 shouldbe regarded as a functional dependencerelationship with unknown constants.

The height of the barrier to tunnelling,as shown in Fig.3, is by definition ofjunction built-in voltage:

Er=q(06,-Y)

The thickness of the tunne 1 1 ingis approximatelly, for an abrupt

(3)

barrier; unction:

wr"" (v) =

Introducing (3)obtain:

T=up

(0b, - Y) (4)

into 2), lre

Irysc8 (o )

2e6

q

N + N.ad,

N N.ad

For other junctions, the voltagedependence of WSCn may have a differentfunctional form. However, for smallforward voltages, WSCR (V) is not verysensitive to V and (4) suffices.

2

h

and (4)

,@

4\/ n\E

h

ne

(s)

,/J_,r-It-r, I- ^^-T-

;F

T .297K

rF

/

and the tun 11in9 probabiJ-ity

285

TEMPERATURE ( K )

lo20toBto18

lotT

roEg

Ez

€zltoz,

o-DtDr)mvo{HC)

{czzFI

zoo5D6)FI

3

G|

Eq

FHv,ztdoFzlrlEEfozoFt=v,(9zJJtu22=F

t.tz t.13 t.l4 t.ts t.l6

SILICON BANDGAP (eV)

Fig.5 Temperature dependence ofVt for a typical device.

under consideration has al_so been studied.The ,mode1 summarized in eqs. Lr 6 and 7predicts that JoT depends exponential ly,with a minus sign, on the bandgap (through4il, and that VT is essentiallyindependent of T. In fact, Fig.6 showsmeasurements for a typical sample. Theexponential dependence of JoT on thesilicon bandgap unequivocally conFirms thetunnelling nature of the current underinvestigation and rules out itsinterpretation as a recombination currentinsicle the SCR, since Jot is not thermallyactivated. AdditionaIIy, VT changes lessthan 1Ot when the temperature spans arange of l-60 K, confirming the temperatureindependence predicted by eqn. 5.

Discussion

The model defined by eqs. t, 6, and 7 caneasily be incorporated in device modellingcodes. Although the model identifies WSCnand NaNd,/ (Na+Nd) as the key parametei-s'controlling tunnelling, the constantsinvolved are expected to be somewhatprocess dependent. For one thing, heavycompensation of the p++ emitter occurs inthe high doping range yielding gradualjunctions. This is understood by observingthe dependence of the reduced doping levelon the epitaxial donor concentration whichis plotted in Fig. 7. Since the slope ofthe doping distribution at the junction ofa real device is expected to be processdependent, the j unct ion potent ial shap€ rand the constants in eqs. 3, 6 and 7 willprobably be sensitive to the details ofthe process. Nevertheless, the modeldefined by eqs. 1, 6, and 7 may be areasonable starting point f or theprediction of tunnelling in practicaldevices.

0ON0R CONCENTRATTON (cni3)

Fig. 7 Reduced doping level versus donorconcentration in the base.

The tunnelling current identified in thiswork may Iimit the maximum gain achievabl-ein scaled bipolar devices. Thislimitation may be avoided by fabricatingthe devices by techniques, like MBE, thatallow the preparation of a dopant-free SCR1131. A trade-off between the maximumtolerable tunnelling current and theintrinsic delay in electron transportacross the SCR will exist. Our dataindicates that a 4OO fi scR yieldsnegligible tunnelling through it, but thismay be too wide for high-speed devices.

Conclusions

Tunnelling assisted by midgap states hasbeen shown to occur in forward-biased p-njunctions. A simple model suitable forincorporation in device codes has beendeveloped. This tunnelling mechanism maylimit the gain in scaled bipolartransistors.

References

f.. T. .H. Ning, 16th Conf . on Solid StateDevices and Materials, Kobe 1984, p. 205.2. J. M. C. Stork and R. D. I saac, IEEETrans. Electr. Dev. ED-30, 1,527 (L983).3. A. G. Chynoweth, W. L. Feldmann, and R.A. Logan, Phys. Rev. L2L, 684 (1961).4. A. Cuthbertson and P. Ashburn, IEEETrans. Electr. Dev. ED-32, 242 (l-985).5. E. Hackbarth, G. P. Li, and T. C. Chen,44th Device Research Conf., Boston L986.6. J. del A1amo, S. Swirhun, and R. M.Swanson, Int. Electron Device Meeting,Washington D. C.1985, p.2gA.7. J. del Alamo and R. Ivl. Swanson, ;I.Electrochem. Soc. L32, 30L1 (1985).8. C. B. Duke, "Tunneling in SoIids",Academic Press 1969.9. K. Seeger, "seBicond,uctor Physics",Springer-Verlag L982 (2nd edition).1O. P. N. Butcher, K. F. Hu1me, and J. R.Morgan, Solid-St. Electr. 5, 358 (1962).11. E. O. Kane, Solid-St. Electr.28,3(Le8s).L2. J. R. Lowney, Solid-St.r87 (198s).13. P. M. Solomon, Proc. I(1e82).

l.l7

JoT and

Electr. 28,

EEE 7O, 489

-i-, -I* T

,/F---tfrr--l-Ft-r

//'-

286