IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. ED-31, NO. 7, JULY 1984
`
`853
`
`Comparison of Experimental and Theoretical
`Results on Polysilicon Emitter
`Bipolar Transistors
`
`PETER ASHBURN AND B. SOEROWIRDJO
`
`Abstract-Two types of polysilicon emitter transistors have been
`fabricated using identical processing except for the surface treatment
`prior to polysilicon deposition. The first type was given a dip etch in
`buffered hydrofluoric acid, which was intended to remove any inter(cid:173)
`facial oxide, while the second type was given an RCA clean, which was
`intended to grow an interfacial oxide of known thickness. Detaiied
`electrical measurements have been made on these devices including the
`temperature dependence of the gain over a wide temperature range.
`The transistors given an RCA clean have gains approximately five times
`In addition, the temperature
`higher than those given an HF etch.
`dependence of the gain is different for the two types, with the HF
`devices exhibiting a much stronger dependence at high temperatures
`than the RCA devices. A detailed comparison is made with the theory
`and it is shown that the characteristics of the HF devices can largely be
`explained using a transport theory, while those of the RCA devices can
`be fully explained using a modified tunneling theory.
`
`I. INTRODUCTION
`
`POLYCRYSTALLINE silicon is increasingly being used in
`
`bipolar integrated-circuit processes for producing self(cid:173)
`aligned circuits [l]-[3] with a consequent improvement in
`both switching speed and packing density. When used in
`shallow emitter structures the polysilicon also provides a
`means of obtaining very shallow junctions with very high
`yields. These high yields result from a reduction in anomalous
`diffusion effects [ 4] and in defect generation [ 5] , [ 6] associ(cid:173)
`ated with the high-concentration emitter diffusion [7].
`In addition to the preceding advantages, improved gains can
`be obtained for transistors with polysilicon emitters [8]. Im(cid:173)
`provements from 3 [9] to 30 [10], [11] have been reported
`in the literature , and in a previous publication [8], (12] the
`authors showed that the gain improvement depended strongly
`on the type of surface treatment given prior to polysilicon
`deposition. Two alternative theories have been proposed to
`explain the improved gains obtained for polysilicon emitter
`transistors. The first was proposed by Ning and Isaac [9] ;
`( 13] , ( 14] , and explained the improved gains in terms of a
`shorter hole diffusion coefficient in the polycrystalline part
`of the emitter than in the single-crystal part. The second was
`proposed by De Graaff and De Groot [10] , and explained the
`improved gains by tunneling through a thin interfacial oxide
`layer located between the polysilicon and single-crystal silicon.
`More recently, Eltoukhy and Roulston have reported [15],
`
`Manuscript received November 30, 1983; revised January 31, 1984.
`B. Soerowirdjo was supported by the British Council.
`The authors are with the Department of Electronics, Southampton
`University, Southampton, Han ts, S09 SNH, England.
`
`[16] on a unified theory which incorporates both these
`mechanisms in a form suitable for fast numerical solution.
`Although considerable work has been published on the
`theoretical aspects of polysilicon emitter transistors, relatively
`little has been published on the experimental characteristics
`of these devices.
`In order to rectify this deficiency, in this
`paper, we present the results of detailed electrical measure(cid:173)
`ments on different types of polysilicon emitter bipolar transis(cid:173)
`tor. A comprehensive comparison is made with the theories
`from the literature in order to determine under which circum(cid:173)
`stances the alternative theories are valid. To assist with this
`comparison we have fabricated two types of polysilicon emitter
`transistor with identical processing except for the surface
`treatment prior to polysilicon deposition. Measurements of
`the temperature dependence of the gain are made on these
`devices over a wide range of temperatures, and a comparison
`made with the dependences predicted by the various theories.
`
`11. THEORY
`In a previous paper, Eltoukhy and Roulston [ 15] presented
`a unified theory for current transport in polysilicon .emitter
`bipolar transistors. A complete set of transport and tunneling
`equations was derived and arranged in such a way as to be suit(cid:173)
`In this paper, we present a
`able for fast numerical solution.
`simplified analytical version of this theory which is in a form
`suitable for comparison with experimental results.
`In this
`way, we are more easily able to identify the physical mecha(cid:173)
`nisms that are controlling the operation of the different types
`of polysilicon emitter transistor.
`Fig. 1 shows the band diagram of a polysilicon emitter tran(cid:173)
`sistor with a thin insulating layer between the polycrystalline
`and monocrystalline regions. This band diagram is a simpli(cid:173)
`fied version of that used in the full unified theory, and has
`been derived by making the following additional approxima(cid:173)
`tions. Firstly, it has been assumed that the emitter-base junc(cid:173)
`tion is extremely shallow so that recombination in the single(cid:173)
`crystal part of the emitter can be ignored. This assumption is
`valid as Jong as the emitter depth below the polysilicon-silicon
`interface is small compared with the diffusion length of holes
`in the single-crystal part of the emitter. Crude calculations
`have shown that, for a junction depth of 0.04 µm and for a
`typical emitter profile, recombination in the single-crystal part
`of the emitter provides approximately 13 percent of the base
`current [ 17] . Secondly, we have assumed that there is no
`band bending at the polysilicon-silicon interface. The pub-
`
`0018-9383/84/0700-0853$01.00 © 1984 IEEE
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on May 15,2024 at 03:07:33 UTC from IEEE Xplore. Restrictions apply.
`
`

`

`854
`
`IEEE TRAHSACTIONS ON ELECTRON DEVICES, VOL. ED-31, NO. 7, JULY 1984
`
`E-----•----
`Fn
`
`-
`
`-
`
`-
`
`-
`
`- • EFp
`-
`- -
`-
`- - -
`- - -E ,
`
`~
`Fig. 1. Band diagram for. a polysilicon-emitter bipolar transistor wi :h
`a thin interfacial oxide layer.
`
`lished models [ 18] , [ 19] for conduction in poly silicon assume
`that the bands bend upwards in the vicinity of the grain boun 1-
`ary. This is caused. by the trapping of carriers at the hi1:h
`concentration of defects and . dangling bonds located at tile
`grain boundary. As a result, the silicon adjacent tci the gra in
`boundaries becomes depleted of carriers and a potential barri~r
`is formed . This barrier can be quite large in lightly dopi ,d
`polysilicon, but in heavily doped material it is negligible. Fi)r
`example, for a typical emitter doping concentration of 101!' -
`• e y -t ,
`102 0 cm - 3
`, and an interface state density of 10 1 2 cm - 2
`the band bending is less than 1 meV (20].
`The tunneling current is derived from the one-dimensional
`time-independent tunneling probability, which is given by
`[15]
`
`D(Ex) = exp [- ~ i~' [2mf(qV(x)- Ex)J 1/ 2 dx]
`
`(l)
`
`where V(x) is the barrier height and Ex is the energy compo(cid:173)
`nent of the incident carriers in the x direction. Assuming that
`the potential barrier is approximately rectangular with a heig :it
`Xh, the preceding integral can be evaluated to give
`
`D(Ex) = exp [- 4:
`
`(2m1')1/ 2 6 (qxh + Ex) 1l 2].
`
`( 2)
`
`The major difference between this equation and that de(cid:173)
`rived by De Graaff and De Groot [ 10] is the presence of the
`(1 - Cfl kT) term in the denominator. This term did not arise
`in their analysis because the tunneling probability D(Ex) was
`assumed to be a weak function of Ex and hence the Ex term in
`(3) was neglected [IS] .
`The temperature dependence of the collector current density
`can be written as [10]
`
`q~-
`Egb
`T4
`Jc(T) = canst - - exp - - exp -
`Pb(T)
`kT
`kT
`
`(7)
`
`where Pb (T) is the temperature dependence of the sheet resis(cid:173)
`tance of the base under the emitter and is used to provide the
`temperature dependence of the mobility in the base. Egb is
`the bandgap in the base. Using this equation together with (4)
`gives the following expression for the temperature dependence
`of the gain:
`
`To.s
`hFE(T) = canst -(-) (I - ChkT) exp -
`Pb T
`
`(AEge - AEgb)
`kT
`
`(8)
`
`where AEge and AEgb are the bandgap narrowing in the emit(cid:173)
`ter and base, respectively.
`In the case where the insulating iayer at the polysilicon(cid:173)
`silicon interface is absent, the base current is determined by
`hole transport in the polycrystalline and monocrystalline
`regions of the emitter. The analysis for this type of device can
`be simplified if it is assumed that there is only one grain of
`polysilicon between the polysilicon-silicon interface and the
`metal contact. This assumption is supported by cross section
`TEM observations (S] which indicate that the polysilicon
`grains in this type of device are columnar in shape and extend
`completely through the polysilicon layer. Under these condi(cid:173)
`tions, the full unified theory reduces to the analysis of Ning
`and Isaac (9] . The base current density is therefore given by
`
`qDp2n[e [
`q~
`Dp2 Lp1]-l
`JB = -~ - 1 +-- - - exp--
`Dp1 W2
`kT
`W2 Nde
`
`(9)
`
`At this point in the analysis, the first two terms in the Tayl ar where the subscripts 1 and 2 refer to the polycrystalline and
`series expansion of (I + Ex/qxh) 1!2 are taken to give
`monocrystalline regions of the emitter, respectively. Using
`(9) and (7), the temperature dependence of the gain can be
`written as
`
`(3)
`
`The hole tunneling current density can then be derived as
`follows:
`
`where
`
`and
`
`( 4)
`
`(5)
`
`(6)
`
`Lp 1 is the hole lifetime in the polysilicon and is given by
`
`Lp 1 =(Dp17p 1 ) 112 .
`
`(10)
`
`(11)
`
`In heavily doped silicon, the hole lifetime 7 pl is limited by
`Auger recombination [21] , [22] as follows:
`
`1
`7pl = C N2
`n de
`
`(12)
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on May 15,2024 at 03:07:33 UTC from IEEE Xplore. Restrictions apply.
`
`

`

`ASHBURN AND SOEROWIRDJO: COMPARISON OF RESULTS ON POLYSILICON EMITTER TRANSISTORS
`
`855
`
`polysilicon
`
`silicon
`
`. . ...
`. .
`
`i• •
`
`0 O
`
`O 0,
`0
`..2.....o--><-_,,---;_O
`()It . . •
`•
`Ot, 0
`0
`
`00
`
`where Cn = 1.6 - 3 X 10··31 cm6 • s- 1 is the Auger coefficient.
`r pl is therefore independent of temperature, and the tempera(cid:173)
`ture dependence of Lp 1 is determined by Dp 1 as follows:
`
`(13)
`
`It is well known that, at high doping concentration, the
`mobility in single-crystal silicon is approximately independent
`of temperature [23]. However, for heavily-doped polysilicon,
`Seto [24] has observed that the hole mobility decreases sig(cid:173)
`nificantly with increasing temperature, and can be described
`as follows:
`
`µP 1 (T) = const r-o.s exp - q:~
`
`(14)
`
`where <l>gb is the barrier height at the polysilicon grain bound(cid:173)
`ary. For heavily doped polysilicon, this barrier height is less
`than 1 meV, and hence the exponential term in (14) can be
`ignored. Making this approximation and using (10), (11 ), and
`(13) gives the following expression for the gain as a function
`of temperature:
`
`hFE = const
`
`1 + A To.7 5
`Pb (T)
`
`exp -
`
`(1:i..Ege - D..Egb)
`kT
`
`(15)
`
`where A is a constant which can be calculated from (10).
`
`Ill. EXPERIMENTAL PROCEDURE
`Two types of polysilicon emitter bipolar transistors have been
`fabricated with identical processing except for the surface
`treatment prior to polysilicon deposition. The first type
`(HF devices) was given a dip etch in buffered hydrofluoric
`acid, which was intended to remove any native oxide which
`might have been present prior to polysilicon deposition. The
`second type (RCA devices) was given a dip etch in buffered
`HF followed by an RCA clean, and this was intended to grow
`an interfacial oxide layer of known thickness. Auger electron
`spectroscopy experiments have shown that a dip etch in buf(cid:173)
`fered HF leaves an oxide layer of less than 2 A [25] , while an
`RCA clean produces an oxide film between 13 and 15 A thick
`[26] . Although there is some doubt about the absolute values
`of these interfacial layer thicknesses, these results do show
`that the interfacial layer produced by the RCA clean is con(cid:173)
`siderably thicker than that produced by the HF etch. This was
`confirmed in our work by the observation that slices given the
`HF etch were hydrophobic while those given the RCA clean
`were hydrophilic.
`The interfacial layer treatments were immediately followed
`by the deposition of approximately 0.4 µm ofundoped LPCVD
`polysilicon. The emitters of the transistors were then formed
`by implanting 1 X 1016 cm -z arsenic into the polysilicon and
`driving-in at 900°C in wet oxygen.
`Detailed electrical measurements were carried out on the
`two types of transistors including measurements of collector
`and base current as a function of base emitter voltage and the
`temperature dependence of the gain and resistance under the
`emitter.
`
`Distance , ~m
`Fig. 2. Arsenic profiles, obtained from Rutherford backscattering
`experiments, for samples given an HF etch and an RCA clean.
`
`The measurements as a function of temperature were carried
`out in an Oxford Instruments cryostat, type DN70, and in an
`oven for temperatures greater than 80°C. Measurements were
`made over the range -115 to + 140°C at intervals of approxi(cid:173)
`mately 7°C. A full set of base and collector current-voltage
`characteristics was taken at each temperature, together with a
`measurement of the resistance under the emitter. The elec(cid:173)
`trical measurements were supplemented with beveling and
`staining to provide the junction depths and Rutherford back(cid:173)
`scattering [27] to provide the arsenic profiles.
`
`IV. RESULTS
`Fig. 2 shows the arsenic profiles, obtained from Rutherford
`backscattering experiments for the two types of surface treat(cid:173)
`ment. The main feature of these profiles is the presence of a
`peak in the arsenic concentration at the polysilicon-silicon
`interface followed by a rapid decrease in the arsenic concen(cid:173)
`tration on entering the single-crystal silicon. The arsenic peak
`has been reported previously [28] and explained by arsenic
`segregation at the grain boundaries. Another feature of the
`profiles in Fig. 2 is that the arsenic has penetrated slightly
`deeper into the silicon for the HF samples than for the RCA
`samples, though the resolution of Rutherford backscattering
`(==300 A) is such that this small difference may not be signifi(cid:173)
`cant. However, similar results have been reported previously
`[28] for samples given stronger drive-ins, and were explained
`by retardation of the arsenic diffusion by the interfacial oxide
`layer.
`The junction depths obtained from bevelling and staining are
`summarized in Table I, along with the results of the resistance
`under the emitter and gain measurements. The value of 0.04 µm
`for the emitter-base junction depths is in reasonable agree(cid:173)
`ment with the arsenic profiles, considering the +0.03-µm error
`associated with the beveling and staining technique. The re(cid:173)
`sults in Table I indicate that, apart from the measured gains,
`the two types of transistor are very nearly identical. The small
`difference in the mean value of the resistance under the emitter
`indicates that the Gummel number [29] in the base of the
`RCA device is slightly higher than that in the HF device, sug(cid:173)
`gesting that the arsenic atoms have penetrated deeper in the
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on May 15,2024 at 03:07:33 UTC from IEEE Xplore. Restrictions apply.
`
`

`

`856
`
`IEEE TRJ .NSACTIONS ON ELECTRON DEVICES, VOL. ED-31, NO. 7, JULY 1984
`
`DEVICE
`TYPE
`
`E/ B JUNCTION
`DEPTH
`µrn
`
`HF
`
`RCA
`
`0.04
`
`0.04
`
`TABLE I
`
`BASEWIDTH
`
`µm
`
`0~ 25
`
`o. 25
`
`RESISTANCE UNDER
`THE EMITTER
`k!l/sq
`
`GAIN
`
`5 . 6 + 3.1
`4. 3 + 2 .1
`
`33 + .2
`146 :;: 15
`
`150
`
`50
`
`0
`
`-50
`
`-100
`
`-120°c
`
`I,
`
`' I
`
`I,
`I
`
`J.o
`
`6-0
`
`5-0
`
`,.o
`1000/T , K-1
`(a)
`
`103 150
`
`50
`
`0
`
`-50
`
`-100
`
`-12o'c
`
`_u
`
`I
`
`'
`
`10-9 ""0-2_....__,(}_, __._o_.6 _ _.__~o'-.a--''----'1. o
`
`VBE , Volts
`
`Fig. 3. Collector and base current as a function of base/emitter voltage
`for a device given an HF etch (device 1), and a comparable device
`given an RCA clean (device 2).
`
`HF device. This correlates well with the arsenic profiles in
`Fig. 2.
`The major difference between the two types of device su m(cid:173)
`marized in Table I is the value of gain obtained. In particu lar
`the RCA device has a gain approximately five times higl er
`than that of the HF device. Similar results have been reported
`previously (8), and hence we can see that the properties of
`polysilicon emitter transistors are strongly influenced by the
`type of surface treatment carried out prior to polysilican
`deposition. In addition, since apart from the interfacial layer
`treatment, the two types of device have had identical process(cid:173)
`ing; we can infer that the physical mechanism controlling the
`gain is different for the two devices.
`In order to facilitate a comparison with the theory, two trnn(cid:173)
`sistors with identical values of resistance under the emit:er
`(4.7 kn/sq) were selected from the large number of transistors
`measured and a more detailed electrical characterization cu(cid:173)
`ried out. Fig. 3 shows a graph of base and collector ven:us
`base-emitter voltage for these two devices. The characterist ics
`for both devices are ideal with an exp qV8 E/kT depender ce
`over four decades of current. As expected, the collector char(cid:173)
`acteristics of the two devices are identical, and the improved
`gain of the RCA device arises because of a decrease in the b:ise
`current.
`Fig. 4(a) shows a graph of current gain as a function of te:n(cid:173)
`perature for two HF devices and for a conventional transistor
`for comparison. The gain was calculated from the ideal part of
`the le and /8 versus V8e characteristics at each temperature.
`Measurements above 140°C were not possible because of
`
`3-0
`
`6-0
`
`5-0
`4-0
`1000/T, K-1
`(b)
`Fig. 4. Temperature dependence of the current gain for (a) two HF
`devices (transistors 1 and 3) and a conventional transistor (dashed
`line) and (b) two RCA devices (transistors 2 and 4) and a conven(cid:173)
`tional transistor (dashed line).
`
`excessive collector leakage and below -115°C because the base
`characteristics were not ideal. The conventional device is a
`BFYSO, general-purpose n-p-n switching transistor, device 1 is
`the HF device presented in Fig. 3, and device 3 is an HF device
`processed at a different time, under different processing con(cid:173)
`ditions. It can be seen that the characteristic for the conven(cid:173)
`tional device is linear, which is as expected for this type of
`device. In contrast, the characteristics for the two HF devices
`are nonlinear, with the gain increasing more rapidly with tem(cid:173)
`perature at high temperatures.
`Fig. 4(b) shows the temperature dependence of the gain for
`two RCA devices and the conventional transistor for compari(cid:173)
`son. The upper characteristic (device 2) is for the RCA device
`presented in Fig. 3 and the lower characteristic (device 4) is
`for a device processed at a different time , under different
`processing conditions. The characteristics for the two RCA
`devices are again nonlinear, but the shapes of the characteristics
`are significantly different than those obtained for the HF
`devices in Fig. 4(a). At low temperatures the characteristics
`of the RCA and HF devices have similar slopes, but at high
`temperatures the rate of increase of gain with temperature
`shows a marked decrease for the RCA devices, whereas the
`opposite occurs for the HF devices. This effect is particularly
`noticeable in device 4.
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on May 15,2024 at 03:07:33 UTC from IEEE Xplore. Restrictions apply.
`
`

`

`ASHBURN AND SOEROWIRDJO: COMPARISON OF RESULTS ON POLYSILICON EMITTER TRANSISTORS
`
`857
`
`,o• ,----,--------,---,-------,,-----,
`
`i
`"' q .,
`...
`>--
`
`'.c a,
`8
`
`-19
`10
`
`-20
`10
`
`10·21
`
`10·22
`
`10·23
`
`,o·24 '---~3..,.o _ _ _ ~3,.,_.5 ___ 1 ---',.o
`1000/T , K
`Graph oflnlco/T4 versus inverse temperature for the RCA and
`HF devices.
`
`Fig. 6.
`
`7 1019
`E
`0
`
`oxide
`
`silicon
`
`···~·~··
`I ! •
`
`••
`..
`•
`
`I
`
`:
`
`•
`
`I EfB
`:Ji..ricl ion
`
`•
`
`1000/T, K-1
`Fig. 5. Temperature· dependence of the resistance under the emitter
`for the RCA and HF devices.
`
`Fig. 5 shows the temperature dependence of the resistance
`under the emitter for the two types of device. These measure(cid:173)
`ments were made on four terminal pinch resistors located
`adjacent to the transistors under test. It can be seen that the
`two types of device have identical characteristics, and that the
`resistance is approximately flat over the temperature range of
`the measurements.
`
`V. COMPARISON OF EXPERIMENTAL RESULTS
`AND THEORY
`
`A. HF Devices
`
`The purpose of this section is to explain the shape of the
`graph of hFE versus 1000/T for the HF devices and hence
`determine which theory is applicable to these devices. The
`first stage of the analysis is to obtain measurements of the
`bandgap narrowing in the base and emitter regions of the
`transistor. From (7), the bandgap in the base Egb can be
`obtained from the slope of a graph of lnlcoPbfT4 versus
`inverse temperature as shown in Fig. 6. The bandgap narrow(cid:173)
`ing in the base AEgb can be obtained by subtracting the band(cid:173)
`gap in the base Egb from the bandgap for undoped silicon
`~xtrapolated to O K (1206 meV) (30]. The value of bandgap
`narrowing obtained from this analysis is AEgb = 74 + 10 meV.
`In order to confirm that the preceding value is reasonable for
`our devices, it is possible to extract the base doping concen(cid:173)
`tration from the measured bandgap narrowing using Slotboom's
`(30] formula and compare this with the theoretical base
`profile in our devices. Using this method, a boron concentra(cid:173)
`tion of 6 X 1018 cm-3 is obtained from our value of 74 meV
`for the bandgap narrowing in the base. Fig. 7 shows the
`theoretical boron profile obtained from the SUPREM process
`simulation program. This gives a peak base doping concentra(cid:173)
`tion of approximately 5 X 10 18 cm-3 which is in good agree(cid:173)
`ment with the value of 6 X 1018 cm-3 obtained from the
`bandgap narrowing data.
`The bandgap narrowing in the emitter can be extracted from
`the temperature dependence of the gain, as plotted in Fig. 4.
`At temperatures below 240 K, the resistance under the emitter
`shown in Fig. 5 is constant, and hence its influence on the
`temperature dependence of the gain can be neglected. In addi(cid:173)
`tion in our devices, the AT0•75 term in (15) is greater than
`unity since Dp 2 /Dp 1 ~ 3 [9] and W2 < Lp 1
`. The bandgap nar-
`rowing in the emitter can therefore be obtained from the
`slope, measured at low temperatures, of a graph of ln hFEI
`
`I
`I
`I
`I
`
`I
`I
`I
`I
`I
`••••••••••••
`I
`1015 ""=-oc-:-.1,--0='=.oo~--=-o.,!--::20~·=·· --="o.,~IJ',--....,0~.50
`
`•
`
`BIC
`Junction
`I
`
`Distance , JJm
`Fig. 7. Theoretical boron profile obtained from the SUPREM process
`simulation program.
`
`T0•75 versus inverse temperature. Using this method a value of
`96 + 13 meV is obtained for the bandgap narrowing in the
`emitter for the HF device.
`Values of bandgap narrowing in the emitter ranging from
`70 to 170 meV (30] -[32] have been reported in the literature
`for various types of bipolar transistors.
`In order to explain
`this wide spread in experimental results, theoretical models
`have been proposed by Van Overstraeten et al. (33], Lanyon
`and Tuft [34], and Slotboom and de Graaff (30] which relate
`the bandgap narrowing to the emitter concentration at high
`doping leve1s. Using these models we obtain an emitter concen(cid:173)
`tration of between 1.8 and 5.0 X 1019 cm- 3 for our value of
`96 meV for the bandgap narrowing in the emitter. The arsenic
`profiles shown in Fig. 2 indicate that the arsenic concentration
`iµ the polysilicon is approximately 3 X 1020 cm - 3 . However,
`Rutherford backscattering gives the total arsenic concentration,
`and we would expect the electrically active concentration to be
`considerably lower. Ryssel et al. (35] have compared the
`total arsenic concentration in ion-implanted polysilicon (1 X
`1016 cm-2 at 90 keV) obtained from Rutherford backscatter-
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on May 15,2024 at 03:07:33 UTC from IEEE Xplore. Restrictions apply.
`
`

`

`858
`
`IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. ED-31, NO. 7, JULY 1984
`
`1 0 0 r - - - - . - - . - - - - , - - - , - -
`
`10
`
`400
`
`100
`
`40
`
`2-0
`
`30
`
`6-0
`
`1000/T ,
`
`30
`
`40
`5-0
`1000/T, K-1
`
`6-0
`
`Fig. 8. Fit obtained between the transport theory (15) and the exp::ri(cid:173)
`mental results (HF device 1) for AT 0 -75 >> 1 (curve a), AT o.75 << 1
`(curve c), and for an intermediate case, (curve b).
`
`Fig. 9. Fit obtained between the modified tunneling theory (8) and the
`experimental results (RCA device 2) for three values of the parameter
`Ch.
`
`ing with the electrically active concentration obtained from
`Hall effect measurements. They found, after a drive-in sched· !le
`of 30 min at 950°C, that only 25-30 percent of the arsenic
`was electrically active.
`In our devices, the emitter drive-in
`was 20 min at 900°C, and hence we would expect an e,en
`smaller proportion of the arsenic to be electrically acti,e.
`Taking into account this information, the value of 96 meV for
`the bandgap narrowing in the emitter is not unreasonable.
`Further confirmation of this point can be obtained by consld(cid:173)
`ering the emitter-base junction depth of 0.04 µm, and compu(cid:173)
`ing this with the mean polysilicon grain size. Cross sectiJn
`TEM observations (5] have shown that, in our devices, the
`mean grain size is between 0.1 and 0.2 µm. We can therefore
`see that the arsenic is unlikely to have penetrated into the
`center of the grains, and hence we would not expect the ek c(cid:173)
`trical activity to be high.
`Having obtained values for the bandgap narrowing in t '.1e
`base and emitter, we are now in a position to investigate hc,w
`the other parameters in (15) influence the temperature depen(cid:173)
`In order to do this y;e need to assume a
`dence of the gain.
`value for the parameter A in (15). Fig. 8 shows a comparison
`of the theoretical and experimental gain versus inverse tem(cid:173)
`perature for AT0 •75 >> 1, AT0•75 << 1, and for an interrr. e(cid:173)
`diate case.
`It can be seen that the best agreement betwe,m
`theory and experiment is obtained for AT0•75 >> 1. Sin;e
`Ning and Isaac [9] have reported that Dp2 /Dp 1 :c!! 3, tr.is
`implies that Lp 1 >> W2 . Assuming an emitter doping concen(cid:173)
`tration of 5 X 1019 cm-3 , as obtained from the bandgap nu(cid:173)
`rowing in the emitter measurements , and using (11) and (1?,)
`with Dp 1 == 0.43 cm 2 • s- 1 [9], [36] we obtain Lp 1 > o. : A
`µm. This value is therefore consistent with the criterion th , t
`ATo.?s >> 1.
`The agreement between theory and experiment shown .n
`Fig. 8 is excellent at low temperatures but somewhat poorer it
`high temperatures. A perfect fit with theory can be obtairn d
`if the r 0 •75 term in (15) is replaced by a r1.o term. This smd l
`discrepancy between the experimental and theoretical results
`indicates that an additional temperature dependent term is
`present which has not been included in (15). The most like:y
`cause of this discrepancy is that the temperature dependence
`of the mobility in the base or in the polysilicon has not bem
`
`In particular, the mobility in the poly(cid:173)
`correctly described.
`silicon quoted in (14) is for majority carriers [24], whereas in
`the emitter ·of our device the holes are minority carriers. Al(cid:173)
`ternatively it might be explained by oversimplification in
`deriving (15), since we have not made allowance for possible
`differences in doping concentration in the polycrystalline and
`single-crystalline parts of the emitter or for the nonuniform
`doping concentration in the base.
`
`B. RCA Devices
`The first stage of this analysis is again to obtain measure(cid:173)
`ments for the bandgap narrowing in the base and emitter. As
`with the HF device, the bandgap narrowing in the base is
`obtained from the temperature dependence of the collector
`current as shown in Fig. 6. The value obtained is 74 + 10 meV
`which, as expected, is identical to that obtained for the HF
`device. The bandgap narrowing in the emitter can be obtained
`from the temperature dependence of the gain as described by
`(8). At low temperatures (1000/T > 4.5) the Ch kT term in
`(8) is small with respect to unity and the resistance under the
`emitter is constant. Consequently the bandgap narrowing in
`the emitter can be obtained from the slope of a graph of
`1n hpE/T0•5 versus inverse temperature. This gives a value of
`101 meV which, as expected, is close to the value of 96 meV
`obtained for the HF devices.
`Having obtained values for the bandgap narrowing in the
`base and emitter we are now in a position to fit our modified
`tunneling theory to the experimental results for the RCA
`device. Fig. 9 shows a comparison of theoretical and experi(cid:173)
`mental gain versus inverse temperature for three values of the
`parameter Ch in (8). It can be seen that an excellent fit can
`be obtained for a value of Ch == 4 ev- 1 . This parameter
`depends upon the interfacial layer thickness fi and the effec(cid:173)
`tive barrier height for holes Xh, and hence from the measured
`value of fi we should be able to obtain an estimate for Xh ·
`Henderson [26] used Auger electron spectroscopy to measure
`the thickness of oxide produced by the RCA clean and ob(cid:173)
`tained a value of approximately 14 A. If we insert this value
`into (5) we obtain 500 meV for the effective barrier height for
`holes. Ng and Card [37] have measured the hole and electron
`tunneling barriers on MOS structures with ultra thin ( <40 A)
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on May 15,2024 at 03:07:33 UTC from IEEE Xplore. Restrictions apply.
`
`

`

`ASHBURN AND SOEROWIRDJO: COMPARISON OF RESULTS ON POLYSILICON EMITTER TRANSISTORS
`
`859
`
`I I
`
`30
`
`,o
`
`I
`
`I
`
`I
`
`I
`
`,0
`
`20
`30
`Oxide Thickness 151 , A
`Fig. 10. Graph, taken from the literature [37], showing the relation(cid:173)
`ship between the effective barrier height for holes and the interfacial
`oxide thickness.
`
`oxide layers, and their data is summarized in Fig. 10. Extrap(cid:173)
`olating their results to an oxide thickness of 14 A gives an
`effective barrier height for holes of approximately 700 meV,
`which is in good agreement with. the value of 500 meV ob(cid:173)
`in our finished
`tained from our electrical measurements.
`transistors it is likely that the oxide layer is thinner than 14 A
`since the devices were subjected to a high-temperature emitter
`drive-in, and Duffill [35] has shown that the oxide thickness
`decreases during high-temperature processing. This would
`have the effect of reducing the value of effective barrier height
`obtained from (5).
`It i~ worth noting at this point that the measurements of
`Ng and Card [37] indicate that the effective barrier height
`for electrons is very much smaller than that for holes. ·Quali(cid:173)
`tatively, this means that the tunneling probability is very
`much higher for electrons than holes, and consequently elec(cid:173)
`trons can penetrate the oxide .barrier more easily. The signifi(cid:173)
`cance of this for polysilicon emitter transistors is that the base
`current is suppressed, but the emitter current is not. This
`explains why very little voltage is dropped across the inter(cid:173)
`facial oxide.
`The modified tunneling theory summarized in (8) and Fig. 9
`predicts , that the gain should decrease with ternperature at
`very high temperatures and for very thick interfacial oxides.
`In our devices, measurements at temperatures >420 K could
`not be made because of excessive collector leakage. However,
`in an attempt to observe this effect we have made measure(cid:173)
`ments on a number of RCA devices processed at different
`times. The characteristic for device 4 in Fig. 4(b) shows a
`very strong decrease iii the rate of increase of the gain at high
`temperatures, and at temperatures above 350 K the gain is
`In addition de Graaff and de Groot
`very nearly constant.
`[10] reported that in some cases the temperature coefficient
`of the gain of their polys