POLYSILICON EMITTERS FOR SILICON CONCENTRATOR
`SOLAR CELLS
`
`J. Y. Gan and R. M. Swanson
`
`Stanford Electronics Laboratories, McCullough 204
`Stanford University, Stanford CA. 94305
`
`ABSTRACT
`
`Polysilicon emitters with a thermally grown interfacial
`oxide are examined in great detail. Both J, and p, are
`measured for the polysilicon emitter (contact) which is
`annealed under various conditions. J,<5 x 10714 A/em?
`and p.~1 x 107° 2-cm? are obtained for both n- and p-
`type polysilicon emitters. The results also indicate that
`the interfacial oxide is only broken by a very small frac-
`tion in order to have p.~1 x 107° Q-cm?. Under such a
`condition, J, is primarily dominated by the recombina-
`tion at the oxide interface.
`
`INTRODUCTION
`
`Polysilicon emitters, because of their important applica-
`tions in high speed bipolar transistors, have been widely
`studied. [t was found that, between the polysilicon and
`the substrate, there always cxists a very thin oxide (in-
`terfacial oxide), which can be as a by-product of the
`cleaning process or is grown intentionally[1,2,3,4,5,6,7].
`In either case, the oxide is not thermally stable; 7.e.,
`it tends to break up during high temperature anneals
`[1,2,3,4,5,6,7]. The breakup of the interfacial oxide has
`direct, impact on both majority and minority carrier
`transport in the polysilicon emitter. This can be seen
`from the dependence of the specific contact resistivity
`(pe) [6] and the the emitter saturation current density
`(J,) [1,2,3,4,5,6,7) upon either the anneal condition or
`the oxide integrity. When the effect of the interfacial ox-
`ide is negligible, such as when the interfacial oxide loses
`its integrity in high temperature anneals, the p, is low
`(~2 x 1077 Q-cm?) [6] and the J, is high (~1 x 10717)
`[1,3]. Conversely, when both carriers are mostly blocked
`by the interfacial oxide, like for an intact chemical ox-
`ide (~15 A), the p, is high and the J, is low. Typically,
`pe21x 104 Q-cm?, and the corresponding J,~1 x 10-4
`
`A/cm? have been observed on the n-type polysilicon
`emitter [1,5,6].
`
`With the extent of the oxide breakup between these two
`extremes, various combinations of p, and J, can be ob-
`tained. It is of particular interest to know the p, and the
`corresponding J, when only a small fraction of the in-
`terfacial oxide is broken. Forsilicon solar cells operated
`under high concentrated sunlight, the emitter recombi-
`nation is one of the main recombination that limit the
`solar cell performance [8]. This is because the diffused-
`emitter, which is formed with dopant diffusion into the
`base (substrate), has high J, (=4.5 x 10713 A/em?).
`Consequently,
`the cell performance can be improved
`if the diffused-emitter is replaced with the polysilicon
`emitter, of which the J, is less than 4.5 x 107! A/cm?
`and the p, is low enough.
`
`In this paper, detailed calibrations for the polysilicon
`emitter with a thermally grown interfacial oxide is pre-
`sented. Compared with the the chemical oxide,
`the
`thermal oxide appears to be more robust to the high
`temperature anneal. Consequently, this lends more con-
`trollability on the oxide breakup process.
`In addition,
`two separate anneal steps were used in order to avoid
`any possible interference from the dopant diffusion. It
`has been reported that the interfacial oxide tends to
`break up more rapidly with the existence of dopant[3,5].
`Moreover, J, also depends on the surface dopant con-
`centration and the emitter depth [9].
`In order to see
`the effect of the oxide breakup on J, and p, alone, it
`is therefore desirable to separate these two processes.
`The first-step anneal, which serves to break the oxide,
`was performed at higher temperature than that for the
`second-step anneal which aims for the dopant drive-in.
`
`EXPERIMENTAL
`
`Procedures
`
`245
`
`0160-8371/90/0000-0245 © $1.00 1990 IEEE
`
`HANWHA1005
`
`HANWHA 1005
`
`

`

`
`
`cantact
`
`polystlicon-substrate
`contact
`
` metal-substrate
`
`
`Figure 1: The device structure for the specific contact
`resistivity measurement.
`
`In this work, the J, and p, were measured indepen-
`dently for each type of polysilicon emitter (or contact).
`J. was measured with photoconductivity decay method
`which has been reported elsewhere [10]. Because high
`bulk lifetime is needed for such a measurement,
`the
`samples were prepared with high resistivity (>100 Q-
`em) FZ wafers.
`In contrast, p. was measured on de-
`vices made on CZ wafers of which resistivity is 0.28 and
`0.5 Q-cm for p- and n- type respectively. The device
`for p,-measurement is shown in Fig. 1 schematically.
`Except for the pattern-definition steps which were not
`required for J,-measurement samples, the two sets of
`wafers were always processed together. The process se-
`quence is shown as followings:
`
`bw . Field oxidation (2000 A)*.
`Nm
`. Polysilicon-substrate contact opening™.
`3. Thin thermaloxide growth at 800 °C*. (10-minute
`oxygen flow)
`. Polysilicon deposition (6000 A) and first-step an-
`neal (1030 °C)*.
`. First-step auneal *,
`. Polysilicon finger definition.
`. Undoped glass deposition (2000 A).
`. Metal-polysilicon and metal-substrate contact open-
`ing.
`9. Dopedglass deposition on both sidesof the wafer’.
`10. Drive-in; 900-1000 °C*.
`11. Doped glass stripping and Al-1% Si alloy deposi-
`tion.
`12. Metal finger definition.
`
`a O
`
`IDtr
`
`cally, the p, of the metal-polysilicon contact is less than
`1 x 107-* Q-cm*. With such low p, and large contact
`Note that the steps marked with “x” are the steps for
`area between the metal and the polysilicon, the metal-
`J,-measurement samples.
`In addition, the J,-samples
`polysilicon contact resistance can be ignored as well. As
`are divided into two sets; one set of samples had an
`aresult, the series resistance derived froin V,3/T is essen-
`additional 400 °C-30 min forming gas anneal (FGA),
`tially determined by two terms, the bulk resistance of
`the other did not. The growth of thin oxide was carried
`the silicon substrate and the polysilicon-substrate con-
`out with 260 sccm oxygen and 5 slm argon mixture.
`tact resistance. Such a series resistance can be expressed
`Typically, oxide thickness less than 20 A was measured
`as a function of the size of the polysilicon-substrate
`with ellipsometer. In step 2, a series of square contacts
`square contact.
`of size between 1 and 64 pmare openedfor polysilicon-
`
`~ ce4Pe )
`substrate contacts.
`R=Cot a,
`(1)
`or
`
`p- Measurement
`
`As shownin Fig. 1, the voltage difference, Vij, between
`probe A and B is caused by several components of the
`series resistance along the current loop. They are the
`resistance imposed by the bulk of the polysilicon and
`the wafer, the metal-polysilicon contact resistance, and
`the polysilicon-substrate contact resistance. The po-
`tential drop across the polysilicon film is conceivably
`negligible because the film is heavily doped and very
`thin. The metal-polysilicon contact resistance is actu-
`ally measured on the similar device (four-lerminalresis-
`tor [11]) next to the device shownin the figure. Typi-
`
`246
`
`R,= Rx=e.+Cpd;
`2)
`where C is a constant that accounts for the contact ge-
`ometry factor, p is the substrate resistivity, d is the
`width of the square contact, and p, is the specific con-
`tact resistivity of the polysilicon-substrate contact. The
`contact size dependence of the last term in Eqn. (2) re-
`sults from the crowding effect as the current flows to-
`ward the polysilicon-substrate contact.
`It
`is obvious
`that, from Eqn. (2), the p, can be derived from the plot
`of R, versus d which is obtained by measuring theseries
`resistance of contacts of different sizes.
`
`

`

`P-type polysilicon contacts
`Secoond-step anneal: 900 °C-4 hrs
`
`@ first-step anneal: 1050 °C-2 hrs
`
`H
`
`first-step anneal: 1050 °C-30 min
`
`
`
`
`
`R,(167+G-em?)
`
`pe.
`
`(Q-cm?) 30 min
`
`First-step anneal: 1050 °C
`Second-step anneal: 900 °C-4 hours
`
`18 min
`
`60 min
`
`120 min
`
`Contact Size (um)
`
`First-Step Anneal Time
`
`Figure 2: The A, versus the contact size for p-type
`polysilicon contacts of different first-step anneals.
`
`Figure 3: p, of both n- and p- type polysilicon-substrate
`contacts is shown as a function of the time of the
`first-step anneal.
`
`RESULTS AND DISCUSSIONS
`
`Fig. 2 shows two R,-plots for the p-type polysilicon con-
`tacts. One has the first-step anneal at 1050 °C for
`30 minutes, the other has 1050 °C for 2 hours. Each
`data point is the average of 16 measurements across the
`wafer and the error bar is drawn with 4 standard devi-
`ations. Since devices were made on the wafers of about
`the same resistivity, the slopes of the linear-fit lines are
`almost identicai. Note that, the devices that received a
`30-minute first-step anneal have higher intercepts than
`those that received «. 2-hourfirst-step anneal. This indi-
`cates that more oxide breakup occurs when the anneal
`time increases. Another characteristic of the interfacial
`oxide breakup can also be seen from the spread in the
`data.(the lengthof the error bar). The magnitude of the
`error bar is not due to the measurement uncertainty;
`instead, it results from the random distribution of the
`oxide breakup. When only a very small fraction of the
`interfacial oxide is broken, it is possible that the inter-
`facial oxide on each contact may not be broken equally.
`Consequently, this gives rise to the fluctuation of the
`resistance measured. On the other hand, the fluctua-
`tion will be subdued when more oxide is broken, This
`explains the different appearance of error bars observed
`between these two sets of data.
`
`To see how the J, and the p, is affected by the first-
`step anneal, one set of samples (and device wafers) were
`first annealed at 1050 °C for various lengths of time
`(first-step anneal). After the doped-glass deposition,
`they were all annealed at 900 °C for 4 hours. The p,
`measured fromthe device and the J, measured from the
`photoconductivity decay are presented in the following
`figures.
`
`In Fig. 3, p- is plotted against the timeof the first-step
`anneal for both types. Note that the p, decreases as the
`anneal time increases.
`In.addition, p, measured from
`both n- and p- type devices are moreor less the same,
`except those annealed with 120 minutes. The similarity
`of p. between the n- and the p- type contact also indi-
`cates that the intcraction between the dopant and the
`interfacial oxide is negligible; since such an interaction,
`if exists, would cause the p, of the p-type to be lower
`[6].
`
`The effects of the first-step anneal on J, are presented
`in Fig. 4 and 5 for the n-type and the p-type polysil-
`icon emitters respectively.
`In each figure, two sets of
`data are shown:
`the one with an FGA, and the one
`without. For the samples with an FGA, the J, is sig-
`nificantly less than that without FGA for the same an-
`neal condition. This indicates that the recombination
`
`247
`
`

`

`N-type polysilicon emitters
`First-step anneal: 1050 °C
`Second-step anneal: 900 °C-4 hrs
`
`J,(107MA/cm?)
`
`—e— without FGA 9
`
`20
`
`40
`
`60
`
`80
`
`100
`
`—— with FGA
`——_with FGA
`
`P-type polysilicon emitters
`First-step anneal: 1050 °C
`Second-step anneal: 900 °C-4 hrs
`
`
`
`J,(10-4A/em?}
`
`——*— without FGA 5
`
`20
`
`40
`
`60
`
`80
`
`100
`
`1290
`
`140
`
`Fist-Step Anneal Time (min)
`
`Fist-Step Anneal Time (min)
`
`Figure 4: The J, versus the first-step auneal time for
`n-type polysilicon emitters.
`
`Figure 5: The J, verens the first-step anneal time for
`p-type polysilicon emitters.
`
`at the oxide interface is, at least, one of the dominant
`mechanisms for the emitter recombination.
`It is also
`
`important to note that the J, of the p-type polysilicon
`emitter (Fig 5) appears to be much lower than what has
`been previously reported [1]. This may have to do to
`the use of thermal oxide as the interfacial layer in our
`samples. Either the thermal oxide is most robust during
`the high temperature anneal or the thermal oxide has
`lower interface defect state density (D,,) compared to
`the chemical oxide. Another interesting thing in these
`figures is that the J, of the sample that. received a 15-
`minute anneal is higher than the sample that received
`a 30-minute ammeal. Moreover, such a behavior occurs
`in both the n- and the p- type samples. The possible
`explanation is that the interfacial oxide quality is im-
`proved by the high temperature anneal. This is not too
`surprising because the oxide was grown around 800 °C,
`in which higher Dj is expected [12,13]. Other than that,
`the J, of both types appears to increase slowly with the
`anneal time. In addition, because the dominant recom-
`bination is likely at the oxide interface, it implies that
`the oxide must only be broken by very a small fraction,
`whichis in consistent with what was observedin Fig. 2.
`
`The effect of the second-step anneal (dopant drive-in) on
`both the J, and the p, were also investigated. All sam-
`ples (and devices) received a first-step anneal at 1050
`°Cfor one hour, followed with various treatineuts in the
`second-step anneal. The results are summarized in Ta-
`ble 1. The label on the right most columnrepresents the
`second-step anneal condition of that sample. As shown
`in the table, the J, increases with the temperature and
`time of the second-step anneal. In the meantime, the p;
`decreases as the drive-in temperature increases. Consid-
`ering the significant increase of J, caused by inercasiug
`the drive-in temperature from 900 to 1000 °C, it is ben-
`cficial to use the two-step anneal to optimize J..
`
`In Fig. 3, the decrease of p, with the annealtime is what
`would be expected from the gradual breakup of the in-
`terfacial oxide. This is in consistent with the increase
`of J, shown in Fig. 4 and Fig. 5, since J, depends on
`the oxide integrity in the opposite way. This is cspe-
`cially clear when both p, and J, are plotted together.
`Fig. 6 is the p. versus J, plot for all the results obtained
`in this experiment. The iuset is the list of the label
`
`
`Table 1: The effect of the second-step anneal ou Jy and pe.
`
`P-type
`
`
`
`Joe (X10-M A/em?)|p. (OQ-cm?)
`900 °C-4 hour
`
`3.5
`|
`18
`
`8.2
`3.8
`1000 °C-30 min
`
`
`10.5
`1000 °C-60 min
`
`
`
`
`
`248
`
`

`

`previously. On the other hand, the slow decrease of the
`fe in this regime is because the majority carrier trans-
`port is completely dominated by the channel resistance.
`The increase of the oxide breakup can be thought as
`more parallel resistors added together; consequently, a
`significant change of p, will not be observed.
`
`The data shown in Fig. 6 are quite useful for silicon
`solar cells. First, most of J, are below 3 x 107'4 A/cm?
`which is about one order magnitude lower than that of
`diffused-emitters. Sccond, mostof p, arc also lower than
`5 x 10-5 Q-cm?. This means that the potential drop
`across the contact is negligible as long as the current
`density is less than 100 A/cm*.
`It is interesting to see
`how much of the cell efficiency can be improved with
`polysilicon emitters of characteristics given here. With
`the cell modeling, which is similar to the one reported
`by Sinton [8], more than one absolute percent of the
`cell efficiency is obtained for a backside contact solar
`cell with polysilicon emitters with characteristics shown
`here (J,=2x 10°“ A/cm? for the n-type, J,=3.5x 10714
`A/ein? for the p-type, and p.=1 x 107° Q-cm? for both
`types).
`
`CONCLUSIONS
`
`To conelude, we showed that polysilicon emitters with
`low J, and a, can be obtained with a very small frac-
`tion of interfacial oxide breakup. Typically, po~1 x 107%
`Q-em?® and J,<5 x 107M A/cm? are obtained for both
`n- and p- type polysilicon emitters. From the cell mod-
`eling, polysilicon emitters with these qualities can be
`quite useful for silicon concentratorsolarcells. Because
`only small fraction of interfacial oxide is broken, the J, is
`primarily dominated by the recombination at the oxide
`interface and in the single-crystal diffusion region. Con-
`sequently, an interfacial oxide of good quality is impor-
`tant to obtain low J,. Our results also indicates that p,
`drops drastically when the oxide starts to breakup. This
`is explained by the transition of the dominant majority-
`carrier transport from the tunneling to the channel re-
`sistance.
`
`
`‘This work was supported by the Electric Power Re-
`search Institute.
`
`s°aeaaaae+a*" J
`
`1, 1050/15-900-240
`P, 1080/15-900/240
`n, 1050/30-900/240
`p, 1050/80-900/240
`n, 1050/60-900/240
`p, 1050/60-900/240
`n, 1060/120-900/240
`p, 1050/120-900/240
`1, 1050/60-1 000/30
`p, 1050/60-1 000/80
`p, 1000/240-900/240
`p, 1000/240-900/240
`p, 1050/15-1 000/30
`
`, (lo-! A/fem?)
`
`
`
`pe{Q-em?)}
`
`Figure 6: p.-J., of this work.
`
`for cach point. It specifies the type of polysilicon emit
`ter, thefirst. step anneal (temmperature/minute), and the
`second-step anneal (temperature/minute). As shown in
`the figure, the data runs from the upper-left corner to
`the lower-right corner whichis similar to that reported
`by Crabbé [6] and is the typical trend when both p. and
`J, ave regulated by the extent of oxide breakup. Nev-
`ertheless, from left to right,
`two distinct regimes are
`shown in the figure.
`In one regime, J, is almost con-
`steut while p. drops drastically.
`In the other regime,
`p. decreases very slowly and .J, inereascs in muchfaster
`pace. In the first regime, that the J, is inert to the ox-
`ide breakup is because the extent of the oxide breakup
`is very small such that the J, is still dominated by the
`recombinations at the oxide interface and in the singly-
`erystal outdiffusion. On the other hand, the drastic
`change of the p, can be attributed to the transition of
`the majority carrier transport from the tunneling to the
`channel resistance. Whenthe interfacial oxide is intact,
`tumucling is the only process available for the majority
`carriers to pass through theinterfacial oxide. This usu-
`ally results in a high contact resistance because of low
`tnuneling probability. As the oxide starts to break up,
`additional highly conductive channels are created duc
`to the intimate contacts formed between the polysilicon
`and the substrate in the broken regions [3,4,5'. A dras-
`tic change of p.
`is observed when more currents flow
`through those channels.
`In the other regime, J, starts
`to increase because the recombiuationin the polysilicon
`becomes comparable to the recombination mentioned
`
`249
`
`

`

`
`
`
`
`[11] Simon $. Cohen and Gennady SH. Gildenblat, ed-
`itors.
`“Metal-Semiconductor Contacts and De-
`vices,” Volume 13 of VESI Electronics Micorstruc-
`ture Science, Academic Press, 1986.
`
`Reda R. Razouk and Bruce E. Deal, “Dependence
`of Interface State Density on Silicon Thermal Ox-
`idation Process Variables,”
`J. Electrochem. Soc.,
`126(9):1573, September 1979.
`[13] M. Revitz, S. J. Raider, and R. A. Gdula. J. Vae.
`Sci. Technol., 16(2):345, 1979.
`
`References
`
`(1)
`
`Y. Kwark, R. A. Sinton, and R. M. Swanson, “Low
`J, Contact Structures Using SIPOS and Polysili-
`con Films,” In 18" LEEE Photovoltaic Specialists
`Conference, page 787, 1985.
`
`“The SIS
`H. C. de Graaff and J. G. de Groot,
`Tunnel Emitter: A Theory Emitters with Thin In-
`terface Lasyers,”
`IEEE Trans. Electron Devices,
`ED-26(11):1771, November 1979.
`
`G. L. Patton, J. C. Bravman, and J. D. Plummer,
`“Impact of Processing Parametcrs on Base Current
`in Polysilicon Contacted Bipolar Transistors,” In
`IEDM Tech. Dig., page 30, 19865.
`
`J. C. Bravman, G. L. Patton, and J. D. Plum-
`mer, “Structure and Morphologyof Polycrystalline
`Silicon-Single Crystal Silicon Interfaces,” J. Appl.
`Phys., 57(8):2279, April 1987.
`
`G. R. Wolstenholme, N. Jorgensen, P. Ashburn,
`and G. R. Booker, “An Investigation of the Ther-
`mal Stability of the Interfacial Oxide in Polycrys-
`tallinc Silicon Emitter Bipolar Transistors hy Com-
`
`paring Device Results with High Resolution Elec-
`tron Microscopy Observation,”
`J. Appl. Phys.,
`61(1):225, January 1987.
`
`in
`“Carrier Transport
`Emmanuel F. Crabbé,
`Polysilicon Emitter Contacts for Scaled Bipolar
`Transistors,”
`PhD thesis, Stanford University,
`1988.
`
`P. Ashburn and B. Soerowirdjo, “Comparison of
`Experimental and Theorectical Results on Polvysil-
`icon Emitter Bipolar Transistors,”
`[EEE Trans.
`Electron Devices, ED-31(7}:853, July 1984.
`
`R. A. Sinton and R. M. Swanson, “An Optimiza-
`tion Study of Si Point-Contact Concentrator Solar
`Cells,” In 19" TEEE Photovoltuie Specialists Con-
`ference, page 1201, 1987.
`
`R. R. King R. A. Sinton and R. M. Swanson,
`“Studies of Diffused Phosphorus Emitters: Satura-
`tion Current, Surface Recombination Velocity, and
`Quantum Efficiency,”
`JEEE Trans. Electron De-
`vices, 37(2):365, February 1990.
`
`[10]
`
` “Measure-
`D. E. Kane and R. M. Swanson,
`ment of the Emitter Saturation Current by a
`Contactless Photoconductivity Decay Method,”
`In 18% TEEE Photovoltaic Specialists Conference,
`page 578, 1935.
`
`250
`
`