1710
`
`IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 45, NO. 8, AUGUST 1998
`
`18% Efficient Silicon Photovoltaic Devices
`by Rapid Thermal Diffusion and Oxidation
`
`Parag Doshi and Ajeet Rohatgi, Fellow, IEEE
`
`This
`|Study
`
`18.6
`
`18.0.
`
`al7.3
`nO
`
`16.0
`
`014.1
`(100 cm?
`ime-Si)
`BB]
`
`1
`
`1997
`
`Float Zone (FZ)
`
`Multicrystalline (mc-Si)
`‘
`
`1
`
`fl
`
`1991
`Year
`
`1993
`
`1995
`
`Czochralski (Cz) —
`
`Web (DW)
`
`By
`
`17+
`
`16
`
`+
`
`15
`
`14
`
`13
`
`Efficiency(%)
`
`12
`1985
`
`1987
`
`\
`
`1989
`
`Fig. 1. Progress of RTP-diffused solar cells on various silicon materials. All
`cells were formed without any high-temperature furnace processing (except
`for 17.3% dendritic web cell which had a CFO). The number in brackets
`represent the reference number.
`
`Abstract— For the first time, cells formed by rapid thermal
`processing (RTP) have resulted in 18%-efficient 1 and 4 cm2
`single-crystal silicon solar cells. Front surface passivation by
`rapid thermal oxidation (RTO) significantly enhanced the short
`wavelength response and decreased the effective front surface
`recombination velocity (including contact effects) from 7:5 105
`to about 2 104 cm/s. This improvement resulted in an increase
`of about 1% (absolute) in energy conversion efficiency, up to
`20 mV in Voc, and about 1 mA/cm2 in Jsc: These RTO-induced
`enhancements are shown to be consistent with model calculations.
`Since only 3 to 4 min are required to simultaneously form the
`phosphorus emitter and aluminum back-surface-field (BSF) and
`5 to 6 min are required for growing the RTO, this RTP/RTO
`process represents the fastest technology for diffusing and ox-
`idizing  18%-efficient solar cells. Both cycles incorporate an
`in situ anneal lasting about 1.5 min to preserve the minority
`carrier lifetime of lower quality materials such as dendritic-web
`and multicrystalline silicon. These high-efficiency cells confirmed
`that RTP results in equivalent performance to cells fabricated by
`conventional furnace processing (CFP). Detailed characterization
`and modeling reveals that because of RTO passivation of the
`front surface (which reduced Joe by nearly a factor of ten), these
`RTP/RTO cells have become base dominated (Job  Joe), and
`further improvement in cell efficiency is possible by a reduction in
`back surface recombination velocity (BSRV). Based upon model
`calculations, decreasing the BSRV to 200 cm/s is expected to give
`20%-efficient RTP/RTO cells.
`
`Index Terms— Diffusion, oxidation, photovoltaics, rapid ther-
`mal processing, solar cells.
`
`I. INTRODUCTION
`
`manufacturers do not even employ oxide passivation since
`the addition of lengthy oxidation steps in a furnace would
`only retard the throughput further. Rapid thermal processing
`(RTP) directly addresses these issues since the processing time
`for diffusion and oxidation is reduced from hours to only
`minutes or seconds. In addition, an emitter and BSF can be
`formed simultaneously without cross-contamination in a cold-
`wall RTP system. It is important to note, that transfer of
`RTP technology to industry will require the development of
`high-throughput RTP machines, however, several groups are
`pursuing this goal and multiwafer RTP machines have recently
`been built [2].
`Fig. 1 shows that several researchers have fabricated RTP-
`diffused solar cells [3]–[9]. Although fair progress in RTP cell
`efficiency has evolved over the years, until now, these efficien-
`cies were lower than those made by their furnace counterpart.
`In previous studies, we demonstrated high-efficiency RTP cells
`of 17.1, 16.8, 14.8, and 15.1% on FZ, Cz, multicrystalline,
`and dendritic web silicon, respectively [10]. These cells were
`formed by rapid, simultaneous diffusion of a phosphorus
`emitter on the front and formation of an aluminum BSF
`on the back. No high-temperature furnace processes were
`employed. To maintain a high minority-carrier bulk lifetime
`a short in-situ anneal or slow-cool was utilized in the simulta-
`neous RTP diffusion cycle for materials such as dendritic-web
`and multicrystalline silicon (mc-Si) which are susceptible to
`quenching-induced lifetime degradation [11]. When compared
`with cells made by conventional furnace processing (CFP),
`0018–9383/98$10.00 ª
`
`LOW COST and high efficiency are the keys to large-scale
`
`acceptability of photovoltaic (PV) systems. Since PV
`modules cost about four dollars per watt, which can produce
`electricity at a rate of about 25 cents per kilowatthour, a
`factor of two to three in cost reduction is required to make
`PV attractive for peak load applications. This can be done by
`reducing manufacturing costs and increasing production output
`to employ the advantages of economies of scale. Reference
`[1] emphasizes that
`to meet
`the production output of 25
`MW/yr close to 50 000 cells (each producing 1.5 W) must
`be produced per day. This translates into one cell every 1.7
`s. To meet the goals of 50 or 100 MW/yr it is important
`to reduce the overall cell processing time. In industry, the
`emitter diffusion and BSF formation (if employed) steps
`frequently bottleneck the cell fabrication process. Most PV
`
`Manuscript received March 26, 1997; revised January 17, 1998. The review
`of this paper was arranged by Editor P. N. Panayotatos. This research was
`supported by Sandia National Laboratories under Subcontract AA-1638.
`The authors are with the University Center of Excellence for Photovoltaics
`Research and Education, School of Electrical and Computer Engineering,
`Georgia Institute of Technology, Atlanta, GA 30332 USA.
`Publisher Item Identifier S 0018-9383(98)05260-5.
`
`1998 IEEE
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on June 22,2024 at 19:30:41 UTC from IEEE Xplore. Restrictions apply.
`
`

`

`DOSHI AND ROHATGI: 18% EFFICIENT SILICON PHOTOVOLTAIC DEVICES
`
`1711
`
`OQ,
`
`N,
`
`FG
`
`5°C/min
`
`(a) CFO cycle
`
`1
`
`5
`
`1
`
`15
`Time (min)
`(a)
`
`N,
`
`O,
`
`N,
`
`cis
`
`1
`
`105
`
`3
`
`135
`
`(b) RTO cycle
`
`\
`
`20
`
`185
`Time (s)
`(b)
`
`265
`
`1
`
`~325
`
`1000
`
`900
`
`800
`
`700
`
`t+
`
`Stabalize
`
`600 +
`
`500 |.
`
`400 +
`
`300
`
`0
`
`Temp.°C)
`
`1000 ;_
`900
`o
`S 800
`700
`600
`& 500 ¢
`400 +
`300
`
`a=
`
`0
`
`Fig. 2. Temperature cycle for growing (a) conventional furnace oxide (CFO)
`and (b) rapid thermal oxide (RTO).
`
`|
`
`Sheet Res. =100-120 Q/sq.
`SiO, Thickness = 12 nm
`After RTO
`
`Before RTO
`
`1E+20
`
`1E+19
`
`1E+18
`
`1B+17
`
`1E+16
`
`1E+15
`
`1E+14
`
`ElectronConcentration(em”)
`
`Qo
`
`0.05
`
`O01
`
`0.15
`
`02
`
`0.25
`
`0.3
`
`0.35
`
`0.4
`
`Depth (um)
`
`Fig. 3. Spreading resistance profiles of diffused emitters before and after
`the RTO cycle.
`
`The RTO cycle serves not only to form a high-quality oxide
`in a total time of 5 min instead of over 2 h, but also to reduce
`the surface concentration
`and increase the junction depth
`(Ns)
`as depicted by the spreading resistance profiles shown in
`Fig. 3. In the past, our RTP emitters were made by a heavier
`21
`—3
`doped phosphorus spin-on concentration of 10
`cm [12],
`N, = 2x 107
`-3
`[15] resulting in emitters with
`cm
`and
`m, whereas the emitters in this study were made
`0.
`20
`Ns, < 10
`—3
`by the 10
`cm spin-on resulting in lower
`cm
`—~3
`and a deeper
`in the 0.18–0.26 m range [see Fig. 3].
`xy
`The decrease in surface concentration is beneficial (provided
`good surface passivation is obtained) because it results in
`reduced front-surface-recombination-velocity (FSRV)
`for
`oxide passivated emitters [16]. Cell fabrication was finally
`completed with the evaporation of a ZnS/MgF antireflection
`2
`coating and 15 min contact anneal at 400 C Unlike our
`previous work [12], in this study, PECVD AR coatings were
`not used so that parasitic absorption within SiN [17] would
`
`bb
`
`RTP cells matched CFP cells in the long wavelength internal
`quantum efficiency (IQE) response [12]; however, in the short
`wavelengths, RTP cells suffered from a lack of adequate
`front surface passivation and heavy doping effects (Auger
`recombination and bandgap narrowing) in the emitter. In
`this paper, we present substantial enhancements in the short
`wavelength response by growing a front passivating oxide by
`rapid thermal processing (RTO) on lower doped RTP emitters.
`The use of an RTO for PV devices was discussed by
`Schindler et al. [8]. An RTO-passivated cell with an effi-
`ciency of 16.6% was fabricated, however a CFP emitter was
`employed. Recently, Sivoththaman et al. [9] fabricated fully
`RTP-processed 100 cm cells on mc-Si, which utilized an RTO
`to achieve efficiencies up to 14.1%. Considering the large area
`of the device, this result is practical except that (as in this
`study) photolithographically-defined evaporated contacts were
`used to permit a lowly doped emitter. Compared with screen-
`printed contacts, evaporated contacts give nearly 2% greater
`cell efficiency because of very low grid shading, superior
`contact quality, and better short wavelength response [13].
`In addition, since RTP cooling rates were not controlled,
`Sivoththaman’s mc-Si cells may have been susceptible to
`quenching-induced lifetime degradation.
`This paper shows that high-throughput RTP cells fabricated
`with an RTO deliver the same performance as cells with a con-
`ventional furnace oxide (CFO). We have been able to bridge
`the efficiency gap between RTP and CFP cells we had reported
`earlier [12]. By implementing RTO passivation, RTP cells have
`for the first time reached efficiencies above 18%—making the
`RTP/RTO process one of the fastest technologies to fabricate
`high-efficiency photovoltaic devices.
`
`II. EXPERIMENTAL
`Before fabricating RTP cells with front oxides, a passiva-
`(J
`tion study involving emitter saturation current density
`measurements of SiO and SiN-passivated RTP emitters was
`conducted.
`was measured by the PCD technique [14] on
`Joe
`samples simultaneously diffused on both sides by phosphorus
`spin-on dopants. The CFO was formed by inserting wafers
`into the furnace at 850 C, allowing 5 min for temperature
`stabilization, growing a 10–15 nm thick oxide for 10 min,
`ramping down to 400 C at 5 C/s for 90 min, and annealing
`in forming gas for 30 min [see Fig. 2(a)]. The SiN films
`were deposited to a thickness of 65–70 nm in a PECVD
`system operating at 13.6 MHz. Deposition conditions included
`a temperature of 300 C, RF power of
`30 W, pressure of
`0.9 torr, and SiH NH flowrate ratio adjusted to give a SiN
`4/
`3
`refractive index of 2.14–2.17 (measured at 632.8 nm).
`+
`+
`Simple n pp
`cells were also fabricated. After initial
`wafer cleaning, aluminum evaporated onto the back and
`phosphorus dopants (with a concentration of 10
`cm ) spun
`onto the front were RTP diffused to simultaneously form the
`emitter and BSF. Next, a CFO or RTO was grown followed
`by evaporation of back and front contacts, which were
`defined by photolithography. The simultaneous RTP diffusion
`was done at about 890 C for only 30 s to achieve an emitter
`Q
`sheet resistance of 90–120
`/sq.; and RTO was done at
`900 C/165 s [see Fig. 2(b)] to grow a 12- to 13-nm oxide.
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on June 22,2024 at 19:30:41 UTC from IEEE Xplore. Restrictions apply.
`
`

`

`1712
`
`IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 45, NO. 8, AUGUST 1998
`
`-«- RTP/CFO
`* CFP
`
`SiN pa
`
`CFO pass
`
`&
`
`aad
`
`Process and Cell ID
`RTP/CFO (2 Q-cm FZ; 2-200x-2D)
`CFP* (2 Q-cm FZ; F13-1)
`[* CFP process includes growth of a CFO.
`
`See | FF | Eff
`Voe|
`(mV) |(mAsem*))
`(we)|
`35.9
`|0.805| 17.9
`[618]
`36.6
`{0.798} 18.1
`620]
`
`40
`
`50
`
`70
`60
`Sheet Resistance (Q/sq.)
`
`80
`
`90
`
`100
`
`of.
`
`300
`
`Both Cells Tested and Verified by:
`(fH) Sandia National Laboratories
`
`1
`
`1
`
`1
`
`400
`
`500
`
`600
`
`800
`700
`Wavelength (nm)
`
`900
`
`1000
`
`1100
`
`1200
`
`300
`
`250
`
`3
`
`i]
`
`—_
`
`3
`
`Joe(fA/em?)
`
`100
`
`50
`
`0
`
`30
`
`Fig. 4. Emitter component of reverse saturation current density (Joe) as a
`function of RTP emitter sheet resistance and surface passivation. Joe was
`measured by the PCD technique.
`
`Fig. 5. Comparison between RTP and CFP cells both passivated by a CFO.
`Nearly identical performance is achieved by both processes because of CFO
`passivation.
`
`not perturb short wavelength IQE response. Details of the
`fabrication of CFP cells has been discussed elsewhere [12].
`
`100
`
`III. RESULTS
`The passivation characteristics of SiO and PECVD SiN
`2
`J
`were first studied by
`measurements. To be relevant to
`oe
`any contact scheme including screen printing, the study was
`carried out on a wide range of emitter sheet resistances
`Q
`from 30 to 100
`/sq. Fig. 4 shows the measured
`as a
`Joe
`function of the RTP emitter sheet resistance. The furnace
`grown oxide (CFO) gave a factor of two to four times lower
`(depending upon the sheet resistance) than SiN passivation
`Joe
`Q
`on the same emitter. At 90 /sq.,
`was reduced significantly
`Joe
`2
`2
`from 130 fA/cm for SiN passivation down to 30 fA/cm
`Q
`for CFO passivation. At 40
`/sq. (which can accommodate
`screen-printed contacts [13]), the reduction was less impressive
`2
`2
`decreasing
`from only 200 fA/cm for SiN to 100 fA/cm
`Joe
`for CFO. It is important to note that the true
`of the cell
`Jo e
`will be greater since the effect of contacts are not present
`in the PCD measurements. It is better, therefore, to judge
`the impact of front surface passivation by short wavelength
`IQE measurements as will be shown in the next section.
`To demonstrate the effectiveness of front oxides on RTP
`cells, the well-established CFO technology was first applied
`to RTP cells. Fig. 5 shows that CFO passivation of RTP cells
`produces virtually the same results as cells fabricated entirely
`by CFP. These RTP/CFO cells eliminated the huge loss in
`short wavelength IQE response we had observed in the past
`[12] and gave virtually the same cell performance.
`Use of a CFO, however, mitigates the attractiveness of an
`RTP process because it requires a lengthy furnace process as
`shown in Fig. 2(a), whereas growth of an oxide by RTP main-
`tains the high-throughput nature of the entire RTP processing
`sequence. Thus, an RTO cycle was developed to provide the
`same high-quality oxide as a CFO but in a time of only 5–6
`min as shown in Fig. 2(b). Cells fabricated according to this
`RTP/RTO process gave efficiencies up to 18.6% which repre-
`sents the highest reported efficiency for any RTP-diffused cell.
`Table I shows that numerous RTP cells employing an RTO
`near or above 18% efficiency have been fabricated. Average
`2
`of (at least) fourteen 1 cm cells on each wafer give above
`
`>S
`=
`ma]So=
`
`ProcessandCell ID
`
`CFP* (1.3 Q-cm FZ; S$5-9)
`RTP/RTO (1,3 Q-cm FZ; X-12)
`CEP process includes growth of @
`
`ional
`
`FF
`
`Eff
`Voc!
`Sse
`(nV) |
`(%)
`(mAfem’)
`36.3 | 0.809 | 18.7
`637)
`36.9 | 0.809 | 18.6
`[623
`furnace oxide.
`
`300
`
`400
`
`500
`
`600
`
`700
`800
`Wavelength (nm)
`
`900
`
`1000
`
`1100
`
`1200
`
`Fig. 6. Highest efficiency RTP/RTO cell compared to CFP cell (both)
`fabricated on 1.3
`-cm FZ silicon. Comparable short wavelength response
`between both types of cells indicates similar quality of RTO and CFO surface
`passivation.
`
`18% for FZ and 17.5% for Cz silicon. Figs. 6 and 7 show that
`virtually identical performance is achieved by the RTP/RTO
`process compared with lengthy CFP (which includes growth of
`a CFO) on both FZ and Cz substrates. In comparison with cells
`without an RTO, these figures also depict the large increase in
`the short wavelength IQE response because of improved front
`surface passivation and reduced heavy doping effects, resulting
`in performance improvements of about 1 mA/cm in
`and
`Js
`1% in absolute efficiency. These enhancements were found to
`be consistent with model predictions, as will be shown in the
`following PC-1D simulations.
`
`IV. PC-1D SIMULATIONS AND DISCUSSION
`In order to improve the understanding of these record-high-
`efficiency (1 and 4 cm ) RTP/RTO cells and provide guidelines
`for further improvements, it was necessary to perform model
`calculations. PC-1D for Windows [18] was used to solve
`the one-dimensional semiconductor transport equations and
`compute solar cell performance using the default material
`recombination parameters set by PC-1D. (These include the in-
`1x 101°
`trinsic carrier concentration
`cm , the band gap
`narrowing model of Klaassen et al. [19], mobility model used
`Cha = 2.2x107%
`in Cuevas et al. [20], and Auger coefficient
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on June 22,2024 at 19:30:41 UTC from IEEE Xplore. Restrictions apply.
`
`

`

`DOSHI AND ROHATGI: 18% EFFICIENT SILICON PHOTOVOLTAIC DEVICES
`
`1713
`
`TABLE I
`HIGH-EFFICIENCY RTP/RTO SILICON SOLAR CELLS
`
`ID
`
`963 1\X-12
`
`15 cell average
`on 9631\X
`
`963 1\C-2*
`
`14 cell average
`on 9631\C
`
`963 1\HS-2
`
`9615\RO2-3*
`
`9615\RO2-6*
`
`Res.
`(Q-cm)
`13
`
`13
`
`2.8
`
`2.8
`
`0.9
`
`0.6
`
`0.6
`
`Material
`
`FZ
`
`FZ
`
`Cz
`
`Cz
`
`HEM mc-Si
`FZ
`
`Area
`(cm’)
`1
`
`Voc
`(mV)
`623
`
`Isc
`__(mA/em’)
`36.9
`
`FF
`
`0.809
`
`Eff.
`(%)
`18.6
`
`1
`
`1
`
`1
`
`1
`
`4
`
`4
`
`619
`
`607
`
`608
`
`593
`
`633
`
`633
`
`36.6
`
`37.5
`
`37.3
`
`34.1
`
`35.9
`
`35.7
`
`0.801
`
`18.2
`
`0.790
`
`18.0
`
`0.772
`
`17.5
`
`0.790
`
`0.789
`
`16.0
`
`17.9
`
`0.793
`
`17.9
`
`961 1\PR-7*
`
`1.3
`
`FZ
`
`FZ
`
`4
`
`623
`
`36.4
`
`0.780
`
`17.7
`
`Scellaverageon |
`FZ
`9611\PR
`*Tested and verified by Sandia National Laboratories.
`
`623
`
`36.6
`
`4
`
`100
`
`90
`
`80
`
`70
`
`60
`
`50
`
`40
`
`30
`
`20
`
`10
`
`0
`
`[ea]
`=
`
`CFP
`RTP/RTO
`RTP (No Pass.)
`
`Elf
`% |
`17.9
`18.0
`RTP (No Pass.) (Czi-3)
`16.8
`* CFP process mcludes growih of @ conventional furnace oxide,
`
`IQE(%)
`
`Cells Tested and Verified by:
`
`(a) Sandia National Laboratories
`
`]
`
`0.776
`
`17.7
`
`a
`
`RTP/RTO (Cell PR-7)
`FSRV
`
`UnpassivatedRTP (Cell P-7)
`FSRV = 7.5x10° cm/s
`
`300
`
`400
`
`500
`
`600
`
`800
`700
`Wavelength (nm)
`
`900
`
`1000
`
`1100
`
`1200
`
`Fig. 7. Highest efficiency RTP/RTO cell compared to CFP cell (both)
`fabricated on 2.8
`-cm Cz silicon.
`
`of Wang et al. [21]) The purpose of the PC-1D simulations
`was threefold: 1) quantitatively assess the impact of RTO
`passivation on reducing the FSRV of cells fabricated in this
`study; 2) using the estimated change in FSRV values, show that
`the enhanced cell performance resulting from RTO passivation
`is consistent with the model calculations; and 3) provide
`guidelines for further improvement of RTP/RTO devices.
`
`A. Estimation of Front Surface Recombination
`Velocity for RTP Cells
`To analyze the short wavelength behavior of RTP cells
`and calculate performance improvements resulting from RTO
`passivation, it is necessary to first assess FSRV values. This is
`generally difficult because the actual FSRV that dictates short
`wavelength response is affected by the presence of contacts.
`Therefore, we decided to assess the FSRV with and without
`contacts by a combination of IQE and
`measurements.
`Joe
`In order to obtain the effective FSRV with the contacts, the
`
`380
`
`400
`
`420
`
`440
`
`480
`460
`500
`Wavelength (nm)
`
`520
`
`540
`
`560
`
`580
`
`Fig. 8. Matching of measured (data points) and PC-1D calculated (solid
`curve) short wavelength IQE response to estimate the spatially-averaged front
`surface recombination velocity (FSRV).
`
`measured short wavelength IQE response was matched with
`PC-1D-calculated IQE using the measured emitter profiles
`in Fig. 3. Fig. 8 shows that using the “after RTO” profile
`from Fig. 3 and matching the short wavelength response of
`the RTO cell gave a best-fit effective FSRV value of about
`20 000 cm/s. It is important to note that any value between
`15 000 and 25 000 matched the data with little error. Matching
`the significantly lower short wavelength IQE response of the
`unpassivated cell in Fig. 8 (using the “before RTO” profile in
`7.5 x10°
`Fig. 3) gave an effective FSRV value of about
`cm/s.
`Although the method of matching short wavelength IQE to
`estimate FSRV places significant dependence on the accuracy
`of IQE measurements, it has advantages over more traditional
`means such as calculating FSRV from
`measurements from
`Joe
`J.
`PCD techniques. Difficulty in calculating FSRV from
`arises because PCD samples, diffused and passivated on both
`sides without the grid, do not account for the effect of front
`contacts which is the true condition of a finished device. To
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on June 22,2024 at 19:30:41 UTC from IEEE Xplore. Restrictions apply.
`
`

`

`IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 45, NO. 8, AUGUST 1998
`
`18.5
`
`18 +
`
`175
`
`es
`a 17+
`.?)
`3
`3 16.5
`Fs
`
`16
`
`155 4
`
`15
`
`1
`
`100
`
`, Surface Cone.
`—~ 1E+19
`~= 2E+19
`4E+19
`
`6E+19
`
`—~ 1E+20
`2E+20
`+ 4E+20
`
`|
`
`1,000
`
`S175
`12
`
`17.0
`
`“167
`
`60
`
`134
`
`Powe = 2.8 Qc
`Tank = 200 ps
`P, = 100 Q/sq.
`
`pode
`
`1
`
`1
`
`10,000
`FSRV (cm/s)
`
`100,000
`
`1,000,000
`
`Fig. 10. PC-1D simulations for the 2.8
`-cm Cz cells showing the effect
`of emitter profiles and FSRV on cell performance. (The sheet resistance was
`fixed at 100
`/sq.)
`
`Q
`/sq. was chosen
`of 100
`The constant sheet resistance
`(ps)
`because model calculations show little gain in increasing
`beyond this value. (We have also found empirically in
`Ps
`100 + 20 2
`numerous experiments in our laboratory that
`/sq.
`gives optimal performance for homogeneous emitters prepared
`for evaporated contacts.) Three different base resistivities are
`analyzed since cells fabricated on FZ and Cz were 1.3 and
`Q
`Q
`2.8
`-cm, respectively; and 0.2
`-cm was included to exhibit
`the case where the emitter strongly affects cell performance
`(i.e.
`low
`All other parameters for the base region
`Jop):
`were fixed with a bulk lifetime of 200 s and back-surface-
`Lb
`recombination-velocity of 10 000 cm/s (which is applicable to
`a relatively poor BSF such as the 1 mm evaporated aluminum
`BSF used in this study).
`Figs. 10 and 11 show all possible combinations of FSRV
`Q
`and (Gaussian 100
`/sq.) emitter profile on cell performance.
`These figures can be used to explain the observed enhance-
`ments in cell performance resulting from RTO passivation.
`For samples with and without RTO passivation,
`is about
`Ns
`1x 1079
`6 x 1019
`and
`cm , respectively (when the profiles
`in Fig. 3 are extrapolated to the surface). As stated earlier,
`Ns
`decreases following RTO because of the additional diffusion
`during the RTO cycle and slight consumption of the surface
`region to grow the SiO Fig. 10 shows that in the case of 2.8
`2+
`2
`N, =1x 107
`-cm Cz material, using the
`curve with FSRV
`= 7.5x 10°
`N, =6x 1019
`cm/s for the unpassivated cell and
`curve with FSRV 20 000 cm/s for the RTO passivated cell,
`the calculated efficiency increases by 1% (absolute). Table II
`shows that this is in good agreement with the I–V measure-
`Q
`ments on 2.8
`-cm Cz cells. Also, notice that the measured
`increase of 12 mV in
`is consistent with the simulations.
`Voc
`Q
`Fig. 11 depicts the simulation for the 1.3
`-cm FZ cells.
`Using the appropriate curves in Fig. 11 and the same decrease
`in FSRV because of RTO passivation, an increase of about
`1.2% is predicted for the FZ cells. Table III shows that the
`2
`18 mV increase in
`and 1.0 mA/cm increase in
`Joc
`Voc
`are in excellent agreement with the calculated predictions.
`Q
`The calculated gain of 1.2% for 1.3
`-cm cells is slightly
`Q
`greater than the 1.0% predicted for 2.8
`-cm cells because
`higher base doping reduces the
`component of the total
`Jon
`J, = Joe + Jop
`, and results in a greater increase in
`Voc
`
`B. Comparison Between Measured and Simulated
`RTO-Induced Enhancements in Cell Parameters
`To study the effect of emitter design on RTP cell perfor-
`mance and explain the observed RTO-induced improvement
`in cell performance, devices were modeled as a function of
`N.
`surface doping concentration
`, junction depth
`, and
`FSRV. The sheet resistance
`was held constant at 100 /sq.
`(p
`and a Gaussian profile was assumed for modeling purposes.
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on June 22,2024 at 19:30:41 UTC from IEEE Xplore. Restrictions apply.
`
`25°C
`
`1048
`
`Emitter profile
`from fig. 3:
`After RTO
`Before RTO
`
`1714
`
`Joe(fA/om?)
`
`100
`
`1,000
`
`10,000
`FSRV (cm/s)
`
`100,000
`
`1,000,000
`
`Fig. 9. Joe computed using (1) and the measured emitter profiles in Fig. 3.
`
`Jo e
`, PC-1D is used to vary
`predict FSRV from the measured
`Jo e
`FSRV until the calculated
`(1) matches the measured
`Joe
`_
`
`(1)
`
`J.
`
`“~~
`
`Ip (Xie)
`exp(qV/kT) — 1°
`V
`in
`In the PC-1D simulation, the cell is biased to any voltage,
`the dark, and the minority hole current,
`is calculated at the
`Jp
`(je)
`edge of the depletion region on the emitter side
`using the
`measured emitter profile. Since
`is a function of FSRV, (1)
`Jp
`J.
`can be iterated until the measured
`is obtained. Fig. 9 shows
`the calculated
`(without the metal grid) as a function of
`Joe
`FSRV determined by (1) for the profiles studied in this paper.
`For an oxide-passivated RTP emitter with a sheet resistance
`J.
`2
`Q
`of 90
`/sq., Fig. 4 showed that
`is about 30 fA/cm which
`translates into an FSRV of about 2000 cm/s. Although this
`value is in good agreement with published values of FSRV
`obtained from PCD analysis [20], [22], it is significantly lower
`than the effective FSRV value of about 20 000 cm/s which
`7.5 x10°
`agreed with the measured short wavelength IQE. The
`cm/s obtained by matching the IQE of the unpassivated cell
`2 x10°
`is also greater than the value of
`cm/s based upon
`Joe
`measurements of bare silicon samples reported by Cuevas et
`al. [16]. Thus, matching the short wavelength IQE gives an
`estimate of the spatially-averaged (or effective) FSRV which
`accounts for the variation in FSRV between the passivated and
`contacted regions. The fact that the spatially-averaged FSRV
`was found to be greater than the FSRV of the passivated region
`is also consistent with the successful PCD measurements on
`grid covered samples by ISFH in Germany [23]. According to
`Fig. 9, the spatially-averaged FSRV values of 20 000 and
`7.5
`x 10°
`cm/s for RTP emitters correspond to spatially-averaged
`values of 115 fA/cm for RTO passivation and about 1000
`Joe
`2
`fA/cm for the unpassivated case.
`
`

`

`DOSHI AND ROHATGI: 18% EFFICIENT SILICON PHOTOVOLTAIC DEVICES
`
`1715
`
`For all cells in this study, (2) gave a
`value greater than
`So
`2
`1000 fA/cm which is an order of magnitude greater than
`J.
`the spatially-averaged
`of about 115 fA/cm for RTO-
`passivated cells (determined in Section IV-A). (Without the
`RTO,
`and
`are comparable.) Thus, because of high-
`Jop
`Joe
`quality RTO passivation of emitters, RTP/RTO cells now suffer
`from base recombination.
`Further analysis reveals that a high back surface recombi-
`nation velocity (BSRV) is responsible for the high
`By
`So
`matching the measured long wavelength IQE of RTP/RTO
`cells (made with a 1- m evaporated BSF,
`in this study)
`to PC-1D calculated IQE, the BSRV was found to be very
`2
`high ( 10 000 cm/s). Furthermore, PCD measurements on
`processed FZ and Cz materials used in this study gave lifetimes
`greater than 200
`s
`m Since this makes
`Luk /W
`the
`ratio high, these RTP/RTO cells are strongly
`limited by back surface recombination. Therefore, research is
`in progress to reduce BSRV by screen-printed aluminum BSF
`technology which has been shown to reduce BSRV value to
`200 cm/s because of the substantially increased depth of the
`+
`p
`region [25], [26]. Other possibilities include a boron BSF
`+
`which is capable of increasing the doping level of the p
`region [27] or dielectric passivation (via RTO and/or PECVD
`silicon nitride) in combination with point or grid rear contacts.
`Q
`PC-1D model calculations on 1.3
`-cm material, show that
`screen-printed aluminum BSF’s with a BSRV of 200 cm/s
`could increase RTP/RTO cell efficiencies beyond 19%.
`
`(Lurk> 800 ps).
`
`pb
`
`V. CONCLUSIONS
`The fabrication, characterization, and analysis of the first
`18% efficient RTP cells was presented. Improved energy
`conversion efficiencies were attained by RTO passivation of
`the emitter which significantly enhanced the short wavelength
`response and decreased the FSRV (including contact effects)
`7.5x 10°
`2x 104
`from about
`to
`cm/s. As a result, cell efficiency
`increased by about 1% (absolute) with 10–20 mV increase in
`(depending upon base resistivity) and about 1 mA/cm
`Voc
`increase in
`Although less improvement would be expected
`Isc.
`for standard screen-printed cells (which have heavily doped
`emitters and depend less on surface passivation), these results
`are still useful for clever cell designs involving a selective
`emitter with RTO passivation. The cells presented in this work
`were diffused and oxidized in a total time of only about 10
`min indicating the potential for substantially increasing the
`throughput of high-temperature processes compared with CFP.
`Compared with cells made by CFP (which includes growth
`of a conventional furnace oxide), these RTP/RTO cells gave
`virtually identical performance using the same cell structure
`and base material. Detailed modeling and analysis was also
`performed to show that the RTO-induced enhancements are
`consistent with PC-1D simulations and to provide guidelines
`for further improvements. RTO passivation of the emitter
`has shifted the dominant recombination mechanism to the
`(Jos > Joe)
`base
`, so the next generation of high-efficiency
`RTP/RTO cells require reducing back surface recombination.
`>
`Based on model calculations,
`19%-efficient RTP/RTO cells
`Q
`are expected by reducing the BSRV to 200 cm/s on 1.3
`-cm
`silicon.
`
`IV
`
`19
`
`18.5 +
`
`g@
`
`18+
`>
`2 175 +
`‘2
`
`bo}
`
`16.5 +
`
`16
`
`155 -
`
`100
`
`Surface Conc.
`—~ 1E+i9
`2E+19
`
`-- 4E+19
`+ 6E+19
`1E+20
`—* 2E+20
`—+- 4E+20
`
`L.
`
`Poase = 1.3 Qm
`Toatk = 200 pas
`Pp; = 100 O/sq.
`
`10
`
`te
`
`10,000
`FSRV (cm/s)
`
`1
`
`100,900
`
`1,000,000
`
`Fig. 11. PC-1D simulations for the 1.3
`-cm FZ cells showing the effect
`of emitter profiles and FSRV on cell performance. (The sheet resistance was
`fixed at 100
`/sq.)
`
`TABLE II
`COMPARISON BETWEEN MEASURED AND SIMULATED
`IMPROVEMENT IN CELL PERFORMANCE RESULTING FROM RTO
`PASSIVATION OF 1 cm2 RTP CELLS ON 2.8
`-cm Cz SILICON
`
`ID
`
`9631\C-2
`Cz11-3
`Actual
`improvement
`
`Calculated
`improvement
`
`Passi-
`Voc
`Ix¢
`vation | GnV) | (mAscm’)
`RTO
`607
`37.5
`None
`595
`35.8
`
`12
`
`12
`
`1.7
`
`13.
`
`FF
`
`0.790
`0.789
`
`0.001
`
`Eff.
`(%)
`18.0
`16.8
`
`1.2
`
`| 0001 | 10
`
`TABLE III
`COMPARISON BETWEEN MEASURED AND SIMULATED
`IMPROVEMENT IN CELL PERFORMANCE RESULTING FROM RTO
`PASSIVATION OF 4 cm2 RTP CELLS ON 1.3
`-cm FZ SILICON
`
`ID
`
`9611\PR-7
`9611\P-7
`Actual
`improvement
`Calculated
`improvement
`
`Iie
`Passi-
`Voc
`vation | (mV) | (mA/cm’)
`RTO
`36.4
`623
`None
`35.4
`605
`1.0
`18
`
`19
`
`11
`
`FF
`
`0.780
`0.787
`-0.007
`
`0.003
`
`Eff.
`(%)
`17.7
`16.8
`0.9
`
`1.2
`
`is reduced. This effect can be amplified by further
`when
`Joe
`reduction in base resistivity, provided sufficient lifetime is
`preserved. For example, similar calculations revealed that the
`Q
`efficiency of 0.2
`-cm cells with a bulk lifetime of 50
`s
`increases by 1.7% because of RTO passivation of the emitter.
`
`C. Guidelines for Further Improvement of RTP/RTO Cells
`To increase the performance of RTP/RTO cells further, it
`is necessary to determine the efficiency-limiting mechanisms.
`Detailed IQE analysis of RTP/RTO cells revealed that
`is
`So
`much greater than
`This was determined by extracting
`Joe
`the effective diffusion length
`from the near-infrared
`(Les)
`(860–980 nm) inverse-IQE response [24] and using the simple
`formula
`
`Na Les
`
`(2)
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on June 22,2024 at 19:30:41 UTC from IEEE Xplore. Restrictions apply.
`
`

`

`1716
`
`IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 45, NO. 8, AUGUST 1998
`
`ACKNOWLEDGMENT
`The authors would like to thank D. Ruby et al. of Sandia
`National Laboratories for cell testing and verification and S.
`Kamra of Georgia Institute of Technology, Atlanta, for help
`with cell fabrication.
`
`REFERENCES
`
`[1] K. Mitchell, R. King, T. Jester, and M. McGraw, “The reformation of
`Cz Si photovoltaics,” in Conf. Rec. 24th IEEE Photovoltaic Specialists
`Conf., 1994, pp. 1266–1269.
`[2] Product information regarding the “RTP solar cell furnace,” and “high-
`throughput RTP solar cell processor model VS2400,” Vortek Industries,
`Vancouver, B.C., Canada.
`[3] A. Usami, M. Ando, M. Tsunekane, K. Yamamoto, T. Wada, and Y.
`Inoue, “Shallow junction formation for silicon solar cells by light-
`induced diffusion of phosphorus from a spin-on source,” in 18th Europ.
`Comm. Photovoltaic Sol. Energy Conf., 1985, pp. 797–803.
`[4] A. Usami, M. Tsunekane, T. Wada, Y. Inoue, S. Shimada, N. Nakazawa,
`and Y. Meada, “New junction formation process for polycrystalline sili-
`con solar cells by light induced diffusion from a spin-on source,” in 18th
`Europ. Comm. Photovoltaic Sol. Energy Conf., 1985, pp. 1078–1083.
`[5] R. Campbell and D Meier, “Simultaneous junction formation using a
`directed energy light source,” J. Electrochem. Soc., vol. 133, no. 10, pp.
`2210–2211, Oct. 1986.
`[6] B. Hartiti, A. Slaoui, J.C. Muller, P. Siffert, B. Wagner, R. Schindler,
`A. Eyer, and A. R¨auber, “Optical thermal processing for silicon solar
`cells,” in 11th Europ. Comm. Photovoltaic Sol. Energy Conf., 1992, pp.
`420–422.
`[7] B. Hartiti, A. Slaoui, J.C. Muller, and P. Siffert, in “Multicrystalline
`silicon solar cells processed by rapid thermal processing,” in Conf.
`Record 23rd IEEE Photovoltaic Specialists Conf., 1993, pp. 224–228.
`[8] R. Schindler, I. Reis, B. Wagner, A. Eyer, H. Lautenschlager, C.
`Schetter, and W. Warta, “Rapid optical thermal processing of silicon
`solar cells,” in Conf. Record 23rd IEEE Photovoltaic Specialists Conf.,
`1993, pp. 162–166.
`[9] S. Sivoththaman, W. Laureys, P. De Schepper, J. Nijs, and R. Mertens,
`“Rapid thermal processing of conventionally and electromagnetically
`cast 100 cm2 mu