IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 44, NO. 9, SEPTEMBER 1997
`
`1417
`
`Modeling and Characterization of High-Efficiency
`Silicon Solar Cells Fabricated by Rapid
`Thermal Processing, Screen Printing, and
`Plasma-Enhanced Chemical Vapor Deposition
`
`Parag Doshi, Jose Mejia, Keith Tate, and Ajeet Rohatgi, Fellow, IEEE
`
`Abstract—This paper presents, for the first time, the successful
`integration of three rapid, low-cost, high-throughput technologies
`for silicon solar cell fabrication, namely: rapid thermal processing
`(RTP) for simultaneous diffusion of a phosphorus emitter and
`aluminum back surface field; screen printing (SP) for the front
`grid contact; and low-temperature plasma-enhanced chemical
`vapor deposition (PECVD) of SiN for antireflection coating and
`surface passivation. This combination has resulted in 4 cm2 cells
`with efficiencies of 16.3% and 15.9% on 2
`-cm FZ and Cz,
`respectively, as well as 15.4% efficient, 25-cm2 FZ cells. Despite
`the respectable RTP/SP/PECVD efficiencies, cells formed by con-
`ventional furnace processing and photolithography (CFP/PL) give
`2% (absolute) greater efficiencies. Through in-depth modeling
`and characterization, this efficiency difference is quantified on
`the basis of emitter design and front surface passivation, grid
`shading, and quality of contacts. Detailed analysis reveals that the
`difference is primarily due to the requirements of screen printing
`and not RTP.
`
`I. INTRODUCTION AND BACKGROUND
`
`COST-EFFECTIVENESS is the major obstacle facing
`
`in the past have used photolithography [4] to achieve a
`selective emitter or texturization, defeating the purpose of
`reducing cost by screen printing. In contrast to CFP, which
`generally involves multiple furnace diffusions and oxidations
`at high temperatures, extensive wafer cleaning, and use of
`more chemicals and gases, RTP significantly reduces the
`cell fabrication time, thermal budget [5], and wafer cleaning
`steps. Use of SP contacts greatly simplifies contact formation
`compared with photolithography (PL) which requires spin-on,
`bake, exposure, and development of photoresist followed by
`evaporation, lift-off, and plating of metal. Finally, PECVD
`of SiN provides antireflection (AR) properties and surface
`and bulk defect passivation. In addition, PECVD is a low-
`temperature process and can be done in a batch reactor [6]. In
`fact, ASE Americas, Inc. (formerly Mobil Solar Energy Corp.)
`[7] and R&S Renewable Energy Systems B.V. [3] have been
`successful in implementing PECVD in an industrial setting
`suggesting the cost-effectiveness of the technology.
`It is important to note that before RTP can be considered a
`viable alternative to CFP for photovoltaic devices, RTP sys-
`tems must evolve from single-wafer to batch-mode processing.
`In fact, one such machine which has recently become available
`can process multiple wafers uniformly over an area of 35 cm
`x
`35 cm or 12 100-cm cells at a time [8]. Since an emitter
`and back surface field (BSF) can be formed by RTP in a
`matter of seconds or minutes, high throughputs are already
`possible with existing equipment. Even greater throughputs
`are conceivable if, for example, simple modifications such as
`retrofitting RTP lamps into a continuous belt-line system are
`realized. In contrast, typical CFP diffusion times are at least
`30 min. Therefore, throughputs for emitter diffusion alone are
`typically limited to only 150–250 wafers/h per furnace tube
`and 400–500 wafers/h per IR belt furnace. BSF formation
`would require an additional lengthy step. Thus, current means
`of diffusion generally bottleneck the production line, and RTP
`may provide the solution with shorter processing times and
`greater cell efficiencies.
`In previous publications, 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 [9],
`[10]. These cells were processed by rapid, simultaneous front
`0018–9383/97$10.00 ª
`
`the silicon solar cell industry. This paper presents the
`development and integration of cell process techniques that
`are suitable for industrial production and have the potential
`of becoming the lowest-cost and highest throughput process
`for the fabrication of silicon photovoltaic devices with high
`energy conversion efficiency. We present
`the fundamental
`understanding and engineering of the rapid thermal process-
`ing/screen printing/plasma enhanced chemical vapor depo-
`sition (RTP/SP/PECVD) process, which not only simplifies
`processing, but also produces 15.5%–16% efficient, manu-
`facturable cells on single-crystal silicon. Unlike other high-
`efficiency screen-printed cell designs [made by conventional
`furnace processing (CFP)] reported in the literature [1]–[4],
`these RTP/SP/PECVD cells involve neither oxide passivation
`nor fine-line printing ( 60 m grid fingers), selective emit-
`ter, or texturization technologies. In fact, some investigators
`
`IA
`
`Manuscript received October 2, 1996. The review of this paper was
`arranged by Editor P. N. Panayotatos. This work was supported by Sandia
`National Laboratories under Contract A0-6162 and by the National Institute
`of Standards under Contract 70NANB5H1071-GIT.
`The authors are with the University Center of Excellence for Photovoltaics
`Research and Education, Department of Electrical and Computer Engineering,
`Georgia Institute of Technology, Atlanta, GA 30332 USA.
`Publisher Item Identifier S 0018-9383(97)06124-8.
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on June 22,2024 at 19:31:14 UTC from IEEE Xplore. Restrictions apply.
`
`1997 IEEE
`
`

`

`1418
`
`IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 44, NO. 9, SEPTEMBER 1997
`
`2 Q-cm
`
`FF
`
`|
`
`|
`
`ps
`(Q/sq,)
`
`21
`
`~
`
`17
`
`Voc
`| EM IN
`Jse_[
`(mV)
`(%)
`(mAfem’)
`| 341
`[ 604
`15.2
`0.741"
`33.6 | 606
`46
`(0.767) 15.6_
`32 | 32.0 | 602 rear
`15.0
`143
`29.5
`| 617
`(0,788
`28.5 [ 618
`14.0
`(0.792
`8 | 252 | 615
`119
`'o767!
`
`100
`
`80
`
`= 0
`~~
`ih
`= 49
`
`20
`
`and back diffusion of the emitter and BSF without any high-
`temperature furnace steps; however, photolithography was
`used to define the front evaporated contacts (hereafter referred
`to as the RTP/PL process). Since most solar cell manufacturers
`do not regard photolithography as a cost-effective technique
`for contact formation for terrestrial photovoltaics, we modified
`the RTP process to accommodate screen-printed contacts.
`Since the emitters of these RTP/PL cells were too shallow
`(
`0.14 m) for screen printing, the RTP diffusion had to be
`adjusted for greater junction depths. Thus, development of the
`RTP/SP process required formulating both the diffusion and
`contact firing cycles.
`In a recent study, we demonstrated cell efficiencies in
`the 14%–15% range made by the RTP/SP process [11]. The
`efficiency of these cells were limited to this range because of
`the heavy doping recombination in the 20–30
`/sq. emitter
`Q
`and front surface recombination. Recent improvements [12]
`>
`have resulted in
`1% increase in cell efficiency by lowering
`the concentration of the phosphorus spin-on source to increase
`the emitter sheet resistance
`and by providing front surface
`(p
`passivation with PECVD SiN. In this study, we first examine
`the importance of the emitter design for RTP/SP cell per-
`formance followed by detailed modeling and characterization
`used to quantify the efficiency limiting mechanisms of these
`16%-efficient RTP/SP/PECVD cells.
`
`II. DEVICE FABRICATION
`+
`+
`RTP/SP/PECVD n -p-p
`cell fabrication in this study in-
`volved simultaneous formation of an emitter and BSF using
`phosphorus spin-on dopants on the front and 1 m aluminum
`Lb
`evaporated on the back. Two concentrations of phosphorus
`-3
`20
`21
`spin-on dopants were used: 10
`and 10
`cm . The RTP
`diffusion temperature/time was varied for the desired
`. Prior
`Ps
`to printing contacts, SiN was deposited by PECVD. Deposition
`conditions involved an RF power of 30 W operating at
`13.6 MHz, pressure of 0.9 torr, and deposition temperature
`of 300 C.
`The front grid contact was formed by screen printing
`commercially available silver pastes on top of the SiN and
`then “fired through” the film. Firing screen-printed contacts
`proved to be extremely critical in order to prevent shunting of
`tj; < 0.3p
`these shallow (
`m) RTP junctions while obtaining
`minimum contact resistance. After numerous experiments, an
`RTP firing cycle of 737 C for only 10 s in high-purity air
`was established. In some cases, a 715 C/90 s contact firing
`cycle was developed for furnace use instead of the RTP system.
`The contact quality was virtually independent of which contact
`firing system used. In this work, the back contact was formed
`°
`by evaporation of Al/Ti/Pd/Ag followed by a 400 C/10 min
`contact anneal in forming gas. Work is in progress to replace
`both the Al BSF and back contact evaporations by a single Al
`BSF screen printing step. This would significantly simplify the
`process, while at the same time, improve the long wavelength
`response because of a deeper BSF profile and lower back
`surface recombination velocity (BSRV) [13]. The final step
`was an evaporation of MgF as a second layer AR coating;
`however, it is important to note that this was used because
`
`0
`
`380
`
`1
`
`L
`
`L
`
`1
`
`1
`
`1
`
`480
`
`580
`
`780
`680
`Wavelength (nm)
`
`880
`
`980
`
`1080
`
`Fig. 1. Optimization of RTP/SP emitter sheet resistance (s).
`
`(n = 1.0)
`. In practice, this second
`cells would be tested in air
`layer would not be deposited for cells in a module, which are
`(n = 1.5)
`under glass
`.
`Preparation of CFP/PL cells was as follows. After an initial
`°
`wafer clean, phosphorus diffusion was carried out at 930 C
`for 30 min in addition to a ramp-up and ramp-down rate of
`5 C/min from 500 C. This yields a sheet resistance of about
`15
`/sq. which is then slowly etched back to about 120
`/sq.
`Q
`Q
`After a second cleaning, 1 m of Al was evaporated on the
`bb
`°
`back side followed by a 850 C/30 min drive-in to form the
`back surface field (BSF). During this step, a 10-nm oxide was
`also grown on the front and a 30 min forming gas anneal
`°
`°
`was performed after ramping down to 400 C at 5 C/min
`After evaporation of the back Al/Ti/Pd/Ag contact, a series
`of photolithography steps were done to define the front grid
`contact. Finally individual cells were isolated through a mesa
`etch followed by evaporation of a ZnS/MgF double layer AR
`2
`coating optimized for silicon in air.
`
`III. RESULTS AND DISCUSSIONS
`
`A. Optimization of RTP/SP Emitter Sheet Resistance
`Design of the emitter region is extremely important for
`achieving high-efficiency RTP/SP cells. A low surface concen-
`tration improves front surface passivation but increases contact
`resistance. A shallow junction depth improves short wave-
`length collection efficiency but increases the susceptibility of
`shunting. Given these trade-offs, experiments were performed
`to optimize the
`of the homogeneous RTP/SP emitter. Fig. 1
`Ps
`shows the results of RTP/SP cells as a function of
`in the
`Ps
`range of 8–55
`/sq. Instead of SiN, these cells had ZnS/MgF
`Q
`2
`AR coatings to prevent short wavelength absorption from
`perturbing internal quantum efficiency (IQE) measurements.
`+
`°
`°
`All cells were RTP diffused at 930 C ( 15 C) using the
`21-3
`10
`cm spin-on source except for the two higher
`values
`Ps
`of 46 and 55
`/sq. which were formed using the 10
`cm
`Q
`spin-on source. As expected, substantial improvements in the
`Js
`short wavelength response and
`occurs with increasing
`.
`ps
`However, there is a trade-off because of the degradation of
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on June 22,2024 at 19:31:14 UTC from IEEE Xplore. Restrictions apply.
`
`

`

`DOSHI et al.: MODELING AND CHARACTERIZATION OF HIGH-EFFICIENCY SILICON SOLAR CELLS
`
`1419
`
`TABLE I
`HIGH-EFFICIENCY RTP/SP/PECVD CELLS
`
`Run #
`(Cell 1D)
`
`622-4
`8fz3-4
`10fz4-4
`10fz5-4
`
`1312z2-1
`14124-4
`101z2-4
`
`10fz1-1
`12tz5-1
`29-4
`
`7ize
`
`|
`
`633.—S
`
`«|S
`
`Jsc
`Voc
`(mV)
`(mA/cm’)
`4 cm? float-zone silicon (2 92-cm)
`0.770
`612
`34.0
`344
`616
`0.761
`0.785
`612
`34.0
`0.774
`612
`343
`0.767
`610
`0.764
`615
`34.2
`0.765
`32.3
`629
`4 cm? Czochralski silicon (3 Q-cm)
`33.9
`605
`0.774
`0.760
`33.5
`599
`32.2
`592
`0.779
`25 cm’ float-zone silicon (0.2 Q-cm)
`—|s0.77a_‘[ 15.4?
`31.4
`* Tested and veritied by Sandia National Laboratories.
`> Base resistivity = 0.2 Q-cm.
`
`FF
`
`Eff
`(%)
`
`16.0°
`16.1?
`16.3
`16.2°
`16.0
`16.1
`15.6°
`
`15.97
`15.2
`14.87
`
`70
`
`60
`S
`~ 50
`ws
`E
`oe 40
`4
`+ 30
`@
`5 20
`= 10
`oO
`0
`
`0.778
`
`40
`
`63
`
`767
`
`65
`
`FF
`
`0.754
`
`0.792
`
`0.788
`
`IQEso9
`
`19
`
`14
`10
`
`0.80
`
`0.78
`
`0.76
`
`10
`
`S
`8a
`0.745
`on
`
`fem
`
`0.72
`
`0.70
`
`0
`
`40
`20
`30
`10
`Sheet Resistance (Q/sq.)
`Fig. 2. Competition between enhanced short wavelength response and de-
`graded FF with increasing s.
`
`50
`
`60
`
`(Fig. 2). FF decreases with
`fill factor (FF) with increasing
`Ps
`higher
`because of increased contact resistance and ohmic
`Ps
`losses in the emitter sheet. Fig. 2 depicts a fairly linear
`Ps
`dependence for both the increasing IQE response at 400 nm
`and decreasing FF. Considering this competition in terms of
`cell efficiency, the optimum
`for the homogeneous RTP/SP
`Ps
`emitter was found to be in the 40–50
`/sq. range. Of course,
`Q
`the ideal emitter for screen-printed contacts is a selective
`Ps 2
`emitter in which a very high
`( 100
`/sq.) is tailored in
`Q
`the field region (between grid fingers) to achieve higher short
`Ps S
`wavelength response, while a very low ( 15 /sq.) remains
`Q
`underneath the contacts to limit contact resistance losses and
`prevent shunting.
`
`2
`2
`B. RTP/SP/PECVD Cells on 4 cm Cz and 25 cm FZ Silicon
`Based on the optimization of emitter sheet resistance, many
`RTP/SP/PECVD runs were performed. Table I shows several
`>
`2
`of the many
`16%-efficient 4 cm FZ cells that were fab-
`ricated by the RTP/SP/PECVD process—demonstrating the
`2
`reproducibility of the process. The same process on 4 cm Cz
`silicon cells gave an efficiency of 15.9%. Preliminary work
`2
`on larger area 25 cm cells have resulted in 15.4% efficient
`FZ cells. These efficiencies are quite attractive for commercial
`cells. In addition the front and back diffusion times are much
`shorter than commercial cells.
`
`IV. DEVICE MODELING AND CHARACTERIZATION
`(RTP/SP/PECVD VERSUS CFP/PL)
`To improve efficiencies further, it is important to iden-
`tify what
`limits current RTP/SP/PECVD cells to about
`16%. Based on a comparison between our high-throughput
`RTP/SP/PECVD process and the standard CFP/PL process
`used in the laboratory, we will identify all of the efficiency
`limiting mechanisms in the RTP/SP/PECVD approach through
`a detailed modeling and characterization study. Fig. 3
`shows that compared with an optimized CFP/PL cell, the
`RTP/SP/PECVD cell gives
`2% lower cell efficiency because
`AJsc = 2.5
`and FF. In this experiment,
`of a decrease in
`Isc
`AFF = 0.024
`2
`mA/cm and
`. We will show that this 2%
`efficiency gap can be explained on the basis of four categories:
`
`CEP/PL
`
`1
`
`2 Q-cm FZ
`4 cm’ cells
`
`RTP/SP/PECVD (Measured)
`
`RTP/SP/PECVD (Absorption removed)
`
`Process
`
`|Voc| FF | Eff.
`ps | Jsc
`(%)
`(Qisq.)|(mAsen?)| (mV)
`CFP/PL
`120 | 36.6 | 620] 0.798 | 18.1
`RTP/SP/PECVD] 40 | 34.1 |612} 0.774| 16.2
`
`500
`
`600
`
`800
`700
`Wavelength (nm)
`
`900
`
`1000-1100
`
`100
`
`80
`
`60
`
`40
`
`L
`
`i
`
`20
`
`0
`400
`
`InternalQuantumEfficiency(%)
`
`Fig. 3. Comparison of CFP/PL and RTP/SP/PECVD cells.
`
`reflectance, quality of
`short wavelength response,
`contacts, and long wavelength response.
`
`(grid)
`
`A. Short Wavelength Losses of the RTP/SP/PECVD Cell
`Resulting from SiN Absorption and Emitter Recombination
`Comparing the IQE of the two cells in Fig. 3 reveals
`that most of the difference between the RTP/SP/PECVD
`and CFP/PL cells lies in the short wavelength response. To
`compute the amount of short circuit current density lost in the
`short wavelength, the IQE (between 400 and 800 nm) of the
`two cells were multiplied with the photocurrent density of the
`g* Nph
`AM1.5 global solar spectrum (
`, where
`is the unit of
`Nph
`is the photon flux)
`electronic charge and
`800 nm
`
`qd
`
`AJec|short =¢q
`
`A=400 nm
`800 nm
`
`[Npn(A)|AML.sG . IQE, ()]
`
`=q
`
`A=400 nm
`
`[NpnQ)lani.se -1QE()].
`
`(1)
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on June 22,2024 at 19:31:14 UTC from IEEE Xplore. Restrictions apply.
`
`

`

`1420
`
`IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 44, NO. 9, SEPTEMBER 1997
`
`RTP/SP Emitter
`(40 Q/sq.)
`
`.
`
`CFP/PL Emitter
`(120 O/sq.)
`
`0.1
`
`0.3
`0.2
`Depth (um)
`
`0.4
`
`0.5
`
`AJsc = -0.2
`-0.1 %
`An
`
`NN
`
`n=2.12;
`film thickness = 61 nm
`
`107"
`
`107°
`
`19)
`
`~~
`=2
`5 10
`_—s
`= 10°
`vvo
`5
`Oo 10"
`
`fot=he 10".
`
`420
`
`460
`440
`Wavelength (nm)
`
`480
`
`500
`
`10
`
`0
`
`15
`
`10
`
`N
`
`+
`
`5
`
`0
`
`400
`
`Absorptance(%)
`
`Fig. 4. Calculated parasitic absorption in the SiN antireflection coating used
`in the RTP/SP/PECVD cell.
`
`2
`This gave a difference of 1.1 mA/cm . (Note that (1) is
`unaffected by reflectance since the summation is taken over
`the IQE. Reflectance loss is considered separately in the next
`section.) Three mechanisms explain this short wavelength loss:
`1) parasitic absorption in SiN; 2) front surface recombination;
`and 3) heavy doping recombination in the emitter. To char-
`acterize the absorption component, spectroscopic ellipsometry
`measurements were performed to measure the extinction coef-
`ficient as a function of wavelength. This data was then used in
`the CAMS AR coating software to compute the absorptance,
`, as a function of wavelength. Fig. 4 shows the computed
`A(A)
`n
`absorptance for a SiN film which had a refractive index ( )
`of 2.12 at 632.8 nm. To calculate the absorption-induced
`Isc
`loss, the absorptance from Fig. 4 was multiplied with the
`(q* Nn)
`photocurrent
`of the AM1.5G spectrum and IQE (at
`each wavelength),
`JeclAbsors = 1), Npn(A)|Ami.sa AVA)IQE()).
`
`d
`attributed to SiN absorption was found to be
`The loss in
`Isc
`2
`only 0.2 mA/cm for the
`film..
`n=
`2.12
`2
`2
`The remaining 0.9 mA/cm of the observed 1.1 mA/cm
`loss in the short wavelength must come from front sur-
`face recombination and heavy doping effects such as Auger
`recombination and bandgap narrowing in the emitter. To
`verify this, emitters were first characterized by spreading
`resistance analysis (SRA), and then the influence of the
`emitter profiles on device performance was modeled by the
`PC-1D semiconductor device simulator [14]. The measured
`profiles in Fig. 5 show that
`the 40
`/sq. RTP/SP emitter
`Q
`compared with the 120
`/sq. CFP/PL emitter has an or-
`Q
`der of magnitude greater surface electron concentration. The
`combination of higher surface concentration and difference in
`surface passivation (oxide for the CFP/PL cell and SiN for the
`RTP/SP/PECVD cell) gives rise to more severe heavy doping
`effects and a greater front surface recombination velocity
`(FSRV) for the RTP/SP emitter. PC-1D was effective in
`decoupling these two loss mechanisms. To extract the FSRV,
`the measured IQE was matched with the PC-1D-calculated
`IQE (Fig. 6). For the RTP/SP/PECVD cell,
`the IQE was
`
`(2)
`
`Fig. 5. Comparison of emitter profiles. Profiles measured by spreading
`resistance analysis.
`
`100
`
`80
`
`60
`
`40
`
`20
`
`4
`
`0
`
`FP/PL Cell
`(120 2/sq)
`FSRV = 9,000 cm/s
`
`RTP/SP Cell
`(40 Q/sq)
`FSRV = 150,000 cm/s
`
`Surface Recomb,
`AJsc = -0.6
`Ay = -0.4 %
`
`Heavy Doping
`AJsc = -0.3
`Aq = -0.3 %
`
`400
`
`450
`
`500
`Wavelength (nm)
`Fig. 6. PC-1D simulations to fit short wavelength response and calculate
`emitter surface recombination and heavy doping losses.
`
`550
`
`600
`
`first corrected by removing the effect of SiN absorption
`(as shown in Fig. 3) using the data in Fig. 4. Fig. 6 shows
`+
`that the FSRV was found to be about 9000 ( 1000) cm/s
`+
`for the CFP/PL emitter and about 150 000 ( 25 000) cm/s
`for the RTP/SP emitter. This translates into a loss of 0.6
`2
`and 4 mV in
`(determined by reducing the
`mA/cm in
`Isc
`Voc
`FSRV value from 150 000 to 9000 in the PC-1D simulation
`of the RTP/SP cell). Finally, giving the RTP/SP cell
`the
`from 40
`/sq.
`CFP/PL emitter profile (i.e., increasing the
`Q
`Ps
`2
`/sq.) gave an additional increase of 0.3 mA/cm in
`to 120
`Q
`and 4 mV in
`because of the reduction in heavy doping
`Js
`Voc
`2
`effects. Thus, absorption (0.2 mA/cm ), surface passivation
`2
`2
`(0.6 mA/cm ), and heavy doping effects (0.3 mA/cm ) account
`for the 1.1 mA/cm gap in short wavelength IQE between
`the RTP/SP/PECVD and CFP/PL cells. In addition, SRA
`characterization in conjunction with PC-1D modeling was in
`very good agreement with cell performance and IQE data.
`The question of what steps can be taken to reduce this
`gap in the short wavelength IQE arises. Our first attempt
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on June 22,2024 at 19:31:14 UTC from IEEE Xplore. Restrictions apply.
`
`

`

`DOSHI et al.: MODELING AND CHARACTERIZATION OF HIGH-EFFICIENCY SILICON SOLAR CELLS
`
`1421
`
`Photolithography (Evaporated) Contacts —»
`
`0.798
`Reduced
`Sheet Loss
`
`Pe
`
`Current Screcn-Printed Contacts
`
`0.80
`
`0.79
`
`0.78
`
`0.7€
`
`FillFactor
`
`Grid Shading 6% vs. 3%:
`PECVD ARC vs, ZnS/MgF):
`
`AJsc
`-11
`-0.4
`
`An
`-0.5%
`-0.2%
`
`RTP/SP + SiN (n=2.12)/MgF, AR Coating
`
`40
`
`35
`
`30
`
`25
`
`20
`
`15
`
`10
`
`5 0
`
`Reflectance(%)
`
`400
`
`500
`
`600
`
`800
`700
`Wavelength (nm)
`
`900
`
`1000
`
`1100
`
`Fig. 7. Reflectance comparison between RTP/SP and CFP/PL cells.
`
`was to fabricate RTP/PL cells with a conventional furnace
`oxide (CFO). This resulted in cells as high as 17.9% in
`efficiency (confirmed by Sandia National Laboratories) with a
`* Voc
`Js
`of 35.9 mA/cm ,
`of 618 mV, and FF of 0.805—nearly
`closing the gap with the 18.1% CFP/PL cell (in Fig. 3)
`fabricated on the same material. Such a process, however,
`required a lengthy furnace process (to grow a passivating
`oxide) which reduces the attractiveness of RTP. A higher
`throughput alternative is to provide rapid thermal oxide (RTO)
`passivation. Recent success, by others [15], with an RTO
`to enhance the short wavelength response of RTP/PL cells
`has been encouraging. We have also been successful in RTO
`passivation of RTP/PL cells and have observed improvements
`of about 1% (absolute) in cell efficiency when compared with
`unpassivated cells [16]. This has resulting in
`18%-efficient
`RTP cells (with photolithographically-defined contacts) on Cz
`and FZ silicon [16]. Attempts are underway to screen print
`RTO passivated cells.
`
`IV
`
`B. Reflectance Losses Resulting from Screen-Printed
`Grid Shading and AR Coatings
`The second major component of the efficiency difference
`between RTP/SP/PECVD and CFP/PL cells was found to be
`the difference in reflectance. Fig. 7 shows that the reflectance
`of the RTP/SP/PECVD cell is considerably greater than that of
`CFP/PL cell. Multiplying these reflectances with the spectrum
`and IQE,
`
`Xx
`
`AJsc|Refl = a>, NoenQ)|amMise
`R
`x [Re(AIQE(A) — Ri AVIQE,(\)]
`2
`reveals that about 1.5 mA/cm is lost because of the difference
`in reflectance. Most of the current loss is due to the difference
`in grid shading factor, which in this case is about 6% for SP
`cells and 3% for PL cells. Currently, our SP grid contains
`finger widths of 125 m compared with only about 25 m for
`the PL grid. Grid shading in the SP cell therefore accounts
`for about 1.1 mA/cm (3% of 36.6) of the
`difference
`Js
`
`(3)
`
`4
`Conductivity (x 10° S/cm)
`Fig. 8. Grid simulations used to compare SP and PL contacts.
`
`2
`
`3
`
`5
`
`5
`
`6
`
`7
`
`indicating that substantial improvement can be realized if fine-
`line ( 60 m) printing, demonstrated by others [1]–[4], can be
`2
`implemented at low cost. The remaining 0.4 mA/cm is due to
`(n = 2.12)
`the difference between SiN-based
`and ZnS-based
`(n = 2.36)
`AR coatings. This difference would be eliminated
`if a SiN refractive index of 2.3–2.4 is used, however, a penalty
`in terms of increased SiN-absorption would ensue. The lower
`index was used because recent work [17] has indicated that
`the emitter saturation current density (and therefore FSRV)
`is minimized at a SiN index of 2.1. Work is in progress to
`optimize PECVD AR coatings to deliver the best combination
`of minimum reflectance, low absorption, and optimum surface
`passivation.
`
`IA
`
`C. Ohmic Losses Related to the Quality
`of Screen-Printed Contacts
`The screen-printed cell in Fig. 3 gives about 0.024 lower
`FF, which can be attributed to the inferior quality of SP
`contacts compared with evaporated PL contacts. Nevertheless,
`SP is justified on economic grounds by most photovoltaic
`manufacturers. To determine what components explain the
`FF loss so that it can be minimized, we first characterized
`the specific contact resistance
`and conductivity
`of
`(a
`(pc)
`the SP contacts and then used these parameters in our grid
`modeling software to analyze the ohmic losses throughout the
`grid and emitter sheet. To measure the contact parameters,
`contact resistance and conductivity measurement structures
`were fabricated in accordance with the TLM method [18].
`og = 2x 10°
`These values were found to be about
`S/cm
`Pe= 10
`2
`and
`m -cm and then used as inputs for the grid
`Q
`model calculations (Fig. 8) to evaluate our (eight-finger) SP
`grid design. Calculations revealed that about 0.01 is lost in FF
`because of the factor of three lower conductivity of SP contacts
`compared with pure silver PL contacts, and another 0.01 is
`lost because of the contact resistance (which is negligible for
`PL contacts). There is an additional component resulting from
`the ohmic power loss in the emitter sheet. This term can be
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on June 22,2024 at 19:31:14 UTC from IEEE Xplore. Restrictions apply.
`
`

`

`1422
`
`IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 44, NO. 9, SEPTEMBER 1997
`
`TABLE II
`SUMMARY OF LOSS MECHANISMS FOR THE RTP/SP/PECVD
`PROCESS COMPARED WITH THE CFP/PL PROCESS
`
`Source of Loss
`
`CFP/PL
`
`RTP/SP
`
`100
`
`80
`
`g 60
`i)
`o°™ 40
`
`20
`
`RTP/SP/PECVD
`a BSRV = 2,000 cm/s
`
`vas
`
`CFP/PL.
`BSRV = 10,000 cm/s
`
`N
`
`AJ
`2
`An = +0.1 %
`
`0
`
`800
`
`850
`
`950
`900
`1000
`Wavelength (nm)
`
`1050
`
`1100
`
`Fig. 9. PC-1D simulations to fit long wavelength response and estimate back
`surface recombination velocity (BSRV).
`
`Contact
`Conduc- Resistance
`livity
`02%
`ee
`O20
`
`Sheeta
`“
`AR Coatiiig
`0.2%
`
`Surface
`Recomb.
`
`Litter
`Doping
`fest
`
`.
`pe"
`Absorption
`(SiN)
`0.1%
`
`Grid
`Shading
`0.5%
`Short Wavelength Response
`O Reflectance
`Quality of Contacts
`
`Fig. 10. Components of the 2% efficiency difference between RTP/SP/
`PECVD and CFP/PL cells.
`
`calculated by [19]
`Psheet =
`
`2
`
`2
`(ps)S* - 1000
`
`(4)
`
`1
`75 Amp)
`S
`2
`is the
`is the power loss in mW/cm and
`where
`Peheet
`spacing (in centimeters) between fingers. Although the 40
`/sq. RTP/SP emitter has a factor of three lower
`, there is
`Q
`Ps
`a factor of 2.5 greater spacing for the SP grid, making
`P.sheet
`dominate in favor of the PL grid. Thus, the combination of
`contact resistance, conductivity, and emitter sheet loss explains
`the observed loss in FF.
`
`Ade or
`AFF
`(mA/cm?)
`
`-0.6
`0.3
`-0.2
`
`-0.4
`
`-0.01
`-0.01
`-0.004
`
`AEff
`(%)
`
`0.4
`-0.3
`-0.1
`
`-0.5
`-0.2
`
`-0.2
`-0.2
`-0.1
`
`1. Short Wavelength Response
`FSRV (cm/s)
`150,000
`9 000
`Emitter Doping
`120 Q/sq.
`40 O/sq.
`k#0
`NA
`SIN Absorption
`2. Reflectance
`3%
`6%
`ZnS/MgF2
`SIN/MgF2
`3._ Quality of Contacts
`Contact Res.
`Negligible | 10 ma-cm?
`2x10°
`6x10°
`Conductivity (S/em)
`2.5 mm
`1.0 mm
`Finger Spacing
`4, Long Wavelength Response
`+02
`2000
`BSAV (cm/s)
`+01
`10,000
`Total Calculated Losses: AJs.=2.4 AFF=0.024 An=1.9%
`AJsc=2.5 AFF=0.024 An=1.9%
`Actual Losses:
`
`Grid Shading
`AR Coating
`
`|
`
`|
`
`[|
`
`of both cells was
`or back surface recombination, the
`(tT)
`Th
`measured by the PCD technique [20] on cells etched down
`to bare silicon and immersed in HF to passivate the surface.
`It was found that
`was 275
`s for the CFP/PL cell and
`350 s for the RTP/SP/PECVD cell. Since for both cases, the
`=
`bulk diffusion length to cell thickness ratio is high, both cells
`are surface dominated. To estimate the effective back surface
`+
`recombination velocity (BSRV) at the p-p
`interface, PC-1D
`was used to match the calculated long wavelength IQE to
`the measured data (Fig. 9). The BSRV was estimated to be
`about five times lower for the RTP/SP/PECVD cell (BSRV
`~
`2000 cm/s) compared with the CFP/PL cell (BSRV
`10 000
`cm/s). The difference in BSRV is attributed to the difference
`in aluminum BSF profiles which were formed at different
`alloy temperatures. Since the BSF alloy temperature for the
`RTP/SP/PECVD cell was 930 C compared with 850 C for
`the CFP/PL cell, the higher temperature BSF will have a
`greater junction depth (in accordance with the Al/Si phase
`diagram [21]), thus reducing the BSRV. Others have also
`shown substantial improvement with increased aluminum BSF
`alloy temperatures by both CFP [22], [23] and RTP [23]
`technologies. Table II summarizes all of the efficiency loss
`mechanisms of the RTP/SP/PECVD process.
`
`IIe
`
`D. Improved Long Wavelength Response of the
`RTP/SP/PECVD Cell Resulting from Reduced
`Back Surface Recombination
`Although most of the difference between the two cells
`lies in the short wavelengths, there is an additional small
`difference between the two cells in the long wavelength
`response (Fig. 3). Multiplying this difference with the solar
`spectrum and IQE (as in (1), but over the wavelength range
`of 800–1100 nm) reveals that the RTP/SP/PECVD cell gives
`0.2 mA/cm greater current density in the long wavelength.
`To decouple whether the difference is due to bulk lifetime
`
`V. CONCLUSION
`integration
`In summary, we present
`the first successful
`of three rapid, low-cost, high-throughput technologies with
`manufacturable cell efficiencies in the range of 15%–16% on
`single-crystal silicon. We have shown that emitter design is
`critical for these cells. A
`of 40–50 /sq. was determined as
`Q
`ps
`the optimum, given the trade-off between short wavelength
`response and fill factor. Detailed modeling and characteri-
`zation of these novel cells was presented to decouple all of
`the RTP/SP/PECVD efficiency-limiting mechanisms (Fig. 10)
`and to provide fundamental understanding and guidelines for
`further improvements. On the basis of this in-depth analysis,
`
`Authorized licensed use limited to: Sidley Austin LLP. Downloaded on June 22,2024 at 19:31:14 UTC from IEEE Xplore. Restrictions apply.
`
`

`

`DOSHI et al.: MODELING AND CHARACTERIZATION OF HIGH-EFFICIENCY SILICON SOLAR CELLS
`
`1423
`
`the performance of these RTP/SP/PECVD cells can be raised
`intelligently and systematically to reduce the efficiency gap
`between these and CFP/PL cells. In an upcoming publication
`[16], we demonstrate that
`the gap between RTP/PL and
`CFP/PL cell efficiencies is eliminated by applying high quality
`rapid thermal oxide (RTO) passivation; thus, the difference in
`performance is related to screen printing losses and not RTP.
`Research is in progress to implement rapid thermal oxidation
`for surface passivation, low-cost selective emitters to reduce
`emitter recombination while preserving fill factor, and deeper
`aluminum BSF’s by screen-printed aluminum to reduce back
`surface recombination. The success of the RTP/SP/PECVD
`process suggests that use of these promising technologies in
`industry would result in significant cost reduction and high-
`efficiency cells.
`
`ACKNOWLEDGMENT
`The authors would like to thank Dr. G. E. Jellison, Jr., of
`Oak Ridge National Laboratories for spectroscopic ellipsom-
`etry measurements and Dr. D. Ruby and coworkers at Sandia
`National Laboratories for cell testing and verification.
`
`REFERENCES
`
`[1] J. Nijs, E. Demesmaeker, J. Szlufcik, J. Poortmans, L. Frisson, K. De
`Clercq, M. Ghannam, R. Merterns, and R. Van Overstraeten, “Latest
`efficiency results with the screenprinting technology and comparison
`with the buried contact structure,” in Proc. 24th IEEE Photovoltaic
`Specialists Conf., Piscataway, NJ, 1994, pp. 1242–1249.
`[2] H. Nakaya, M. Nishida, Y. Takeda, S. Moriuchi, T. Tonegawa, T.
`Machida, and T. Nunoi, “Polycrystalline silicon solar cells with V-
`grooved surface,” in Tech. Dig. Int. Photovoltaic Solar Energy Conf.,
`Nagoya, Japan, 1993, pp. 91–92.
`[3] L. Verhoef, P.-P. Michiels, R. J. C. van Zolingen, H. H. C. de Moor,
`A. Burgers, R. Steeman, and W. Sinke, “Low-cost multicrystalline
`silicon solar modules with 16% encapsulated cell efficiency,” in Proc.
`24th IEEE Photovoltaic Specialists Conf., Piscataway, NJ, 1994, pp.
`1547–1550.
`[4] K. Fukui, H. Yamashita, M. Takayama, K. Okada, K. Masuri, K. Shira-
`sawa, and H. Watanabe, “15 cm  15 cm high efficiency multicrystalline
`silicon solar cell,” in Proc. 22nd IEEE Photovoltaic Specialists Conf.,
`Piscataway, NJ, 1991, pp. 1040–1044.
`[5] R. P. S. Thakur et al., “Thermal budget considerations in rapid isother-
`mal processing,” Appl. Phys. Lett., vol. 64, pp. 327–329, 1994.
`[6] C. Leguijt, P. L¨olgen, J. A. Eikelboom, A. W. Weeber, F. M. Schuur-
`mans, W. C. Sinke, P. F. A. Alkemade, P. M. Sarro, C. H. M. Mar´ee, and
`L. A. Verhoef, “Low temperature surface passivation for silicon solar
`cells,” Sol. Energy Materials Sol. Cells, vol. 40, pp. 297–345, Aug. 1996.
`[7] R. Gonsiorawski and G. Czernienko, “Method of fabricating solar cells
`with silicon nitride coating,” U. S. Patent 4 751 191, assigned to Mobil
`Solar Energy Corporation, issued June 14, 1988.
`[8] Product information regarding “RTP solar cell furnace,” Vortek Indus-
`tries, Vancouver, B.C., Canada.
`[9] P. Doshi, A. Rohatgi, M. Ropp, Z. Chen, D. Ruby, and D. L. Meier,
`“Rapid thermal processing of high-effici