The in-line RTP furnace

For any state-of-the-art solar cell production line, wafer throughput is a key figure. A short-term goal is to produce 1200 solar cells per hour per line [162]. With the single-wafer RTP systems available today, these requirements are not met unless a large number of systems is used at once.

Burn-out zone RTP unit Diffusion zone Cooling zone

THL

Cooled isolation Isolation

Mercury lamp Gas inlet

Gas outlet Pyrometer

Wafer on string Reflector Quartz channel

Fig. 1.3:Schematic drawing of a cross section through the in-line RTP furnace. The wafer runs through the furnace from the left to the right.

For this reason, we aimed at designing new RTP furnaces for PV applications. There are two possible ways to achieve high-throughput RTP systems.

One way is to enlarge the process chamber in order to process several wafers simultaneously.

This approach has been pursued within the European FLASH project for example [35]. The system currently being built features a process chamber of 60 60 cm²in size and will enable us to process 25 wafers of the 100 100 mm² size at once. This can be called the real RTP approach because the system will feature all the characteristics of a single-wafer system.

The second approach is to use an in-line furnace and to reduce the time being spent by the wafer at high temperature. Some people use a resistance or infrared heated metal belt furnace and simply increase the belt speed or reduce the furnace length [169, 46]. For such systems the typical time at high temperature is in the range of 2 to 6 minutes [30, 66]. Of course, from the viewpoint of conventional processes (POCl3 diffusion for instance) such processes might be called fast or even rapid. However, some of the RTP specific advantages like low cross contamination, cold walls, short wavelength irradiation and high heating and cooling rates are not supported by such systems. For this reason we aimed at building a real in-line RTP furnace keeping many advantages of single-wafer RTP.

Description of the in-line RTP system

In collaboration with the company Centrotherm a new in-line RTP system was designed and built. A schematic drawing of the system is shown in Fig. 1.3. The overall furnace length is approximately 5 m. The wafer runs through the furnace on ceramic strings working according to the walking beam principle (see section below). The strings move in a channel formed of

quartz plates which represent the actual process chamber. The entire chamber and the transport system contain no metal parts that could lead to wafer contamination during processing. In addition, the inside of the channel is kept at slightly higher pressure which helps to avoid cross contamination from outside the chamber. Heating is performed by tungsten halogen lamps (THL) located outside the chamber. The process temperature is monitored and controlled via thermocouples attached to the inner side of the quatz chamber.

The furnace basically consists of four zones. The first zone is designed as a burn-out zone, where organic components of dopant sources deposited on the wafer surface can evaporate and burn. This zone is followed by a specially designed RTP unit featuring all the key characteristics of a single-wafer RTP reactor. For example, its walls are water-cooled and coated with a light reflector. Wafers can be heated up and cooled down homogeneously with very fast and pyrometrically controlled temperature ramps up to 200 K/s. The RTP zone can be regulated independently from all other zones and was implemented in order to have all kinds of freedom in the design of thermal cycles. Mercury UV lamps, mainly emitting high-energy photons with wavelengths between 200 and 320 nm, can be switched on optionally¹. Next to the RTP unit, a rather conventionally designed 1.2 m long segment is attached. Within this segment an approximately 1 m long zone of constant temperature forms the plateau of a thermal cycle (e.g. diffusion plateau). The medium isolated walls ensure that the wafer temperature and the temperature of the surroundings do not deviate too much whitin this zone. However, since the THLs typically run at 40 to 60 % of their maximum power, the situation in the chamber is far from thermal equilibrium. In addition, the thermocouples regulating the lamp power are irradiated directly by the lamps. Hence, the actual wafer temperature can deviate significantly from the temperature of the thermocouples. Measurements suggest a deviation by approximately 50 C. Like in the RTP unit, mercury UV lamps are implemented in the diffusion zone as well. The last zone represents a cooling zone featuring well-cooled walls to ensure rapid wafer cooling. Normally, the lights in this zone are switched off.

The walking string drive

The wafers are moved through the furnace by a novel patented drive [7, 8] based on two revolving pairs of strings (see Fig. 1.4). The strings are made of flexible ceramics which can be operated at elevated temperatures up to 1400 C. The strings move as follows to transport the wafer through the furnace. When, for example, the wafer lies on the outer pair, this pair moves about 30 cm in the forward direction. In the meantime, the inner pair moves back to its turning point. Next, the inner pair accelerates in the forward direction and takes up the same speed as the outer pair. Then it lifts vertically up to the same position of the outer pair ensuring that the wafer lies on both pairs of strings for a short moment. Now, the outer pair moves down, reverses its speeds and goes through the same cycle as described for the inner pair. This version of the drive ensures a fully continuous wafer transportation.

1The UV lamps were implemented in order to benefit from the proposed photon-enhanced diffusion of P in Si.

However, as shown in chapter 2, we could not verify the observation of this effect.

Fig. 1.4: Picture of the walking string transportation system implemented in the novel in-line RTP furnace.

The walking string transportation system offers the following advantages over currently used metal belts:

The low thermal mass of the strings allows us to establish diffusion processes with steep heating ramps and short cooling times because no massive metal belt has to be heated up and cooled down (see 3.1.7).

The strings are made of ceramics and move back and forth only 30 cm which leads to significantly reduced cross-contamination of wafers [9].

Due to its high flexibility, this furnace allows us to apply process schemes already known from single-wafer processing units, but in a high-throughput in-line equipment. The maximum drive speed of currently 1800 mm/min belongs to a diffusion time of approximately 30 s. This enables a maximum throughput of roughly 600 wafers/h of the size 125 125 mm² with our single-track prototype furnace. A production type version would feature at least four tracks, thus enabling a maximum throughput of 2400 wafer/h. The corresponding rapid diffusion processes and the resulting P emitters are presented and characterized in section 3.1.7. In section 4.3.4 theses P emitters are used for the manufacturing of Cz solar cells and in section 4.6.7 for the manufacturing of solar cells from EFG silicon. We will show that a diffusion time of 18 s is sufficient for the diffusion of screen-printable emitters.

Chapter 2

On the photon-enhanced diffusion of phosphorus in silicon

Several authors proposed an enhanced diffusion of P in Si under illumination with visible and UV photons. In this chapter we try to verify this observation using UV photons from excimer UV lamps and visible photons from tungsten halogen lamps. Several P sources of different chemical composition are tested on their response to diffusion with and without illumination.

Much effort is invested to avoid all problems concerning temperature measurements in RTP units. Within the measurement error we f ind no evidence of any photon-enhancement of the diffusion process. In addition, we show that the proposed enhancement might be explained also by erroneous temperature measurements.

2.1 Introduction and literature survey

A frequently asked question about RTP is: why is RTP so fast compared to conventional pro-cessing in a quartz tube furnace (CFP)? Of course, the significantly higher heating and cooling rates used in RTP reduce the overall process time drastically compared to CFP. Also, the plateau temperatures applied in RTP are generally higher than the ones feasible in CFP. This makes it possible to achieve the same kinetic results with shorter plateau times. Additionally, in the case of diffusion of phosphorus in silicon, highly P-doped SiO2 layers (PSG) are deposited on the wafer surface prior to RTP, whereas in CFP formation of the PSG takes place in the furnace by a chemical reaction of POCl3 and O2 with the Si surface. However, especially for the diffusion of phosphorus, several researchers have reported that RTP is faster than CFP, even if heating and cooling times are neglected and identical plateau temperatures and P sources are employed. Quite often, high energy photons from the tungsten halogen lamps (THL), present exclusively in RTP, are suggested as a possible cause of the enhancement. If so, the use of extra light sources emitting short wavelength photons (e.g. UV photons from excimer lamps) could reduce the diffusion temperature significantly. For photovoltaics, this could be advantageous because some of the widely used multicrystalline Si materials have shown to degrade at elevated diffusion temperatures. They might benefit from reducing the diffusion temperature below

15

some critical temperature (e.g. 900 C) yet maintaining the RTP specific short process times and, possibly, high throughput. In the following a brief literature overview on the observations on the enhanced diffusion is given and, as far as available, on the possible explanations for the acceleration mechanisms.

Hartiti et al. [58] have compared RTP and CFP using a phosphorous spin-on dopant (SOD).

According to their results, 25 s of RTP yields lower sheet resistances than 15 min of CFP at the same temperature. Unfortunately they did not mention whether the 15 min of CFP took place at the plateau temperature or included the time necessary for pushing and pulling the wafers into and out of the furnace, respectively, which would decrease the actual time spent at the plateau temperature significantly.

Doshi [28] has also investigated RTP and CFP diffusion from a P spin-on source under identical temperature/time conditions of 880 C/10 min. The sheet resistance of the RTP sample was only 28 sq compared to 100 sq for the CFP sample. Remarkably, the P profile of the RTP sample was not only deeper than that of the CFP sample but also showed much higher near-surface concentrations. Again, it is not clear whether the 10 min CFP included loading and unloading of the wafers and whether the peak temperature was attained.

Singh and co-workers have reported on the enhanced diffusion of P in Si in many publications (e.g. [164, 109, 165, 166, 186, 168]. The experiments were mainly carried out in an RTP reactor equipped with a bank of tungsten halogen lamps on just one side. Using spin-on and spray-on P sources, the resulting sheet resistance was lower when the wafer surface with the dopant faced the halogen lamps than when it was directed away from them. Singh and co-workers propose that the short wavelength photons ( 800 nm) of the THL spectrum promote quantum photoeffects which cause the enhancement [166]. The claim that because of the use of halogen lamps with a color temperature of 2000 to 3000 K, RTP features a high flux density of visible and UV photons, whereas CFP only includes photons in the infrared range of the light spectrum since furnace and wafer are at the same temperature. Hence, they say, RTP is faster than CFP due to this fundamental difference in the utilized radiation spectrum. Additionally, they claim that extra illumination with UV photons from deuterium or mercury lamps further accelerates the diffusion. According to their understanding, visible and UV photons are a cause of electronic excitation of atoms and molecules. In the electronic exited states there is an increase in bond lengths compared to the ground state which results in a decrease in bond energies and hence enhances diffusion. In contrast, the low energy infrared photons present in CFP only cause rotational and vibrational excitation [166]. It has to be noted that Singh et al. have so far not given any explanation of the effect that would take into account the specific mechanisms of P diffusion in Si like vacancy-P or interstitial-P pairs. Nevertheless, they also stated that the diffusion of P is in fact enhanced in the Si bulk which they concluded from the observation that a CFP pre-diffused P profile is driven in deeper during a subsequent RTP step when the wafer surface is irradiated with light that contains more UV photons.

The diffusion of P in Si under light irradiation at low temperatures was investigated by Ishikawa et al. [73, 72]. The authors observed enhanced P (and B) diffusion by the irradiation of light from tungsten halogen lamps for spin-on and CVD deposited phosphosilicate glasses. For example, from the P profiles diffused for 20 to 120 min at 760 C in an RTP arrangement, they

have deduced the concentration dependent diffusion coefficient. Apparently, this coefficient corresponded to the one published for diffusion at even 900 C in a conventional furnace. It should be mentioned that, during the illuminated diffusion, the wafer was lying on another wafer which served as a boat. The temperature was measured using a thermocouple attached to the bottom of the boat. Ishikawa and co-authors suggested that the reason for the enhancement may be excess self-interstitials generated at the interface of the doped oxide and the silicon surface.

Noël et al. [123, 124] (see also [50]) reported on the impact of ultraviolet light from mercury lamps during rapid thermal diffusion of P. The process temperature was measured with a calibrated pyrometer pointing towards the uncoated back surface of the samples. According to their results, diffusion from a highly concentrated spin-on source is enhanced by extra UV light incident on the coated surface. The enhancement was demonstrated by lower sheet resistances and deeper profiles compared to the arrangement without extra UV light. In addition, different lamp configurations (front, back and double-sided heating) were investigated. Like Singh and co-workers, they observed that photons from the tungsten halogen lamps cause accelerated diffusion. The sheet resistance was lower and the P profile was deeper when the wafer surface covered with the dopant faced the halogen lamps than when it was directed away from them.

However, in contrast to Singh, they observed non-accelerated diffusion in case of RTD of conventionally pre-diffused profiles. Thus, they concluded that the acceleration was due to an effect in the deposited spin-on dopant glass or at the oxide/silicon interface rather than to an effect in the Si volume itself.

A comparison of the kinetics of P diffusion from a spin-on glass source during RTP and CFP diffusion was performed by Mathiot et al. [108]. For conventional diffusion (900 C, 15 min), the resulting P profile could be simulated well using a standard diffusion simulation tool whereas in the case of RTD (850 C, 90 s) the simulation failed because it was much deeper than the predicted one. The authors claimed that this would confirm the enhanced behavior caused by the SOD-RTD process. Further, they concluded that the enhancement was neither due to RTP itself, because they managed to simulate RTD of pre-diffused profiles, nor to the presence of the SOD film itself (e.g. because of interfacial stress effects). It was suggested that the enhancement was due to a noninstantaneous dissociation of P-Si-self-interstitial pairs injected into the Si substrate due to the presence of the highly doped SOD source. According to the authors these transient phenomena might be further strengthened by additional irradiation with UV light.

From the cited work one might be tempted to think that photon-enhanced diffusion is a well established fact. However, it has to be mentioned that the observed enhancement may in fact be caused simply by thermal effects rather than by photon or other non-thermal effects. It is widely accepted that determination of the absolute wafer temperature is extremely difficult in RTP equipment owing to the fact that the wafer is not in thermal equilibrium with its surrounding. Much of the work and experiments describing diffusion behavior of semiconductors is accompanied by uncertainty of the sample temperature. For example, the wafer temperature might be different in an arrangement with extra UV irradiation because of inaccurate temperature measurement related to an inappropriate pyrometer calibration.

Furthermore, in some experiments, thermocouples sometimes faced the incident radiation and, at other times, the opposite direction (see for instance [109]). Later in this work it will be shown that in this case the associated error in temperature can easily reach up to 100 C. In [102], Lojek, one of the pioneers in the field of RTP, remarked that the frequently stated hypothesis about atypical diffusion in RTP systems was just a consequence of unknown or incorrectly determined processing temperature.

In support of this statement, Nagabushnam et al. [115] presented experimental results in-dicating the absence of any photon-enhanced diffusion at least at high temperatures (1000 to 1050 C). They studied the implanted dopant movement in Si during rapid thermal annealing, once with the implanted side facing the lamps and once with the wafer back side facing the lamps. The sheet resistance and the atomic concentration profile were found to be identical for both annealing conditions. It is noteworthy that the authors guaranteed that both wafers were subjected to the same thermal cycle as they used an RTP system equipped with wafer emissivity independent pyrometry. However, they conceded that at temperatures much lower than 700 C their calculations showed that there might be a photon-enhanced diffusion stemming from carrier recombination-assisted diffusion.

The experiments presented in this chapter were designed to verify the proposed accelerated diffusion of phosphorus in silicon by the irradiation of short wavelength photons (visible and/or UV) during RTP diffusion and to gain some knowledge of the prerequisites for the enhancement (e.g. kind of P source). However, the goal of this work was not to provide a microscopic understanding of the mechanisms involved. Also, we did not aim to prove that diffusion in RTP is generally enhanced compared to diffusion in CFP equipment.