Resource consumption and efficiency
Resource flows in carrier/co-product metal systems for CdTe raw material production
The CdTe system is a good example of a system in which the key raw materials are produced as by-products of other systems. Approximately 40% of Te produced in the world today is used in CdTe PV applications (USGS 2021). To produce the Te and Cd, the prior production of Cu, Zn, and Pb cannot be avoided. The Cu system is required for Te production and the Zn system for Cd production. The Zn and Pb systems are linked, and the Pb system is also needed for the recycling of Te. The quantities and overall recoveries for relevant metals produced in the system to manufacture one CdTe module, as predicted using the simulation model, are shown in Table 1.
Table 1 Total mass and recovery of metals produced Table 1 shows that, to produce the CdTe needed for one PV module, also taking into account Cd and Te needed for non-PV uses, the system represented by this simulation would automatically also produce 104 kg Cu, 74 kg of Zn, and 35 kg of Pb due to the interconnectedness of the metal production systems. Without production infrastructure for the carrier metals, it would not be possible to bring PV modules to market. With the strong forecast growth in PV deployment (IRENA 2019), carrier- and co-metal production requirements could be challenging to meet. For CdTe PV, it has been shown that meeting even conservative demand forecasts would be limited by the periodic availability of Te, rather than its scarcity, which is strongly dependent on the supply chain and production methods of Cu (Fthenakis 2012; Bustamante and Gaustad 2014).
Resource flows and efficiencies in the mono-Si system
Figure 4 shows a Sankey diagram representing the closed, steady state Si balance for the system, with line widths proportional to the elemental Si content of each stream. It provides a visualisation of the locations and relative magnitudes of Si-containing streams, including losses, in the life cycle for a case in which 50% of the kerf residue and 95% of EOL wafers are recycled.
Considering the streams exiting the wafer cutting process, the magnitude of the loss of high-grade, expensive SG-Si as kerf becomes clear and highlights the opportunity to increase material efficiency through kerf recycling at MG-Si quality. A second option, the vertical line between wafering and the Czochralski process, is shown for recycling kerf at the higher SG-Si quality. This option will be highlighted again in the carbon footprint analysis. Based on the configuration of our simulation, considerable amounts of Si also leave the system as microsilica, a useful byproduct used as an additive in refractories and concrete (Ciftja et al. 2008), and in residues from various other processes. By identifying and quantifying losses throughout the life cycle, a more realistic view of RE is obtained.
For the scenario shown in Fig. 4, the recoveries of materials as a percentage of the quantity entering the assumed recycling process are 86.9% Si from wafers, 70.3% Ag, 82.2% Cu, 98.9% Al (including module frames), 94.1% Sn (as SnO2), and 94.0% Pb (as PbO2). Note that these values have been updated from a previous version (Bartie et al. 2021a)—we have removed the assumption that 10% of recycled Si wafers can be re-used directly, and have included Al recovery from module frames.
Effects of closed-loop Si recycling on PV power potential in the mono-Si system
The use of NNs as surrogates for the simulation model allows for the effects of parameter ranges to be analysed relatively easily. Figure 5 shows the combined effects of closed-loop EOL and kerf Si recycling, at constant primary quartzite consumption, on the nominal PV power that could be generated from all the Si available in the system at a PCE of 21.7%. As one would intuitively expect, increased circularity increases power generation potential without the need for increased primary resource consumption. As described in Sect. 2.1, this effect can be quantified realistically using the thermodynamic process simulation approach. Three scenarios are shown as points in Fig. 5 at a constant quartzite consumption of 100 kt—the reference scenario with no recycling, a 95% EOL recycling scenario, and one in which 50% of the kerf residue is additionally recycled. While the simulation model allows for the curved surface to be generated for any quantity of quartzite consumption, 100 kt is used in Fig. 5 to allow for a horizontal surface representing the total nominal PV power generation capacity deployed by the EU and UK in 2020 (IEA 2021), to be shown as a tangible reference.
For the consumption of 100 kt of quartzite with no recycling, the equivalent nominal PV power generation potential is 9.7 GWDC. With 95% EOL recycling, this value increases by 88%, less than 95% because of accounting for losses in the system. Adding the 50% kerf recycling results in a 137% increase from the reference scenario. In this simulation, an increase greater than 100% is achieved because primary quartzite consumption is not displaced by the equivalent amount of recycled Si but is added to the available Si in the system, representing growth in PV deployment. The 16.7 GWDC of nominal PV power reference can be achieved via any combination of raw material consumption, EOL recycling rate, and kerf recycling rate on the horizontal plane at that value. However, potential trade-offs with environmental impact and economic viability must also be evaluated. More detailed results can be found in Bartie et al. (2021a).
Using exergy to identify sources of resource inefficiency
The exergy flows that occur during CdTe and mono-Si PV manufacturing are shown in Fig. 6. Of the exergy inputs (material and energy streams combined), 41% and 24% are lost irreversibly in the CdTe and mono-Si systems, respectively, which equate to specific exergy dissipations of 19.5 and 15.5 kWh/m2 of these modules produced, respectively. Module sizes are based on the specifications of commercial units—0.72 m2 for CdTe (First Solar 2018) and 1.96 m2 for Si (Frischknecht et al. 2015). Reducing the amounts of Al used for module frames (assumed for both systems) or producing frameless modules, reducing the use of adhesives and polymer foils, and the incorporation of more renewable energy sources into electricity grid mixes are highlighted as potential opportunities for RE optimisation under the operating conditions specified in the models presented here.
This type of analysis can be done for any process or combination of processes in the life cycle. In the CdTe system, for example, a Zn fuming furnace was introduced to connect the Pb and secondary Cu systems and this resulted in a 4% increase (from 53 to 57%) in overall system exergy efficiency (Bartie et al. 2020). At the same time, however, this resulted in a 7% increase in GWP and a 9% increase in AP, highlighting the interaction and trade-offs between RE and environmental impacts that need to be optimised. It should be noted that although this system is based on best available techniques, it has not yet been optimised. Therefore, there is a high probability that efficiencies could be further improved through innovation while also reducing the magnitudes of any trade-offs. Figure 7 shows the contribution of subsystems to the total exergy cost for the production of Te and Cd. Exergy cost is expressed as exergy (in kWh) dissipated per tonne of metal produced.
For both Te and Cd, more than half of the total exergy dissipation originates from energy-intensive electrochemical refining processes. When electricity is used, its exergy (which equals its energy) is completely dissipated, regardless of how it was produced. However, its embodied environmental impact is strongly influenced by how it was produced (e.g. the mix of fossil and non-fossil resources used), which depends strongly on where it was produced. Therefore, the most effective way to improve the net sustainability of these processes would be to locate them where electricity grid mixes are made up of predominantly renewable energy sources and not in locations where carbon-based electricity grids are still the norm. This is discussed further in Sect. 3.4.
Environmental impact
Impacts and allocation challenges in the CdTe system
To produce all the quantities in Table 1, the total system GWP has been estimated at 733 kg CO2-equivalent (kgCO2e) and the total AP at 7.7 mol H+ equivalent (mol H+-eq.) using the ILCD midpoint v1.09 (ILCD 2011) life cycle impact assessment method. To be able to state the environmental impact associated with individual products in this large system, the overall impacts need to be distributed in an appropriate way. As mentioned, LCA guidelines recommend that, if it cannot be avoided, allocation should be based on physical relationships between the products and their environmental impacts or on economic value, the former the preferred option. In multi-metal systems such as that presented here, allocation cannot be avoided as subdivision of the production processes is not possible (Ekvall and Finnveden 2001). Following these guidelines, the distributions of overall impacts to the system’s products were calculated by quantity produced, exergy cost, exergy content, and economic value for comparison (see Table 2).
Table 2 Total emission distribution between outputs by mass, exergy cost, exergy content, and value of system products As is evident from Table 2, the results are generally inconsistent—it is difficult to decide which set of distributed impacts is most likely to be representative of reality. Similar challenges have been reported by others (e.g. Stamp et al. 2013; Bigum et al. 2012). Furthermore, various additional calculation approaches are recommended in LCA guidelines to account for EOL impacts (i.e. cut-off/recycled content, EOL recycling/avoided burden), which are applied differently for open loop and closed loop recycling, and in attributional and consequential LCAs (Nordelöf et al. 2019). Detailed descriptions are beyond the scope of this paper but suffice it to say that these add further complexity and can be counterintuitive (Guinée and Heijungs 2021). Difficulties also arise when attempting to compare results with those of other researchers, as the studied systems are often not directly comparable (Farjana et al. 2019). In this study, only some of the allocated values agree with those published by e.g. Nuss and Eckelman (2014), Van Genderen et al. (2016), and Ekman Nilsson et al. (2017), and only if mixed allocation methods are used. A hybrid allocation method could, therefore, be implemented in some way, but would likely have to be based on somewhat arbitrary and subjective assumptions.
For the current system, as defined in the simulation: the price of Cd is only 3% of that of Te. To produce the CdTe semiconductor, however, Cd is clearly just as important as Te. In this case, mass-based allocation would be more appropriate. Looking at the overall system in which much larger amounts of Cu, Zn, and Pb are produced with Cd and Te for applications other than PV, value-based allocation would probably make more sense as the producer’s objective would be to maximise profit. Because Te is significantly more expensive than all the other metals, a portion of the environmental impact would be allocated to it—in this case an order of magnitude more than with the other allocation factors. Such a small quantity is produced; however, that its impact is virtually negligible relative to that of the system (0.2% based on economic allocation). Similarly, and even though eleven times more Cd than Te leaves the system as a product, its allocated impact is even smaller (less than 0.1% for mass, exergy, and economic allocation). Allocation based on exergy cost gives impacts several orders of magnitude higher for both Cd and Te, but it is unclear how a sensible choice between the allocation factors would be made. Subjective or arbitrary decisions would have to be made in this scenario to generate an uncertain result that would likely carry low credibility.
The simplest and clearest way to avoid having to choose between various EOL and allocation methods and/or combinations of them is to make use of detailed process models such as those presented here and in other recent work (Abadías-Llamas et al. 2019; Bartie et al. 2020; Hannula et al. 2020; Fernandes et al. 2020). The flowsheet models contain all the necessary detail to determine the absolute emissions from every process in the system as and when they really occur, eliminating the need to divide the overall emission between outputs.
Effects of circularity on carbon footprint in the mono-Si system
Following the same approach as for nominal power generation, Fig. 8 shows the CO2-equivalent emissions per nominal kW power generated for the German electricity mix and quantifies how increased circularity could increase sustainability. With no recycling, emissions amount to 659 kgCO2e/kWDC based on the assumptions in our simulation, decreasing by 13% with 95% EOL recycling and an additional 1% by adding 50% kerf recycling. The decreases are mainly due to reductions in Scope 2 emissions—in the system as defined here, EOL recycling bypasses the Siemens process, which is the most energy-intensive process in the life cycle, while kerf recycling only bypasses MG-Si production. An additional 3% decrease in emissions could be achieved by recycling kerf at SG-Si quality, in which case the recyclate would also bypass the Siemens process (see Fig. 4). This analysis highlights and quantifies the effects of recyclate quality on sustainability and the potential benefits innovation and development of high-quality recycling processes could bring, albeit that the potential environmental footprint of such upcycling processes has not been considered here.
The locations of the energy-intensive processes have a strong influence on emissions. Although it is assumed in Fig. 8 that the entire life cycle is co-located on the German electricity grid, it is instructive to point out that moving it to Australia, for example, would result in a 32% increase in overall CO2-equivalent emissions, while moving it to Brazil would result in an 26% decrease. There are, of course, other factors at play, such as where material resources are geographically located, production costs at different locations, transport costs, trade regulations, etc. No one conclusion should be viewed in isolation, but rather as part of the overall system that needs to be optimised.
Combined effects of Si wafer thickness and EOL recycling on carbon footprint in the mono-Si system
Over the last decade, improvements in cell and module efficiencies have resulted in a 50% reduction in the amount of Si needed to generate a Watt of power and this trend is expected to continue. The average thickness of the most frequently used Si wafers is currently between 170 and 175 μm, accounting for about 72% of the weight of a standard mono-Si cell (VDMA 2021). This value is expected to decrease to between 150 and 160 μm by 2031 (VDMA 2021), further reducing the consumption of Si for PV systems. Figure 9 shows the variation in CO2-equivalent emissions with EOL and kerf recycling rate for wafer thicknesses of 150 and 175 µm. The reduction from 175 to 150 µm (without recycling) results in a 5% reduction in emissions. However, combined with an EOL recycling rate of 95%, emissions decrease by 15%. Compared to recycling alone (Fig. 8), the contribution of a 25 µm reduction in wafer thickness to decreasing carbon footprint is relatively small. This analysis of the effects of wafer thickness highlights one of the advantages of the process simulation approach—the ability to change process parameters and generate new process inventory data to assess the impacts of expected technology developments.
Technoeconomic assessment and the effects of carbon taxation in the mono-Si system
Figure 10 shows the impacts carbon taxation on MSP and how it is influenced by closed-loop recycling. A carbon tax of $100/tCO2e increases MSP by 24% (from $67/m2 to $83/m2) when no closed-loop EOL recycling takes place, and by 20% at a 95% recycling rate. As soon as recycling is introduced, an upwards step change in MSP occurs due to the fixed costs of the recycling process. As recycling rate then increases, MSP decreases but can only break even with the original MSP when the carbon tax is higher than approximately $75/tCO2e. Below this level, recycling cannot compensate fully for its cost. At a tax level of $100/tCO2e, however, the minimum recycling rate needed to break even is a relatively high 85%. These effects follow the same pattern for the LCOE, but are less significant. A $100/tCO2e tax results in a 6.6% increase in LCOE, from 7.81 to 8.33 c/kWh. However, the effects of balance-of-system items such as land, concrete, support structures, and others on the overall carbon footprint have not been included in our LCOE calculations yet. With these included, the effect of carbon tax on LCOE would be larger.
Both recycling and carbon taxation aim to reduce environmental impacts and lead to increased cost. The linking of process, environmental, and technoeconomic models allows for analyses such as this; however, that shows that there are conditions under which increased recycling could reduce costs to below their original values despite the carbon tax.
The location dependence of the embodied carbon footprint of PV energy in the mono-Si system
As highlighted throughout this paper, infrastructure location plays a pivotal role in life cycle sustainability. Figure 11 shows the ratio of power generated over the lifetime of a mono-Si PV system and its manufacturing carbon footprint as a function of closed-loop EOL recycling rate for different locations. It is assumed that manufacturing takes place in the country indicated that the PV system has a 30-year lifetime with a 0.5%/year degradation rate and a PCE of 21.7%. The average annual insolation is assumed to be 1,500 kWh/(m2.year).
Figure 11 clearly shows how manufacturing location influences carbon footprint. Moving manufacturing from China or Australia to Germany, for example, would have a positive effect by increasing the ratio of energy generated and CO2 emitted by more than 30% in the case without recycling. In China and Australia, 71% and 77% of electricity were generated from fossil resources in 2020, respectively (IEA 2022). Countries like Brazil and Norway, on the other hand, respectively, generated 75% and 98% of their electricity using wind, solar PV, and hydropower in the same year (IEA 2022). To maximise energy output per embodied carbon footprint, it is clear that infrastructure development should occur away from countries still largely dependent on fossil fuels for power generation.
Increased circularity lowers the embodied carbon footprint of PV energy, but this effect is weaker the lower the carbon intensity of the relevant electricity grid. The reason is that the strongest effect of Si recycling is its contribution to avoiding electricity consumption in energy-intensive processes and hence avoiding Scope 2 emissions. The carbon intensity of Norway’s electricity grid is an order of magnitude lower than those of the other countries, and as a result, this life cycle’s Scope 2 emissions are lower than its Scope 1 emissions, the latter constant regardless of location. In this case, the increase in Scope 1 emissions brought about by increased recycling is higher than the simultaneous decrease in Scope 2 emissions (explained in Sect. 3.2.2), resulting in a slightly negative slope for Norway in Fig. 11. To take full advantage of the Si circularity effect, it would make sense to focus on using secondary Si in locations where electricity grids are most carbon-intensive.
These comparisons are based on recently published emission factors for the generation, supply, and distribution of electricity from the ecoinvent (version 3.8, 2021) database (Wernet et al. 2016). It should be noted that these are higher than the carbon intensities of electricity generation reported by the European Energy Agency (EEA) and the International Energy Agency (IEA), among others. The EEA reports a carbon intensity of 0.311 kgCO2e/kWh for Germany (EEA 2021), for example, compared to the 0.558 kgCO2e/kWh from the ecoinvent database (Treyer 2021). While the former is more recent (2020 vs. 2018), it does not include emissions associated with supply chains and electricity losses during transmission across networks. As a result of the time lag, the values used to generate the results presented here (and used for environmental impact assessments in general) may not fully reflect recent progress in reducing the carbon intensities of electricity consumption.