The vast majority of photovoltaic (PV) solar cells produced to date have been based on silicon wafers, with this dominance likely to continue well into the future. The surge in manufacturing volume over the last decade has resulted in greatly decreased costs. Multiple companies are now well below the US$1 W−1 module manufacturing cost benchmark that was once regarded as the lowest possible with this technology. Despite these huge cost reductions, there is obvious scope for much more, as the polysilicon source material becomes more competitively priced, the new ‘quasi-mono’ and related controlled crystallization directional solidification processes are brought fully online, the sizes of ingot produced this way increase, wafer slicing switches to much quicker diamond impregnated approaches and cell conversion efficiencies increase towards the 25 per cent level. This makes the US Government's ‘SunShot’ target of US$1 W−1 installed system cost by 2020 very achievable with silicon PVs. Paths to lower cost beyond this point are also explored.
The last decade has seen a remarkable evolution in mainstream silicon solar cell technology, documented by greatly increased production volumes and greatly reduced costs. The present state of the art is discussed, and some of the potentially key developments over the coming decade are reviewed and possible directions for the longer term outlined.
2. The last decade
The growth of the silicon photovoltaic (PV) industry over the last decade has exceeded most expectations. Production volumes have increased over 60-fold, with costs reducing to only a fraction of their past values. Manufacturing cost of US$1 W−1 for silicon PV were achieved for market leaders during 2011. PVs are increasingly seen as a viable way of meeting growing energy demands on a large scale without adding to the carbon burden.
Despite these impressive gains, there remains plenty of scope for further improvement. The cost and price of polysilicon source material has reduced quickly, the directionally solidified ingots of silicon produced from this material are becoming larger and better quality, the yield of wafers from this material is expected to almost double over the next decade and cell efficiency continues to improve, as discussed in more detail later.
3. Present costs
Silicon manufacturing can be broken down into four stages: purifying silicon from a metallurgical grade (MG), produced mainly for the steel and aluminium industries, to a much higher purity polysilicon, then forming ingots of silicon of good crystallographic quality from this polysilicon and slicing the ingots into wafers, then processing the wafers into cells and, finally, packaging the cells into a PV module. A breakdown of present silicon module manufacturing costs  for these four stages is estimated in figure 1, for the situation at the middle of 2011. With margins on module sales tightly squeezed during 2011, average costs and prices are not dissimilar, with average module selling prices of US$1.30 W−1 in mid-2011 , reducing to US$1.00 W−1 by the end of 2011. Some vertically integrated companies incorporate all four of the above processing stages within their operations, whereas other companies specialize in one or more stages. Progress is expected in all four areas, with likely developments, primarily in the first three, reviewed below.
For companies that do not produce their own polysilicon, the costs of purified polysilicon account for almost a quarter of total module manufacturing costs in mid-2011. Vertically integrated PV cell or module companies producing their own polysilicon historically have had a significant advantage because of the recent large margins on polysilicon sales, but this advantage steadily reduced during 2011, as these margins reduced.
Over the last decade, a significant transition occurred when the PV demand for polysilicon overtook the microelectronics demand. The transition created a supply shortage with consequent sharp price increases, from which the industry has only recently recovered. The transition also encouraged new entrants into the polysilicon supply market, particularly from Asia (GCL, OCI, LDK, Renesolar), and also encouraged the expansion of capacity by established players (Hemlock, Wacker, REC). It also triggered a large effort to commercialize intermediate quality upgraded metallurgical grade (UMG) silicon because of the lower capital costs and potentially shorter lead times for building new facilities, with UMG silicon now marketed by some companies (Elkem, Silicor Materials).
Polysilicon prices reduced steadily in 2011 as demand and supply came more into balance with polysilicon selling at US$30 (kg)−1 by the end of the year, compared with manufacturing costs of below US$20 (kg)−1 for the lowest cost manufacturers. At the 6 g W−1 requirement for completed modules, these manufacturing costs correspond to less than 12 c W−1 contribution to total module costs. Manufacturing costs of UMG are at present slightly higher than standard best practice with the standard Siemens process , although future developments, such as ingot and cell processing approaches that relax the required silicon quality for high performance, could alter such relativities.
5. Silicon ingots
Polysilicon is traditionally converted into mono-crystalline ingots either by Czochralski (CZ) crystal growth (figure 2) or into larger multi-crystalline ingots by simpler directional solidification (DS) of molten silicon within a crucible (figure 3). CZ ingots are typical 20 cm in diameter and 150 kg in weight, while DS ingots are much heavier, now up to 600 kg. The size of both ingot types will increase over the next decade with this particularly significant for DS silicon, where ingot weight is expected to exceed 1 t in the not too distant future .
The central regions of the DS ingots are generally of the best quality due to contamination of regions adjacent to the crucible wall by impurities from this wall, the sinking of precipitates to the bottom of the melt, segregation of impurities to the top of the ingot (the last region to solidify), dislocations caused by stress and the changes in crystallography during solidification. The larger the lateral dimensions of the ingot, the larger the ratio of good to poor material.
Ingots are cut into bricks, as indicated in figure 4. The dominant DS ingot technology is now referred to as Generation 5, where ingots are sliced into 5×5 (25) bricks, each of 156×156 mm cross section (figure 4a). Generation 6 equipment is now on the market (same brick cross section), with the next jump expected to be to Generation 8 technology (8×8 bricks), sometime over the coming decade . This evolution not only reduces costs by increasing throughput, it also increases the yield of good-quality material.
The reducing cost of forming bricks is being matched by reductions in the cost of sawing bricks into increasingly thin wafers. Large gains are expected over the decade by switching from slurry-based wire sawing to sawing with diamond impregnated wire. Advantages include potential cutting speeds two to three times higher, plus prospects for recycling the silicon now lost as kerf loss . This yield of wafers sliced from the bricks is expected to improve from typical 2011 values of 30–50 cm−1 over the decade .
6. Quasi-mono silicon
In the past, manufacturers had the choice of mono-crystalline wafers produced using the CZ process or multi-crystalline wafers made by DS. A very recent development has been refinement of the latter DS approach to produce quasi-mono-crystalline and other controlled crystallization wafers.
As suggested in figure 5, modifying the DS process so that the melt is seeded by a single crystal region at the melt bottom produces large ingots with a mono-crystalline central region, surrounded by a multi-crystalline region around the edge of the crucible. When processed into cells, wafers sliced from the ingot have properties bridging the range between multi-crystalline and mono-crystalline .
Although suddenly taking the industry by storm, this seeded approach has a long history. One of the first papers describing cell results on multi-crystalline silicon also reported experiments with seeded crystallization . Even prior to this, Crystal Systems had proposed extending a technique developed for sapphire to silicon, with good results soon demonstrated . After joint work with Crystal Systems, BP Solar stimulated the recent interest in the quasi-mono material through publication of their work on this approach in 2008 . By the first half of 2011, both ALD and GT Advanced Technologies announced the imminent availability of equipment for producing this material, while JA Solar and Suntech announced cells and modules based on the approach (figure 6).
Since multi-crystalline regions form around the perimeter of the ingot as shown in figure 5, going to bigger ingots also will increase the yield of mono-crystalline material. While this approach has the potential to improve cell efficiencies obtained from DS silicon, there may be other advantages. The lack of grain boundaries in central regions of the ingot means that the segregation of impurities to the last regions to solidify is more effective, possibly allowing the use of lower UMG grades without compromising efficiency loss. Reasonable quality ingots have been produced using similar approaches, even using raw MG silicon .
Since the quasi-mono technology is so young, processing refinements are expected to progressively improve the yield and quality of mono-crystalline material, which has the potential to be better than past CZ mono-crystalline material due to lower oxygen content. This work has also encouraged experimentation with a wider range of controlled crystallization approaches, which has also led to improved wafer quality.
7. Ag-free processing
The recent large increase in Ag price (increased from about US$15 (troy oz)−1 (US$480 (kg)−1) in 2010 to over US$40 (troy oz)−1 (US$1290 (kg)−1) in 2011) has been at least partly caused by the rapidly escalating amount of Ag being used in PV manufacturing, now accounting for close to 10 per cent of all Ag use [10,11]. Uptake for PV is more than offsetting the decrease in demand for photography (with silicon the culprit in both cases). Ag already accounts for a substantial fraction of wafer to cell processing costs (up to a third).
Ag forms a major component in the screen-printing pastes used in the cell metal contacts of the standard cell structure used almost universally by the industry (figure 7). About 50–100 mg W−1 of Ag is required, with cell performance inextricably linked to the amount used in the normal structure [10,11]. As there seems to be no suitable replacement for Ag pastes, this link may encourage a move away from the standard screen-printing approach over the coming decade. Options for less Ag intensive approaches range from minor changes to the present approach, such as plating thin screened metal with Cu, to radically new designs that are also able to capture more of the ultimate efficiency potential of silicon cells [10,11].
8. Cell efficiency
The evolution of the highest reported silicon cell efficiency from the first cell in 1940 to the recent present is shown in figure 8. All of the improvements in cell efficiency since the early 1980s have come from the author's group, with one important exception discussed below.
The 1970s were an active period of cell development with two notable achievements. One was the reporting of the first successful use of pyramidal surface texturing with the ‘black’ cell (figure 9) reported in 1974, boosting cell performance by 10–15% relative . The second was the development of the screen-printing processes used in modern cells, at about the same time . Despite the time that has since elapsed, the present mainstream cell technology is very much locked into the improvements of this decade. The standard commercial cell in figure 7 uses no technology or design features not known in the 1970s, apart from the use of silicon nitride antireflection (AR) coatings, first reported in this context in 1984 . The present standard commercial cell is essentially a ‘black’ cell with screen-printed contacts, demonstrating similar energy conversion efficiency (17–18% efficiency range).
Since then, the performance of the best laboratory cells has improved by 40–50% relative. Over the coming decade, there is likely to be more exploitation of these gains in commercial sequences, possibly encouraged by the increasing cost of Ag in standard sequences.
As mentioned previously, all of the post-1970s efficiency improvements have come from the author's group with one notable exception, the first cell to exceed 22 per cent efficiency. This was a highly innovative rear junction cell developed at Stanford University and subsequently commercialized by SunPower (figure 10). Very similar or even higher efficiencies are now being reported in production, with the resulting modules being the highest efficiency on the market (the first commercial modules to exceed 20% efficiency). Interestingly, since the finger metallization on the rear of the cell runs the whole cell length, much more metal is required than on a conventional cell. Cu is already used since Ag would have been prohibitively expensive, even at the more modest Ag prices of the past.
High efficiency has allowed SunPower to carve out a market niche in the utility-scale market by taking advantage of the combination of cell efficiency and sun tracking. Notwithstanding, the rear junction cell design is probably not suited to becoming the industry standard. This is because premium quality silicon is required for the design to work well. In particular, the design is not suited for use with multi-crystalline silicon, the industry standard and the material where large ongoing cost reductions are most likely, as previously outlined.
The rear junction cell nonetheless has an important place in the history of cell development, since it was the first cell to have both front and rear surfaces well optimized or ‘passivated’ (low surface recombination rates for photogenerated and other excess carriers). The cells that earlier took cell efficiency past 18, 19, 20 and 21 per cent had only the front surface fully optimized, with rear surfaces similar to both the ‘black’ cell and the present mainstream commercial product. The subsequent cell designs taking efficiency past 23, 24 and then 25 per cent  had both surfaces well optimized, finally resulting in the passivated emitter, real locally diffused (PERL) cell in figure 11. The main difference is in the voltage output possible from the cell. While cells with the front surface optimized can give open-circuit voltages up to 650 mV, both surfaces need to be optimized to reach voltages above 700 mV.
9. Cell development
It is anticipated the mainstream solar industry will follow the same evolutionary path as followed by laboratory cells, first refining front surface design, then, once gains here have been exhausted, optimizing rear surface properties. This is already happening to some extent. Several manufacturers are now reporting cell efficiencies above 19 per cent by incorporating some of the high-efficiency front surface design features shown much earlier to be effective in improving laboratory cells, such as the selective emitters discussed later.
The author's team pioneered the use of laser processing as a low-cost way of implementing the team's high-efficiency designs. The buried contact solar cell was the first high-efficiency cell to be introduced into commercial production, beginning in the early 1990s and, at its peak, accounting for 15–20% of total European cell production . An improved version of this technology is the laser doped, selective emitter (LDSE) in figure 12.
In the LDSE sequence, the cell is processed in much the same way as for standard cells, up until after the application of the silicon nitride AR coating, except for a lighter top surface diffusion. A laser is then used to define the area of the final metallization lines by selectively removing the AR coating. At the same time, dopant diffuses into the scribed areas driven by the localized heating, with dopants sourced either from the AR coat, a layer under or on top of it, or from a liquid jet guiding the laser.
After the laser step, metal is plated to the scribed areas by light-induced plating, with plated Cu forming the key conductor within the plated stack. This sequence results in a cell with a selective emitter but with fine linewidth metal automatically aligned to the heavily doped regions of the emitter. Independently confirmed efficiencies up to 19.4 per cent have been reported for large-area cells largely processed on standard lines . Ongoing refinement of such sequences is likely to take standard cell efficiencies above 20 per cent on CZ wafers. The lower oxygen levels in the mono-crystalline regions of quasi-mono ingots may allow even higher efficiency.
Working on the rear surface with similar sequences opens up the potential of much higher voltage output, as well as the potential for a significant boost in current, owing to improved rear reflection and light-trapping properties of the cells. Preliminary progress in this area has recently been reported, with independently confirmed efficiency of 20.3 per cent recently confirmed using commercial wafers . By the end of the decade, it may be possible for commercial cells to converge on the performance of present laboratory cells by this evolutionary approach.
A more revolutionary approach to producing high-performance cells has been adopted by Sanyo with its heterojunction with intrinsic thin layer (HIT) cell approach in figure 13 . The HIT cell relies on hydrogenated amorphous Si (a-Si : H) technology, using the higher bandgap a-Si : H layers at the front and rear to provide both the junction and the required surface passivation. The cell design is also bifacial, responding to light striking either surface. The recent expiry of relevant patents has broadened interest in this approach. Although the approach gives good results on good-quality material through improved voltage, the top a-Si : H layer absorbs a significant amount of incoming light. On low-quality substrates incapable of high voltage, there may be little to gain and something to lose, with advantages less compelling. Although there are savings due to avoidance of Al pastes on the rear, the current collecting requirements on the Ag paste fingers on both top and rear surfaces add to the Ag intensity of the technology compared with the standard approach.
There is also much apparent interest in other cell designs that are contacted from the rear, less reliant on good material than the SunPower rear junction approach, such as the emitter wrap through (EWT) and metal wrap through (MWT) cells shown in figure 14. A recent European PV roadmap in fact suggests the whole industry might transfer to such designs by 2016 . The advantage is the simplification of cell encapsulation, although this comes at the expense of increased processing complexity with little or no gain in cell performance. The Ag intensity is also not significantly different from standard technology [10,11]. Advantages may not be sufficiently compelling to stimulate such a major change in cell technology.
10. Further into the future
The earlier sections of this paper highlight the present state of the art and the improvements expected in increasing the quality and reducing the cost of silicon wafers. Since the cost of these wafers is one key differentiator between silicon and competing ‘second-generation’ thin-film PV technologies, material cost advantages of the latter are expected to decrease with time. This is likely to be compounded by escalation in the price of scarce materials required for the presently most popular commercial thin-film technologies, such as Te and In, similar to that recently seen with Ag, should these reach the production volume of present Si technology. Substitution of these metals is likely to be much more challenging than for Ag.
This raises the question as to how these future low-cost, high-quality silicon wafers might best be used to advantage in the long term. The author has outlined elsewhere his view  that PV must ultimately evolve to high-efficiency, ‘third-generation’ technology targeting the thermodynamic limits on solar conversion (74%) rather than the single junction, Shockley–Queisser limit (33%) [10,11]. Do these increasingly low-cost silicon wafers provide a route to this third generation?
Of all approaches targeting the third-generation efficiency space, only those involving tandem stacks of cells (figure 15) have demonstrated an efficiency improvement over conventional devices to date. Since silicon's bandgap is slightly on the low side of optimal for a single junction cell and its performance is further inhibited by Auger recombination at full sun intensity, the limiting efficiency of a silicon cell, at about 29 per cent, is lower than the optimum for an unconstrained choice of cell material at 33 per cent. The best experimental Si device has efficiency of 25 per cent, while the best from other material has an efficiency of 28.3 per cent  (both 85% of their respective limits).
For an unconstrained choice of two cells in a tandem stack, the efficiency limit increases to 45 per cent. Constraining the bottom cell choice to silicon is not a severe handicap, giving a limiting efficiency value of 42.5 per cent. Going to a three-cell stack, having silicon as the lowermost cell is again not a large disadvantage, giving 47.5 per cent limiting efficiency compared with 50.5 per cent for the unconstrained choice. A four-cell stack with silicon as the lowermost cell has a limiting efficiency above 50 per cent. Adding more cells to the stack progressively improves the efficiency limit for the Si-based tandem. However, the limit falls further and further, in relative terms behind the corresponding limit for an unconstrained choice involving the same total number of cells.
One way that silicon wafer-based technology could evolve into a third-generation technology is by serving as a clean, low-cost substrate for the deposition of thin subsequent layers of a tandem stack. The most promising candidates for the uppermost cells at present would be the III–V compounds, which have produced the best tandem results to date. Since these have smaller atomic spacing than Si, a viable approach needs to be found to match their crystal structure to Si. In the longer term, other semiconductors apart from the III–V may be of interest in such application. For example, there is current interest in exploring the properties of I2–II–IV–V4 tetrahedrally coordinated semiconductors for PVs , with opportunities for atomic spacing matches to silicon within this group.
With the passage of time, silicon seems to be consolidating its position as the leading PV material. Moreover, its technology is evolving rapidly, with enormous resources worldwide contributing to this development.
The situation resembles in some ways the contest in bicycle racing between the peleton and the breakaway. A company with unique technology corresponds to the breakaway, be the technology a different thin-film material or a radically different cell design. The breakaway has to rely largely on its own resources to continue the development of its unique technology. For the mainstream silicon technology, there is a huge industry of equipment and material suppliers all working hard to improve the products being offered, as well as helping to raise the whole industry towards best practice. In bicycle racing, the breakaway might stay out in front long enough to hold out to the finish line. In the solar industry, there is no finish line, just a future of ongoing product development and reducing costs. The breakaway cannot stay out from the peleton indefinitely, but must either assemble its own rival peleton or get caught and go to the back of the pack.
This combined effort is expected to continue the recent price and cost reductions in silicon PVs over the coming decade, with a recent roadmap  suggesting a further 60 per cent cost reduction over this period. Such reductions will see the rapid recent uptake of PVs maintained. The recent very positive experience with large quantities of PVs distributed through the German power grid  should further encourage such uptake, as should the very positive field experience with durability .
One contribution of 15 to a Discussion Meeting Issue ‘Can solar power deliver?’.
- © 2013 The Author(s) Published by the Royal Society. All rights reserved.