Manufacturing silicon: computers’ and solar panels’ crucial ingredient

by Katie Singer

Could we discuss the ecological impacts of manufacturing silicon, that substance on which our digital world depends?¹ Silicon is a semiconductor, and tiny electronic switches called transistors are made from it. Like brain cells, transistors control the flow of information in a computer’s integrated circuits. Transistors store memory, amplify sound, transmit and receive data, run apps and much, much more.

One smartphone (call it a luxury, hand-held computer with portals to the Internet) can hold more than four billion transistors on a few tiny silicon chips, each about the size of a fingernail.

Computer chips are made from electronic-grade silicon, which can have no more than one impure atom per billion. But pure silicon is not found in nature. Producing it requires a series of steps that guzzle electricity² and generate greenhouse gases (GHGs) and toxic waste.

Silicon’s story is not easy to swallow. Still, if we truly aim to decrease our degradation of the Earth and GHG emissions, we cannot ignore it.

Step One  

Silicon production starts with collecting and washing quartz rock (not sand), a pure carbon (usually coal, charcoal, petroleum coke,³ or metallurgical coke) and a slow-burning wood. These three substances are transported to a facility with a submerged-arc furnace.⁴

Note that transporting the raw materials necessary for silicon production—between multiple countries, via cargo ships, trucks, trains and airplanes—uses oil and generates greenhouse gases.⁵

Step Two

Kept at 3000F (1649C) for years at a time, a submerged-arc furnace or smelter “reduces” the silicon from the quartz. During this white-hot chemical reaction, gases escape upward from the furnace. Metallurgical-grade silicon settles to the bottom, 97-99% pure—not nearly pure enough for electronics.⁶

If power to a silicon smelter is interrupted for too long, the smelter’s pot could be damaged.⁷ Since solar and wind power is intermittent, they cannot power a smelter.

Typically, Step Two takes up to six metric tons of raw materials to make one metric ton (t) of silicon. A typical furnace consumes about 15 megawatt hours of electricity per metric ton (MWh/t)⁸ of silicon produced, plus four MWh/t for ventilation and dust collection; and it generates tremendous amounts of CO₂.⁹

Manufacturing silicon also generates toxic emissions. In 2016, New York State’s Department of Environmental Conservation issued a permit to Globe Metallurgical Inc. to release, per year: up to 250 tons of carbon monoxide, 10 tons of formaldehyde, 10 tons of hydrogen chloride, 10 tons of lead, 75,000 tons of oxides of nitrogen, 75,000 tons of particulates, 10 tons of polycyclic aromatic hydrocarbons, 40 tons of sulfur dioxide and up to 7 tons of sulfuric acid mist.¹⁰ To clarify, this is the permittable amount of toxic waste allowed annually for one New York State metallurgical-grade silicon smelter. Hazardous waste generated by manufacturing silicon in China likely has significantly less (if any) regulatory limits.

Step Three

Step Two’s metallurgical-grade silicon is crushed and mixed with hydrogen chloride (HCL) to synthesize trichlorosilane (TCS) gas. Once purified, the TCS is sent with pure hydrogen to a bell jar reactor, where slender filaments of pure silicon have been pre-heated to about 2012F (1100C). In a vapor deposition process that takes several days, silicon gas atoms collect on glowing strands to form large polysilicon rods—kind of like growing rock candy. If power is lost during this process, fires and explosions can occur. A polysilicon plant therefore depends on more than one source of electricity—i.e. two coal-fired power plants, or a combination of coal, nuclear and hydro power.¹¹

A large, modern polysilicon plant can require up to 400 megawatts of continuous power to produce up to 20,000 tons of polysilicon per year (~175 MW/hours per ton of polysilicon).¹² Per ton, this is more than ten times the energy used in Step Two—and older plants are usually less efficient. A single plant can draw as much power as an entire city of 300,000 homes.

Once cooled, the polysilicon rods are removed from the reactor, then sawed into sections or fractured into chunks. The polysilicon is etched with nitric acid and hydrofluoric acid¹³ to remove surface contamination. Then, it’s bagged in a chemically clean room and shipped to a crystal grower.

Step Four

Step Three’s polysilicon chunks are re-melted to a liquid, then pulled into a single crystal of silicon to create a cylindrical ingot. Cooled, the ingot’s (contaminated) crown and tail are cut off. Making ingots often requires more electricity than smelting.¹⁴ The silicon ingot’s remaining portion is sent to a slicer.

Step Five

Like a loaf of bread, the silicon ingot is sliced into wafers. More than 50 percent of the ingot is lost in this process. It becomes sawdust, which cannot be recycled.¹⁵

Step Six

Layer by layer, the silicon will be “doped” with tiny amounts of boron, gallium, phosphorus or arsenic to control its electrical properties. Dozens of layers are produced during hundreds of steps to turn each electronic-grade wafer into microprocessors, again using a great deal of energy and toxic chemicals—and water.¹⁶ I cannot recommend enough viewing “The Semiconductor Water Problem,” a video from www.Asianometry.com.

Questions for a world out of balance 

In 2013, manufacturers began producing more transistors than farmers grow grains of wheat or rice.¹⁷ Now, manufacturers make 1000 times more transistors than farmers grow grains of wheat and rice combined.¹⁸

After I learned what it takes to produce silicon, I could hardly talk for a month. Because I depend on a computer and Internet access, I depend on silicon—and the energy-intensive, toxic waste-emitting, greenhouse gas-emitting steps required to manufacture it.

Of course, silicon is just one substance necessary for every computer. As I report in letter #3¹⁹, one smartphone holds more than 1000 substances, each with their own energy-intensive, GHG-emitting, toxic waste-emitting supply chain.²⁰ One electric vehicle can have 50-100 computers.²¹ When a computer’s microprocessors are no longer useful, they cannot be recycled; they become electronic waste.²¹

Solar panels also depend on pure silicon. At the end of their lifecycle, solar panels are also hazardous waste. (In another letter, I will outline other ecological impacts of manufacturing, operating and disposing of solar PV systems.)

I’d certainly welcome solutions to silicon’s ecological impacts. Given the magnitude of the issues, I’d mistrust quick fixes. Our first step, I figure, is to ask questions. What’s it like to live near a silicon smelter? How many silicon smelters operate on our planet, and where are they? If we recognize that silicon production generates greenhouse gases and toxic emissions, can we rightly call any product that uses it “renewable,” “zero-emitting,” “green” or “carbon-neutral?”

Where do petroleum coke, other pure carbons and the wood used to smelt quartz and produce silicon come from? How/could we limit production of silicon?

How does our species’ population affect silicon’s production and consumption? I’ve just learned that if we reduced fertility rates to an average of one child per woman (voluntarily, not through coercion of any kind), the human population would start to approach two billion within four generations.²³ (At this point, we’re nearing eight billion people.) To reduce our digital footprint, should we have less children? Would we have less children?

What would our world look like if farmers grew more wheat and rice than manufacturers make transistors? Instead of a laptop, could we issue every student a raised bed with nutrient-dense soil, insulating covers and a manual for growing vegetables?

What questions do you have about silicon?

REFERENCES

1. Without industrial process designer Tom Troszak’s 2019 photo-essay, which explains how silicon is manufactured for solar panels (and electronic-grade silicon), I could not have written this letter. Troszak, Thomas A., “Why Do We Burn Coal and Trees for Solar Panels?” https://www.researchgate.net/publication/335083312_Why_do_we_burn_coal_and_trees_to_make_solar_panels “Planet of the Humans,” Jeff Gibbs and Michael Moore’s documentary, released on YouTube in 2020, also shows how silicon is manufactured for solar panels. https://planetofthehumans.com/
2. Schwarzburger, Heiko, “The trouble with silicon,” https://www.pv-magazine.com/magazine-archive/the-trouble-with-silicon_10001055/ September 15, 2010.
3. Stockman, Lorne, “Petroleum Coke: The Coal Hiding in the Tar Sands,” Oil Change International, January,2013; www.priceofoil.org
4. Silicon processing: from quartz to crystalline silicon solar cells; https://www.researchgate.net/publication/265000429_Silicon_processing_from_quartz_to_crystalline_silicon_solar_cells; Daqo new Energy: The Lowest-Cost Producers Will Survive (NYSE:DQ), 2017, https://seekingalpha.com/article/4104631-daqo-new-energy-lowest-cost-producers-will-survive.
5. “Greenhouse gas emissions from global shipping, 2013-2015; https://theicct.org/sites/default/files/publications/Global-shipping-GHG-emissions-2013-2015_ICCT-Report_17102017_vF.pdf
6. Chalamala, B., “Manufacturing of Silicon Materials for Microelectronics and PV (No. SAND2018-1390PE), Sandia National Lab, NM, 2018. https://www.osti.gov/servlets/purl/1497235; Polysilicon Production: Siemens Process (Sept. 2020), https://www.bernreuter.com/polysilicon/production-processes/; Kato, Kazuhiko, et. al., “Energy Pay-back Time and Life-cycle CO2 Emission of Residential PV Power System with Silicon PV Module,” Progress in Photovoltaics: Research and Applications, 6(2), 105-115, John Wiley & Sons, 1998; https://onlinelibrary.wiley.com/doi/abs/10.1002/(SICI)1099-159X(199803/04)6:2%3C105::AID-PIP212%3D3.0.CO;2-C
7. Schwarzburger, 2010; Troszak, “The effect of embodied energy on the energy payback time (EPBT) for solar PV;” https://www.researchgate.net/publication/335612277_The_effect_of_embodied_energy_on_the_energy_payback_time_EPBT_for_solar_PV/figures
8. Kramer, Becky, “Northeast Washington silicon smelter plans raise concerns,” The Spokesman-Review, 11.1.17.
9. Thorsil Metallurgical Grade Silicon Plan; Helguvik, Reykjanes municipality (Reykjanesbaer), Reykjanes peninsula, Iceland, Environmental Impact Assessment, February, 2015.
10. New York State Dept. of Environmental Conservation – Facility DEC ID: 9291100078 PERMIT Issued to: Global Metallurgical Inc.; http://www.dec.ny.gov/dardata/boss/afs/permits/929110007800009_r3.pdf
11. “Polysilicon Market Analysis: Why China is beginning to dominate the polysilicon market,” 2020, https://www.bernreuter.com/polysilicon/market analysis/; also, Bruns, Adam, 2009.
12. Bruns, Adam, “Wacker Completes Dynamic Trio of Billion-Dollar Projects in Tennessee: ‘Project Bond’ cements the state’s clean energy leadership,” 2009, www.siteselection.com.
13. Schwartzburger, 2010.
14. Dale, M. and S.M. Benson, “Energy balance of the global photovoltaic (PV) industry-is the PV industry a net electricity producer?” Environmental Science and Technology, 47(7), 3482-3489, 2013.
15. The Society of Chemical Engineers of Japan (ed.), “Production of silicon wafers and environmental problems,” Introduction to VLSI Process Engineering, Chapman & Hall, 1993.
16. https://www.youtube.com/watch?v=Dq04GpzRZ0g
17. Hayes, Brian, “The Memristor,” American Scientist, 2011.
18. https://marginalrevolution.com/marginalrevolution/2019/01/claims-about-transistors.html
19. www.OurWeb.tech/letter-3/
20. Needhidasan, S., M. Samuel and R. Chidambaram, “Electronic waste: an emerging threat to the environment of urban India,” J. of env. health science and engineering, 2014, 12(1), 36. https://doi.org/10.1186/2052=336X-12-36; https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3908467/
21. www.OurWeb.tech/letter-5/
22. Needhidasan, S., 2014.
23. Hickey, Colin, et al. “Population Engineering and the Fight against Climate Change.” Social Theory and Practice, vol. 42, no. 4, 2016, pp. 845–870., www.jstor.org/stable/24870306.

 

For more comprehensive information about the hidden costs of manufacturing silicon for transistors and solar panels, see Tom Troszak’s paper: https://www.enseccoe.org/data/public/uploads/2021/11/d1_energy-highlights-no.16.pdf

In response to comments and questions posted on Facebook about this letter, industrial process designer, researcher and writer Tom Troszak prepared the following notes:

Dear Katie:

Your readers’ assertions are incorrect, and I reckon their math is off by at least a factor of 45 times:15.83 kWh / IC chip (32MB) / 0.35 (reader estimate)= 45.22 times. In reality, 15.83 kWh / IC chip is a very low boundary estimate (see reference below). In the real world, the actual value is probably at least double that figure (see reference). So your reader’s estimate is probably low by closer to 90X. See my comments below.

T.K.: Dear Katie Singer,
I am sorry, but I find this article rather poor.
For starters, silicon is found “in nature.” It is the second most abundant element in the crust of the >earth, after oxygen. It is literally what sand is made of

Tom Troszak: Dear T.K., you’re incorrect. Sand is made of silicon dioxide (SiO2). Separating the silicon(S) and oxygen(O) (reduction process) requires a carbon(C) reductant, wood chips (bulking agent) and about 20 MWh of fossil electricity per tonne(t) of mgSI that is reduced in the smelter. The process produces a final result of ~97% Si (and CO). The CO converts to CO2 in the atmosphere. After crushing of the final Si product (10% loss) the embodied energy (EE) rises to ~24.44 MWh / t for mgSi.
https://www.researchgate.net/publication/335083312_Why_do_we_burn_coal_and_trees_to_make_solar_panels.

T.K.: But perhaps I am being pedantic here, as I know you are talking about the highly pure silicon that >does indeed not occur in nature at this level of purity.

Troszak: It is not a question of purity. Silicon (metalloid) does not exist in nature at all, at any level of purity. It is only found in combination with other elements like oxygen. (See above).

T.K.: What I really have a problem with is the lack of any context for your numbers.
So producing a metric tonne of silicon takes 175 MWh (“MW/hours” doesn’t make any sense).

Troszak: This is Incorrect. 175 MWh is the base value for the electricity needed for producing 1 ton of polysilicon — in the Siemens process — that converts about 2 t of mgSi into 1 t of polySi. “MW/hours” is a unit of energy equal to the work done by a power of a million watts in one hour. Or, simply – power(MW) times the length of time in hours(h) — MW x h = MWh. So if a polysilicon plant draws 400MW, and produces 2.28 tonnes of polySi per hour — then 400 MW / 2.28 (t / h)
= 175.43 MWh / tonne. And that 400 MW plant has a nameplate capacity of 20,000t/year. so — 2.28 t/hour * 8760 hours / year
= 19,972.8 t polySi / year – yep looks close!
— If it takes 2 t of mgSi to make 1 t of polySi, then add 40 MWh / t for the EE of the incoming raw mgSi material.
400MW / 2.28 (t / h) + 40 MWh / t (20 MWh/t x 2 t) = 215.43 MWh/t (polySi from reactor)
— And if 10% of the polySi output from the reactor is lost to contamination and fines, then the total EE of polysilicon becomes 215.43 / 0.9 = 239.37 MWh/t (polySi ready to ship).
— And after that polysilicon has been processed into ingots, and sliced into wafers (not IC chips or solar cells, just raw wafers), the total EE becomes 2,130 kWh / kg single crystal (sc) silicon / 0.095 kg (yield ratio – kg (mgSi) : kg (scSi) wafer) = 22,421 kWh / kg — silicon wafer (before processing into IC chip or solar cell)

T.K.: But how does that compare to anything? How much do we get out of a metric tonne of Si?

Tom Troszak: If you are referring to solar cells, the answer depends on the efficiency of the resulting cells, modules, arrays, DC storage losses (if batteries are included), and inverter and transformer losses when converting to AC. If you are referring to IC chips, see below.

T.K.: According to a (very) cursory search, a typical microchip has 2g of Si in it. So we get 500,000 chips >out of a t. That would put a chip at 0.35 kWh

Tom Troszak: This is incorrect. There are many many losses and conversions along the way. That estimate is low by a factor of 45 – 90 times. (see below)

T.K.: Even if it’s coal fired electricity (worst-case scenario), that’s only around 350g of CO2, or maybe 2 >miles of driving a car (I’m a bit sloppy with my numbers here, it’s about the order of magnitude).

Tom Troszak: Or by my estimate, roughly 90X sloppy.

T.K.: Similarly, “solar panels use silicon”. Yes. But how much?

Tom Troszak: At 15 % cell efficiency, 1 MW of cells requires 6,666 m2 of cell area. At current best practice, this requires about 9.6 t of incoming polysilicon per MWp (DC) of cells produced (made from 19.2 t of mgSi), of which about 3 tons ends up in one megawatt (1 MW) or 6,666 m2 of cells.

Or roughly of 52 kWp (DC)peak (347 m2 of cell area) per tonne of the original mgSi from the silicon smelter. When you factor in performance ratio (0.68), conversion to AC (0.85), that’s an effective output of 30 kWp per ton of mgSi silicon. If you add Li batteries (RTE=0.55), you end up with an effective output of about 16.5 kWp (AC + Li STORAGE) per tonne of silicon smelted. Ouch.

T.K.: And how many kWh of electricity are produced from 1g of silicon in a solar panel?

Troszak:So each gram of silicon (1g) in a solar panel is only a teeny fraction of the raw silicon (19.2g) that must be smelted from ore, (see above) so that figure cannot be used to estimate the total environmental impact of PV production. That is a flaw common to all LCA.

— Assuming 1,620 hours per year of sunlight (US, average), and a 25 year module lifetime, the lifetime energy production of 1 MW (or any unit) of solar PV is 1620 hours * 25 years * 0.68 PR * 0.85

= 23,409 MWh maximum possible lifetime AC output. If you add Li batteries (0.55 RTE) you’ll only get 12,874.95 MWh total.

T.K.: And how much coal is avoided due to that 1g of silicon?

Troszak: Sadly, none. The silicon processing alone (see above) requires a minimum of 25,495 MWh / MW(DC) peak. So when converted to AC, in their entire 25-year lifetime, the PV modules produce a bit less than the fossil energy needed to produce and process the silicon alone. If you include the EE of the glass, aluminum, and other components, that just increases the energy deficit. If you add any amount of Li batteries to the PV system, the embodied energy (EE) needed to make the batteries alone – that just pushes the lifetime system energy balance that much deeper into negative return. If you actually *use* the batteries, their actual round-trip return is only about 50%. If we add backup generators to the PV/grid system, then all that EE and standby fuel just pushes the deficit deeper. If we include the amount of energy need to build and maintain PV factories, then PV can never produce a surplus of energy. So, by the time the PV modules are integrated into the grid, the amount of fossil energy expended on them is already several times what you’ll ever get back.

“Since CIGS and sc-Si both run an energy deficit even before the inclusion of storage, they cannot support any level of storage(…)Some PV technologies (CIGS and sc-Si) are barely in the electricity surplus region, so the requirement of any amount of storage pushes these technologies into electricity deficit(…)CIGS and sc-Si cannot support any amount of storage, since they are already operating at a deficit(…)sc-Si and CIGS, both of which are already operating at an energy deficit…”

Can we afford storage? A dynamic net energy analysis of renewable electricity generation supported by energy storage
https://pubs.rsc.org/en/content/articlelanding/2014/ee/c3ee42125b#!divAbstract

Dale, M., & Benson, S. M. (2013). Energy balance of the global photovoltaic (PV)
industry-is the PV industry a net electricity producer?. Environmental science & technology, 47(7), 3482-3489.
https://gcep.stanford.edu/pdfs/EST_energybalancePVIndustry.pdf

T.K.: These are the kinds of numbers that would make this a really interesting article!

Troszak: I agree! Thanks for asking! Here is an interesting reference on the embodied energy of computer chips. Again, the most important thing to remember is that only a tiny smidgen of the original silicon from the smelter ends up in a single IC chip, but ALL of the energy that goes into the next 200 processing steps that are needed to convert that wafer into little transistorized brains still counts. This is really a worthwhile read: “The result of the calculation is a mass input of 1200 g of fossil fuels to produce a 2-g ram DRAM chip, and 440 g during the use phase…The production chain yielding silicon wafers from quartz uses 160 times the energy required for typical silicon, indicating that purification to semiconductor grade materials is energy intensive.The lower bound of fossil fuel and chemical inputs to produce and use one 2-gram microchip are estimated at 1600 g and 72 g, respectively. Secondary materials used in production total 630 times the mass of the final product, indicating that the environmental weight of semiconductors far exceeds their small size. This intensity of use is orders of magnitude larger than that for “traditional” goods. Taking an automobile as an example, estimate of life cycle production energy for one passenger car range from 63 to 119 GJ (42). This corresponds to 1500-3000 kg of fossil fuel used, thus the ratio of embodied fossil fuels in production to the weight of the final product is around two.”

2,130 kWh / kg silicon / 0.095 kg (yield ratio – kg mgSi : kg wafer)
= 22,421 kWh / kg silicon wafer (before processing into IC chip)
57 (MJ / IC chip) * 0.27 (MJ / kWh)
= 15.83 kWh / IC chip (32MB)

https://pages.jh.edu/aandreo1/495/Archives/Bibliography/Processing/EnergyCosts/EnergyAndMaterialsUseInMicrochips_EST.pdf

________________

Katie Singer writes about the energy, extractions, toxic waste and greenhouse gases involved in manufacturing computers, telecom infrastructure, electric vehicles and other electronic technologies. She believes that if she’s not aware that she’s part of the problem, then she can’t be part of the solution. She dreams that every smartphone user learns about the supply chain of one substance (of 1000+) in a smartphone. Her most recent book is An Electronic Silent Spring. She currently writes about nature, democracy and technology for Meer.com. Visit www.OurWeb.tech and www.ElectronicSilentSpring.com.

This article was originally published by Meer.com.