# Recycling Techniques for Silicon Kerf Loss Topic radar · 68 documents · 2021-01-01–2026-11-10 · updated 2026-10-12 Canonical: https://slate.greyb.com/radar/recycling-techniques-for-silicon-kerf-loss/innovations ## In one minute ### The waste is a pure silicon core wrapped in an oxide shell Slicing an ingot into wafers throws away about a third of the silicon. CEA puts the loss at 35% of ingot mass, and it leaves the saw as powder a couple of microns across, with a median particle size near 1.6 µm and 30.4 m² of surface per gram [1] [2]. That surface is the whole technical problem. The particles have a core-shell structure: high-purity silicon inside, a silicon-oxide skin outside, formed during sawing [3]. Every process in this field either strips that skin or accepts it, and that split runs through the chemistry below. 1. [agris.fao.org](https://agris.fao.org/search/en/records/6970eb64bd789e57df2ea76f#1) 2. [icarus.eu.com](https://www.icarus.eu.com/tag/cea/) 3. [sciencedirect.com](https://www.sciencedirect.com/science/article/abs/pii/S092583882500091X#1) ### The only committed capital in the field is a tariff bet in Tennessee Every other project here is a pilot or a project report. Highland Materials is the single announced build with a balance sheet behind it, and its numbers are large enough to reset what commercial scale means in this field: 16,000 tonnes a year of solar-grade polysilicon initially, ramping to 24,000 tonnes, on a brownfield site at Phipps Bend, Tennessee. ### The published work comes from one European consortium, and the wave is its closing Almost all of the research literature this field has produced in the last two years traces to ICARUS, a project of eighteen European partners running four pilot lines, which ended in November 2025. The output arrived in a burst rather than a trickle: the carbothermic kerf-and-pot-scrap paper at SiliconPV in December 2025, SINTEF's own record of the same campaigns in March 2026, and the consortium's summary paper in April 2026. ### Four pilot modules were built, and their capacities describe different plants A European project called ICARUS ran four processing lines and reports each one's rating. Read together, they do not form a plant. ### The price it has to beat has rebounded, and it is still short The commodity the recycled product displaces has moved back up. Chinese N-type dense polysilicon was quoted at RMB 41.44 per kilogram in November 2026, roughly 5.8 dollars, after trading near RMB 32.3 per kilogram, about 4.75 dollars, in July. Inventory above 570,000 tonnes and November output near 95,000 tonnes keep the recovery fragile. ### The organisations that proved the process hold almost none of the patents The patent record in this field is Chinese and Korean, plus one early family from Total and SunPower. None of SINTEF, ReSiTec, LuxChemtech or Norsun holds a family. ROSI is the exception among the European pilot partners, with families filed in six to eight jurisdictions, including a dilute hydrogen peroxide wash for kerf slurry and a granulation route filed in France and as a WO. ### Leaching clears the surface cheaply, and the melt is where silicon is lost The standard sequence is an acid wash followed by a high-temperature step. In 1 M hydrochloric acid, 91% of the iron, 77% of the nickel and 75% of the phosphorus come out at ambient temperature. Adding an inductive melt raises removal to 99% iron, 92% phosphorus, 85% calcium, 79% nickel and 78% aluminium. ### What has actually been produced is drying, powder and metallurgical silicon The outputs with tonnage behind them are narrow. A continuous dryer for kerf filter cake reached below 1% moisture at 876 to 2,600 tonnes a year, and the submerged arc campaigns produced silicon close to the traded grade metal silicon #553, quoted at 1,585 dollars per tonne, roughly 1.59 dollars per kilogram. The highest-purity output is ROSI's granule line at 4N, running at 1 kg/h. ## Overview Recovered kerf now reaches 4N granules at a kilogram an hour, and refined material at 6N to 8N has been blended into an industrial Czochralski pull at 8% of the charge. The technical question is largely answered. The commercial one has moved against the field: the programme set its target at a maximum cumulative recycling cost of 10 dollars per kilogram, while Chinese mono-grade polysilicon traded near 5.8 dollars per kilogram in November 2026. A recycler selling at half the market price sells below the 4.39 euros per kilogram unit cost the most detailed business case modelled. Everything else here follows from that arithmetic. ## Research ### Carbon removal now takes seconds, and boron is the remaining wall Burning carbon off the powder traditionally oxidises the silicon with it. Two chemical routes avoid that. A piranha treatment of sulphuric acid and hydrogen peroxide removes surface carbon and improves crystallinity, but the reagent ratio sets the silicon bill: at equal parts, recovery falls to about 80%. An aerosol reactor goes further, taking carbon below detection in air at 900 °C with a six-second residence time and little silicon oxidation. ### The pilots mostly produce below solar grade, and that shapes what gets built The highest-purity stream is ROSI's granule line: 4N silicon at 1 kg/h, with carbon at 110 ppm and oxygen at 15 ppm. The smelting route lands lower still, producing metal close to the traded grade called silicon #553. ### One route has put recycled silicon back into a crystal, at a blend rate Refined kerf has been fed into an industrial Czochralski process at 6N to 8N purity, modelled as replacing 8% of the solar-grade silicon in a wafer line. A semi-industrial pilot built the rest of that chain, taking kerf and quartz scrap through submerged arc smelting, plasma treatment and directional solidification. ## Market ### No purchase has been announced No offtake agreement, qualification announcement or published transaction price for recycled kerf appears in the public record. Market analyses name the binding constraints as customer qualification, compliance capability, material traceability and confirmation of the final disposal route, ahead of scale. ### The vendors with a market position earn most from everything except wafer silicon The commercial energy in this field has shifted off the silicon. ROSI redirected its operating business to end-of-life panel recycling in 2023, citing low silicon prices and limited domestic production. LuxChemtech's chemical line converts residues to silica, water glass and green hydrogen and recovers roughly €17 of silver per kilogram of cell scrap. ### The published market sizing covers a bigger subject than kerf Vendor forecasts put semiconductor silicon waste recycling near 2.0 to 2.1 billion dollars in 2025, rising to roughly 3.1 to 3.2 billion by 2032 at about 6.2% to 6.5% a year. Both definitions sweep in dummy and monitor wafers, broken wafers, sludge and kerf slurry together. ### What would turn this from capability into a market Three things would do it, and none is a chemistry result. ## Strategy ### Chinese patenting concentrates on the cheap end of the process Kunming University of Science and Technology is the most persistent filer in the wider record. It has a two-stage acid leach using a mixed acid with a cationic surfactant, a method that purifies slurry online before it is ever pressed into cake, and an earlier route to 4N to 5N silicon paste and pellets. Around it sits a ring of materials and equipment firms filing on separation, washing and drying: Jiangsu E-Ontech, Anhui Huasun and Dongxu Technology. ### The newest filings point away from the wafer The most recent patent in this record makes silicon carbide powder from wire-cut sludge, claiming over 99.9% removal of metallic impurities and a 30% raw material cost saving against the conventional Acheson process. Its closest earlier document is the Czochralski re-entry patent from Total and SunPower, which is the field's founding ambition and its least used route. ### The silicon producers are filing on their own waste now GCL New Energy's route presses sludge into a cake, granulates it, dries it and heats it to 1450 to 1600 °C to make recycled silicon blocks. State Power Investment Group's Qinghai Huanghe arm has filed on recovering silicon from retired modules. These are the companies that would otherwise buy recovered powder. ### Inside Europe, two tracks are running at different speeds Norway's academic line is the most continuous single thread in the field. NTNU and SINTEF have published a connected run from acid leaching and inductive melting, through vacuum melting of kerf agglomerates, to the submerged arc campaigns with quartz pot scrap. One group, one place, from beaker to furnace. ### New capacity is announced where the saws are, or behind a tariff Highland Materials is building in the United States and describes the project as the first large-scale facility turning kerf into polysilicon, with its cost and emissions figures offered as company claims rather than audited results. India's ingot and wafer build-out targets at least 80 GW by June 2028 against roughly 2 GW operating in September 2026, and the guidance reaching plant designers there is to confirm kerf handling obligations with state pollution boards before committing to a site. ### The next layer of work is specification, and almost nobody is doing it A 2026 review argues that recycled-content use in photovoltaics depends on clearer specifications, verification and reporting, including explicit distinctions between pre-consumer and post-consumer feedstock and defined impurity and blending limits. ## Areas - Wafer Manufacturing Waste · 33 documents - Market Project Landscape · 19 documents - Purification Processing · 8 documents - Feedstock Sourcing · 5 documents - End Use Applications · 3 documents