Upcycling Polysilicon into Ferrosilicon Powders

The photovoltaic industry spends an enormous amount of energy turning quartz into silicon that is pure to six, nine, sometimes eleven nines. Then, at four separate points in its own value chain, it throws some of that silicon away — as pot scrap, as head and tail crops, as broken and off-grade wafers, as sawing kerf.

The expensive half of making ferrosilicon is not the melting. It is the reduction — stripping oxygen off silica in a submerged arc furnace, at thousands of kilowatt-hours a tonne — and the silicon that comes out of it carries a tail of aluminium, calcium and titanium inherited from the ore and the reductant. The solar industry's discards have had that step done to them already, to a far higher standard, and charged to somebody else's cost sheet.

What arrives, though, is not furnace feed. It is thin, fine, light and coated — forms a melt shop was not designed for, carrying contaminants a quartz-fed furnace never sees. This article sets out the streams, what each is worth in a high-specification powder, and where each one bites.

1. Where the scrap comes from

“Solar silicon scrap” is not one material. It is at least six, and they differ enormously in how much work they need before they are fit to charge.

The silicon scrap streams, and what each one brings with it

Scroll the table sideways to see every column

Two distinctions matter more than all the others.

Pre-metallisation versus post-metallisation. A broken wafer that never reached the cell line is almost pure silicon with an oxide skin. A broken cell carries a silicon nitride anti-reflective coating, silver finger and busbar paste, and — critically — an aluminium back-surface field. Several grammes of aluminium per cell, charged into a melt sold on a low-aluminium specification, is not a trace contaminant. It is the whole specification. These two materials look similar on a weighbridge and behave completely differently in a furnace.

Bulk versus surface. Chunk and crop material is contaminated at the surface, which is a small fraction of its mass. Kerf is almost all surface. The same nominal purity means something quite different in the two cases, because what dominates the melt result is oxide and adhering metal on the outside, not the lattice on the inside.

2. Why it beats quartz as a feedstock

Conventional ferrosilicon is made by carbothermic reduction: quartz and a carbon reductant in a submerged arc furnace, with iron units added to set the grade. It is a reduction reaction, and it is expensive in exactly the way reduction reactions are.

• The electrical energy for 75% ferrosilicon is commonly quoted in the range of 8,000–9,000 kWh per tonne, varying with furnace design, raw materials and operating practice.
• The reduction releases carbon dioxide stoichiometrically, from the reductant itself, before any account is taken of the electricity.
• The feedstock sets the impurity floor. Aluminium, calcium and titanium enter with the quartz and the reductant ash, and no amount of downstream care removes them entirely. A low-aluminium ferrosilicon starts with better quartz and better coke; ladle refining takes some of the aluminium and calcium out afterwards, but it is working on whatever the furnace already gave it.

Charging already-reduced silicon changes the nature of the operation. Melting silicon and alloying it with iron is a mixing problem, not a reduction problem.

• The thermodynamic minimum to take silicon from room temperature to a melt is under 1 kWh per kilogramme — roughly 0.9 kWh/kg, combining sensible heat to 1414 °C with the latent heat of fusion. Real furnace efficiency multiplies that figure, but it stays an order of magnitude below the reduction route.
• There is no reductant, so there is no process CO₂ from reduction. What remains is the carbon intensity of the electricity and of the iron unit.
• The impurity floor moves. With a six-nines silicon unit, the aluminium, calcium and titanium in the finished alloy are no longer inherited from an ore body. They come from the iron you add, the refractory, the slag practice and the atomizing water — all of which are inside the plant's control, and all of which can be engineered.

That last point is the commercially interesting one. It is the difference between buying a specification and controlling one.

3. What a purer silicon unit actually buys the customer

A cleaner feedstock is only worth paying for where the impurity it removes was doing damage. Three places where it does:

Atomized ferrosilicon 45% for welding consumables

The 45% grade is the workhorse of electrode coating and flux-cored wire, and it is worth being explicit about why it is 45% and not 75%. Silicon is wanted in the formulation as a deoxidiser and as a controlled alloy addition; the iron it is diluted with is not waste, it is part of the deposit. A 45% alloy lets a formulator dose silicon in sensible weight fractions of the coating, rather than micro-dosing a 75% powder where a small weighing error moves the deposit chemistry.

Silicon earns its place in the deposit four ways. It strips oxygen out of the weld pool and sends it to the slag as silica. It raises the fluidity and wetting of the pool, which is what gives a well-behaved bead profile and good toe blending. It changes slag viscosity and surface tension, and so affects detachability. And what remains in solution contributes strength — up to the point where too much of it starts to take toughness away, which is why the specification band matters as much as the nominal figure.

Why the atomized form, specifically. Against a crushed or milled powder of the same nominal mesh, atomized particles are rounded rather than angular, and that changes four things a consumable plant can measure:

• Flow and packing. Rounded particles flow through hoppers and feeders and pack to a consistent apparent density. In an extrusion press that means consistent coating weight on the core wire, which means consistent arc behaviour and consistent deposit chemistry down the batch.
• Binder demand. A lower, more consistent surface area takes up less silicate binder for the same consistency, which leaves more of the coating available for the functional ingredients.
• Abrasion. Angular ferrosilicon is hard and it wears extrusion dies and tooling. Rounded particles are gentler on both.
• Oxidation and moisture. Lower specific surface area means less oxide pick-up in storage and less moisture held on the particle. In a low-hydrogen basic electrode, every path that moisture can take into the coating ends up as diffusible hydrogen in the deposit, and the feedstock powders are one of those paths.

What the feedstock change adds on top of that. Everything above is true of any atomized 45% powder. What a high-purity silicon unit adds is control over the elements that ride in with it:

• Aluminium is the one that matters most. In submerged arc and flux-cored deposits, uncontrolled aluminium ties up oxygen in a way that changes inclusion population and nucleation behaviour, and it affects slag detachability. Consumable formulators add aluminium deliberately when they want it; they do not want it arriving uninvited in the ferrosilicon.
• Carbon sets the ceiling on what the powder can be used for. Low-carbon and stainless deposits cannot tolerate a feedstock that contributes carbon, and the carbothermic route is, by definition, carbon-rich at the point of production. A route with no reductant in it does not have that problem to solve.
• Phosphorus and sulphur drive hot cracking and degrade toughness. Both enter the quartz route with the ore and the reductant.
• Titanium and boron are potent at very low concentrations. In micro-alloyed deposits they are added in tens of parts per million and controlled tightly; a feedstock that contributes an unknown and variable amount of either makes that control impossible.

There is also a structural point about consistency that is easy to miss. A tapped alloy has the composition the furnace gave it that day. An alloy made by melting weighed quantities of a high-purity silicon unit into selected iron has the composition it was designed to have, and the silicon band can be held tight batch to batch because nobody is relying on a reduction reaction to land in the right place. For a consumable that has been qualified to a standard and cannot be requalified casually, that repeatability is worth more than a decimal place on the assay.

A note on dopants, because it is the first question a metallurgist asks about wafer scrap: p-type wafers are boron-doped and n-type wafers phosphorus-doped, but at doping levels of the order of 10¹⁶ atoms per cubic centimetre, the boron content works out at well under 0.1 ppm by weight. It is metallurgically irrelevant in a ferrosilicon melt. It is still worth measuring on incoming material for boron-sensitive grades — not because the doping is a problem, but because measuring it is how you find out if something else came in with the load.

Ferro silicon 15% for dense media separation

DMS ferrosilicon is a different kind of product from everything else in the catalogue, and the difference changes what “high specification” means.

In a dense medium circuit, finely divided ferrosilicon is suspended in water to make a fluid of a chosen density, and ore is separated by whether it floats or sinks in it. Magnetite covers the low densities; ferrosilicon is what gets you into the range that diamond, iron ore and base metal plants need. The medium is then recovered magnetically, cleaned and returned to the head of the circuit, and the same powder goes round again — hundreds of times.

That circulation is the whole commercial story. A DMS plant does not consume ferrosilicon the way an electrode plant does; it loses it. Losses come from three places: medium that leaves adhering to the product and the reject, medium the magnetic separators fail to recover, and medium that corrodes away into solution and fines. Purchase price sets a small part of the cost; loss rate sets most of it. A powder that costs slightly more and is lost more slowly is cheaper.

Three properties decide that loss rate, and they pull against each other:

• Density and silicon content. Around 15% silicon is where the alloy is dense enough to make the medium, corrosion-resistant enough to survive recirculating in process water, and still strongly ferromagnetic. Push the silicon higher and magnetic recovery falls off; drop it lower and corrosion resistance goes. The window is narrow, which is exactly why the silicon band on the certificate needs to be a guaranteed limit and not a typical value.
• Particle shape. Atomized and milled grades behave differently and are not interchangeable. Rounded atomized particles give a lower-viscosity suspension at the same density — which means a sharper cut, less pumping power and less wear on pumps and cyclones. Milled, angular material settles more slowly and holds the medium more stable, at the cost of viscosity and abrasion. Measured side by side, atomized medium also corrodes more slowly and is lost less to adhesion on the product. Which one a circuit wants depends on its cut density and its geometry, and switching between them is a plant decision, not a substitution.
• Size distribution. Fineness sets the balance between medium stability and viscosity, and it also sets how much of the medium is too fine to recover magnetically. Fines are not a side issue in a DMS grade; they are a direct loss mechanism.

Against that, what a cleaner and more consistent silicon unit buys is corrosion behaviour and lot-to-lot sameness. Corrosion in a recirculating medium does not announce itself on an incoming certificate — it shows up months later as a consumption figure that has drifted, and by then the lot is long gone. Consistency is the thing the circuit is actually paying for, and a feedstock whose residual chemistry does not wander from campaign to campaign is the only way to deliver it.

One operational note worth making, since it cuts the other way from the welding grade: a 15% alloy needs only a fraction of the silicon that a 75% grade does. The same tonne of recovered solar silicon goes several times further as DMS medium — but by the same arithmetic, the iron is the overwhelming majority of the product, so iron selection, not silicon purity, is where most of the residual control has to happen.

Powder metallurgy and specialty blends

Here the controls are oxygen content and lot-to-lot repeatability. Neither is improved by feedstock purity alone — oxygen is largely decided by the atomizing and drying practice — but a feedstock whose chemistry does not wander removes one of the two major sources of variation between campaigns.

4. The four things that make this feedstock difficult

Anyone who has tried this at scale will recognise all four. They are the reason the route is not already universal.

Surface area, oxidation and fire

Silicon fines have an enormous surface-to-volume ratio and a thermodynamic appetite for oxygen. Kerf in particular can heat and burn vigorously when charged carelessly, and material that has been through an etching step is more reactive still. The fine fractions are a handling and storage problem before they are a metallurgical one: moisture, segregation, dust control, inerting, and a charging practice that does not expose a large mass of fines to a hot free surface all at once.

This is also where the yield goes. Silicon that oxidises on the way into the melt reports to the slag as silica — the exact compound the whole process exists to avoid making.

Coatings, pastes and organics

Post-cell material brings silicon nitride, silver, aluminium paste, encapsulant and, in module scrap, fluoropolymer backsheet. Aluminium breaks a low-aluminium specification outright. Organics burn off, but they burn off somewhere, and that somewhere needs to be a controlled combustion step with the right off-gas handling, not the furnace freeboard.

The honest answer for most plants is to exclude metallised material from the high-specification stream entirely, and handle it, if at all, on a separate route with its own grade targets. A single mixed load can contaminate a campaign.

Tramp metals from the saw

Diamond wire is a steel core, usually brass-coated, carrying diamond held in an electroplated nickel layer or a resin bond. All of it wears, and what wears off reports into the kerf as iron, copper, nickel and zinc. Iron is harmless in ferrosilicon — it is most of the product. Copper and nickel are not harmless in every end use, and they are not removable by slag practice. Kerf therefore needs incoming analysis for exactly those residuals, not just for silicon assay.

Bulk density and charging

Chips, shards and thin wafers have terrible bulk density, poor flow and a tendency to float and to bridge. Fines cannot simply be shovelled into a melt. Some combination of size reduction, compaction or briquetting, and a submerged or controlled addition practice is necessary — and the compaction step has its own safety considerations for reactive fine silicon, including the order in which drying, binding and pressing are done.

5. The process route in practice

The route that works is mostly incoming control and segregation discipline. The melting is the easy part.

1. Segregate at source. Agree the stream definition with the supplier and keep pre-metallisation and post-metallisation material physically separate from the moment they leave their plant. Mixed loads are not economically recoverable into a high-specification grade.
2. Incoming inspection on every load. Visual and magnetic check, moisture, loss on ignition as a proxy for surface oxide and organics, and a composite chemical analysis by ICP-OES or glow discharge mass spectrometry for the residuals that matter: Al, Ca, Ti, Cu, Ni, Zn, B, P, C.
3. Cleaning and drying. Degreasing and removal of coolant residues where present; controlled drying to a defined moisture limit before any thermal step.
4. Size reduction and compaction. To a chargeable form with a defined bulk density, with the handling sequence set so that reactive fines are not exposed at the wrong stage.
5. Melting and alloying. Induction or arc melting with iron units selected on residual content. Once the silicon unit is six-nines material, the iron is the dirtiest thing in the charge — so low-residual iron is specified as carefully as the silicon ever was.
6. Hold, deslag, verify. Chemistry confirmed on a tapped sample before atomizing, not after.
7. Water atomization, then dewatering, drying, de-dusting and classification into the required mesh fractions.
8. Full product testing. Chemistry, sieve analysis against a named series, apparent density, moisture — and for DMS grades, density and magnetic recovery.

The transport classification does not change because the feedstock did: ferrosilicon with 30% or more but less than 90% silicon remains UN 1408, Class 4.3.

6. The bottom line

The solar industry has, as a by-product of doing something else entirely, created a stream of silicon that is purer than any ferroalloy furnace can produce and already reduced. Charging it displaces a reduction step, its reductant and most of its energy, and it moves the impurity floor of the finished alloy out of the ore body and into the plant's own control — which is where a high-specification powder has to be controlled from.

None of that is free. It is bought with segregation discipline at source, incoming analysis on every load, a handling practice built around reactive fines, and the willingness to turn away a cheap mixed load that would contaminate a campaign. The metallurgy is straightforward. The material control is the product.

AIM ALLOYS LLP has manufactured ferro alloy powders and metal powders at Khushkhera, Rajasthan since 1983, and was India's first BIS-approved manufacturer of atomized ferrosilicon powder. If you are specifying a low-aluminium atomized ferrosilicon, a 15% grade for dense media separation, or a silicon metal powder, and you want to discuss feedstock origin and residual control as part of that specification, our technical team will take the question.

And if the flow runs the other way — if you are a polysilicon, ingot, wafer or cell producer sitting on pot scrap, crops, broken wafers or sawing kerf and wondering what it is worth — that is a conversation we are glad to have. Tell us which stream it is, roughly how much of it arises, and whether any of it has been through metallisation, and we will tell you straight what can be upcycled into a high-specification powder and what cannot. Get in touch.