Short versionAbout 75 tons of ore go in for every ton of finished magnet that comes out. The losses are not evenly spread, and where they fall is most of the economics.
Before the stepsThe yield funnel
Mass is lost at nearly every step. For a bastnaesite deposit, ore runs roughly 7% mixed rare earths, and only about a tenth of that is neodymium. Downstream, every process from casting to machining gives up material. The bars alongside show, in round numbers, how much of each intermediate is needed to yield one metric ton of finished magnet.1
Two things are worth noticing. The first is that nearly all of the mass loss happens before the oxide exists, which is why a magnet plant is a fraction of the size of the mine and separation plant that feed it. The second is that the losses after the oxide are the ones a magnet maker controls, and the machining loss in particular is the difference between a plant that makes money and one that does not.
Short versionDig up rock that is roughly 7% rare earths. Ionic clays skip the rock entirely, which is most of their appeal.
01 · Ore to oxideMining
Ore is extracted from the deposit.
Rare earths are mined like most other metals. Hard-rock deposits such as bastnaesite and monazite are drilled, blasted, and hauled out of an open pit. The ore is not remarkable to look at; the rare earth minerals are a small fraction of the rock, and the rest is waste that will be removed in the next step.
Ionic clay deposits are the exception. The rare earths are held loosely on the surface of fine clay particles, so there is no rock to crush. The clay is already in a form ready for extraction, and its element mix tends to be more favorable for magnet making, with a larger share of the heavy rare earths that magnets need.
Short versionCrush, grind, float. Same minerals, ten times more concentrated, and most of the mined mass left behind as tailings.
02 · Ore to oxideBeneficiation
Ore is crushed, ground, and concentrated.
The ore is crushed and milled to a fine powder to liberate the rare earth minerals from the surrounding rock. The powder is then run through flotation: chemicals are added that make the rare earth minerals attach to air bubbles, so they float to the surface and are skimmed off while the waste rock sinks.
The result is a concentrate. Where the ore ran around 7% rare earths, the concentrate runs around 60%. Nothing has been chemically changed yet; the same minerals are simply present in far greater proportion, and the great majority of the mined mass has been left behind as tailings.
Short versionDissolve everything, then pull the elements apart one small chemical preference at a time. Out the far end comes NdPr oxide.
03 · Ore to oxideSeparation
The concentrate is dissolved and the individual rare earths are split apart.
The concentrate is roasted and leached in acid, which takes the rare earths into solution and leaves the remaining rock behind. What comes out of the leach is a mixed solution of all the rare earths together, and they have to be separated into individual elements before any of them is useful.
The rare earths are chemically very similar to each other, so separation is done by solvent extraction: the solution is passed, stage after stage, against an organic liquid that pulls some elements across slightly more readily than others. Repeated over hundreds of stages, the small preference becomes a clean split. Neodymium and praseodymium come out together as a pair, and are precipitated and calcined into an oxide powder. That oxide is the feedstock for everything that follows.
Short versionStrip the oxygen off the oxide with electricity in a bath of molten salt. This is the single most valuable input to the magnet, and it is where my own work starts.
04 · Metal and alloyMetallization
Neodymium oxide is reduced to neodymium metal by molten salt electrolysis.
An oxide is a metal that has already reacted with oxygen, and it is chemically very stable. To get the metal back, the oxygen has to be stripped off. For neodymium this is done by electrolysis in a bath of molten fluoride salts, running continuously at high temperature.
The oxide is fed onto the surface of the bath, where it dissolves. A direct current is passed between graphite anodes and a tungsten cathode. Neodymium metal forms at the cathode and drips into a crucible below, while the oxygen reacts with the graphite at the anode and leaves as gas. The crucible is periodically tapped and the metal is cast into ingots. A neodymium cell runs around the clock, and the metal it produces is the single most valuable input to the magnet.
Where I come inEverything upstream of this step is chemistry, and it ends with a bag of oxide powder. Everything from here on is metallurgy, and it ends with a magnet. Peculiar Materials works on the second half: the metal, the alloy, and the magnet. I have spent a career on these ten steps and comparatively little time at the mine, and the page is weighted accordingly.
Short versionMelt the alloy and freeze it in a fraction of a second on a copper wheel. We do it this way to get a microstructure that works with the rest of the process, and as few undesirable phases as possible to deal with later.
05 · Metal and alloyStrip casting
Metals are melted into the magnet alloy and cast into thin flake.
The magnet is not made of neodymium alone. Neodymium, iron, boron, and a set of minor additions are melted together under vacuum in an induction furnace to form the magnet alloy. The chemistry set here is the single largest determinant of what the final magnet can be.
The molten alloy is poured through a tundish onto a spinning, water-cooled copper wheel. It freezes in a fraction of a second into a thin flake, well under a millimeter thick, that peels off the wheel. We do it this way to develop a microstructure that works with the rest of the process, and to minimize the amount of undesirable phases we have to deal with later. The big one is free iron (α-Fe). The older book-mold process cast thick ingots that cooled slowly enough to grow it, and soft iron inside a hard magnet is a defect you pay for later. Cooling this fast keeps it out. What does form is columnar grains of the magnetic phase a few microns wide, separated by thin neodymium-rich layers, and that width is no accident: it is about the size of particle the jet mill will make, so milling can turn each grain into a single crystal. The flake is not the magnet's final microstructure. It is the raw material for one, and the real thing gets built in the press and the furnace.
Where SmCo differsSamarium-cobalt is usually book-mold cast rather than strip cast: the alloy is poured into a thick mold and allowed to freeze slowly, because strip casting produces a microstructure too fine for what the finished SmCo magnet needs. The alloy can also be made by co-reduction, reducing samarium oxide directly in the presence of cobalt and skipping the separate metal step altogether. There is also a vacuum problem that neodymium never has: samarium has a much higher vapor pressure than neodymium, so in a hard vacuum it sublimates straight out of the melt. The casting furnace is run with that in mind, and the samarium that boils off is one of the things the charge is sized to cover.
Short versionLet hydrogen crack the flake apart along its own grain boundaries. No hammer required.
06 · PowderHydrogen decrepitation
Hydrogen breaks the flake into a coarse powder along its grain boundaries.
The flake is loaded into a rotating kiln and exposed to hydrogen. Hydrogen is absorbed preferentially by the neodymium-rich layer between the grains, and that layer swells as it takes the hydrogen up. The grains themselves barely change, so the swelling cracks the flake apart along exactly the boundaries that were formed on the wheel.
The flake crumbles, without any mechanical force, into a coarse powder around 200 microns in size. The kiln is then evacuated and heated to drive most of the hydrogen back out. This step is elegant because it uses the microstructure to break the material where we want it broken, and it produces a powder that the next step can mill efficiently.
Where SmCo differsSamarium-cobalt is not hydrogen decrepitated. It takes very high hydrogen pressures to break it down efficiently, so the cast alloy is simply hammer-milled to a coarse powder before it goes to the jet mill.
Short versionNitrogen jets knock the powder down to single-grain particles a few microns across. From here to the furnace it is self-heating and stays under cover.
07 · PowderJet milling
The coarse powder is milled to a fine, uniform powder a few microns in size.
The coarse powder is fed into a fluidized bed jet mill. High-velocity jets of nitrogen accelerate the particles into one another, and they fracture on impact. There is no grinding media and nothing touches the powder but gas, which keeps it clean. A classifier wheel at the top of the mill lets only particles below a set size leave; anything larger is thrown back into the bed to be milled again.
The result is a powder around three microns in size, tightly controlled, in which each particle is a single grain of the magnetic phase. From this point until the material is sintered, the powder is self-heating: with this much surface area, it will oxidize and warm on contact with air. It is kept under a protective atmosphere throughout.
Where SmCo differsSamarium-cobalt powder is less reactive than neodymium powder. A protective atmosphere is still important, but it is not absolutely critical, for a reason that is easy to miss: oxygen that does get in simply ties up samarium and pulls it out of the system, rather than actively damaging the magnetic performance the way it does in NdFeB. The fix is to blend in a little extra samarium to compensate.
Short versionLine every grain up in a magnetic field, then squeeze. Die pressing for volume; isostatic for the best alignment.
08 · Press and sinterPressing
The powder is aligned in a magnetic field and compacted into a green body.
Each particle of powder is a single grain, and each grain has one direction along which it prefers to be magnetized. To make a strong magnet, all of those directions have to point the same way. The powder is placed in a die, a magnetic field is applied, and the particles physically rotate to line up with it. While they are held in alignment, they are pressed into a solid, fragile shape called a green body, at a little over half of full density. We press to the highest density we can that still yields a good magnet. A lot of people who come to magnetics from other powder-processing fields want to go much higher, only to discover that the improved shape they can make comes at the price of material performance.
There are two ways to do this. In die pressing, the powder is compacted between punches, with the field applied across the direction of pressing; this is the workhorse process for high-volume parts. Pressing along the field is also possible, but it costs magnet performance. In isostatic pressing, the powder is sealed in a flexible bag, aligned in a pulse coil, and then squeezed by fluid pressure from every side at once. Because nothing pushes the particles out of line, isostatic pressing gives the best alignment and the highest-performing material, at the cost of throughput.
Where SmCo differsSamarium-cobalt needs significantly more magnetic field to orient than neodymium powder does, so the presses and pulse coils for it are built accordingly.
Short versionThe furnace turns a fragile green body into a dense block, and a second, cooler heat treatment turns the block into a magnet. This is where the know-how lives.
09 · Press and sinterSintering and aging
The green body is fused into a dense block, then heat treated to develop its magnetic properties.
The green bodies are loaded into a vacuum furnace and heated until the neodymium-rich phase between the grains melts, while the grains themselves stay solid. That thin liquid pulls the grains together, the pores between them close, and the part becomes a fully dense metal block, with the alignment set in the press preserved. It shrinks in a weird way. Along the magnetic axis it shrinks A LOT, about 35%, and perpendicular to that only 10% or so. That difference is what kills a lot of great ideas about sophisticated orientation patterns.
The block does not yet perform as a magnet. A second, lower-temperature heat treatment, called aging, follows. How long it takes is highly dependent on the manufacturer's recipe, anywhere from an hour to several days. Aging re-forms the neodymium-rich phase into a thin, continuous layer around every grain, which is what allows each grain to hold its magnetization independently of its neighbors. The sintering and aging cycle is where the magnetic properties are actually created, and it is the most closely held know-how in the process.
Where SmCo differsSamarium-cobalt tends to run tighter tolerances on the heat treat than NdFeB. It is also more brittle, which makes it more prone to cracking in the quench. And samarium's high vapor pressure follows it into the furnace: hold SmCo in a hard vacuum at sintering temperature and the samarium leaves the surface of the part, so the furnace atmosphere has to be managed to keep it where it belongs.
Short versionSaw and grind to size, and accept that about a quarter of the block leaves as dust.
10 · FinishMachining
Blocks are sliced and ground to finished dimensions.
Sintered blocks are made oversize and in simple shapes, because the shrinkage in the furnace cannot be held to a finished tolerance. The block is sliced into individual parts, usually with a multi-wire saw, and each part is ground to its final dimensions. The material is hard and brittle, so it is cut with diamond abrasives and cannot be machined the way a steel part can.
Every cut removes material, and that material is lost. Between the saw kerf, the grinding stock, and parts that chip or crack, machining typically consumes a quarter of the sintered block. This is the largest single yield loss in the plant after the mine, and it is why part geometry and block design matter so much to the economics of a magnet.
Where SmCo differsThe brittleness that shows up in the quench shows up again here: samarium-cobalt is more prone to cracking in machining than NdFeB, and the yield loss runs correspondingly higher.
Short versionPut the expensive heavy rare earth only where it does any good: down the grain boundaries, not through the grains.
11 · FinishGrain boundary diffusion
Heavy rare earths are diffused into the grain boundaries to raise resistance to demagnetization.
The heavy rare earths dysprosium and terbium make a magnet more resistant to demagnetization at high temperature, which is what motors and generators demand. Historically they were added to the melt, which put them everywhere, including inside the grains where they do little good and reduce the strength of the magnet. They are also scarce and expensive.
Grain boundary diffusion puts them only where they are needed. A source of dysprosium or terbium is applied to the surface of the finished part, and the part is heated. The heavy atoms travel inward along the neodymium-rich boundaries, which are far easier paths than the grains themselves, and enrich the outer edge of each grain they pass. The effect is strongest at the surface and tapers off quickly with depth, so the process works best on thin parts. It delivers the high-temperature performance of a heavily alloyed magnet with a fraction of the heavy rare earth, and no loss of strength. The specific methods of applying the source are the most heavily patented territory in the industry.
Where SmCo differsSamarium-cobalt is not grain boundary diffused. The whole point of this step is to buy NdFeB resistance to demagnetization at temperature, and SmCo does not need to buy it: it performs enthusiastically up to and beyond 300 °C on its own, which is why it is still the magnet of choice when the application runs hot.
Short versionA nickel barrier, applied by chemistry in a tumbling barrel, so the magnet does not slowly consume itself.
12 · FinishCoating
A protective layer is applied to keep the magnet from corroding.
Neodymium magnets corrode readily. The neodymium-rich phase at the grain boundaries, which every step so far has been working to create, is also the part of the magnet that reacts most easily with moisture. Left bare, the surface will oxidize and the magnet will shed material and lose performance.
The finished parts are coated. Electroless nickel is the common choice: the parts are tumbled in a perforated barrel through a sequence of baths, and the nickel bath deposits metal by a chemical reaction rather than an electric current, so the layer builds at the same rate on every surface, including holes and edges, and the thickness is uniform. Other coatings, from zinc to epoxy, are chosen when the application calls for them. Whichever is used, the coating is a barrier and nothing more; it adds no magnetic performance, and it must be intact for the magnet to survive its service life.
Where SmCo differsSamarium-cobalt is generally not coated at all. It is quite corrosion-stable, so the barrier that NdFeB cannot live without is simply not needed.
Short versionOne pulse, a few milliseconds, and the part becomes a magnet. The demagnetization curve is the receipt.
13 · FinishMagnetizing
A pulse of magnetic field turns the part into a magnet.
Until this point, the part is not a magnet. Every grain has its preferred direction lined up with its neighbors, but the grains are magnetized in a mix of directions that cancel each other out, and the part will not stick to a refrigerator. This is deliberate: an unmagnetized part can be machined, coated, handled, and shipped without pulling on everything ferrous around it.
To magnetize it, the part is placed in a fixture and a capacitor bank is discharged through a coil around it. The pulse lasts a few thousandths of a second and produces a field far stronger than the magnet itself will ever generate. Inside the material, every grain snaps into the same direction along the axis set at the press, and stays there. The part is now a magnet, and will remain one for decades. Its performance is described by a single curve, the demagnetization curve, which is the culmination of every step before it.