Skip to content
Peculiar Materials
Insights · Soft magnetic materials

The trouble with Hiperco

Iron-cobalt carries about 20% more flux than silicon steel, and the data sheet is not lying about it. The trouble starts the moment you cut it.

Short versionHiperco 50 is iron with roughly as much cobalt as a body-centered iron lattice will take before the moment per atom turns over, plus 2% vanadium so that it can be rolled at all. On the data sheet it saturates near 2.4 T against about 2.0 T for 3% silicon steel, and that 20% is real: saturation belongs to the atoms and survives anything you do to the sheet. Everything else on the sheet (the permeability, the coercivity, the loss) was measured on a ring that was annealed AFTER it was cut and then never touched again. Your lamination is cut, handled, annealed, coated, stacked, pressed onto a shaft and potted, and every one of those steps is a chance to hand part of the 20% back. Silicon steel arrives finished and forgives you. Hiperco arrives unfinished and does not. The trouble with Hiperco is not the alloy. It is that the alloy's last manufacturing step is yours.

Peculiar Materials LLC · The curves on this page are schematic or representative: the Slater-Pauling hump and the two magnetization curves follow the published shapes and endpoints but are not digitized from any one document, and the edge profiles are drawn to explain a mechanism, not to report a measurement. Numbers in the text are footnoted to data sheets and papers; where a number is my own arithmetic it says so.

What this page skips, so nobody mistakes it for complete: 6.5% silicon steel, amorphous and nanocrystalline ribbon (the real competition at high frequency), powder cores, grain-oriented Supermendur and field annealing, the split of loss into its hysteresis and eddy parts, the measurement standards themselves, and the heat-treat recipe. The last one is on purpose: the mechanism is here, the temperatures are on the data sheet.
1
What it is
Cobalt, iron, and why the data sheet looks like that
2
What happens when you use it
Cold work, two kinds of cut edge, which properties care
3
The heat treat
One coil, two alloys, and the point of no return
4
Versus silicon steel
What the cobalt costs, and when it pays
Scroll to build the figure

Short versionPure iron saturates at about 2.16 T. Add cobalt and it climbs to about 2.45 T near one-third cobalt, then falls again. Hiperco 50 sits at half and half because the ordered 50/50 alloy has the best permeability in the family, and the 2% vanadium is there so the mill can roll it.

1 · What it isAs much cobalt as iron will take

Start with the plot, because the plot is the whole sales pitch. Pure iron saturates at about 2.16 T. Mix in cobalt and the saturation rises, because cobalt's extra electron fills the 3d band in the way that raises the net moment per atom, until, at roughly a third cobalt, it stops helping and starts hurting. That hump is the iron-cobalt stretch of the Slater-Pauling curve, the plot of moment per atom against electron count that organizes the whole transition-metal family, and iron-cobalt sits at its very top: the highest room-temperature saturation of any known alloy.1

You will sometimes hear it described as cobalt with iron dissolved in it. It is the other way round. This is body-centered iron with cobalt substituted onto the lattice, and it stays body-centered until it orders; nothing here is face-centered. The ordering is the point. On slow cooling the 50/50 alloy arranges itself into the B2 structure (iron on the corners, cobalt in the middle, or the reverse), and that ordered state has very low magnetocrystalline anisotropy, which is where the family's high permeability comes from.2

So why is the flagship alloy 49/49 rather than 65/35, where the saturation peaks? Because of the mill, not the physics. The ordered alloy is brittle enough that a plain 50/50 melt cannot be cold rolled into strip at all. The 2% vanadium slows the ordering reaction enough that the strip can be quenched disordered, rolled thin, and ordered later, in a furnace that is not the mill's.2 Hiperco 27, with a little over half the cobalt, gives up almost no saturation but a lot of permeability (its coercivity runs several times higher) in exchange for ductility and price.3

A word on names, because this alloy has more of them than it has compositions. Bell Labs made the 50/50 alloy in the late 1920s and called it Permendur; it was so brittle it had to be quenched to be worked at all. The 2% vanadium fix came in the early 1930s, and that version was 2V-Permendur. Supermendur, in the 1950s, was the same composition at higher purity with a field anneal. Hiperco was Westinghouse's trade name before it was Carpenter's. VACOFLUX 48 and 50 and VACODUR are VAC's; AFK 502 is Aperam's; 1J22 is the Chinese standard designation.4 Every one of those is roughly 49% iron, 49% cobalt and 2% vanadium, and every one of them is ASTM A801 Type 1, UNS R30005.5 The names tell you who rolled it. What they do NOT tell you is interchangeable: the mills differ in purity, in the pinch of niobium or tantalum they add to hold the grain size, and in rolling practice, and a suffix (50A against 50 HS, VACOFLUX against VACODUR) means a different recipe and a different set of properties from the same base alloy. Treat the trade name the way the grade guide treats a magnet grade: a label for a family, with the data sheet doing the actual specifying.

Used in: aircraft starter-generators and other machines sized by torque per kilogram, high-speed rotors, and actuators and relays whose pull is set by the square of the flux density.6

Short versionThe magnetization curve and the loss table on the sheet come from a ring that was cut, annealed with the mill's recipe, and never touched again. The 20% at the top belongs to the atoms and survives everything. The shape of the approach to it does not.

1 · What it isThe data sheet, and what it was measured on

Put the two typical DC curves on the same axes and the appeal is obvious. At the field a motor tooth sees, iron-cobalt is carrying something over 2 T where M-19 is carrying about 1.5, and the ceilings are roughly 2.35 to 2.4 T, depending on whose sheet and at what field, against about 2.0 T.789 Translate that into a machine: at the same current and geometry, more flux through the tooth; at the same flux, a narrower tooth and a lighter core. Force in an actuator scales with B², so 20% more flux is about 44% more pull from the same pole face. Those are the numbers that put iron-cobalt in aircraft generators, where every kilogram of core is weighed.6

Now read the fine print, or rather what is not in it. The curve was measured on a ring sample (a toroid, wound with its own primary and secondary) that was cut, cleaned, annealed by the recipe on the back of the same sheet, and then measured, by the standard ring method. Nobody stamped it into a stator. Nobody pressed it onto a shaft, glued it, bolted it, or wound a coil tight against its teeth. Every number on the sheet is the value of the alloy in the best condition it can ever be in. The rest of this page is about how far your part falls from that condition, and what it takes to climb back.

In practiceI have spent a career reducing oxides to metal, casting it into alloy, pulverizing, pressing, sintering, grinding, gluing and testing (hard magnets mostly, but the lesson transfers). The one constant across hard and soft magnetics is that the number on the sheet belongs to the test piece. The part has to earn it.

Short versionHiperco arrives cold rolled, which is to say already wrecked magnetically, on purpose: the cold-worked, disordered state is the only one soft enough to roll. Cutting adds more damage of the same kind. The anneal is what turns strip into Hiperco; everything you do before it is reversible, and nothing after it is.

2 · What happens when you use itCold work: the strip is born damaged

Magnetic properties divide into two camps. Saturation and Curie temperature are structure-insensitive: they depend on which atoms are present and in what proportion, not on how they are arranged, and no amount of hammering changes them. Permeability, coercivity and hysteresis loss are structure-sensitive: they are set by how easily domain walls move, and domain walls are stopped by dislocations, grain boundaries, inclusions and internal stress. Cold work is the manufacture of dislocations and internal stress. So the as-rolled strip in the figure saturates at exactly the same 2.4 T as the annealed one, and takes something like ten times the field to get anywhere near it (the shape is mine, drawn to make the point). The curve has been sheared over.

Why is iron-cobalt so much worse at this than silicon steel? Magnetostriction: the coupling between a material's strain and its magnetization, which runs both ways. Iron-cobalt's saturation magnetostriction is about 60 to 70 parts per million on the three mills' sheets;10 polycrystalline 3% silicon steel's works out to roughly 6 or 7, by my arithmetic from the published single-crystal constants.11 A stressed iron-cobalt lattice sees that stress as an anisotropy the domains have to fight, and because the alloy's crystal anisotropy is so low to begin with, a few megapascals of internal stress is already comparable to it.12 That one number explains most of the rest of this page: why the edges matter, why the fixture matters, why the shrink fit matters.

The strip is supplied in this cold-worked state because it has to be. The vanadium buys enough time to quench the alloy disordered and roll it, but the ordered, soft-magnetic version is the brittle one. VAC says it in one line: annealing does not improve the alloy's workability, it makes it harder to work, because of the embrittlement.8 So the sequence is fixed by metallurgy rather than by anyone's preference: shape it while it is hard, stampable and magnetically useless, then anneal it into a material that is magnetically excellent and that you can no longer shape.713

Short versionA punched edge is damaged by strain; a laser-cut edge is damaged by heat. The hard damage is a fraction of a millimeter deep; the magnetic penumbra runs millimeters. In silicon steel you usually ship that damage, and laser is the worse of the two. In Hiperco the anneal that follows cutting removes most of it, the ranking compresses, and the one published head-to-head on the alloy puts laser-plus-anneal on top.

2 · What happens when you use itDie cut against laser cut

Look at the two cross-sections. Punching drags the metal: the top corner rolls over, the punch burnishes a vertical band, the sheet fractures across the rest, and a burr hangs off the bottom, typically a few percent of the sheet thickness tall.14 Behind that edge is a band of plastic strain, dense with dislocations, reaching in a few tenths of a millimeter, roughly the sheet thickness, and a wider halo of elastic stress that lowers permeability for several millimeters beyond it.15 In silicon steel the classic experiment is to cut the same sheet into narrower and narrower strips and watch the loss climb as the damaged fraction grows: a 5 mm punched strip can lose half again as much as a wide one.15 The burr has a second cost: it bridges the insulation between laminations and shorts them, and in a stacked core that is worth several percent of the loss on its own.14

A laser does not deform; it melts. The edge is square and nearly burr-free, but it carries a recast skin of metal that melted and froze back onto the edge, an oxide film about a micron thick, a bead of dross underneath, and a heat-affected zone in which the microstructure has been run through a temperature excursion it was never meant to see, with residual thermal stress to match.16 Measured in silicon steel, the laser's magnetic damage reaches as far as the punch's or farther, and in most of the published comparisons the laser-cut edge is the worse of the two.16 Laser cutting is the prototype tool because it needs no die, not because it is kinder.

Then the Hiperco twist. In silicon steel the stress-relief anneal is optional and often skipped, so the damaged band ships. In Hiperco the anneal is NOT optional. You are going to recrystallize the whole lamination anyway, and recrystallization does not know which dislocations came from the mill and which from your punch. In silicon steel, an anneal after cutting brings the different cutting methods to nearly the same loss;17 in iron-cobalt, the one published head-to-head I can find (punching, wire EDM and laser, with the anneal before or after the cut) found that laser cutting followed by the anneal gave the best magnetic result of the routes tried.18 Carpenter's own process note says stamping gives the highest magnetic performance of its stack-building methods.19 Those two do not agree, and the honest reading of both is that after a proper anneal the cutting method is a second-order effect and the anneal is the first-order one. What survives the anneal at a laser edge is thin: the recast skin and the oxide, which do not recrystallize into anything useful. What survives at a punched edge is the burr, which the anneal does not remove. Wire EDM leaves a recast skin of its own, and surface flaws that have been known to start cracks in a finished part.20 So choose the cut on burr, tolerance, gauge and die cost, deburr whatever you choose, and never skip the anneal. One warning on the die: the strip arrives cold rolled to something over 1,000 MPa yield,21 so expect it to be harder on tooling than silicon steel.

Short versionAcross a cut edge, saturation is flat, permeability dips, coercivity and loss climb, and the band is about a thickness wide. The anneal flattens most of it. What it cannot flatten is a thin laser skin, which is why laser-cut Hiperco needs an edge allowance in the design rather than an apology afterward.

2 · What happens when you use itWhich properties care, and how far in

The profiles in the figure say which property is damaged, and how far from the edge. Saturation is a straight line: the edge is still iron and cobalt. Permeability near the edge drops to a fraction of bulk and recovers over about a thickness; coercivity does the mirror image; loss, which at a given frequency and flux is largely a coercivity story, follows coercivity. After the anneal the green curves are nearly flat. What remains is thin: a skin of oxide and recast at a laser edge, which you design around with an allowance, and a burr at a punched edge, which you remove before stacking, because the anneal will not.17

Now the arithmetic, which is mine. Suppose the hard-damaged band is one thickness per edge, the silicon-steel rule of thumb applied to iron-cobalt on my say-so.22 A stator tooth 2 mm wide cut from 0.15 mm strip has two edges, so 0.3 mm of its 2 mm, 15%, is damaged before the anneal. A 1 mm tooth is 30% damaged. Thin strip and fine teeth are exactly the combination iron-cobalt gets specified for, in high-speed, high-frequency machines, so the damaged fraction is not a corner case. It is the design. In silicon steel you would ship it. In Hiperco the anneal gives it back, if the anneal is done, and done right, which is the next section.

Short versionThe final anneal does two jobs at once: it erases the cold work, and it sets grain size and the degree of order, which fix permeability on one side and yield strength on the other. Push it toward coarse grains and you get the stator alloy. Hold it back and you get the rotor alloy. Same coil.

3 · The heat treatOne coil, two alloys

What the anneal does, in order. First it recovers and then recrystallizes the cold-worked structure, replacing the dislocation forest with new, strain-free grains; that is what returns the permeability. Second, as the new grains grow, grain boundaries become scarcer, and grain boundaries pin domain walls, so the further the anneal is pushed the higher the permeability and the lower the coercivity. Third, on the way back down, the alloy orders into the B2 structure below roughly 730 °C, and the rate of cooling through that range decides how completely; order is what gives the 50/50 alloy its low anisotropy and its brittleness, both.2 The atmosphere matters too, because the alloy oxidizes and picks up carbon and nitrogen that pin walls, so the anneal runs in dry hydrogen or vacuum, and the parts go in clean.7 The data sheet gives the temperatures, times and atmosphere. This page does not, on purpose.

The catch is that a coarse-grained, fully ordered alloy is also a weak one. Coercivity falls with grain size and yield strength rises with it, and the two laws, Mager's for the one and Hall-Petch for the other, pull in opposite directions through the same variable.23 A rotor turning at tens of thousands of rpm is held together by its yield strength, so the rotor designer stops the anneal early and accepts the permeability penalty, and the stator designer pushes it and accepts the strength penalty. The mills sell both ends. Carpenter's Hiperco 50A has no grain refiners and is the magnetic end; Hiperco 50 HS, with niobium to hold the grains fine, is the strength end.7 VAC's VACODUR 49 data sheet says it plainly: stator and rotor laminations can be stamped from the same strip and then annealed differently, with the yield strength set anywhere from about 210 to 400 MPa, against about 18,000 maximum permeability at the soft end and 8,000 at the strong one, and a tantalum-zirconium cousin pushes the strength end to 800 MPa at a permeability a magnet designer would not call soft.24 The lesson is more general than the catalog: the alloy you specified is not finally decided until the furnace door closes. Design the anneal with the part, not after it.

In practiceI spent years on Formula 1 MGU-K and MGU-H systems, machines that live between 50,000 and well over 100,000 rpm, and the rotors I built were surface-mount permanent-magnet designs. Whether the stacks in them were iron-cobalt, and what their anneal looked like, is under NDA and will stay there. The shape of the problem is public: at those speeds the rotor's yield strength is on the critical path, and every unit of permeability bought with a harder anneal is paid for in strength you no longer have. The omissions here are deliberate.

Short versionThe silicon-steel route is stamp, stack, ship. The Hiperco route is cut, clean, anneal flat, coat, bond, and assemble without stress, and nothing after the anneal may deform the part. Plan the fixtures, the coating step, the witness rings and the lead time before you order the strip.

3 · The heat treatPlanning around the point of no return

Walk the two rows. Fully processed silicon steel comes off the mill annealed and already wearing its insulation, a coating of a class that survives a stress-relief anneal.25 It is stamped on a progressive die at a few hundred strokes a minute,26 interlocked or welded into a stack in the die, sometimes stress-relief annealed and usually not,27 and that is a stator. Hiperco comes off the mill cold rolled and hard, carrying at most a thin separator (on VAC's strip it becomes a magnesium-oxide film under a micron thick in the anneal; on Carpenter's the sheet names a magnesium-oxide anti-stick layer or an oxide you grow afterward).87 It is cut. It is cleaned, since anything on the surface goes into the furnace with it. It is annealed, either as loose laminations between flat plates or as a stack that was interlocked in the die, because thin strip does not come out of a furnace flat on its own.28 THEN the insulation is finished, usually by growing an oxide in a second, cooler treatment, and loose laminations are bonded into a stack.7 Welding a stack after the anneal is a laser cut you did on purpose.

And it grows. The 49% cobalt alloys lengthen permanently in the anneal, by up to about 0.2%, and not equally in the two directions, so a round core can come out slightly elliptical;828 a precision bore therefore has to be finished after the anneal, and VAC's own advice for that finishing is "as gentle as possible," which is the whole dilemma in four words.8 After the anneal: no restrike to flatten, as little machining as the growth allows, no shrink fit that loads the yoke, no winding pulled tight against a tooth, no potting compound whose cure shrinkage squeezes the stack. Each of those is a small cold work, and chapter 3 showed what small cold work does to this alloy; in silicon steel a shrink fit alone has been measured to add a third to the yoke's loss and cut its permeability several-fold.29 The engineering answer is to design the stress out: locate the stack on the shaft without loading it, bond with low-shrinkage adhesives, clamp at a few points rather than press, and give the laser-cut edge its allowance.

Two more lines for the plan. Put a witness ring, a toroid cut from the same strip, into every furnace load, because the data sheet number is a ring number and the only way to know what your laminations became is to anneal a ring beside them and measure it. And buy the lead time early: the atmosphere furnace, the fixtures and the coating step are weeks, not days, and they belong on the order with the strip, not after it.

In practiceMagnet bonding and adhesion have been a focus of my career, and a bonded Hiperco stack is the same problem with one extra rule: the adhesive's cure must not load the laminations. A glue line that shrinks as it sets is a clamp you cannot remove, and the stack will tell you about it in the permeability, not in the pull test.

Short versionHalf of every kilogram of Hiperco is cobalt, and cobalt is priced like a battery metal, which lately it is. Before anybody melts anything, the cobalt alone puts Hiperco's raw-material floor well above a finished kilogram of silicon-steel coil. The strip price is not published. The floor is enough to decide most projects.

4 · Versus silicon steelWhat the cobalt alone costs

My arithmetic, raw content only. Iron at the US hot-rolled band price is about $1.30 a kilogram this month, and half that at the world export price.30 Three percent silicon, at ferrosilicon prices, adds about six cents.31 A finished kilogram of non-oriented electrical steel coil in the common gauges is somewhere between $2 and $5.40 in the United States, and I give the range because the two public price indices disagree with each other by a factor of two and a half.32 Hiperco 50 is 49% cobalt, and cobalt metal on 19 October 2026 was about $35,700 a metric ton, call it $36 a kilogram,33 so the cobalt in a kilogram of Hiperco is about $17.50 before the iron, the vanadium, the vacuum melt, the rolling to a few tenths of a millimeter, the slitting and the yield losses at every step. The mills do not publish the strip price and I am not going to guess at it. What is public is the multiplier: a peer-reviewed study of aerospace starter-generator laminations puts cobalt-iron at about ten times the cost of silicon iron, and Arnold, selling thin-gauge silicon steel against it, says its product is about a third the price of cobalt-iron.34 So the raw cobalt alone is already three to nine times the price of a finished kilogram of silicon steel, depending on which index you believe, and whatever the strip costs, it moves with cobalt, which through 2025 and 2026 moved with Congolese export policy: a ban in February 2025, quotas from October, a price that more than doubled over the year, and then a fall of a third in a single month this autumn.35

Two cost consequences beyond the kilogram. Thin strip: at high frequency the laminations get thin, and thin strip costs more per kilogram in any alloy, because the last rolling passes are the slow ones.36 And yield: a stamped stator throws a large fraction of the coil away as slot and bore scrap. In silicon steel that scrap is scrap. In Hiperco it is half cobalt, and a program that does not reclaim it is paying for the cobalt twice.

Short versionOn the sheet, Hiperco wins on saturation, permeability and temperature and holds its own on loss. Off the sheet it loses on strength after the anneal, delivered condition, cutting freedom, growth, tolerance of stress and cost. It pays when the machine is sized by flux or by mass in a fixed envelope, or runs hotter than silicon steel can. Otherwise, silicon steel, and spend the cobalt money on copper.

4 · Versus silicon steelWhen it pays, and when it does not

While Hiperco's data sheet beats silicon steel's in the columns most people read, those columns matter when the machine is flux-limited, and in the columns that matter when it is cost-limited or process-limited, silicon steel wins, and most machines are those machines. Loss is the column where the sheet surprises people. At the same gauge, Carpenter's Hiperco 50A ring posts a lower 60 Hz loss than M-19 does on its own sheet (about 1.8 against 3.2 W/kg at 1.5 T in 0.36 mm strip), and at 400 Hz it takes a thinner silicon steel to match it.79 The catch is in the word ring. That is a well-annealed, untouched toroid, and your stator is neither. And at kilohertz, where the eddy term rules and the resistivity (about 40 µΩ·cm against 45 to 55 for 3% silicon steel) is slightly against it, thin silicon steel and amorphous ribbon pull ahead; Arnold's own line is that cobalt-iron is not recommended above about 1,200 Hz.37

Temperature is the other column where the physics is on Hiperco's side. Its Curie point is about 940 °C against about 745 °C for 3% silicon steel,738 and nobody designs at the Curie point; the practical limit is the winding insulation and the coating. But a core at a few hundred degrees has kept nearly all of its saturation if it is iron-cobalt and given up a noticeable slice if it is silicon steel, which is why the aircraft starter-generator programs that run their cores toward 500 °C are iron-cobalt programs.39 The payoff, when it pays: a published more-electric-aircraft starter-generator design came out about 20% lighter in the stator on cobalt-iron than on the best silicon steel it was compared with, and Carpenter's own aerospace pitch is 20 to 30% in power density.6

So: use Hiperco when the design is sized by flux or by mass at a fixed envelope (aircraft generators, high-speed rotors where a thinner back iron is a lighter spinning mass, actuators whose pull is set by B²), or when the core runs hotter than silicon steel's practical range, and ONLY with a process plan that treats the anneal as the last manufacturing step, with the fixtures, the coating, the witness rings and the lead time bought before the strip. Everywhere else, silicon steel, and put the cobalt money into copper, cooling or a better magnet. The trouble with Hiperco is that it makes you the mill.

Saturation against cobalt content 1 / 9

References

Physics is not cited. Every data-sheet value, every measurement from the cutting literature, every price and every policy event is. Where a number is my own arithmetic the text says so, and the footnote says what it rests on. The historical dates in the names paragraph (Permendur in the late 1920s, the vanadium version in the early 1930s, Supermendur in the 1950s) are decade-level on purpose; the footnote there is for the names, not the dates.

  1. University of Birmingham, School of Metallurgy and Materials, "Magnetic materials: properties" (background series, table 3), undated. Room-temperature saturation and Curie points: iron 2.15 T and 770 °C; iron with 35% cobalt 2.45 T and 970 °C; iron with 3% silicon 2.00 T and 740 °C; cobalt 1.76 T. The hump's position is given variously as 28 to 35% cobalt across the literature (Schoen et al., arXiv:1701.02177, find 2.42 T at about 35% in thin films; Gerasimov et al., arXiv:1907.06613, put the maximum near 28%), which is why the text says "roughly a third" and the figure is schematic.
  2. Couto and Ferreira, "Phase transformations and properties of Fe-Co alloys", J. Materials Engineering 11(1), 1989; and US patent 6,685,882, "Iron-cobalt-vanadium alloy", 2004, background section. The B2 (CsCl-type) ordering of the equiatomic alloy, its ordering temperature of roughly 730 °C, the brittleness that comes with order, and the role of about 2% vanadium in slowing the ordering so the alloy can be quenched disordered and cold worked. Couto and Ferreira also note that ordering has little effect on saturation. The ordering temperature is a material constant and is the only temperature on this page; no anneal temperature is.
  3. Carpenter Technology, alloy finder, "Hiperco 27", and Aperam, "IMPHY AFK 1" data sheet, June 2020. Carpenter's own description: saturation nearly identical to Hiperco 50, lower permeability and higher coercive force, without the brittleness. Aperam's 27% grade sheet gives coercivity several times that of its 49% grade and an annealed elongation of 13 to 17% against 4 to 5%. Both are the mills' own claims.
  4. Couto and Ferreira, "Phase transformations and properties of Fe-Co alloys", J. Materials Engineering 11(1), 1989 (Permendur, the 2% vanadium addition of White and Wahl, and Supermendur); Goodfellow, Hiperco 50 coil listing (synonyms Supermendur, Permendur, FeCo49V2); ASM Alloy Digest, "Hiperco 50" (Westinghouse), January 1967, abstract; Li et al., "1J22 FeCo alloy under stress", SMM27, 2025, for the 1J22 designation. The lineage of the names, at decade level; the ASM entry is paywalled and used only for the Westinghouse attribution and the date. The claim that the grades are the same base alloy rests on ASTM A801 and on the mills' own composition tables (Carpenter, VAC and Aperam, cited above); the claim that they are not interchangeable rests on the same tables, which differ in the grain-refining additions and in the guaranteed properties.
  5. ASTM International, ASTM A801-21, "Standard specification for wrought iron-cobalt high magnetic saturation alloys (UNS R30005 and K92650)". Type 1 is the 49% cobalt, 49% iron, 2% vanadium alloy; Type 2 is the 27% cobalt alloy. The scope says the material is supplied for fabrication into parts that will later be given a final heat treatment to develop the magnetic properties, which is the point of this page in one sentence. Carpenter, VAC and Aperam each cite A801 for their 49% grades.
  6. Golovanov, Gerada et al. (University of Nottingham), "The influence of stator material on the power density and iron loss of a high-performance starter-generator for more electric aircraft", ICEMS 2018; and Carpenter Technology, aerospace and avionics page. The Nottingham study is a design comparison, not a flight part: a cobalt-iron stator of 3.7 kg against 4.6 kg on the best silicon steel compared, which is the "about 20% lighter" in the text, with roughly a quarter of the iron loss. Carpenter's "20 to 30%" power-density figure is the manufacturer's own claim for aircraft generators and APUs. Neither source supports a general rule; both support the claim that the alloy is chosen where mass is weighed.
  7. Carpenter Electrification, "Hiperco 50A alloy" data sheet (E199), and Carpenter Technology alloy finder pages for Hiperco 50 and Hiperco 50 HS. Saturation 24 kG (2.4 T); Curie 938 °C; resistivity 40.1 µΩ·cm; density 8.11 g/cm³; core loss on 0.36 mm strip 1.75 W/kg at 1.5 T and 60 Hz and 30 W/kg at 400 Hz; yield 1,276 MPa with 1% elongation as cold rolled, 202 MPa with 6.7% after the magnetic anneal; laminations stamped from cold-rolled strip and final annealed in dry hydrogen or vacuum after cleaning; insulation by a magnesium-oxide anti-stick layer or an oxide grown after the anneal; 50A has no grain refiners, 50 HS carries niobium for grain refinement and strength. All of it is the manufacturer's own data, measured on rings. The data sheet carries the heat-treat temperatures, times, cooling rates and atmosphere; this page omits them deliberately.
  8. Vacuumschmelze, "Soft magnetic cobalt-iron alloys VACOFLUX and VACODUR", product brochure, 2021 edition. Saturation polarization 2.30 T and flux density 2.35 T at 40 kA/m for the 49% grades; Curie 950 °C; saturation magnetostriction +70 ppm; the 49% grades are supplied only in the hard, cold-rolled condition; the final anneal is mandatory; workability is not improved by annealing but made worse by embrittlement; a permanent, anisotropic length increase of up to 0.2% in the heat treatment; any processing after the final anneal affects the magnetic properties and final fine machining should be as gentle as possible; the strip's insulation layer becomes magnesium oxide under a micron thick in the anneal, with an optional oxidizing treatment for more. The brochure also carries the recommended heat treatments, which this page does not reproduce.
  9. AK Steel (now Cleveland-Cliffs), "Selection of electrical steels for magnetic cores", product data bulletin, 2007 (university-hosted copy); and the DI-MAX M-19 fully processed data reproduced in MIT OpenCourseWare 6.685. M-19 at 0.36 mm, fully processed: 1.0 T at about 106 A/m, 1.5 T at about 1,110 A/m, 2.0 T at about 31,000 A/m; core loss 1.45 W/lb (3.19 W/kg) at 1.5 T and 60 Hz; resistivity 43 to 50 µΩ·cm across the M grades; yield 269 to 358 MPa fully processed; fully processed means annealed by the producer and ready for use, with an optional user stress-relief anneal. The 2.0 T ceiling in the text is the saturation of 3% silicon iron from the Birmingham table and DOE material summaries, not a value on the AK sheet, which does not state one.
  10. Vacuumschmelze brochure above (+70 ppm); Aperam, "Iron cobalt alloys, IMPHY AFK", September 2022 (65 ppm for AFK 502 R); High Temp Metals, Permendur 2V / Hiperco 50A technical data (60 ppm). Three sheets, three numbers, all within the 60 to 70 ppm the text gives.
  11. Hubert and Daniel, "Multiscale modeling of the magneto-mechanical behavior of grain-oriented silicon steels", J. Magn. Magn. Mater. 320 (2008), table 1. Single-crystal magnetostriction constants for 3% silicon iron: λ100 = 23 ppm and λ111 = −4.5 ppm. The polycrystalline average in the text, about 6.5 ppm, is my arithmetic from the standard isotropic average (two fifths of λ100 plus three fifths of λ111); no source I found quotes that number directly, and textured sheet will differ from it. The order-of-magnitude gap to iron-cobalt is the point, not the decimal.
  12. Banu, Ferrara, Pasquale, Fiorillo et al. (INRiM), "Temperature and frequency dependence of magnetic losses in Fe-Co", IEEE Access 11, 2023. Saturation magnetostriction about 70 ppm and a crystal anisotropy of only a few hundred J/m³, so that an internal stress of about 5 MPa produces a magnetoelastic energy of the same order as the anisotropy. That is the sense in which "a few megapascals is already comparable" is used.
  13. European Commission, CORDIS, FeCoProGen final report (Clean Sky, 2016). Iron-cobalt strip is sold as rolled so it can be stamped; after the anneal the parts become brittle and cannot withstand stamping or punching; box annealing deforms them; glued stacks in that project showed about 15% more loss than the reference. A single project's findings, not a survey.
  14. Hamzehbahmani, Anderson, Jenkins and Lindenmo, "Experimental study on inter-laminar short circuit faults in laminated cores" (Durham University repository). Burr height typically 2 to 10% of the sheet thickness; burrs on both edges form current loops; up to about 18% extra loss measured in the packs tested. Measured on silicon steel, not iron-cobalt; the mechanism is material-independent.
  15. Paltanea, Manescu, Antoniac, Nemoianu and Gavrila, "Mechanical and magnetic properties variation in non-oriented electrical steels with different cutting technology: a review", Materials 17(6), 2024 (open access); and Gmyrek, Szulakowski, Vaschetto and Cavagnino, "Changes in the Steinmetz coefficients of punched soft-magnetic sheets", IEEE ECCE 2022. Hardness at a sheared edge returns to bulk beyond about one sheet thickness; the plastically deformed zone is a few tenths of a millimeter; grain-structure damage reaches 0.1 to 0.2 mm while the stress that affects permeability reaches several millimeters; and the strip-width series relayed in the review (for example 3 W/kg at 120 mm rising to 4.9 W/kg at 5 mm for one punched steel). All silicon steel; the primaries (Schoppa, Naumoski, Weiss and others) are behind paywalls and are quoted here as the review relays them.
  16. Xiang, Cheng and Wu, "Effects of laser cutting parameters on the magnetic properties of 50W350 non-oriented silicon steel", Materials 16(4), 2023; Sundaria et al. (Aalto), "Effect of laser cutting on core losses in electrical machines", IEEE Trans. Ind. Electron. 67(9), 2020; Wu, Wang, Yue and Li, "Effect of processing methods on the magnetic properties of non-oriented electrical steel", TMS 2024. Laser-cut edges carried higher loss than sheared edges in every parameter set tested (worse by 4 to 13% at 1 T and 50 Hz); laser degradation fitted to about 5 mm from the edge in a stator, against a punched zone of a few tenths of a millimeter of hard damage; an oxide layer about a micron thick at the laser edge. All silicon steel. The same literature shows that an optimized laser can approach wire EDM, so "the worse of the two" is the usual result, not a law.
  17. The Paltanea review above, relaying Shi et al.: after a stress-relief anneal the losses of punched, laser-cut, wire-EDM and water-jet samples converged to nearly the same value, and the hardness at a punched edge fell back toward bulk. Silicon steel, one laboratory; the text uses it for the shape of the result, that the anneal removes most of the difference between cutting methods, and for nothing finer than that. The burr is the exception the anneal does not remove, which is the Durham study's point.
  18. Li, Gill, Wang, Lambourne, Oh and Chen, "Effect of laminate cutting and annealing treatment on the magnetic properties of Fe49Co49V2 alloy", IEEE Trans. Magn. 57(8), 2021. Punching, wire EDM and laser cutting compared on ring samples at 400 Hz, with the anneal before or after the cut; laser cutting followed by the anneal gave the best magnetic performance of the routes investigated. I could reach only the abstract, so the size of the differences is not on this page; it is the one open head-to-head on the alloy itself that I could find, and it is one study.
  19. Carpenter Technology, "Soft magnetic expertise from melt to mass production lamination stacks", company blog (also carried by Magnetics Magazine, July 2021). Stamping offers the lowest high-volume cost and the highest magnetic performance of the stack-building methods named (booklet EDM, laser cutting, notch and progressive-die stamping), but gives up tolerance control because the alloys grow during processing; cutting, insulation, bonding and machining all adversely affect magnetic performance. The manufacturer's own claim, and it points the other way from the Li study above; the text says so.
  20. James (NASA Goddard), "Mechanical tensile testing of Hiperco 50A at cryogenic temperature", NASA technical presentation, 2019. Specimens tested at 77 K failed at wire-EDM-cut pin holes, with failure propagating from surface flaws left by the EDM, and the recommendation that flight parts be screened for cracks. A cryogenic tensile result, used here only for the point that EDM is not a damage-free edge.
  21. Vacuumschmelze, VACOFLUX 48 data sheet, July 2022. Yield 1,230 MPa and 1% elongation as cold rolled; 190 MPa and 2% after the final anneal; coercivity about 35 A/m and maximum permeability about 21,000 after the anneal; magnetostriction +70 ppm. Carpenter's 50A sheet gives 1,276 MPa and 202 MPa for the same two conditions.
  22. My own assumption, and the footnote that undercuts the page. The "one thickness per edge" band is a silicon-steel rule of thumb for the hard (plastic) damage at a punched edge, taken from the hardness profiles in the Paltanea review; the magnetically affected zone in silicon steel reaches several millimeters, so the 15% and 30% figures understate the as-cut damage to permeability and overstate nothing. No public measurement of the band width in iron-cobalt was found, and the alloy's stress sensitivity suggests its magnetic halo is wider, not narrower. Treat the arithmetic as an illustration of a fraction, not a measurement.
  23. Sourmail, "Evolution of strength and coercivity during annealing of FeCo based alloys", Scripta Materialia 51 (2004); and Hailer, "Effect of heat treatment on magnetic and mechanical properties of an Fe-Co-V-Nb alloy", master's thesis, Virginia Tech, 2001. Coercivity falling with grain size (a Mager-type inverse relation) and yield strength following Hall-Petch in the same alloy; grain growth accelerating above the ordering temperature; and, in Hailer's words, strength gained at the expense of magnetic performance. The mechanism is physics; the laws are quoted from these sources rather than derived.
  24. Vacuumschmelze, VACODUR 49 data sheet and VACODUR S Plus data sheet, July 2022. VACODUR 49: yield adjustable from 210 to 400 MPa by the final anneal, with maximum permeability about 18,000 at the magnetically optimized end and 8,000 at the mechanically optimized end, and the sheet's own statement that stator and rotor laminations can be stamped from the same strip and annealed differently. VACODUR S Plus: yield to 800 MPa at a maximum permeability near 1,100. Manufacturer's data, on rings.
  25. ASTM A976 insulation classes as summarized on a stamper's classification sheet; thyssenkrupp Steel, "stabolit insulation coatings", September 2022. C-3 organic coatings do not survive a stress-relief anneal; C-4 and C-5 do, within stated limits; mills apply them to fully processed coil. The ASTM text itself is behind the ASTM store.
  26. Sko-Die, "Electrical steel laminations" (a lamination stamper's capabilities page), and Schuler Group, "Systems for the manufacture of electric motor laminations", brochure. Progressive-die presses at more than 250 strokes a minute (Sko-Die) and stroke ranges of 80 to 1,000 a minute (Schuler), both the vendors' own figures.
  27. Cogent Power, "Electrical steel, non oriented, fully processed", brochure, 2002 (university-hosted copy). Fully processed grades are normally used without a stress-relief anneal; only the uncoated or C-5-coated product should be annealed, in an inert or slightly oxidizing atmosphere. An old brochure, but the practice has not changed.
  28. Vacuumschmelze, US patent 8,887,376, "Method for production of a soft-magnetic core having CoFe or CoFeV laminations", and US patent 11,261,513, "Strip of a cobalt iron alloy, laminated core and method". The first describes annealing laminations between flat steel or ceramic plates with the flatness checked, an oxidizing anneal to form the insulating oxide, a dimensional change of about 0.2% along the rolling direction and 0.1% across it that can make a round core elliptical, and finishing by erosion after the anneal. The second gives the ordered annealed state as not suitable for stamping or punching, a growth of 0.03 to 0.20% in the anneal that rises with the degree of cold rolling, and a magnesium-oxide coating that keeps laminations from welding to each other in the furnace. Patents describe what the applicant claims to have done, not what the industry does; the brochure above says the same things in the manufacturer's own voice.
  29. Matsui and Yun, "Effect of shrink fitting on magnetic properties of a stator core", MATEC Web of Conferences 207, 2018; and Sundaria et al. (Aalto), "Effects of manufacturing processes on core losses of electrical machines", IEEE Trans. Energy Conversion 36(1), 2020. A wire-cut silicon-steel core's relative permeability fell from about 6,000 to about 1,500 after shrink fitting, with 50 Hz loss up by about a third (Matsui and Yun); shrink-fit yoke losses up 6 to 28% in the Aalto compilation. Silicon steel, with its single-digit magnetostriction; the text applies the shape of the result to iron-cobalt and nothing finer.
  30. SteelBenchmarker, price history, 12 October 2026. US hot-rolled band $1,310 per metric ton FOB mill; world export hot-rolled band $490 per ton. A proxy for the iron, not a price of electrical-steel feedstock; the point is the order of magnitude.
  31. ScrapMonster, ferrosilicon 75% price, 21 September 2026. About $1,470 a ton FOB China and $1,640 at Rotterdam. Three percent silicon at that price is about six cents in a kilogram of steel, by my arithmetic. The 2% vanadium in Hiperco is left out of the figure for want of a price I could cite; at that fraction it does not change the picture.
  32. IMARC Group, electrical steel pricing report (June 2026: United States about $5,360 a ton), and Expert Market Research, electrical steel price trends (second quarter 2026: United States about $1.99 a kilogram). Two market-research indices that do not separate grade or gauge and disagree with each other by about two and a half times; neither is a mill price list. The text gives the whole range for that reason, and the ratio to raw cobalt is quoted as three to nine rather than as a number.
  33. Trading Economics, cobalt, read 21 October 2026. $35,705 a metric ton on 19 October 2026, down about 10% on the month and 19% on the year. The London Metal Exchange's own delayed three-month close on the same days was about $34,600. This is the number that moves fastest on the page; by the time you read it, it has moved.
  34. Cossale, Krings, Soulard, Boglietti and Cavagnino, "Practical investigations on cobalt-iron laminations for electrical machines", IEEE Trans. Ind. Appl. 51(4), 2015; and Arnold Magnetic Technologies, Arnon electrical steel product page. "About ten times the cost of SiFe" is the peer-reviewed figure for laminations; "about a third the cost" of cobalt-iron is Arnold's claim for its thin-gauge silicon steel, which is itself dearer than ordinary coil. Both are multipliers, not prices, and no mill publishes a strip price. Online merchant listings for Hiperco-type strip run from roughly $100 to $300 a kilogram, and I treat those as indicative only, which is how the figure labels them.
  35. Fastmarkets, "DRC quota system looks to force uncertainty for cobalt", 14 October 2025, and "Dried-up feedstock pipeline, cobalt prices soaring", 6 January 2026; Nasdaq / Investing News, "Cobalt market 2025 year-end review", 13 January 2026; PricePedia, "Cobalt: is the price decline set to continue?", 6 October 2026. Export ban from 22 February 2025, extended twice, replaced by quotas from 16 October 2025 (about 87,000 tons a year base through 2027); metal price from about $24,000 a ton at the start of 2025 to about $53,000 at its end, above $55,000 through most of 2026, then down more than 30% in a month to below $38,500 by early October. Policy and price history, as reported; none of it is a forecast.
  36. Constantinides (Arnold Magnetic Technologies), "Thin gage silicon-iron", conference presentation, about 2010. Thin-gauge silicon iron is more expensive than thick gauge, and below about 0.13 mm punching becomes a controlled tear. A vendor's presentation with no multiplier given; the text claims only the direction.
  37. Arnold Magnetic Technologies, Arnon electrical steel product page. "Co-Fe materials are not recommended for applications over 1200 Hz," in the vendor's own words, on a page selling the alternative. The resistivities in the text are from the Carpenter 50A sheet (40.1 µΩ·cm) and the AK Steel bulletin and Sko-Die grade table (43 to 55 µΩ·cm across the M grades).
  38. thyssenkrupp Steel, powercore typical physical properties: saturation polarization 2.03 T and Curie temperature 745 °C. That page is for the grain-oriented product; the Curie point and saturation are set by the roughly 3% silicon and carry over to non-oriented grades of similar composition, which is why it is used here alongside the Birmingham table's 740 °C. None of the non-oriented data sheets I opened states a Curie temperature at all.
  39. Liu (University of Dayton Research Institute), "High-performance soft magnetic laminates for aerospace power systems", AFRL-PR-WP-TR-2005-2021, December 2004. Iron-cobalt laminates for internal starter-generators on aircraft engines, at speeds to 60,000 rpm and operating temperatures approaching 550 to 600 °C, with rotor hoop stresses of order 85 ksi; and the observation that the niobium grade showed no grain growth where the plain grade did. A research program's targets, not a production part's ratings.

Related: What is a magnet? for where iron-cobalt sits in the family, How electric motors work for where the core goes, and Reading a B‑H curve for the axes used here.

Work with Peculiar Materials

Technical and financial diligence on magnetics, magnet manufacturing, and the rare‑earth supply chain — from oxide to finished magnet.

Get in touch