Choosing a magnet

Short versionOne plot answers most of the material question: maximum energy product on the vertical, temperature on the horizontal. Every family draws a line on it, every line ends somewhere, and where it ends is usually more important than where it starts.

1 · The plotEnergy product against temperature

(BH)_max, the maximum energy product, is the largest rectangle that fits under the magnet's demagnetization curve. It is the closest thing the industry has to a single figure of merit: roughly, the work a unit volume of magnet can do. If you have not met it before, it is one of the four numbers explained on Reading a B-H curve.

Datasheets print it at 20 °C. Almost nothing runs at 20 °C. So the useful question is not "which grade has the biggest number" but "which curve is still above my requirement at my temperature, and does it stop before I do."

How the curves are drawnRemanence falls roughly linearly with temperature. Energy product goes as the square of remanence, so a grade losing 0.11% of B_r per degree loses about 0.22% of its energy product per degree. Every curve here is that square law, run from a published typical room-temperature value and stopped at the material's conventional maximum working temperature.¹

1 · The plotStart at the top

N52 is about as much energy product as anyone sells in a sintered neodymium-iron-boron (NdFeB) magnet: roughly 52 MGOe, about 414 kJ/m³. It is the grade that shows up in headlines, in holding magnets, and in every specification written by someone who has decided that more is better.

Look where it stops. Conventional practice puts a plain N-class grade at a maximum working temperature around 80 °C, and several producers rate their N50 and N52 lower still — 60 or 70 °C.² That is a warm room and a poorly ventilated enclosure. It is not a motor.

The line does not stop because the material vanishes at 81 °C. It stops because past there, in a typical magnetic circuit, the magnet starts losing output it never gets back. Which brings us to the ladder.

Short versionThe letter on a neodymium grade is a coercivity class, and coercivity is bought with heavy rare earth, and heavy rare earth costs remanence. Every rung you climb up the temperature ladder costs you energy product. That trade is the whole chart.

2 · The NdFeB ladderHighest grade in each class

N, M, H, SH, UH, EH, AH: minimum intrinsic coercivity of 12, 14, 17, 20, 25, 30 and 35 kOe respectively, which the industry maps onto nominal maximum working temperatures of 80, 100, 120, 150, 180, 200 and 230 °C.²

Plotted here is the strongest grade commonly available in each class — N52, N50M, N48H, N45SH, N42UH, N38EH, N35AH — each running from room temperature to its rated ceiling. Notice that they are not parallel: the hotter classes fall more slowly, because the same dysprosium and terbium that buy coercivity also flatten the temperature coefficient.³

Read it as: the price of 150 °C is about seven points of energy product against the best room-temperature grade. The price of 230 °C is about seventeen.

2 · The NdFeB ladderThe number that is not a material property

"Maximum working temperature" is the most widely quoted and least defensible number on a magnet datasheet. It is not a property of the material. It is the temperature at which a magnet of assumed geometry, sitting at an assumed working point, in an assumed magnetic circuit, with an assumed tolerance for loss, starts to lose flux permanently.

Change the geometry and the number changes. A long thin magnet magnetized along its length will happily run hotter than its rating; a thin disc in an open circuit will fail below it. Producers publish different ceilings for the same class — one manufacturer's chart tops out at N44SH, N40UH and N35EH where others sell N45SH, N42UH and N38EH.⁴

Treat the rating as a shipping label, not a limit of science. The real limit is where your load line crosses the knee at your hot spot, and that is the subject of the next section.

In practiceEvery serious argument I have had about a magnet specification eventually turned out to be an argument about a temperature nobody had measured. The spec says 150 °C because the last spec said 150 °C. The thermal model says 118 °C at the magnet's hottest corner under the worst duty cycle. Somebody instruments a real rotor and finds 104 °C. That gap between 150 and 104 is a whole rung of the ladder, and a rung is worth real money.

Short versionGrain boundary diffusion puts the heavy rare earth only where coercivity is actually generated — at the grain surfaces — instead of dissolving it through the whole grain. Same coercivity for a fraction of the dysprosium, or much more coercivity for the same remanence. It moved the entire frontier up and to the right.

3 · Grain boundary diffusionThe frontier moves

Coercivity in a sintered NdFeB magnet is decided at the surfaces of the grains, where reverse domains get their start. Alloying dysprosium (Dy) or terbium (Tb) into the melt hardens those surfaces — and also the grain interiors, where the heavy rare earth does nothing useful except couple antiparallel to iron and drag remanence down with it.

Grain boundary diffusion (GBD) coats the finished magnet in heavy rare earth and drives it in along the grain boundaries with a heat treatment. It started as a way to cut the dysprosium bill on existing grades. Then the industry noticed the more interesting half: at the same heavy rare earth content you get much more coercivity, at no cost in remanence, so grades that could not previously exist suddenly could.

The dashed line is what that unlocked. Grades in the neighborhood of N54SH, N52UH, N48EH and N45AH are quoted as feasible in mass production — which is roughly a full class of temperature capability for free.⁵

3 · Grain boundary diffusionWhere it runs out

GBD is a diffusion process, so it obeys diffusion's rules: it works inward from the surface, and it gets exponentially less enthusiastic with depth. Published process limits land between about 5 and 12 mm of finished thickness depending on how the heavy rare earth is applied, and that is a hard planning constraint, not a detail.⁵ A thick rotor magnet may simply not be a GBD candidate.

Two more consequences worth knowing before you specify it. The diffusion happens after the magnet is machined, so you cannot grind a GBD magnet down afterward without grinding off part of what you paid for. And the coercivity is not uniform through the thickness — the datasheet number is a bulk measurement of a deliberately non-uniform part.

Where it matters most: traction motors, robot joint actuators, and anywhere UH or EH used to be the answer.

Short versionSmCo 33E is not a stronger magnet than neodymium. It is a magnet that is barely bothered by heat, and that is still in business at 350 °C, where the entire neodymium family has been over for a hundred degrees.

4 · Samarium cobaltSmCo 33E, the 2:17 flagship

Sintered samarium cobalt tops out around 33 MGOe at room temperature in the Sm₂Co₁₇ material — call it two-thirds of N52 — and is typically rated to 300 or 350 °C depending on grade.⁶

Then look at the slope. Sm₂Co₁₇ loses about 0.03% of remanence per degree against sintered neodymium's 0.11%, nearly four times flatter.³ Its curve is almost a straight, patient decline where the neodymium curves are in a hurry.

Here is the honest part, and it is not the part the sales sheets lead with: in the range where both materials exist, neodymium usually still wins on energy. A GBD EH grade at 200 °C is comfortably above SmCo 33E at 200 °C. You do not choose samarium cobalt because it is stronger hot. You choose it because it is still there, and because of what its coercivity is doing while it gets there.

Short versionAlnico's line is nearly flat and runs off the right-hand side of the chart. It is also down at the bottom. Alnico buys temperature and stability by giving up almost all of its energy — and, more dangerously, almost all of its coercivity.

4 · AlnicoFlat, hot, and weak

Cast alnico 5 is about 5.5 MGOe and rated to 525 °C; alnico 8 and 9 trade remanence for coercivity and run to 550 °C.⁷ Its reversible temperature coefficient of remanence is around −0.02%/°C, the best of any production magnet material, and its coercivity coefficient is very slightly positive.³ Over a wide temperature swing, alnico barely notices.

That combination — poor energy, superb stability, high Curie temperature — is why alnico survived the rare-earth era in exactly the places where stability is the product: instrument magnets, sensors, magnetrons, guitar pickups, and holding applications that have been in production since before anyone had heard of neodymium.

The catch is coercivity, and it is a big catch. We will come back to it in the next section, where it is the most instructive failure mode on the page.

4 · The whole pictureWhat the chart is telling you

Ferrite joins at the bottom for completeness: about 3.5 to 4.3 MGOe, rated to 250 °C or so, and roughly an order of magnitude cheaper per kilogram than anything else here.

Four readings worth carrying away. Below about 100 °C, nothing beats neodymium and the only question is how little coercivity you can get away with. Between 100 and 230 °C, you are climbing the neodymium ladder, and GBD is how you climb it without paying full price. Above 230 °C there is no neodymium at all, at any price, and samarium cobalt owns the room. Above 350 °C there is alnico, and there is the question of whether you really need a permanent magnet.

Plot your ownThese are typical values for grade families. For real digitized curves — 130-plus grades from published catalogs, with the load line and the knee drawn at your temperature — use the field calculator.

Short versionIn a demagnetizing environment you are not buying energy product. You are buying intrinsic coercivity at the hot operating point, with the worst-case field applied, in the geometry you actually built. The room-temperature datasheet number is close to irrelevant.

5 · Demagnetizing environmentsWhere the magnet actually sits

A magnet in service does not sit at the top of its curve. It sits at an operating point set by its own geometry and its magnetic circuit, called the load line or permeance coefficient, P_c. A long magnet magnetized along its long axis has a steep load line and sits high. A thin pancake in an open circuit has a shallow one and sits far down the curve, already partly self-demagnetized before anything else happens to it.

Everything the outside world does — armature reaction from motor current, a fault or short circuit, a repelling magnet in the same assembly, a magnetizer firing next door — pushes that operating point further left. Heat pulls the knee to the right to meet it.

The ruleReversible loss comes back when you cool down. Irreversible loss does not. The dividing line is the knee: stay above it and the magnet recovers, drop below it and you have permanently sold part of the magnet you paid for.

5 · Demagnetizing environmentsWhat heat does to the knee

This is an N45SH at room temperature and the same magnet at 150 °C, its rated ceiling. Remanence has dropped about 13%, which is annoying but recoverable. Intrinsic coercivity has dropped from 20 kOe to under 6, because coercivity falls at roughly −0.55%/°C — five times faster than remanence.³

The knee has come from somewhere off the left of the plot to somewhere uncomfortably close to the load line. Add the demagnetizing field from a peak current event and the operating point steps left into it.

Which is why the specification that matters in a demagnetizing environment is not "N45SH". It is "H_k at 150 °C, at P_c = 1.2, with 400 kA/m of armature reaction applied, with less than 2% irreversible loss." Almost nobody writes it that way. Everybody should.

5 · Demagnetizing environmentsWhy samarium cobalt wins here

Same plot, SmCo 33E added at the same 150 °C. Its remanence is lower — it was always going to be — but its coercivity coefficient is about −0.20%/°C against sintered neodymium's −0.60%/°C,³ so it arrives at 150 °C having lost a quarter of its coercivity rather than three quarters. The knee is still off the left-hand edge. There is nothing to hit.

That is the actual argument for samarium cobalt, and it is a much better argument than the energy-product chart made for it. You are not buying flux. You are buying the certainty that a peak current event at the worst moment of a hot duty cycle does not permanently cost you 4% of your motor's torque constant.

In practiceI have spent years on rotors for Formula 1 energy recovery systems — surface-mount permanent magnet machines, not interior. Essentially everything interesting about that work is under NDA, so take this as the shape of the problem rather than an account of anybody's solution. The shape is this: the magnet is on the outside of the rotor, so it sees the winding's field with almost nothing in between; the duty cycle is brutal and brief; the thermal path out of a magnet spinning in its own retaining sleeve is terrible; and the failure you are designing against does not announce itself. It shows up as a machine that is quietly 3% down on torque and never recovers. The published envelope alone tells you why coercivity is the currency: the MGU-K runs between roughly 50,000 and 100,000 rpm, the MGU-H ran well over 100,000, and the 2026 regulations delete the MGU-H while raising the MGU-K from 120 to 350 kW.

5 · Demagnetizing environmentsTwo materials that fail backwards

Ferrite demagnetizes when it gets cold. Its intrinsic coercivity has a large positive temperature coefficient, around +0.27%/°C, so coercivity falls as the magnet cools and a knee climbs into the working part of the curve.³ A ferrite motor that is happy all summer can take permanent loss on a −30 °C cold start.⁸ If you are specifying ferrite, the qualification test that matters is the cold one.

Alnico has almost no coercivity at all. Alnico 5 is around 640 Oe — roughly one thirtieth of an SH-grade neodymium magnet.⁷ It can be knocked down by a stray field, by being pulled off its steel circuit, or by being stacked against another magnet the wrong way around. Alnico assemblies ship with keeper bars for a reason, and alnico designers generally magnetize the part in its final circuit.

The flip side is genuinely useful: because alnico is easy to demagnetize, it is also easy to re-magnetize, in place, without disassembly. A material with a recoverable failure mode is not the same thing as a bad material.

5 · Demagnetizing environmentsWhat to do about it

In rough order of how much they cost you:

• Measure the real hot spot before buying a temperature class. The cheapest coercivity is the coercivity you discover you did not need.
• Raise the load line: more length in the direction of magnetization, less area. Geometry is free; grade is not.
• Fix the circuit — flux barriers, bridges, keepers, a steel return path — so the magnet never sees an open circuit, including during assembly and service.
• Limit the fault, not just the duty. Size for the peak current the inverter can actually deliver into a fault, not the rated current.
• Buy a higher coercivity class. It works, and it is the most expensive line on the page, which is the subject of section 4.
• Move to samarium cobalt. Correct above about 200 °C, and often correct below it for reasons of stability rather than strength — at two to four times the material cost.

One habit worth adopting: specify the maximum permissible irreversible loss, as a percentage, at a stated temperature and applied field. It is the only requirement on this list that a supplier can actually be held to.

Short versionNeodymium magnets corrode because the phase that makes them work is the phase that rusts. The coating is not cosmetic — it is a structural part of the product, it eats your dimensional tolerance, and it fails at the edges first.

6 · Corrosive environmentsWhy NdFeB rusts and why it matters

A sintered NdFeB magnet is mostly Nd₂Fe₁₄B grains held together by a thin neodymium-rich phase at the grain boundaries. That phase is what makes the magnet magnetically hard. It is also strongly anodic relative to the grains, so in the presence of moisture it corrodes preferentially — and once the boundary phase goes, the grains simply fall out.

This is why magnet corrosion does not look like rust on a bracket. It looks like the part getting smaller and shedding dark powder. And the reaction produces hydrogen, which the alloy absorbs and which cracks it further — the same mechanism the industry uses on purpose, as hydrogen decrepitation, to break ingot into powder in the first place.⁹ Your magnet is being manufactured in reverse.

6 · Corrosive environmentsThe coating menu

Roughly in ascending order of what they survive and what they cost:

Published salt-spray and pressure-cooker numbers for these vary by an order of magnitude between vendors, and are not comparable across sources.¹⁰ Use them to rank options, never to predict service life.

6 · Corrosive environmentsWhat nobody tells you about coatings

• The coating is a dimension. Fifteen microns per face is thirty microns on the part, on a magnet whose ground tolerance might be fifty. Specify whether the drawing is before or after plating and watch people's faces.
• Edges and corners are where coatings thin and fail. A 0.3 mm chamfer buys more real corrosion life than a fancier coating does.
• Plating is a wet process on a porous part. Trapped bath chemistry and absorbed hydrogen are a real failure mode; baking after plating is not optional.
• Coatings are electrically and thermally in the way. A plated magnet in a high-frequency machine is also a shorted turn's worth of conductive shell.
• A coating is also an adhesive surface. Nickel, epoxy and parylene bond very differently, and the coating choice is often really a bonding choice.

In practiceMagnet bonding and adhesion has been a running theme of my career, and the pattern repeats: a magnet assembly fails, everyone blames the adhesive, and the adhesive was fine. The failure was at the coating-to-magnet interface, or on a surface that was passivated for corrosion and therefore excellent at not being stuck to. If a bonded joint is load-bearing, the coating is part of the joint design, and it should be chosen with the adhesive supplier in the room. Corrosion protection and adhesion pull in opposite directions more often than either supplier will volunteer.

Short versionSamarium cobalt is generally supplied bare. There is no iron-rich grain boundary phase to attack, so the usual reason for coating a rare-earth magnet does not apply.

6 · Corrosive environmentsWhy SmCo is not coated

Samarium cobalt contains relatively little iron, and cobalt is a great deal nobler than iron. Suppliers ship samarium cobalt uncoated as a matter of course, and say so in their literature.¹¹ In a corrosive environment where NdFeB would need epoxy over nickel and an argument about salt spray hours, SmCo often needs nothing.

It is still worth coating in four cases, and they are not corrosion cases: galvanic isolation when the magnet is bonded against dissimilar metal in an electrolyte; particle containment, because samarium cobalt is brittle and chips, and loose magnetic debris in a machine is its own failure mode; appearance and handling; and adhesion.

Two caveats. At sustained high temperature in air, the mechanism that matters is not aqueous corrosion but oxidation — long dwells above roughly 300 °C will work on the surface of a 2:17 magnet over thousands of hours. And ferrite does not corrode at all, being already an oxide, while alnico is usually painted or plated for appearance rather than survival.

Short versionCoercivity is the expensive axis, not remanence. And the cheapest material almost never makes the cheapest part, because for most magnets the money is in the shape, not the alloy.

7 · Cost-sensitive environmentsWhere the money actually goes

For a sintered NdFeB magnet from a large producer, raw material is the dominant line — usually the clear majority of the ex-works cost — and inside that line, the light rare earth (neodymium and praseodymium) is the bulk of the mass while the heavy rare earth is the volatility. Dysprosium and terbium are produced in tiny quantities relative to neodymium, and terbium in particular is priced accordingly.¹²

Which means the single biggest lever on magnet cost is the letter at the end of the grade, not the number in the middle. Going from N45 to N45SH costs real money. Going from N42 to N45 costs almost nothing by comparison. People spend weeks negotiating the wrong one.

Second biggest lever: the block you cut from, not the part you get. A magnet's material cost is set by the mass of sintered block consumed, including kerf and grinding loss — which is why a shape that nests badly can cost more than a higher grade.

7 · Cost-sensitive environmentsThe ferrite question

Ferrite is roughly an order of magnitude cheaper per kilogram and about a tenth the energy product. The naive conclusion — that it is a wash — is wrong in both directions, and which way it is wrong depends on what you are building.

• Where volume and mass are cheap, ferrite wins outright and is not a compromise: loudspeakers, holding, separation, sensors, many pumps, and an increasing number of industrial motors.
• Ferrite gets better at resisting demagnetization as it heats up, which is a genuine engineering advantage and not a consolation prize.
• You need roughly ten times the magnet volume for the same energy, and that volume propagates: bigger rotor, more steel, more copper, more housing, more mass to accelerate. The magnet line goes down and the system line often goes up.
• Ferrite's cost is mostly processing and tooling rather than material, so the savings only appear at volume. In low quantities the tooling amortization can make ferrite the expensive option.
• Check the cold case. See the previous section.

And one thing worth saying plainly, because it is the most common wrong reason to choose ferrite: it is not a way around a Chinese supply chain. Ferrite and alnico are also overwhelmingly made in China, on cost. Alnico in particular is not a cheap material — it is largely cobalt and nickel, and it prices like it.

7 · Cost-sensitive environmentsThe list I actually use

• Stop over-specifying temperature class. Measure, then buy one rung of margin, not three.
• Ask for a GBD grade wherever the part is thin enough. Same coercivity, less heavy rare earth, and the price difference has been moving the right way.
• Design the shape for the block. Rectangular tiles cut from a plaque, with the direction of magnetization sensible, beat a clever arc that wastes half its block.
• Loosen the tolerances that do not do anything. Ground-all-over on six faces, when two faces set the gap, is a large and entirely optional cost.
• Coat for the environment you have, not the one in the standard. Nickel plating a part that lives in dry nitrogen is free money left on the table.
• Do not chase the last two points of energy product. N52 over N48 buys about 8% of flux for a grade with less thermal headroom and fewer qualified suppliers.
• Do not let a single-source grade into the design unless you meant to. A grade that only one producer makes is a commercial decision disguised as an engineering one.

In practiceWhen I run technical and financial diligence on a magnet program, the fastest way to find the money is to lay the grade specification next to the measured thermal data and the actual duty cycle. In most programs, something in that stack is inherited from a previous design nobody has revisited. The magnet is rarely the problem. The specification usually is.