How a thin skin of dysprosium or terbium, soaked in along the boundaries between grains, let the magnet industry cut its heavy rare earth use roughly in half and sell grades that did not exist before. Told only from patents, papers and catalogs that anyone can read.
Short versionPut dysprosium or terbium all through a neodymium magnet and you buy coercivity with remanence. Put it only in a thin shell around each grain and you keep nearly all the remanence and get most of the coercivity. The idea was in print by 2000 and made practical by Shin-Etsu by 2005, but it mostly sat on the shelf until dysprosium's 2011 price spike made an extra furnace step worth paying for. The public record now covers how to coat the magnet, what to coat it with and how the heat treatment works; it does NOT cover the recipes companies actually run in production. The payoff: roughly half the heavy rare earth for a given grade, new grades above and to the right of the old ones, and a hard practical limit of a few millimeters of depth.
Peculiar Materials LLC · Every fact on this page is footnoted to a public document, except the first-person asides marked In practice, which are my own recollection. The microstructure drawings are schematic; the charts are drawn from the cited patents, papers and catalogs, and where I have done arithmetic or drawn a curve between published points, the footnote says so. Disclosure: I worked at MP Materials (it is on my LinkedIn), and I have commercial relationships with companies named on this page. Nothing on this page comes from that work.
What this page skips, so nobody mistakes it for complete: process temperatures, times and coating weights (many of the cited patents publish theirs; I deliberately do not repeat them); hot-deformed and bonded magnets, except where a source forces the comparison; Chinese-language patent filings, which I did not sweep; and the cost of running a diffusion line.
Short versionMost of the production know-how for this process is unpublished, and the process is the subject of a pending lawsuit. So this page uses only what anyone can read, and footnotes every bit of it.
1 · Ground rulesA closely held process
Grain boundary diffusion (GBD) is a finishing step for sintered neodymium-iron-boron (NdFeB) magnets: coat the finished magnet with a dysprosium or terbium compound, heat it, and let the heavy rare earth soak in along the boundaries between the grains. The current IP landscape is a veritable thicket of patents and trade secrets.
It is also closely held. In May 2026 MP Materials sued USA Rare Earth, a former MP engineer and FOM Technologies in the Texas Business Court over grain boundary diffusion.1 USA Rare Earth has denied the claims.2 The case is pending.2
Because this is an area of active, and competitive research, I am going to bend over backwards here to make sure I do not make a single claim that is not supported by publicly available information, and I will cite it inline. There is thankfully still a lot of public information available, which will support a useful discussion, but I will not disclose any technology, or render opinions here.
1 · Ground rulesHow this page works
So the rule here is strict. Every fact traces to a public document (a patent, a peer-reviewed paper, a company catalog or press release, or published news reporting), and every one is footnoted. Where a company has described its approach in a patent, I name the company and the patent. Where I could find nothing public, I say so and stop.
Patents are the gift in all this. A company that patents a coating method has to explain it to get the patent, so the patent record is the one place the industry is obliged to show its work. What patents do NOT show is which of the options described actually runs in a factory, and at what cost. Keep that in mind for the whole page.
In practiceI have spent a career reducing oxides to metal, casting it into alloy, pulverizing, pressing, sintering, grinding, gluing and testing magnets, and that career includes MP Materials. Essentially everything I know about how any particular company runs this process is under NDA, so take this page as the shape of the problem, assembled from the public record, rather than an account of anybody's solution. The gaps are deliberate.
Short versionHeavy rare earth in the middle of a grain costs you remanence and does little for coercivity. Heavy rare earth at the surface of the grain buys almost all the coercivity for almost none of the cost. That asymmetry is the whole trick.
2 · HistoryTwo places to put dysprosium
A sintered magnet is millions of Nd₂Fe₁₄B grains, a few micrometers across, held together by a thin neodymium-rich grain boundary phase. Coercivity, the resistance to being demagnetized, is decided mostly at the grain surfaces, because that is where reversal starts.
Dysprosium and terbium raise the magnetic anisotropy of the crystal they sit in, which is what you want. The catch is that their moments point opposite to iron's, so every atom that replaces neodymium subtracts magnetization. Hitachi Metals' patent says the heavy rare earth and iron moments have "mutually opposite directions"3, and Shin-Etsu's calls the remanence loss unavoidable when dysprosium is alloyed in.4
For decades the way to make a high-temperature grade was to melt the dysprosium into the alloy, which puts it in every grain, all the way through, including the middle where it earns very little. Grain boundary diffusion puts it in a thin outer layer of each grain instead: a core–shell structure, with a dysprosium- or terbium-rich shell around an ordinary Nd₂Fe₁₄B core.56
2 · HistoryThe comparison that sold it
The cleanest public before-and-after is in Shin-Etsu's original patent. A base magnet measured 1,000 kA/m intrinsic coercivity (HcJ) at a remanence (Br) of 1.420 T. Diffused with dysprosium fluoride, it reached 1,500 kA/m at 1.415 T. A magnet with dysprosium alloyed in to reach the same 1,500 kA/m came in at 1.370 T. Terbium fluoride, on a larger magnet of the same base alloy, reached 1,800 kA/m at 1.415 T.7
Five millitesla of remanence lost versus fifty, for the same coercivity gain: about a tenth of the penalty. A second Shin-Etsu patent shows the same tenfold ratio.4 A 2022 study measured the efficiency directly and got roughly 13.5 kOe of coercivity per weight percent of dysprosium when diffused, against 3.8 kOe per weight percent when added to the bulk (on a yttrium-bearing magnet rather than a standard one, so treat the ratio as indicative).8
Short versionIn print by 2000, practical by 2005, and mostly ignored until dysprosium got expensive in 2011.
2 · HistoryFirst on paper
The earliest work that later papers and patents cite is Park, Hiraga and Sagawa at the Sendai rare earth magnet workshop in 2000: dysprosium sputtered onto thin sintered magnets, then heat treated, which roughly doubled coercivity.9 Sagawa is the same Masato Sagawa whose group published the sintered NdFeB magnet in 1984.10 Osaka University's Machida group was sputtering dysprosium and terbium onto small magnets by 2004, and its own patent concedes the route has "a mass productivity problem."11
Shin-Etsu's contribution was to skip the vacuum chamber: coat the magnet with an oxide or fluoride powder and heat it below the sintering temperature. Its first filing has a priority date of 19 October 200412, and the journal papers followed in 2005 and 2006, the second of which carries in its title the name the process still goes by, the grain boundary diffusion process.135
2 · HistoryThen dysprosium got expensive
For most of that decade dysprosium was cheap: Arnold puts the 2003 low at $28.50 a kilogram.14 Spending a coating step, a long furnace cycle and a size restriction to save on a cheap element was a hard sell, and one industry columnist recalls the technique being met with a shrug at first.15
Then 2010 and 2011 happened. U.S. Geological Survey (USGS) tables put dysprosium oxide at about $170 a kilogram in 2009, $310 in 2010 and $1,600 at the end of 2011, after a mid-year spike, before easing to $1,010 in 2012.16 Japanese government figures quoted by the Congressional Research Service have dysprosium metal going from about $250 in April 2010 to $2,840 in July 2011.17 (Every source picks a different peak.18)
That changed the question from "does it work?" to "can it be made cheap?", and the patent record from 2010 onward is mostly about the second question. By 2015 Arnold could write that diffusion was halving the dysprosium in high-coercivity grades across the industry.14
Short versionEvery method is a way to put a controlled dose of heavy rare earth on the surface without wasting it. There are three families: wet coatings, dry contact and vapor.
3 · ApplicationGetting it onto the surface
The basic recipe from Shin-Etsu's 2004 filing is simple: get the compound onto the surface, then heat below the sintering temperature.12 The arguments since have been about the first half.
Wet coatings. Dipping the magnet in a powder suspension was Shin-Etsu's original route19, and remains the most common one in the patents I read. Shin-Etsu later patented electrodeposition, where the magnet is an electrode and the voltage sets the coating weight.19 Ningbo Yunsheng sprays a slurry20, Yantai Shougang (now Yantai Dongxing) uses plasma spraying21, Intermetallics (now Daido Steel) screen-prints22 and TDK paints on a binder paste.23
Dry contact. Hitachi Metals (now Proterial) tumbles magnets with a loose heavy rare earth source and ceramic stirring media in a rotating heated chamber; a related patent buries the magnets in coarse source particles that separate cleanly afterward and can be reused.24 Intermetallics laid down a sticky film and dusted powder onto it ("barrel painting")22, and Yantai Shougang presses powder onto the magnet with tape.21
Vapor. Hitachi Metals and ULVAC heat a dysprosium or terbium source near the magnets, without touching them, so the metal evaporates, lands and diffuses in before a film can build up: evaporation diffusion.2425 Sputtering a film directly is the oldest route of all, and Hitachi Metals and Toyota both have patents that use it as the worked example.2426
3 · ApplicationThe engineering is in the boring parts
Read in date order, the patents tell a scale-up story. Shin-Etsu went from a lab dip in a sonicated alcohol suspension (2005) to electrodeposition (2012 to 2014), to a family of production coaters filed in 2015 (walking beam, pocketed drum, net belt, sponge roll), to room-temperature aerosol deposition (2021).19 Hitachi Metals moved away from wet coatings altogether, first to vapor and then to coarse, reusable alloy particles.24
One surprise: sputtering, which is what most people picture, is rarely the worked example in the granted U.S. patents I read. It shows up mostly as a listed alternative, or as the prior art a patent says it is improving on.27
In practiceMagnet bonding and adhesion was a focus of mine for many years, and this part of the literature covers familiar ground. Many of these patents deal with making a layer stay where it is put, at the intended thickness, through drying and handling, without bonding the parts to each other in the furnace. Daido's silicone grease to stop the coating peeling, and Vacuumschmelze's solid separator particles to stop stacked magnets welding together, address the same kind of problem from two sides.2228
3 · ApplicationWho published which method
The matrix sorts the patents by company and by method. A few of these patents, read in full, describe something other than what they are often taken to describe. JL Mag's U.S. patent describes a suspension coating with soaking preferred, not sputtering.29 Beijing Zhong Ke San Huan dips a slurry onto a block that is only partly sintered, while it is still porous.30 Advanced Technology & Materials (AT&M) calls its terbium-copper plates "targets," but they evaporate rather than sputter.31 Vacuumschmelze (VAC) published two paste applications and then abandoned both.28
Noveon's patents use the similar-sounding phrase "grain boundary engineering," but they blend an additive into the powder before sintering, which is a different process, covered in the two-powder section below.32
The last column is localized diffusion: putting the heavy rare earth only where the motor tries hardest to demagnetize the magnet. Honda and TDK reported about 30% less dysprosium that way with no loss of motor performance33; Shin-Etsu, TDK, Hyundai and Denso all have patents on coating only some faces or regions.19233435
Short versionTerbium buys the most coercivity per atom. Dysprosium is cheaper and more plentiful. Holmium, on the numbers, has little to add. Praseodymium shows up as a carrier.
4 · DiffusantWhich element
The diffusant is whatever you coat the magnet with, and the first choice is the element. What matters is the anisotropy field of the compound each rare earth forms with iron and boron: the field it would take to swing the magnetization off its preferred axis. It is about 6.7 T for Nd₂Fe₁₄B, roughly 15 T for the dysprosium version and 22 T for terbium's, while the saturation polarization drops from about 1.6 T to about 0.7 T for both heavies.363738
The measurements follow. On the same base magnet with the same hydride form, terbium took coercivity from 14.65 kOe to 25.62 kOe where dysprosium reached 18.90 kOe, with remanence nearly untouched by either.39 Swapping dysprosium for terbium in a Hitachi Metals alloy source raised the gain by about 70%, from 450 to 760 kA/m.40 Dysprosium persists because it is more abundant and cheaper.37 Which one travels deeper is genuinely disputed: one study found dysprosium went further, another that terbium diffuses faster.3941
Holmium's iron-boron compound has an anisotropy field only slightly above neodymium's.36 Shin-Etsu's rule, that any rare earth with higher anisotropy than neodymium helps42, says it should help, though the gap in anisotropy field is small. It is described as relatively low-priced43, and it appears in diffusion patents only as one permitted element in a list.44 I found no peer-reviewed paper that uses holmium itself as the diffusion source.
Short versionThe compound matters nearly as much as the element. Fluorides beat oxides, hydrides and low-melting alloys beat fluorides in some studies, and whatever rides along with the heavy rare earth changes the grain boundary.
4 · DiffusantWhich compound
Shin-Etsu's original patent ran the comparison on magnets of one base composition: terbium fluoride added 800 kA/m of coercivity, dysprosium fluoride 500, dysprosium oxyfluoride 470 and dysprosium oxide 400, each for 0 to 5 mT of remanence. Alloying the dysprosium in instead cost 40 to 50 mT.12
The passengers matter. Fluorine absorbed with the rare earth ends up as neodymium oxyfluoride at the grain boundaries45, and oxides already in the magnet trap arriving dysprosium.46 In one dip-coating study, more hydride in the coating meant more coercivity and deeper dysprosium than fluoride gave.47 TDK's patent found dysprosium-iron and terbium-iron compounds beat dysprosium fluoride on the same base.23 Low-melting terbium-aluminum-copper alloys reached about the same gain with roughly 20% less terbium48, and two separate sources (one praseodymium-based to open the boundaries, one carrying terbium) beat one premixed alloy.49
There is also a heavy-free branch: a neodymium-copper eutectic that raised coercivity with no dysprosium at all, shown on powder by the National Institute for Materials Science (NIMS) in 201050, and Toyota's neodymium-praseodymium-copper alloys, on hot-deformed magnets.51 They raise coercivity less than terbium does, but they need no heavy rare earth at all.
In practiceI came to materials science as a chemistry student, so this is the part of the literature I enjoy most. The heavy rare earth never travels alone. The fluorine has to end up somewhere, the oxygen already in the magnet is waiting to trap whatever arrives, and hydrogen tidies up on its way out. Many of the papers comparing diffusant forms end up discussing what happens to the elements that travel with the heavy rare earth.
Short versionDiffusion runs fast along the molten grain boundary and slowly into the grain. You want plenty of the first and only a little of the second.
5 · Heat treatmentTwo speeds
During treatment the neodymium-rich grain boundary phase melts, and the heavy rare earth moves through that liquid with relative ease.3752 Where the liquid touches a grain, heavy rare earth trades places with neodymium in a thin outer layer, and the neodymium moves out into the boundary. That swap is what makes the shell.37
Both halves of the result matter. The National Institute for Materials Science group found the coercivity gain needs the heavy-rare-earth shell AND an intact neodymium-rich layer between grains, and that diffusion into the grain stays near the boundaries.6 A Fraunhofer and TU Darmstadt study argues the shells form by melting and re-solidification near the surface rather than by solid-state diffusion.53 In a 7 mm terbium-treated magnet, one group mapped three zones: thick shells just under the surface, thin shells and continuous boundaries a little deeper, and terbium mostly stranded at triple junctions further in.54
Short versionToo little and the heavy rare earth stays bunched under the surface. Too hot or too long and it floods the grains while the grains grow. The useful window is narrow, and nobody publishes where they actually run.
5 · Heat treatmentThe window
Hotter drives the heavy rare earth deeper, up to a point. Constantinides warns that too high a temperature promotes grain growth.37 In one terbium-nickel series the gain rose with temperature and then reversed, as grains coarsened and, near the surface, turned terbium-rich all the way through, and squareness suffered.55 In a terbium hydride study, at a fixed time the hottest condition gave the most coercivity but lost remanence, because the shells thickened.56 This is the bulk-alloy penalty creeping back in. At the other end, heavy rare earth that never gets far stays in a steep gradient just under the coated face; even a two-step process left it concentrated within about a millimeter in one study.5457
A second, lower-temperature anneal follows, to reshape the grain boundary phase into continuous layers that separate grains magnetically.5212 In magnets that already contain dysprosium, a Toyota and NIMS group found much of the gain only arrives after that anneal.58 Daido and NIMS added a second diffusion step with a neodymium-copper "carrier" that pushes heavy rare earth inward, which improved squareness, the measure of how sharply the demagnetization curve holds before it collapses (in hot-deformed magnets).57
Direction matters too. Diffusing through the pole faces, along the magnetization axis, works better than diffusing across it, though the papers disagree on why.596061
Short versionThe dose falls off steeply with depth. The practical depth is a few millimeters, which caps the thinnest dimension of a fully treated magnet at roughly six.
5 · Heat treatmentHow deep
Constantinides gives a worked example: a 6 mm magnet with a base coercivity of 1,790 kA/m reached 2,270 kA/m at each face and 2,010 kA/m at 3 mm, the center.37 He puts the practical depth at 3 to 4 mm from each surface and the thinnest dimension at about 6.5 mm.37 Arnold's catalog says its process favors magnets under 6 mm thick, or laminated stacks.62 Shin-Etsu showed dysprosium reaching all the way through a 2 mm magnet in 2007.42
Lab papers often see visible shells disappear within a millimeter. That is less a contradiction than a definition problem: "depth" can mean where the shell vanishes, where the element is still detectable, or where coercivity is still raised.63
The consequence designers live with: a diffused magnet is not uniform. General Motors' engineers describe corners and edges enriched in heavy rare earth and very little reaching the center, which complicates getting representative demagnetization curves for motor design.64 Hitachi Metals' catalog says plainly that the coercivity distribution "is not homogeneous,"65 and Shin-Etsu warns its diffused grades vary with size and shape.66 Shin-Etsu even patented cutting slits into a magnet to add surface for the coating.4
Short versionBlend a dysprosium-rich powder into a heavy-free one before pressing, and the shells form during sintering, all the way through the magnet. No thickness limit and no extra furnace step, but less coercivity per gram of dysprosium than diffusion.
6 · Two-powder methodThe alternative: do it in the powder
Before diffusion, the way to put heavy rare earth near the grain boundaries was to do it in the powder. Shin-Etsu describes its own earlier "two-alloy method": a main alloy with no heavy rare earth, blended with a low-melting sintering aid that carries the heavy rare earth, so it concentrates near the grain boundaries as the magnet sinters.67
The current literature calls the same idea the two-powder method: a coarser, heavy-free main-phase powder mixed with a finer dysprosium-bearing powder, which forms core–shell grains during sintering, with no separate diffusion step.68 In a 2024 TU Darmstadt study, the dysprosium ended up in the outer region of the grains; coercivity rose by roughly 120 kA/m per added weight percent of dysprosium, while remanence fell only from about 1.37 T to 1.32 T across 0 to 3 wt%.68
The advantage is geometry. The same paper notes that diffusion can be applied only to thin magnets, while the two-powder method can be applied to any magnet dimension.68 Yantai Zhenghai's U.S. patent, which blends terbium, dysprosium or holmium powder into the alloy powder before pressing, makes the same argument: it describes surface diffusion as limited to thin pieces and its own method as free of any size restriction.69
6 · Two-powder methodWhat it costs, and where it stands
The shells form while the grains are solid, and they come out thicker: about 0.4 µm at 1 wt% dysprosium, growing to about 2.4 µm at 3 wt%, which the authors say means heavy rare earth use is not yet optimal. They describe grain boundary diffusion as the more efficient route.68
Two sources put numbers on the gap. On one base magnet, a 2022 study measured about 3.8 kOe of coercivity per weight percent of dysprosium added as a hydride powder before sintering, against about 13.5 kOe per weight percent diffused from the surface.8 Shin-Etsu reports about 30% more coercivity from its diffusion process than from its two-alloy method, with no reduction in remanence.67
On status, the 2024 study reports demonstration in batches of about 20 kg starting from strip casting, not production.68 Noveon's "grain boundary engineering" patents blend an additive alloy into the powder before sintering, a variant of the same idea.32 The two routes can also be stacked: one group made a magnet by the double-alloy method and then diffused terbium into it, reporting a two-layer terbium and dysprosium shell and a 7.6 kOe coercivity gain with a slight remanence loss.70
Short versionFor a given grade, about half the heavy rare earth. The published reductions run from about 30% to about two-thirds.
7 · OutcomesSame grade, less dysprosium
Arnold's 2015 table gives typical dysprosium content by grade suffix under the old, alloyed approach: 1.4% for M, 2.8% for H, 4.2% for SH, 6.5% for UH and 8.5% to 11% for EH and AH. With diffusion, Arnold wrote, SH grades were coming down from 4%-plus to about 2%, and the 8% to 11% grades to 4% to 5%.14 That is a halving, and it matches the roughly 50% one trade columnist described as the original goal.15
The other public numbers cluster around it. General Motors' engineers describe diffused magnets using about one-half, and in some cases one-third, of the dysprosium or terbium of conventional ones.64 Nissan said the motor in its updated LEAF, launched in November 2012, used 40% less dysprosium than conventional EV motors, thanks to a grain boundary process.71 Honda and TDK reported about 30% from localized diffusion.33
On volume, JL Mag's 2021 Hong Kong prospectus reports 4,111 metric tons of magnets made with GBD in 2020, and cites Frost & Sullivan for the claims that it held about 21% of the world GBD magnet market and that fewer than ten magnet makers worldwide apply the process.72 Note that none of these baselines is the same "conventional" magnet, so do not average them.73
Short versionAt the same coercivity, about a tenth of a tesla more remanence; and coercivity classes the old catalogs never offered.
7 · OutcomesGrades that did not exist before
The fairest comparison is one company's two catalogs, same revision, side by side. In Arnold's, the best conventional 20 kOe grade (N45SH) is 1,350 mT typical; the best diffused grade at the same class (G52SH) is 1,440 mT. At 25 kOe it is 1,310 mT against 1,430; at 30 kOe, 1,270 against 1,385; at 34 kOe, 1,195 against 1,270. The diffused catalog also lists 35 and 37 kOe grades with no conventional counterpart at all.7462
Shin-Etsu's current lineup tells the same story another way: its diffused N54SH-HGF carries 1.43 to 1.49 T at 1,671 kA/m, while its conventional grade at that coercivity, N36SA, carries 1.15 to 1.22 T.66 Constantinides puts it in one sentence: the diffused grades sit above and to the right of the standard ones.37
In practiceGrade names are not directly comparable between makers. When I was running Arnold's plant in Lupfig we developed Recoma 33E for motorsport. By our own numbers it was a 32E, but we judged the "32" grades then on the market to sit closer to 31, so we called ours 33. That is why the chart compares one company's typical values rather than names from different makers.
Short versionThe physics and the menu of methods are public. The recipes that make it work cheaply at volume are not.
7 · OutcomesWhat the public record does not say
Put the whole record together and you can explain how grain boundary diffusion works, who patented which coating, which element and compound does what, why the furnace window is narrow and why the magnet has to be thin. You cannot learn how much of the coating ends up in the magnet rather than on the furnace furniture, what the yield is, how long a batch takes, what a kilogram costs, or which of the patented options a given plant actually runs.
Those are also the details a statement that a supplier "has GBD" leaves open: the thickness it works to, which faces it diffuses through, how uniform the result is, and what it costs. The public record can show whether an answer is consistent with the published work. It cannot confirm it.
A sintered magnet, up close1 / 15
References
Every external fact is cited; general magnet physics is not. Patent descriptions are paraphrased from the patents' own text on Google Patents or the European Patent Office; where I describe what a patent "describes," I have not checked every claim's exact wording, and a description is not a claim. Process temperatures, times and coating weights published in these sources are deliberately not repeated.
Rare Earth Exchanges, "America's rare earth rivalry moves from the mine to the courtroom", 27 May 2026. Supports the court (Texas Business Court), the filing date (22 May 2026), the three defendants (USA Rare Earth, a former MP engineer, FOM Technologies) and that the dispute concerns grain boundary diffusion. News reporting of a complaint, not the complaint itself; I have not read the petition.
Hitachi Metals (now Proterial), US 8,187,392 B2, "R-Fe-B type rare earth sintered magnet and process for production of the same," priority 2 July 2007. The quoted phrase about the heavy rare earth and iron moments, and the statement that remanence falls in proportion to the light rare earth (neodymium or praseodymium) replaced. The patent also describes a vapor-delivered, heavy-rare-earth-enriched outer layer on the grains.
Shin-Etsu Chemical, US 7,559,996 B2, "Rare earth permanent magnet, making method, and permanent magnet rotary machine," priority 22 July 2005. States that alloying dysprosium or terbium in makes a remanence loss unavoidable; compares a diffusion gain of about 400 kA/m for a 5 mT remanence loss with an alloyed design costing about 50 mT; and describes cutting slits into the magnet surface to add area for the coating. The patent does not say whether the alloyed figure was measured or a design target.
K. Hirota, H. Nakamura, T. Minowa, M. Honshima (Shin-Etsu), "Coercivity enhancement by the grain boundary diffusion process to Nd–Fe–B sintered magnets," IEEE Transactions on Magnetics 42(10), 2909–2911 (2006), doi:10.1109/TMAG.2006.879906. Abstract only, read through the OpenAlex index (the full text is paywalled): oxide or fluoride powders of dysprosium and terbium on magnets several millimeters thick, coercivity gains above 400 kA/m with little remanence loss, and heavy-rare-earth-rich shells formed by substitution for neodymium. Its title carries the name the process still goes by.
H. Sepehri-Amin, T. Ohkubo, K. Hono (NIMS), "The mechanism of coercivity enhancement by the grain boundary diffusion process of Nd–Fe–B sintered magnets," Acta Materialia 61(6), 1982–1990 (2013), abstract page, doi:10.1016/j.actamat.2012.12.018. Abstract and introduction only. Dysprosium-vapor diffusion; reports a (Nd,Dy)₂Fe₁₄B shell, that both the shell and the neodymium-rich grain boundary layer are needed for the gain, and that diffusion into the grain is limited to the region near the boundaries.
Shin-Etsu Chemical, US 8,377,233 B2, "Preparation of rare earth permanent magnet material," divisional of US 8,211,327 (priority 19 October 2004). Source of the comparison in figure 3: untreated 1.420 T / 1,000 kA/m; dysprosium fluoride diffused 1.415 T / 1,500 kA/m; dysprosium alloyed 1.370 T / 1,500 kA/m; terbium fluoride 1.415 T / 1,800 kA/m. The dysprosium examples are very small magnets (4 × 4 × 0.5 mm), the best case for diffusion; the terbium example is a larger magnet (10 × 20 × 1.5 mm) of the same base alloy, and the patent does not say how much dysprosium the alloyed magnet held or how much the diffused one absorbed. Also cites Park 2000 and Machida 2004 as prior art.
S. Guo et al., "Comparisons of Dy utilization efficiency by DyHx grain boundary addition and surface diffusion methods in Nd-Y-Fe-B sintered magnet," Materials 15(17), 5964 (2022), open access. About 0.42 wt% dysprosium by surface diffusion gave about the same coercivity as about 1.6 wt% added to the grain boundaries before sintering (15.15 vs 15.51 kOe), with less remanence lost; the authors report about 13.5 versus 3.8 kOe per wt% dysprosium. The base magnet contains yttrium and the "addition" route is a powder blend rather than true alloying, so the ratio is indicative, not general.
K. T. Park, K. Hiraga, M. Sagawa, "Effect of metal-coating and consecutive heat treatment on coercivity of thin Nd-Fe-B sintered magnets," in Proceedings of the 16th International Workshop on Rare-Earth Magnets and Their Applications, Sendai, pp. 257–264 (2000). I have NOT seen this paper. The citation is as given in Shin-Etsu's US 8,377,233; a 2010 Materials Science Forum paper cites the same pages (with a typo in the workshop number). The description of its content (a sputtered dysprosium film on thin magnets, roughly doubled coercivity, "the genesis" of the process) comes from Z. Liu, J. He, R. V. Ramanujan, "Significant progress of grain boundary diffusion process for cost-effective rare earth permanent magnets: a review," Materials & Design 209, 110004 (2021), open access.
M. Sagawa, S. Fujimura, N. Togawa, H. Yamamoto, Y. Matsuura, "New material for permanent magnets on a base of Nd and Fe," Journal of Applied Physics 55(6), 2083–2087 (1984), doi:10.1063/1.333572. Cited only to identify Sagawa; abstract seen.
Japan Science and Technology Agency and Osaka University (inventors K. Machida, S. Suzuki), EP 1 843 360 A1, "Nd-Fe-B magnet with modified grain boundary and process for producing the same," priority 16 December 2004; U.S. counterpart US 7,824,506 B2. Describes the sputtering route's mass productivity problem (quoted) and instead buries the magnet in a heavy rare earth compound with a calcium hydride reductant, so the metal is formed in place. Also cites the group's 2004 conference work, which I have not seen.
Shin-Etsu Chemical (inventors H. Nakamura, K. Hirota, T. Minowa), US 8,211,327 B2, "Preparation of rare earth permanent magnet material," priority 19 October 2004 (PCT WO 2006/043348). The founding powder-coating patent: an oxide, fluoride or oxyfluoride powder on the surface, heat treated at or below the sintering temperature, followed by a lower-temperature aging step. Its examples give the compound comparison in figure 8 (terbium fluoride +800, dysprosium fluoride +500, dysprosium oxyfluoride +470, dysprosium oxide +400 kA/m, the oxide result on a second base magnet with the same starting properties and some examples on different geometries, each for 0 to 5 mT of remanence, versus 40 to 50 mT for alloyed dysprosium). It prefers small magnets and notes that oxide films from water-based machining coolant block absorption. Temperatures and times in it are not repeated here.
H. Nakamura, K. Hirota, M. Shimao, T. Minowa, M. Honshima (Shin-Etsu), "Magnetic properties of extremely small Nd-Fe-B sintered magnets," IEEE Transactions on Magnetics 41(10), 3844–3846 (2005), doi:10.1109/TMAG.2005.854874. Abstract only, read through the OpenAlex index: coating small magnets with Dy₂O₃, DyF₃ or TbF₃ powder and heat treating raised coercivity greatly without significant remanence loss. Often cited as the first journal publication of the powder route.
S. Constantinides, "The important role of dysprosium in modern permanent magnets", Arnold Magnetic Technologies, rev. 150903a (September 2015). Supports the $28.50/kg 2003 low and $3,410/kg early-2011 peak for dysprosium metal (p. 4); Table 1 of typical dysprosium content by grade suffix across the industry, which the caption says reflects 1985 to 2011 practice (not Arnold's own grades); and the statement that diffusion was reducing heavy rare earth content from 4%-plus to about 2% in SH grades and from 8–11% to 4–5% in the highest grades (p. 2). He calls it dysprosium or heavy rare earth diffusion. Disclosure: I worked at Arnold, and have co-authored papers with Constantinides.
S. Trout, "Dysprosium 2.0", Magnetics Magazine, 6 March 2013. A trade column, not peer reviewed. Dates the technique to about 2005, says it was first met with a shrug and adopted as dysprosium prices rose, and describes the aim as cutting dysprosium use by roughly 50%.
U.S. Geological Survey, Minerals Yearbook: Rare Earths, Table 3 in each: 2009, 2010, 2011, 2012. Dysprosium oxide, 99%, US$/kg: about 170 (2009, estimate), 310 (2010, estimate), 1,600 (year-end 2011), 1,010 (2012). The price source changes from Rhodia to Metal-Pages in 2012, so the series is not perfectly consistent. The 2011 chapter describes the mid-year spike.
M. Humphries, Rare Earth Elements: The Global Supply Chain, Congressional Research Service R41347, 8 June 2012. Dysprosium metal about $250/kg in April 2010 and $2,840/kg in July 2011, figures attributed to Japan's Ministry of Economy, Trade and Industry (METI).
My note on the price peak. The sources do not agree on how high dysprosium went: $2,840/kg for metal in July 2011 (METI via CRS), $3,350/kg (J. M. D. Coey, "Perspective and prospects for rare earth permanent magnets," Engineering 6(2), 118–130 (2020), open access), and $3,410/kg in "early 2011" (Arnold 2015). The USGS oxide series is annual and misses the peak entirely. They differ in metal versus oxide, export versus domestic China price, and timing. Treat any single peak number, including mine, as roughly right and not exact; the order of magnitude of the jump is the point.
Shin-Etsu Chemical coating patents, each on Google Patents. Dip in a powder suspension: US 7,488,393 and US 7,520,941 (both priority 23 March 2005; granted 10 February and 21 April 2009). Electrodeposition: US 10,179,955 and US 10,181,377 (priority 31 August 2012). Partial-immersion electrodeposition, coating chosen regions only: US 9,845,545 (priority 19 February 2014). Production coaters (walking beam, pocketed drum, net belt, rotating jig, sponge roll): US 10,790,076, US 10,861,645, US 11,424,072, US 10,916,372, US 11,224,890 (all priority 28 April 2015). Aerosol deposition at room temperature: US 11,915,845 (priority 18 August 2021). A rotor patent listing sputtering as an alternative to powder coating and treating only some faces: US 8,638,017 (priority 18 September 2009). Descriptions are mine, from the patents' text; claim wording not checked on every one.
Ningbo Yunsheng, US 10,741,326 B2, "Method for improvement of magnetic performance of sintered NdFeB lamellar magnet," priority 25 December 2015. Describes spraying a heavy rare earth powder slurry onto the magnet.
Yantai Shougang Magnetic Materials (now Yantai Dongxing Magnetic Materials) patents: US 11,107,627 (plasma spraying, priority 30 December 2017); US 11,270,839 and US 11,315,728 (powder held by tape or adhesive and pressed on, priority 20 July 2018); US 11,948,734 (a laser-drilled organic film used as a stencil to meter the powder, priority 29 November 2019); EP 4 287 220 A1 (lists spray, dip, screen print, sputtering or evaporation, priority 31 May 2022). The most varied set of application methods I found from a single company.
Intermetallics (now Daido Steel) patents: screen printing, US 9,884,368 and EP 2 565 885 (priority 27 April 2010); a silicone-grease paste meant to stop the coating peeling during heating, US 10,475,561 (priority 18 March 2013); "barrel painting," a sticky layer and then powder, applied by vibrating the magnets in a container with coated media, US 8,562,756 (priority 11 January 2008); pole faces coated with other sides masked, US 9,589,714 (priority 10 July 2009). Ownership transfers as shown on Google Patents, which labels current assignee as an assumption, not a legal conclusion.
TDK patents: dysprosium-iron and terbium-iron compound slurry, US 8,394,450 (priority 29 May 2008), whose examples beat a dysprosium fluoride comparison on the same base magnet; hydride slurry, US 9,548,157 (dates not displayed on the page I read); binder paste, US 11,404,207 (priority 3 December 2018); one or both main faces only, sides uncoated, US 10,530,198 (priority 11 September 2015); an extra coat on one end, US 12,074,475 (priority 25 March 2020). US 10,530,198 lists vapor deposition and sputtering as alternatives.
Hitachi Metals (now Proterial) patents. Evaporation diffusion from a separate heavy rare earth body facing the magnets: US 8,206,516 (priority 3 March 2006) and plates stacked with magnets, US 9,721,724 (priority 30 September 2010). A sputtered dysprosium film over a metal underlayer as the worked example: US 8,038,807 (priority 31 January 2006). Tumbling with a loose source and stirring media in a rotating chamber: US 9,293,252 (priority 30 September 2010) and US 9,613,748 (priority 27 June 2011; the source can be used over and over). Magnets buried in coarse source particles in a box, which separate easily afterward for reuse: US 9,484,151 (priority 19 January 2011). Several of these describe wet coating and sputtering as the prior art they improve on.
ULVAC patents: US 8,262,808 and US 8,128,760 (priority 21 December 2006; dysprosium or terbium evaporated beside the magnets, the second from a hydride source); US 9,434,002 (the evaporation source itself, a refractory mesh dipped in molten metal, priority 8 October 2008); US 8,075,954 (a hot-wall chamber with cold magnets shuttled in, priority 18 March 2005).
Toyota Motor, US 9,928,956 B2, "Permanent magnet and manufacturing method therefor," priority 9 July 2012. A sputtered dysprosium film as the worked example; the claim concerns grooves that guide diffusion and the later splitting of the block.
My reading of the company patents cited in this section. In the granted U.S. patents I read, sputtering is the main worked example only in Hitachi Metals' US 8,038,807 and Toyota's US 9,928,956; elsewhere it is listed as one option or described as the prior art being replaced (the JST/Osaka patent calls it hard to mass produce). Chinese-language filings were not swept and may tell a different story.
Vacuumschmelze, US 2013/0153088 A1 (priority 15 December 2011; a hydride paste with solid separator particles so stacked magnets do not weld together) and US 2014/0366991 A1 (priority 12 June 2013; the paste on a support near the magnet rather than on it). Both shown as abandoned on Google Patents. An abandoned application tells you what was tried, not what is used.
JL Mag Rare-Earth, US 9,947,447 B2, "Neodymium iron boron magnet and preparation method thereof," priority 18 December 2015. The description lists brushing, spraying, rolling and dipping a suspended alloy powder and says soaking is preferred. I did not see the claim text, so whether the claim fixes the coating method is unconfirmed.
Beijing Zhong Ke San Huan High Tech, US 10,714,245 B2, "Method for preparing an R-T-B permanent magnet," priority 19 December 2014. A heavy rare earth compound slurry applied by immersion to a pre-sintered, still-porous block.
Advanced Technology & Materials Co., US 10,385,442 B2, priority 7 May 2015. Terbium-copper plates, called targets, sandwich the magnet in a vacuum furnace and supply terbium by evaporation; the patent describes sputtering as the prior art it improves on.
Urban Mining Company (now Noveon Magnetics), US 9,336,932 B1, "Grain boundary engineering," priority 15 August 2014, and later family members including US 11,942,245. These blend an additive alloy into the magnet powder before pressing and sintering. Same goal (put the expensive elements at the grain boundaries), different process.
T. Higashi et al. (Honda R&D) and M. Iwasaki (TDK), "Development of rare earth-saving magnet using localized diffusion method", SAE Technical Paper 2013-01-1757, 8 April 2013. Abstract: dysprosium diffused only at positions chosen from a computed coercivity map, reducing dysprosium use by about 30% without compromising motor performance. The company's own claim; no patent verified.
Hyundai Motor patents: coercivity diffused in increasing amounts toward the hot center of each magnet, US 9,385,563 (priority 12 December 2012); spray, suspension or "barrel painting" of a hydride or fluoride powder, US 10,714,246 (priority 9 January 2015) and, with Star Group and Kia, US 10,490,326 (priority 12 December 2016).
Denso, US 12,519,354 B2, "Rotary electric machine and manufacturing method therefor," priority 17 February 2021. The diffusion source is applied to the two planar faces only of thin magnet segments.
D. N. Brown, D. Lau, Z. Chen, "Substitution of Nd with other rare earth elements in melt spun Nd₂Fe₁₄B magnets," AIP Advances 6, 056019 (2016), open access, Table I. Room-temperature saturation polarization and anisotropy field for R₂Fe₁₄B: Nd 1.60 T and 6.7 T; Pr 1.56 T and 8.7 T; Ho 0.81 T and 7.5 T. The table is attributed to Herbst and Croat (1991) and Buschow (1991), which I have not opened.
S. Constantinides, "Grain boundary diffusion: a technique to reduce heavy rare earth requirement in NdFeB permanent magnets", Magnetics & Materials LLC white paper, 29 October 2023 (the file is labeled v.2). Table 2: anisotropy field 5.3, 11.9 and 17.5 MA/m and saturation polarization 1.61, 0.72 and 0.70 T for the Nd, Dy and Tb compounds (5.3 MA/m is about 6.7 T; the conversion is mine). Also: the liquid grain boundary phase and the core–shell exchange (p. 4); too high a temperature promotes grain growth (p. 4); the 6 mm worked example, 1,790 to 2,270 kA/m at the faces and 2,010 kA/m at 3 mm (p. 5); practical depth 3 to 4 mm and thinnest dimension about 6.5 mm (pp. 5, 7, 9); terbium more effective, dysprosium preferred for abundance and cost (p. 10); "G" grades above and to the right of standard grades (p. 3). Disclosure: I have co-authored papers with Constantinides.
My note on the anisotropy numbers, which undercuts the precision of figure 7. Published room-temperature anisotropy fields for Nd₂Fe₁₄B range from about 6.7 T to 7.6 T depending on the source (6.7 in Brown 2016 and Constantinides; 7.3 in a 1985 single-crystal paper as cited by Xiao 2022; 7.6 in Qin 2025 citing Herbst 1991). I have not opened the primary single-crystal tables (Herbst, Rev. Mod. Phys. 63, 819 (1991); Hirosawa et al., J. Appl. Phys. 59, 873 (1986)). The ranking (Tb, then Dy, far above Pr, Ho and Nd) holds in every source; the second digit does not.
C. Qin et al., "Comparative study of Dy and Tb diffusion behavior in grain boundary diffusion processed Nd-Fe-B sintered magnets," Journal of Materials Research and Technology 35, 3556–3562 (2025), open access. Same base magnet, same hydride form and process: 14.65 kOe base, 18.90 kOe with dysprosium, 25.62 kOe with terbium; remanence 14.25, 14.18 and 14.20 kG. Reports that dysprosium diffused deeper and formed thicker shells. The paper's printed percentage gains do not match its own absolute values, so I quote the absolute values.
Hitachi Metals (now Proterial), US 9,613,748 B2, "RH diffusion source, and method for producing R-T-B-based sintered magnet using same," priority 27 June 2011. In the like-for-like comparison (Table 13, same heavy rare earth content), a dysprosium-based source gave a coercivity gain of 450 kA/m and a terbium-based source 760 kA/m, about 70% more, with no remanence change in either. Other dysprosium sources in the patent gave about 410 kA/m.
H. Nakamura, K. Hirota, T. Minowa, M. Honshima, "Coercivity of Nd-Fe-B sintered magnets produced by the grain boundary diffusion process with various rare-earth compounds," Journal of the Magnetics Society of Japan 31(1), 6–11 (2007), open access. Abstract: dysprosium supplied from the surface penetrated a 2 mm magnet throughout; compounds of rare earths whose iron-boron compound has higher anisotropy than neodymium's raised coercivity, and those with lower anisotropy lowered it.
Xiao et al., Metals 12, 1922 (2022), open access. Describes holmium as relatively low-priced because it is less used; gives no price. Its experiments substitute holmium into the alloy, not diffusion.
Patents naming holmium as one permitted heavy rare earth among several: Hitachi Metals US 8,206,516 and US 8,038,807; JST/Osaka US 7,824,506; Fujian Jinlong (now Golden Dragon) US 11,501,914; Hyundai US 10,714,246. The examples in these use dysprosium or terbium.
Shin-Etsu Chemical, US 7,488,393 B2, "Rare earth permanent magnet," priority 23 March 2005. Absorbed fluorine reacts with neodymium oxide in the magnet to form oxyfluoride at the grain boundaries; too much fluorine reduces coercivity. Its claim is to the resulting magnet (fluorine and heavy rare earth rising toward the surface), not to a method.
T.-H. Kim, S.-R. Lee, H.-J. Kim, M.-W. Lee, T.-S. Jang, Acta Materialia 93, 95–104 (2015), doi:10.1016/j.actamat.2015.04.019, abstract. Oxide phases at the grain boundaries block dysprosium and consume it.
K.-H. Bae et al., "Magnetic and microstructural characteristics of DyF₃/DyHx dip-coated Nd-Fe-B sintered magnets," Journal of Alloys and Compounds 612, 183–188 (2014), doi:10.1016/j.jallcom.2014.05.166, abstract. Coercivity rose with the hydride share of the coating (14.9, 15.4 and 16.5 kOe at 0, 50 and 100%), with deeper dysprosium and better developed shells; the authors attribute the difference to fluorine versus hydrogen.
D. H. Lee et al., "Coercivity and thermal stability enhancement of Nd-Fe-B sintered magnets by grain boundary diffusion with Tb-Al-Cu alloys," Materialia 36, 102161 (2024), abstract. About 20% less terbium for a similar coercivity gain compared with pure terbium.
W. Song et al., Journal of Materials Research and Technology 18, 841–851 (2022), doi:10.1016/j.jmrt.2022.03.016, abstract. The DOAJ record confirms the numbers; authors, volume and pages are from my research notes. Separate praseodymium-aluminum-copper and terbium-aluminum-copper sources gave 1,911 kA/m and better squareness, against 1,832 kA/m for one premixed alloy, from a 1,040 kA/m base.
Toyota Motor, US 9,257,227 B2, "Method for manufacturing rare-earth magnet," priority 26 January 2012. Neodymium-praseodymium-copper or -aluminum modifier alloys infiltrated into a fine-grained hot-deformed magnet, with no heavy rare earth.
J. He et al., "Grain boundary diffusion sources and their coating methods for Nd-Fe-B permanent magnets," Metals 11(9), 1434 (2021), open access. Review: the melting grain boundary phase as the diffusion channel; a second, lower-temperature step to modify the distribution of the grain boundary phase; and a broad survey of coating methods and sources. Its specific temperatures are not repeated here.
K. Löwe, C. Brombacher, M. Katter, O. Gutfleisch, "Temperature-dependent Dy diffusion processes in Nd–Fe–B permanent magnets," Acta Materialia 83, 248–255 (2015), abstract page. Abstract and conclusion: shells near the surface form by melting and solidification of a heavy-rare-earth-rich phase, not by solid-state lattice diffusion; the gain peaks at an optimum temperature. Katter is from Vacuumschmelze.
Y. Lu et al., "Nd-Fe-B magnets: the gradient change of microstructures and the diffusion principle after grain boundary diffusion process," Materials 12(23), 3881 (2019), open access. A 7 mm commercial N48 magnet treated with terbium: thick shells near the surface, thin shells and continuous boundaries from roughly 0.3 to 1 mm, terbium mostly at triple junctions by about 1.5 mm; coercivity highest at the faces and lowest at the center, but above the untreated value throughout.
X. Zhang et al., "Diffusion behavior controlled performance enhancement of Nd-Fe-B magnets diffused by Tb-Ni alloy at various temperatures," Journal of Materials Science: Materials in Engineering 20, 7 (2025), open access. The coercivity gain rose with temperature and then reversed, with grain growth and, in the surface region, grains becoming terbium-rich throughout; squareness fell. The paper is not fully consistent about absolute coercivity at the highest temperature.
Y. Zhao, H. Feng, A. Li, W. Li, "Diffusion temperature dependent microstructure and magnetic properties of TbHx grain boundary diffused Nd–Fe–B magnets," AIP Advances 11, 025208 (2021), open access. At a fixed treatment time the hottest condition gave the highest coercivity but lower remanence, with terbium flooding the grains under the coating and thicker shells; the best overall balance came from a slightly cooler, longer treatment.
T.-H. Kim et al. (NIMS, Toyota Central R&D, Toyota), Magnetics Society of Japan 2019 abstract 25aA-5; journal version Acta Materialia 172, 139–149 (2019), doi:10.1016/j.actamat.2019.04.032. Conference abstract seen, journal paper not: in a dysprosium-containing base, dysprosium collected in the boundary during diffusion moved into a secondary shell during the post-diffusion anneal, which delivered much of the gain.
T. Ma, X. Wang, X. Liu, C. Wu, M. Yan, "Coercivity enhancements of Nd–Fe–B sintered magnets by diffusing DyHx along different axes," Journal of Physics D 48(21), 215001 (2015), doi:10.1088/0022-3727/48/21/215001. Abstract: diffusing along the c-axis gave a larger gain than diffusing across it; the authors credit the anisotropic distribution of the neodymium-rich phase.
T.-H. Kim et al., "Anisotropic diffusion mechanism in grain boundary diffusion processed Nd-Fe-B sintered magnet," Acta Materialia 112, 59–66 (2016), abstract. Deeper penetration parallel to the alignment direction than across it, attributed to orientation-dependent lattice diffusion rather than the boundary phase. This is the disagreement with Ma 2015.
F. Chen et al., "Dependence of the demagnetization behavior on the direction of grain boundary diffusion in sintered Nd-Fe-B magnets," Journal of Magnetism and Magnetic Materials 465, 392–398 (2018), record. Abstract: diffusing parallel to the c-axis gave better squareness.
Arnold Magnetic Technologies, Grain Boundary Diffused Neodymium Iron Boron Magnet Catalog, rev. 210607. States the process favors magnets under 6 mm thick or laminated materials; source of the diffused grades in figure 14 (typical remanence, minimum coercivity). Catalog values are typical and minimum figures, not measurements. Two rows (GB48UH, GB50UH) list coercivities in kA/m that do not match their kOe values, and GB52UH differs between its summary row and its data sheet; all three are left off the chart. Disclosure: I worked at Arnold.
My note on "depth," which undercuts every depth number on this page, mine included. Published depths run from under 0.1 mm to about 4 mm because they measure different things: where a visible shell disappears (tens to hundreds of micrometers in lab studies such as Lu 2019, Zhang 2025 and Kim 2016), where the element is still detectable (millimeters, Nakamura 2007), and where coercivity is still measurably raised (about 3 mm, Constantinides). Source chemistry, temperature, time and the base magnet all move the number, and the chart in figure 11 is my curve through three published points, not a measured profile.
M. P. Thompson et al. (General Motors), "Grain-boundary-diffused magnets: the challenges in obtaining reliable and representative BH curves for electromagnetic motor design," IEEE Electrification Magazine 5(1), 19–27 (2017), doi:10.1109/MELE.2016.2644561, free digital edition. Partial text seen: diffused magnets have heavy-rare-earth-rich corners and edges and very little at the center, unlike conventional magnets; they typically use about one-half, or in some cases one-third, of the dysprosium or terbium of conventional magnets. Not tied to a specific vehicle in the pages I read.
Hitachi Metals (now Proterial), NEOMAX catalog HG-A27-F (April 2016). States that the coercivity distribution of its dysprosium-diffused grades is not homogeneous, and quotes coercivity by position. The extracted table was garbled, so I do not reproduce its numbers.
Shin-Etsu Magnetics, N series product data, retrieved 29 October 2026. Lists a separate grain boundary diffusion type (for example N54SH-HGF, 1.43–1.49 T, at least 1,671 kA/m) beside conventional perpendicular-press and parallel-press grades (for example N36SA, 1.15–1.22 T at 1,671 kA/m); warns that diffused grades vary with size and shape and may not meet the listed values. The page does not explain its suffixes.
Shin-Etsu Magnetics, "Announcement" on its grain boundary diffusion technology (technology section dated 28 March 2007), retrieved 29 October 2026. Describes the earlier two-alloy method (a heavy-free main alloy blended with a low-melting, heavy-rare-earth-bearing sintering aid) and states that diffusion gives about 30% more coercivity than the two-alloy method with no reduction in remanence; no test conditions are given, so this is the company's own claim. The page also lists Shin-Etsu's related U.S., Japanese, Chinese and European patents and states that it does not license them to third parties.
K. Opelt, C.-C. Lin, M. Schönfeldt, J. Gassmann, S. Yoon, O. Gutfleisch, "Diffusion behavior of heavy rare-earths for grain boundary engineering of sintered Nd-Fe-B-based permanent magnets produced by the 2-powder method," Acta Materialia 270, 119871 (2024), open access. Coarse heavy-free and fine dysprosium-bearing powders blended; dysprosium confined to the outer region of the grains; shell thickness about 0.4 µm at 1 wt% to about 2.4 µm at 3 wt% dysprosium; coercivity up roughly 120 kA/m per wt%, remanence from about 1.37 to 1.32 T over 0 to 3 wt%; states diffusion is more efficient but limited to thin magnets while the two-powder method applies to any dimension; demonstrated at about 20 kg batch scale. One research group's results on its own compositions.
Yantai Zhenghai Magnetic Material, US 9,901,984 B2, "Method for preparing R-Fe-B based sintered magnet," priority 30 May 2013. Blends terbium, dysprosium or holmium powder (or a compound) into the alloy powder before pressing, then pressure-sinters. The patent describes surface diffusion as needing thin pieces and its own method as free of the restriction of magnet size; that is the applicant's characterization.
T. Zhou et al., "Sintered NdFeB magnets with Tb-Dy double-layer core/shell structure were fabricated by double alloy method and grain boundary diffusion," Journal of Alloys and Compounds 856, 158191 (2021), doi:10.1016/j.jallcom.2020.158191, abstract. Abstract only: a double-alloy base with dysprosium shells, then terbium hydride diffused in, giving a two-layer shell and a 7.6 kOe coercivity gain with a slight remanence decrease.
Nissan, "New Nissan EV motor cuts rare earth use by 40 percent", press release, November 2012. The new process distributes dysprosium around each grain's boundary, reducing dysprosium use by 40%. The company's own claim; the magnet supplier is not named.
JL Mag Rare-Earth Co., Hong Kong IPO prospectus, 31 December 2021, summary pp. 1, 3, 24. 4,111 metric tons of magnets made with GBD in 2020 and 4,336 in the first nine months of 2021; about 21.3% of the world GBD magnet market and fewer than ten makers worldwide applying the process, both attributed to Frost & Sullivan. These are a company's own and a commissioned consultant's figures; the prospectus gives no percentage saving of heavy rare earth.
My note on the outcome numbers. Arnold's halving compares diffused grades with typical alloyed grades of the same suffix across the industry; GM compares with "conventional" magnets generally; Nissan compares with conventional EV motors; Honda and TDK compare localized with uniform treatment. None states its baseline composition in the sources I read, so the 30% to two-thirds spread is partly real and partly a difference in what is being compared.
Arnold Magnetic Technologies, Neodymium Iron Boron Magnet Catalog, rev. 210607. Source of the conventional grades in figure 14 (typical remanence, minimum coercivity); grades marked X or Z (availability to be confirmed) are left off the chart.