Sinter: Mechanisms, Kinetics, and Applications

• Sinter is a process where particles bond, form necks, and reduce pore volume through thermal, pressure, electric, or chemical means.
• It encompasses various kinetic regimes such as bulk and surface diffusion, characterized by laws like t^(1/4) and t^(1/3) for neck growth.
• Engineered sintering routes—including field-assisted, microwave, and cold sintering—demonstrate its versatility in optimizing material properties.

Sinter and sintering denote a family of particle-coalescence and densification phenomena in which discrete grains, nanoparticles, or powder particles bond, form necks, reduce pore volume, and often densify under heat, pressure, electric field, or chemically assisted transport. In heterogeneous catalysis, the term often denotes thermally driven loss of nanoparticle dispersion; in ironmaking, “sinter” also denotes an agglomerated ferrous burden material; and, in a distinct homonymous usage in machine learning, “Sinter” names a sine-based nonlinear function used in low-rank adaptation (Zhu et al., 2022, Yang et al., 2024, Chakrabarty et al., 17 Nov 2025, Deng et al., 26 Sep 2025).

1. Thermodynamic basis

Sintering is fundamentally driven by reduction of interfacial free energy. In solid-state sintering of ceramic or metallic powders, the relevant energetic terms are surface energy and grain-boundary energy, and the characteristic morphological sequence is neck growth, pore evolution, and shrinkage toward a denser body (Yang et al., 2024). In the earliest stage of contact between roughened crystals, the local driving force can be written through a Gibbs–Thomson relation,

$\mu=\mu_0+\Omega_v \kappa \gamma,$

so the highly concave neck region has lower chemical potential than neighboring convex regions and therefore draws matter inward (Farr et al., 2010).

For supported nanoparticles, the same free-energy logic appears in a different form. A nanoparticle and its dispersed ad-atom ensemble compete through enthalpy and configurational entropy,

$\Delta G = N\Delta h - T (S_{NP} - S_{ad}),$

with positive $\Delta G$ implying that the dispersed ad-atoms are more stable and negative $\Delta G$ implying that the aggregated nanoparticle is more stable (Zhu et al., 2022). This makes sintering a free-energy competition rather than a purely monotonic consequence of heating.

A common simplification is to treat sintering as identical with neck growth alone. The phase-field-micromechanics formulation of sintering instead treats densification as a coupled problem of diffusion, grain-boundary evolution, grain motion, and mechanics, with total free energy

$$F = \int_{\Omega} f(C, n_1,\dots,n_{n_g},\nabla C,\nabla n_1,\dots,\nabla n_{n_g})\,d\Omega$$

and an energy law

$\frac{dF}{dt}=-D\le 0,$

so shrinkage-producing grain motion is derived from the same energetic structure as interfacial evolution (Yang et al., 2024). This suggests that “sinter” is best understood not as one mechanism but as a thermodynamically organized class of morphology-changing processes.

2. Kinetic regimes and neck-growth laws

The kinetics of sintering depend on the dominant transport path. For viscous neck growth between amorphous particles, the paper on caking of amorphous molecular powders uses the Frenkel relation

$$\left( \frac{x}{d} \right)^2 = \frac{\gamma t}{6 d \eta},$$

where $x$ is neck diameter, $d$ is particle diameter, $\gamma$ is surface tension, and $\Delta G = N\Delta h - T (S_{NP} - S_{ad}),$0 is viscosity (Braz et al., 31 Mar 2025). With the paper’s “strong bridge” criterion $\Delta G = N\Delta h - T (S_{NP} - S_{ad}),$1, the characteristic caking time becomes

$\Delta G = N\Delta h - T (S_{NP} - S_{ad}),$2

making smaller particles faster to bridge at fixed $\Delta G = N\Delta h - T (S_{NP} - S_{ad}),$3 (Braz et al., 31 Mar 2025).

For roughened crystals that have just touched, two asymptotic limits produce different early-time neck-growth laws. In the bulk-diffusion-limited case, the surrounding concentration field satisfies Laplace’s equation and the early neck radius obeys

$\Delta G = N\Delta h - T (S_{NP} - S_{ad}),$4

In the surface-diffusion-limited case, the paper predicts a single early pinch-off event and then approximate growth

$\Delta G = N\Delta h - T (S_{NP} - S_{ad}),$5

(Farr et al., 2010). These exponents arise from a slot-cavity approximation to the local post-contact geometry rather than from a late-stage spherical simplification.

In solid-state powder sintering more broadly, the relevant transport channels include surface diffusion, grain-boundary diffusion, volume diffusion, evaporation-condensation, and grain motion. The phase-field-micromechanics model makes this explicit by combining a conserved phase field $\Delta G = N\Delta h - T (S_{NP} - S_{ad}),$6, convected grain-order variables $\Delta G = N\Delta h - T (S_{NP} - S_{ad}),$7, and a mechanical velocity field $\Delta G = N\Delta h - T (S_{NP} - S_{ad}),$8, with

$\Delta G = N\Delta h - T (S_{NP} - S_{ad}),$9

and

$\Delta G$0

In this formulation, densification is not attributed to diffusion alone; convection-like grain motion contributes directly to shrinkage (Yang et al., 2024).

3. Engineered sintering routes

Sintering is not confined to classical furnace densification. The reported literature spans field-assisted, microwave, cold, solvent-assisted, and hybrid routes.

Route and system	Key conditions	Reported outcome
FAST of highly doped Si nanoparticles (Schwesig et al., 2010)	35 MPa, 100 K/min, 860–1160 °C, 3 min, 0.5–1 kA through the compact	Up to about 97% density while retaining nanocrystallinity
915 MHz microwave sintering of alumina (Marinel et al., 2020)	Hybrid SiC-susceptor heating, 1400 °C for 1 h after 27 °C/min then 7 °C/min ramps	99.6% of theoretical density for samples larger than 30 cm$\Delta G$1
Cold sintering of Te-doped CoSb$\Delta G$2 (Serrano et al., 2023)	150 °C, 90 min, 750 MPa, glacial acetic acid; then Ar post-anneal	86% relative density after CSP, around 92% after post-annealing above 500 °C
CSP-assisted SFO composite magnets (García-Martín et al., 2023)	190 °C, 2 h, 2.5 bar, glacial acetic acid; then 1100 °C for 2 h in air	Relative density of about 92% and coercivity up to 3.0 kOe
Water-vapor-assisted sintering of Ag nanoparticle inks (Bourassa et al., 2019)	Moist oven below 120 °C, including 80 °C and 120 °C cases	Resistivity much lower than dry oven; about 3 times bulk Ag resistivity reported

These routes differ in their dominant local driving mechanisms. In field-assisted sintering of conductive Si nanopowders, densification is linked to current percolation, localized Joule heating, particle mobilization, path breakup, and re-formation of current paths rather than to slow diffusion alone (Schwesig et al., 2010). In the microwave alumina system, the key mechanism is hybrid heating: SiC susceptors first absorb microwave power, then radiatively heat the alumina until the ceramic’s own dielectric loss becomes significant (Marinel et al., 2020). In cold sintering, a transient liquid assists dissolution, rearrangement, and reprecipitation at temperatures far below conventional ceramic or ferrite sintering temperatures (Serrano et al., 2023, García-Martín et al., 2023). In printed Ag inks, water vapor is proposed to assist removal or displacement of organics from interparticle gaps and to promote contact formation and neck growth without invoking melting of $\Delta G$3 nm particles (Bourassa et al., 2019).

A recurrent implication is that sintering route and microstructure are inseparable. The same nominal target—higher density—can emerge from distinct local transport paths, and those paths determine whether nanostructure is preserved, whether secondary phases form, and whether the final functional response improves or degrades.

4. Coalescence, redispersion, and counterintuitive regimes

A common misconception is that heating always drives irreversible aggregation. For supported Pd/CeO$\Delta G$4, Cu/TiO$\Delta G$5, and Ag/TiO$\Delta G$6, the reported behavior can reverse: low-loading supported nanoparticles redisperse upon heating and re-sinter upon cooling under constant oxidizing conditions (Zhu et al., 2022). In Pd/CeO$\Delta G$7, particles initially about 1.41 nm disappear at 500 °C in O$\Delta G$8, reappear after cooling to 200 °C at about 2.05 nm, and disappear again on reheating; APXPS supports a dispersion $\Delta G$9 sintering interpretation rather than oxidation $\Delta G$0 reduction (Zhu et al., 2022). The paper attributes this to configurational entropy of dispersed ad-atoms overwhelming aggregation enthalpy at high temperature and low surface concentration.

Another simplification is to use static adhesion energy as a predictor of catalyst sintering resistance. Deep-potential simulations of Cu nanoparticles on alumina show that this can fail. Cu$\Delta G$1 binds more strongly to $\Delta G$2-Al$\Delta G$3O$\Delta G$4(0001) than to $\Delta G$5-Al$\Delta G$6O$\Delta G$7(100) at 0 K, yet diffuses several times faster on $\Delta G$8-Al$\Delta G$9O$$F = \int_{\Omega} f(C, n_1,\dots,n_{n_g},\nabla C,\nabla n_1,\dots,\nabla n_{n_g})\,d\Omega$$0(0001) at 800 K because surface Al atoms move out of plane to maintain contact with the nanoparticle and relax back as it moves away (Xu et al., 21 Jan 2025). In direct coalescence simulations, nine Cu$$F = \int_{\Omega} f(C, n_1,\dots,n_{n_g},\nabla C,\nabla n_1,\dots,\nabla n_{n_g})\,d\Omega$$1 particles fully merge into Cu$$F = \int_{\Omega} f(C, n_1,\dots,n_{n_g},\nabla C,\nabla n_1,\dots,\nabla n_{n_g})\,d\Omega$$2 on $$F = \int_{\Omega} f(C, n_1,\dots,n_{n_g},\nabla C,\nabla n_1,\dots,\nabla n_{n_g})\,d\Omega$$3-Al$$F = \int_{\Omega} f(C, n_1,\dots,n_{n_g},\nabla C,\nabla n_1,\dots,\nabla n_{n_g})\,d\Omega$$4O$$F = \int_{\Omega} f(C, n_1,\dots,n_{n_g},\nabla C,\nabla n_1,\dots,\nabla n_{n_g})\,d\Omega$$5(0001), whereas coalescence on $$F = \int_{\Omega} f(C, n_1,\dots,n_{n_g},\nabla C,\nabla n_1,\dots,\nabla n_{n_g})\,d\Omega$$6-Al$$F = \int_{\Omega} f(C, n_1,\dots,n_{n_g},\nabla C,\nabla n_1,\dots,\nabla n_{n_g})\,d\Omega$$7O$$F = \int_{\Omega} f(C, n_1,\dots,n_{n_g},\nabla C,\nabla n_1,\dots,\nabla n_{n_g})\,d\Omega$$8(110) remains strongly inhibited over the same time scale; even at an initial interparticle spacing of 30 Å, coalescence can still occur on $$F = \int_{\Omega} f(C, n_1,\dots,n_{n_g},\nabla C,\nabla n_1,\dots,\nabla n_{n_g})\,d\Omega$$9-Al$\frac{dF}{dt}=-D\le 0,$0O$\frac{dF}{dt}=-D\le 0,$1(0001) within 10 ns at 800 K (Xu et al., 21 Jan 2025).

In additive manufacturing, sintering can appear as an intermediate defect-forming state rather than a final densification step. In directed energy deposition of Ti6242, a loosely sintered powder layer forms ahead of and around the melt pool, reduces wetting, and promotes lack-of-fusion porosity; the sintered layer can become more than three times the track height in extreme cases (Chen et al., 2020). The operative issue is not simply whether powder sticks, but whether partially heated powder sinters before full incorporation into the melt.

Hydrogen-based direct reduction of iron oxide provides another reaction-coupled example. Starting from approximately spherical 10 nm magnetite particles, the reduction sequence $\frac{dF}{dt}=-D\le 0,$2 is accompanied by self-assembly and sintering into elongated grains about 100–350 nm long and 20–50 nm wide, with the strongest linkage to agglomeration occurring during the FeO $\frac{dF}{dt}=-D\le 0,$3 Fe transition (Zheng et al., 2023). Here sintering changes transport pathways as much as transport drives sintering.

5. Modeling and multiscale representation

Sintering has been modeled from continuum thermodynamics to atomistic dynamics. In the phase-field-micromechanics model of sintering, thermodynamic consistency is enforced through

$\frac{dF}{dt}=-D\le 0,$4

with grain motion derived from variational forces rather than prescribed ad hoc (Yang et al., 2024). The model reproduces parabolic stress distribution along grain boundaries, system-size-independent shrinkage strain in particle chains, and monotonic free-energy decay (Yang et al., 2024).

For early-stage caking in amorphous powders, the same bridge-formation problem is recast as ranked bond percolation on a DEM-generated contact network (Braz et al., 31 Mar 2025). The order parameter is the largest bridged cluster fraction $\frac{dF}{dt}=-D\le 0,$5, the fluctuations are

$\frac{dF}{dt}=-D\le 0,$6

and the percolation threshold $\frac{dF}{dt}=-D\le 0,$7 is taken from the fluctuation peak (Braz et al., 31 Mar 2025). The reported threshold decreases for low size dispersion, reaches fastest caking around $\frac{dF}{dt}=-D\le 0,$8, and increases again at larger $\frac{dF}{dt}=-D\le 0,$9, so the caking time is non-monotonic in particle-size dispersion (Braz et al., 31 Mar 2025). This is a network-level representation of a process that remains microscopically Frenkel-like.

At the atomistic scale, the alumina nanoparticle study benchmarks four empirical interatomic potentials and shows that similar bulk accuracy does not guarantee similar sintering predictions (Roy et al., 2022). Among the tested models, the Coulomb-Buckingham potential best matches the experimental melting range $$\left( \frac{x}{d} \right)^2 = \frac{\gamma t}{6 d \eta},$$0–$$\left( \frac{x}{d} \right)^2 = \frac{\gamma t}{6 d \eta},$$1 K by predicting $$\left( \frac{x}{d} \right)^2 = \frac{\gamma t}{6 d \eta},$$2 K, whereas the charge-transfer CTIE model predicts $$\left( \frac{x}{d} \right)^2 = \frac{\gamma t}{6 d \eta},$$3 K yet yields the fastest nanoparticle sintering kinetics (Roy et al., 2022). The paper argues that truncated Coulomb interactions can be especially problematic for free surfaces and necks, and therefore for nanoparticle sintering itself.

Planetary materials require yet another scale. Thermal-evolution models of chondritic planetesimals treat sintering as hot pressing of initially porous material under self-gravity and radiogenic heating (Henke et al., 2011). The improved binary matrix–chondrule treatment shows that matrix-dominated precursor material sinters at much lower temperature than chondrule-dominated precursor material: in sample models, matrix porosity collapses around 680 K, whereas chondrule-dominated material compacts around 960 K (Gail et al., 2014). This produces a compact interior and a residual porous outer shell that strongly affects heat conduction and inferred burial depths of meteorites (Henke et al., 2011, Gail et al., 2014).

6. Extended technical usages of “sinter”

In blast-furnace ironmaking, sinter is a material class rather than a process descriptor. It is one of the three ferrous burden materials, alongside pellets and iron ore lumps, and the image-analysis study characterizes it as uneven, highly irregular, and porous relative to pellets (Chakrabarty et al., 17 Nov 2025). Across four industrial size ranges—6–8, 8–10, 10–15, and 15–20 mm—the cumulative distributions of three shape descriptors are nearly size-independent for sinter: aspect ratio $$\left( \frac{x}{d} \right)^2 = \frac{\gamma t}{6 d \eta},$$4, circularity $$\left( \frac{x}{d} \right)^2 = \frac{\gamma t}{6 d \eta},$$5, and average contact eccentricity normalized by projected diameter $$\left( \frac{x}{d} \right)^2 = \frac{\gamma t}{6 d \eta},$$6 (Chakrabarty et al., 17 Nov 2025). Median values are reported around $$\left( \frac{x}{d} \right)^2 = \frac{\gamma t}{6 d \eta},$$7, $$\left( \frac{x}{d} \right)^2 = \frac{\gamma t}{6 d \eta},$$8, and $$\left( \frac{x}{d} \right)^2 = \frac{\gamma t}{6 d \eta},$$9, with the third descriptor highlighted as especially relevant for DEM through its connection to rolling resistance (Chakrabarty et al., 17 Nov 2025).

In machine learning, Sinter is a distinct homonym unrelated to thermal particle coalescence. The LoRAN paper defines it as

$x$0

with fixed amplitude and frequency hyperparameters and no additional trainable parameters (Deng et al., 26 Sep 2025). It is applied elementwise after the low-rank update $x$1, so that $x$2 with $x$3 Sinter (Deng et al., 26 Sep 2025). The reported default hyperparameters are $x$4 and $x$5, and the paper presents it as a structured perturbation that introduces curvature into LoRA’s low-rank affine update space without changing parameter count (Deng et al., 26 Sep 2025). Empirically, it is the strongest activation on the SAMSum summarization ablation and nearly matches full fine-tuning on MRPC, although the paper’s own 20 Newsgroups activation table reports Tanh at 78.24 accuracy versus Sinter at 77.78, so uniform dominance across every isolated table entry is not supported by the tabulated classification result (Deng et al., 26 Sep 2025).

Taken together, these usages show that “sinter” is now a polysemous technical term. In materials science it denotes a class of thermodynamically driven coalescence and densification processes, in ironmaking it names a specific agglomerated burden material, and in contemporary parameter-efficient fine-tuning it names a sine-based nonlinear transformation. The shared feature across these otherwise unrelated meanings is structural transformation: in one case of particles and pores, in another of burden morphology, and in the last of low-rank update geometry.