Compositionally Graded Alloys (CGAs)

Updated 8 July 2026

Compositionally graded alloys are materials with deliberate spatial variations in chemical composition that enable tuned local properties and optimized performance.
They are produced via techniques like sputtering, additive manufacturing, and diffusion processes to create continuous composition spreads for high-throughput screening and dissimilar-metal joining.
High-throughput characterization and computational methods link composition paths with microstructural evolution, phase stability, and functional optimization in these advanced materials.

Searching arXiv for recent and foundational papers on compositionally graded alloys to ground the article. Compositionally graded alloys (CGAs) are alloys in which chemical composition varies spatially within a single specimen or component, enabling spatially varying alloy compositions and local tuning of properties. In the literature, CGAs appear as continuous composition spreads in sputtered thin films, graded regions produced by laser directed energy deposition and related additive-manufacturing routes, diffusion-generated gradients in bulk samples and wires, and compositionally modulated multilayers. They are used both as functional materials and as experimental platforms for high-throughput alloy discovery, dissimilar-metal joining, and performance-driven structural design, with representative systems including Fe–Pt magnetic films, CoCrFeNi stacking-fault-energy gradients, SS316L/Monel400 and SS316L/C300 transitions, Al/Cu gradients, AlMgSi wires, Cu–Co diffusion couples, and Si/Ge multilayers (Hong et al., 2021, Hanagan et al., 14 Feb 2025, Yang et al., 2024, Ben-Artzy et al., 2021, Chen et al., 2013).

1. Scope, architectural forms, and design objectives

A central premise of CGAs is that the combination of two or more alloys within a single part can yield substantial enhancement in performance and functionality. In structural contexts, CGAs are proposed for components in which local loading conditions, corrosion conditions, or thermal fields vary across the geometry; in discovery-oriented contexts, a continuous composition gradient allows rapid screening of coupled composition–processing–microstructure–property relationships without sample-to-sample procedural drift (Hanagan et al., 14 Feb 2025, Hong et al., 2021).

The field spans several distinct architectural forms. One form is the wafer-scale composition spread, in which a thin-film library contains a continuous one- or two-dimensional gradient and is probed by spatially resolved metrology. A second form is the additively manufactured bulk gradient, in which composition is changed layer by layer or segment by segment during deposition. A third form is the diffusion-generated gradient, produced by welding, hot pressing, co-drawing, or epitaxial growth followed by interdiffusion. More recent work extends the concept from two-terminal gradients to multi-terminal CGAs, where several performance-optimized terminal alloys are connected within one continuous compositional network (Allen et al., 2024).

CGA form	Representative systems	Primary objective
Thin-film composition spread	Fe–Pt; Al–Co–Cr–Fe–Ni	Coercivity mapping; corrosion screening
Additively manufactured bulk gradient	CoCrFeNi; SS316L–Monel400; SS316L–C300; Al/Cu	Property-gradient design; joining; crack avoidance
Diffusion-generated or modulated gradient	Al–Cu–Li–(Mg); Cu–Co; AlMgSi wires; Si/Ge multilayers	Precipitation studies; multifunctionality; thermal transport control

This range of embodiments makes CGAs simultaneously a materials platform and a design methodology. The same gradient can serve as an overview route for a usable component, as in dissimilar-metal transitions, or as a combinatorial specimen for identifying an optimum composition and then transferring that composition to a homogeneous film or bulk alloy for deployment (Hong et al., 2021).

2. Fabrication routes and processing strategies

Magnetron sputtering remains a standard route for thin-film CGAs. Compositionally graded Fe–Pt thin films were prepared on stationary 100 mm Si substrates by magnetron sputtering a base target of Fe on which a piece of Pt is asymmetrically positioned; the composition gradient was controlled through the size and position of Pt foil pieces, with graded films deposited at room temperature and thickness of approximately 100 nm (Hong et al., 2021). In a corrosion-screening context, combinatorial Al–Co–Cr–Fe–Ni thin-film libraries were produced by magnetron co-sputtering on Si wafers, creating continuous two-dimensional composition spreads with 177 uniquely identified spots after annealing (Sur et al., 2023). A related library for scanning droplet cell testing used Al $_{0.7-x-y}$ Co $_x$ Cr $_y$ Fe $_{0.15}$ Ni $_{0.15}$ thin films on a 3-inch Si/SiO $_2$ wafer, diced into 177 pieces of near-uniform composition (Sur et al., 2023).

Directed energy deposition and related additive-manufacturing methods are central to bulk CGAs and graded joins. In CoCrFeNi, a monotonic stacking-fault-energy path was fabricated by laser directed energy deposition using pure Fe, Ni, and Cr elemental powders, with ten composition segments selected from an autonomously planned path (Hanagan et al., 14 Feb 2025). SS316L to Monel400 functionally graded materials were fabricated by DED through spatially tailored compositions obtained by adjusting the feed rates of SS316L, Monel400, and Ni powders (Yang et al., 2024). SS316L to C300 maraging steel gradients were built by DED using dual powder hoppers and thirteen discrete composition layers, chosen using metallurgical considerations to ensure a smooth transition in properties and microstructure (Ben-Artzy et al., 2021). Al/Cu gradients were fabricated by laser engineered net shaping with independent Al, Cu, and TiB $_2$ feeders and a 450 nm blue laser, which enabled deposition on a Cu substrate and promoted intermixing (Abere et al., 2024).

Diffusion-based processing provides a complementary route when a long, smooth gradient is required. In Al–Cu–Li–(Mg), two base alloys were joined by linear friction welding, the initial gradient was enlarged by diffusion at $515^\circ\mathrm{C}$ for 14 days, and hot rolling extended the gradient to approximately 10 mm (Ivanov et al., 2018). In Cu–Co, diffusion couples were produced by hot pressing pure Cu and Cu–2 wt% Co, followed by annealing at $1000^\circ\mathrm{C}$ for 15 days and channel-die hot compression to spread the concentration profile (Geuser et al., 2015). Composition graded AlMgSi wires were obtained by co-drawing commercially pure Al with an AlMgSi alloy followed by diffusion annealing, yielding a radial solute gradient and, after aging, a radial gradient in precipitate density and hardness (Yang et al., 2020). In Si/Ge multilayers, composition modulation emerged through Ge surface segregation and partial interdiffusion during epitaxial growth, producing a graded Si $_{1-x}$ Ge $_x$ 0 profile rather than an ideal superlattice with atomically sharp interfaces (Chen et al., 2013).

Across these routes, the processing logic is consistent: control the spatial derivative of composition, then use thermal treatment, cooling rate, or both to direct phase formation. The specific implementation differs sharply among sputtered wafers, DED builds, and diffusion couples, but all seek a coupled control of composition path and microstructure path.

3. High-throughput characterization and combinatorial workflows

A defining feature of CGA research is the use of spatially resolved, high-throughput metrology. In Fe–Pt, compositional variation was mapped by EDX using one-dimensional line scans and two-dimensional $_x$ 1 point maps with 5 mm spacing, while coercivity was mapped by scanning polar MOKE with pulsed magnetic fields up to 4 T; hundreds of loops per wafer were obtained in approximately 3 hours, and scanning XRD supplied spatially resolved diffraction patterns along the gradient (Hong et al., 2021). This combination of EDX, MOKE, and XRD enabled direct correlation of local composition, structure, and magnetic response.

Corrosion-oriented CGA libraries use analogous spatial workflows. In Al–Co–Cr–Fe–Ni combinatorial films, 177 wafer positions were sampled by XRF for composition and by XRD for phase identification, then screened electrochemically in deaerated $_x$ 2 using metrics such as the impedance modulus at 10 mHz, critical current density, passive current density, and linear polarization resistance (Sur et al., 2023). The scanning droplet cell methodology automated OCP monitoring, potentiostatic EIS, cathodic reduction, and LSV, with a per-composition test time of approximately 26 minutes (Sur et al., 2023). In this framework, $_x$ 3 serves as a rapid discriminator of passive-film quality, while $_x$ 4 and $_x$ 5 provide complementary measures of passivation performance.

Time-resolved nanoscale characterization has also been adapted to graded specimens. In Al–Cu–Li–(Mg), in-situ SAXS scanned a 70 $_x$ 6m section across the Mg gradient with 500 $_x$ 7m spatial resolution for 3 days at room temperature, allowing cluster formation kinetics to be measured as a function of both composition and time (Ivanov et al., 2018). In Cu–Co, synchrotron SAXS with a beam size of approximately 200 $_x$ 8m monitored precipitate size, volume fraction, and number density along the composition gradient during isothermal aging at $_x$ 9, $_y$ 0, and $_y$ 1 (Geuser et al., 2015).

These workflows convert a single graded specimen into a dense dataset over composition space. That architecture is especially important when the relevant output is not a single endpoint property but a kinetic trajectory—clustering, precipitation, phase ordering, self-healing passivation, or coercivity evolution after annealing.

4. Structure–property relationships across graded compositions

The strongest empirical justification for CGAs is the ability to resolve structure–property relationships continuously rather than composition by composition. In Fe–Pt, coercivity emerges near 37.5 at.% Pt, peaks at approximately 1.6 T at 55–60 at.% Pt after $_y$ 2, and drops at higher Pt content; scanning XRD shows that tetragonal distortion and ordered $_y$ 3 FePt appear for $_y$ 4, while the region of maximum coercivity coincides with lattice parameters characteristic of well-ordered $_y$ 5 FePt (Hong et al., 2021). The relationship was summarized graphically as

$_y$ 6

In dissimilar-metal joining, the decisive issue is often not optimization of a peak property but suppression of failure modes. For SS316L–Monel400, hot cracking arises from low Fe–Cu solubility, Cu segregation to interdendritic regions, and formation of a low-melting-point Cu-rich liquid film. Introducing Ni into the gradient region reduces Cu microsegregation, smooths the solidification path, and suppresses compositionally driven hot cracking; a hybrid Scheil-equilibrium criterion treats the appearance of a secondary Cu-rich FCC phase at high solid fraction as the crack-susceptibility indicator (Yang et al., 2024). In SS316L–C300, the selected thirteen-step path avoided deleterious intermetallic phase fields, EBSD showed a transition from martensitic/BCC to FCC across the gradient, and no deleterious intermetallic phases were detected across the interface (Ben-Artzy et al., 2021).

In Al/Cu gradients, microstructural control is dominated by rapid solidification and gradient-induced mass transport. Pure Al deposition on Cu produced columnar dendrites and acicular Al $_y$ 7Cu $_y$ 8-rich microstructures, whereas TiB $_y$ 9 additions and an actively varied print-layer composition suppressed the dendritic growth mode and produced an equiaxed microstructure in the graded part (Abere et al., 2024). The phase assemblage remained governed by rapid-solidification kinetics, with Al $_{0.15}$ 0Cu $_{0.15}$ 1 and Al $_{0.15}$ 2Cu appearing while other equilibrium intermetallics were kinetically suppressed.

CGA libraries also expose subtle local-order effects. In Al–Cr-containing corrosion-resistant compositionally complex alloys, the best passivation was associated with surface enrichment of Cr(III) and Al(III) and with more random Ni–Al distribution in the first nearest-neighbor shell; greater Ni–Al clustering, expressed through more negative Warren–Cowley parameters, correlated with poorer passivation (Sur et al., 2023). In Al–Cu–Li–(Mg), Mg accelerated Cu-rich clustering with no measurable lower threshold down to at least 0.02 at.% Mg, while larger precipitates of about 2 nm appeared above 0.1 at.% Mg and the effect saturated above approximately 0.2 at.% Mg (Ivanov et al., 2018). In AlMgSi wires, the radial solute gradient produced a hardened outer region and a conductive core, yielding an improved combination of electrical conductivity and torsion strength compared to the predictions based on a classical rule of mixture (Yang et al., 2020).

Thermal transport offers a different but related example. In epitaxial Ge/Si multilayers, partial interdiffusion of Ge into Si spacers lowered cross-plane thermal conductivity below both the perfect-superlattice case and the homogeneous-alloy limit, because interface scattering and alloy disorder scatter different parts of the phonon spectrum (Chen et al., 2013). This result is notable because it reverses the usual assumption that the sharpest possible interface is necessarily optimal.

5. Computational path planning, graph design, and topology optimization

As CGA design moved into higher-dimensional composition spaces, path planning became an explicit computational problem. In CoCrFeNi, a monotonic stacking-fault-energy gradient was designed using fixed-node rapidly-exploring random trees, with the feasible composition space restricted by CALPHAD-based phase-fraction and solidification-range constraints:

$_{0.15}$ 3

The path objective penalized lack of monotonicity in SFE and included a small path-length term,

$_{0.15}$ 4

so that the designed gradient moved from minimum to maximum SFE while avoiding infeasible regions (Hanagan et al., 14 Feb 2025).

A separate development replaces offline surrogate training with on-the-fly path planning. Using RRT with direct CALPHAD calls, followed by a lever-rule-based “Bad Phase Purge,” this approach avoids the exponential setup cost of surrogate modeling in high dimensions and was reported to be as much as 106 times more efficient than surrogate modeling (Price et al., 2024). The path-purification step transforms a mixed-phase path into one containing only “good” phases using the already computed equilibrium phase fractions and phase compositions:

$_{0.15}$ 5

Graph-based design generalizes the problem from two-terminal to multi-terminal CGAs. In the ATLAS framework, the alloy design space is represented as a labeled property graph whose nodes are alloy compositions and whose edges are feasible compositional steps. After partitioning into maximally connected constrained subgraphs, multi-terminal CGA design is posed as a Minimum Steiner Tree Problem, and TreeMAP conformally maps the resulting compositional tree onto a discretized structure such as a gas turbine blade (Allen et al., 2024). This formulation directly addresses the limitation of prior approaches that only connect a single start alloy to a single end alloy.

Topology optimization extends CGA design from path selection to full-field composition design under manufacturing constraints. TOBACO represents the spatial composition field by a band-limited coordinate neural network, with the maximum allowable gradation enforced implicitly through bandwidth control. The relevant manufacturing constraint is

$_{0.15}$ 6

and Bernstein’s inequality is used to relate the gradient bound to a bandwidth limit, eliminating the need for explicit per-element gradation constraints (Chandrasekhar et al., 14 Aug 2025). This yields mesh independence, high-resolution design extraction, and end-to-end differentiability for elastic and thermo-elastic topology optimization.

6. Applications, model limitations, and unresolved issues

The application space for CGAs now spans magnetic films, corrosion-resistant compositionally complex alloys, dissimilar-metal joints, thermally engineered multilayers, conductors with coupled strength and conductivity targets, and semiconductor contact layers. In high-Al-content AlGaN transistors, a reverse Al-composition graded contact layer reduced contact resistance to $_{0.15}$ 7, while devices with $_{0.15}$ 8 and $_{0.15}$ 9 reached $_{0.15}$ 0 and breakdown voltage greater than 220 V for $_{0.15}$ 1 (Razzak et al., 2019). In Fe–Pt, the graded-film workflow was explicitly positioned as a step toward integration of coercive FePt films into collectively fabricated devices (Hong et al., 2021).

At the same time, the literature emphasizes that phase-feasible composition paths are not sufficient by themselves. In the CoCrFeNi SFE gradient, equilibrium CALPHAD predicted a single FCC phase throughout the path, yet Fe-rich sections transformed martensitically after quenching, demonstrating a limitation of using equilibrium CALPHAD calculations for phase stability predictions (Hanagan et al., 14 Feb 2025). In surface segregation modeling of a Co–Cu–Fe–Mo–Ni quinary alloy, low-rank machine-learning potentials predicted strong Co enrichment at the surface despite Co having a relatively high surface energy, indicating that chemical potential and local ordering can dominate surface-energy intuition (Ferrari et al., 2022). Molecular-dynamics simulations of an equiatomic CoCrCuFeNi-like alloy likewise showed that nuclei formed during homogeneous crystallization are Fe- and Co-rich and Cu-poor over a wide range of sizes, imprinting short-range order and nanometer-scale compositional gradients very early in microstructure development (Mishra et al., 2023).

These results suggest that CGA design is constrained not only by equilibrium phase fields and manufacturability envelopes, but also by segregation, metastability, short-range order, solidification path, and transformation kinetics. A plausible implication is that robust CGA design increasingly requires coupled treatment of composition path, process path, and local environment effects rather than composition alone. The present literature accordingly combines CALPHAD, Scheil-type solidification models, DFT-trained surrogates, graph algorithms, neural representations, and high-throughput spatial metrology to resolve where a gradient is feasible, where it is useful, and where predictive models remain incomplete (Yang et al., 2024, Hanagan et al., 14 Feb 2025, Chandrasekhar et al., 14 Aug 2025).