Inside the virtual test cell: how neutronics models are shaping Europe’s fusion neutron source

Deep beneath the planned IFMIF-DONES facility in Granada, Spain, a beam of deuterons will slam into a curtain of liquid lithium, generating one of the most intense neutron fields ever engineered by human hands. The neutrons produced there, at fluxes approaching 10^15 per square centimetre per second, will hammer away at samples of steel and other materials destined for the walls of future fusion power plants. Before a single brick of the test cell that will house this neutron storm is finalized, engineers must first build it inside a computer. A new study published in Frontiers in Nuclear Engineering by Yuefeng Qiu of the Karlsruhe Institute of Technology and colleagues across Germany, Spain and Slovenia now documents, for the first time in full, the painstaking evolution of these computational models of the IFMIF-DONES test cell, the shielded chamber at the heart of the facility where fusion materials will earn their credentials.

IFMIF-DONES, short for International Fusion Materials Irradiation Facility–Demo Oriented Neutron Source, exists for one purpose: to qualify the structural and functional materials that will make a demonstration fusion reactor, or DEMO, buildable, licensable and safe. A 125-milliampere, 40-megaelectronvolt deuteron accelerator strikes a flowing liquid lithium target, and the resulting d-Li fusion reactions bathe test modules in a neutron spectrum closely resembling the one that will batter a reactor’s first wall. The test cell itself is a heavily shielded, atmosphere-isolated enclosure containing the lithium target and the High Flux Test Module, or HFTM, where thousands of miniature material specimens will be irradiated for years. Every critical design question, from displacements per atom in the steel to nuclear heating, gas production, activation and radiation dose rates in neighbouring rooms, depends on an accurate geometric model of this chamber.

The tool of choice for such analyses is the Monte Carlo method, in which codes like MCNP track millions of individual neutrons through a mathematical replica of the facility. But Monte Carlo codes do not read engineering CAD drawings directly. The team’s workflow begins with a laborious pre-processing stage: simplifying the CAD geometry, homogenizing volumes, and hunting down interferences between components using tools such as SpaceClaim. The cleaned-up ‘neutronic CAD model’ is then converted into constructive solid geometry, or CSG, the surface-based representation that Monte Carlo codes understand, using conversion programs like McCad, SuperMC or the newer GEOUNED. Crucially, the conversion must also describe the void spaces around solids, invisible in the original drawings but essential for particle transport. Every model is then stress-tested by firing particles inward from a surrounding sphere, with a strict quality gate: less than one particle lost per million, and computed volumes cross-checked against the CAD originals to flag outliers.

The first DONES test cell model, designated ‘mdl6.9’, was inherited from the earlier IFMIF/EVEDA design phase and adapted to the new baseline. The changes were substantial. DONES uses a single accelerator rather than the two originally foreseen, so one deuteron source was deactivated in the McDeLicious neutron source routine, leaving a dummy beam duct behind. The medium- and low-flux test modules of IFMIF, including the Tritium Release and Creep Fatigue Test Modules, were dropped, leaving only the HFTM, then a ‘thin’ version roughly five centimetres thick holding three rows and eight columns of specimen capsules. The quench tank, which catches the high-speed lithium jet, sat outside the test cell to suppress tritium production. Shielding consisted of magnetite heavy concrete, four metres thick on the downstream and lateral walls and 3.5 metres upstream, manually layered in the model so that different variance-reduction ‘importance’ values could be assigned to each layer to tame statistical noise.

Engineering reality soon overtook that baseline. The target assembly adopted a ‘bayonet back-plate’ design sealed with a gasket, the quench tank was moved inside the test cell to cure lithium-flow instabilities and mitigate neutron streaming, and the HFTM was redesigned with a thicker 4-by-8 capsule array to hold more samples. The result was ‘mdl8.3.0’, the first model truly oriented to DONES rather than its predecessor, with detailed target and HFTM geometry built directly from the engineering CAD data, a shielding plug blocking the second beam duct, and the lithium inlet and outlet enclosed in a shielded interface cell. A comprehensive suite of nuclear analyses on damage doses, gas production and heating followed. Later, the team restructured the model around a modular ‘skeleton’ concept, with space-holders for major components that can be swapped in and out as designs mature, dramatically reducing the modelling effort needed for each design update.

That modular version, ‘mdl9.2.1’, also embraced one of the most striking architectural changes to the facility: the inner monolithic concrete shielding was divided into eleven removable biological shielding blocks, or RBSBs, each designed with dogleg joints to frustrate neutrons seeking a straight path out. The bayonet back-plate was reverted to an integrated configuration after it became clear that no gasket material could survive the brutal irradiation at the target region. Detailed piping layouts were filled into the lower shielding plugs, and a neutron beam tube was added to guide particles from the test cell to a neighbouring room for complementary experiments. Materials were standardized too: EUROFER steel compositions aligned with DEMO neutronics activities, and reference ordinary and heavy concrete with certified production data, borrowed from ITER calculations, was adopted for the shielding.

The current baseline, ‘mdl9.2.7’, goes further still. The RBSB engineering was refined, the inner steel liner thickened for better heat removal, the upper shielding plugs modified to accommodate HFTM piping, and the test cell cover plate filled with shielding to plug streaming paths through the block gaps. Heavy concrete remains inside the RBSBs, prized for both shielding power and thermal conductivity, while the surrounding bucket structure switches to cheaper ordinary concrete. Most ambitiously, the entire test cell geometry has been integrated into the 2022 baseline model of the DONES main building, allowing analysts to assess radiation impact on the accelerator area, the access cell above the test cell and downstream experimental rooms, and even to run large-scale safety evaluations of public dose through sky-shine and direct radiation pathways.

But how much geometric detail does a Monte Carlo model actually need? The team attacked this question with a series of sensitivity studies centred on damage dose, measured in displacements per atom, the key metric for materials qualification. Because DPA varies by orders of magnitude across the HFTM, neither the average nor the peak alone is a fair comparator; the researchers therefore devised the ‘volume-DPA product’, or VDP, which weights dose by the volume experiencing it. Three model variants were compared against the full model: two retained detailed target and HFTM geometry within progressively smaller envelopes, while a third reduced everything to a crude block of lithium, a EUROFER backplate and a uniform mixture of steel, sodium and EUROFER. The two detailed variants agreed with the full model to within one percent on VDP and two percent on VDP10, but the crude block diverged badly, over-moderating the neutrons and misrepresenting the target structures.

That finding licensed a validated lightweight model for accelerated simulations and design optimization loops. It preserves the essentials, the lithium target, EUROFER backplate, HFTM body, capsules and reflectors, while homogenizing the helium cooling channels into the structural material at their exact volume fractions of 31 percent helium and 69 percent SS316LN steel. Benchmarked against the full model with the McDeLicious d-Li source and the FENDL-3.1d nuclear data library, the simplified version matched total neutron flux within two percent for the central four capsule columns, where the precious material samples actually sit. Deviations grow toward the lateral capsules, which serve mainly as reflectors, so the authors caution against using the lightweight model for dedicated analyses of the outer columns. In the opposite direction, the team also built the first HFTM model with explicit specimen-level payload geometry, the CLCv2.0 capsule loading configuration reaching an 85 percent EUROFER volume fraction, and found that flux and damage doses barely changed compared with the homogeneous mixture, a reassuring consequence of the high neutron energies and long mean free paths involved.

Finally, the researchers demonstrated a mesh-based alternative to CSG. Using a fast tetrahedral meshing workflow, exporting the CAD model to STL, meshing it with TetGen in minutes, and post-processing into Abaqus format for MCNP6, they produced an unstructured mesh HFTM of 157,000 elements and slotted it into the test cell model in place of the CSG version. The flux contours from the two approaches overlap almost perfectly, proving the method viable for rapid design iteration, though at a cost: the mesh model needed roughly 13,500 CPU-hours for a billion particle histories against 950 hours for CSG, a penalty the team attributes to one-to-one matching of mesh sets and expects to slash by merging same-material sets. Together, the model lineage, the validated simplifications and the mesh demonstration give the fusion community a traceable, quality-controlled computational backbone for DONES, one that will help decide, years before first beam, whether the materials of a fusion power plant can truly endure.

Subject of Research: Neutronics geometry model development for the IFMIF-DONES test cell

Article Title: Neutronics model developments for the IFMIF-DONES test cell

Article References: Qiu, Y., Serikov, A., Alvarez, I., Mota, F., Čufar, A., Becerril, S., & Castellanos, J. (2026). Neutronics model developments for the IFMIF-DONES test cell. Frontiers in Nuclear Engineering, 5, Article 1783403. https://doi.org/10.3389/fnuen.2026.1783403

Image Credits: AI Generated

DOI: 10.3389/fnuen.2026.1783403

Keywords: IFMIF-DONES, neutronics, Monte Carlo simulation, MCNP, fusion materials, test cell, High Flux Test Module, damage dose, CAD-to-MC conversion, unstructured mesh, radiation shielding, DEMO