Skip to content

Design of a diamond-based in-vessel soft x-ray camera system for the SPARC tokamak

FieldValue
JournalReview of Scientific Instruments
Published2026-10-01
AuthorsStanley J. Normile, Didier Vezinet, Blair Frandeen, Conor Perks, F. Bombarda
DOI10.1063/5.0350758

The proposed SPARC soft x-ray tomography system uses four in-vessel cameras and 60 CVD single-crystal diamond Schottky photodiodes, with 15 detectors in each camera. The design predicts minimal photodiode degradation at 1 × 10^15 n cm^-2, which should allow operation into SPARC’s third campaign. The design represents the first tomography camera to use diamond photodiodes as an array, following single-detector demonstrations on FTU and JET.

Four cameras supply the overlapping views needed for tomographic inversion. The maximum observable poloidal mode number depends on camera overlap, while radial resolution depends on lines of sight within each camera. At least two cameras are needed to observe the m = 2 tearing mode.

The authors place four cameras across the upper, midplane, and lower port plugs in one poloidal plane, following their earlier placement study. This arrangement provides good coverage of the vacuum vessel and places the greatest overlap in the brightest plasma region.

Each camera uses 15 photodetectors because SPARC’s cryostat passthroughs support 60 detectors. Soft x-ray emission from the plasma is optically thin and line integrated, so detector views are combined to reconstruct the spatial emission distribution.

The authors modeled the spectral emissivity of a representative SPARC Campaign 1 D-T L-mode plasma with the TOFU Python package. The model produced a low-energy emissivity spike at the outer plasma edge. Impurity line radiation causes this spike, which harms reconstruction and can corrupt the Bremsstrahlung measurement used to infer plasma pressure. The detector and filter design therefore had to suppress low-energy light without discarding useful signal.

The detector uses a platinum Schottky contact on CVD single-crystal diamond. The authors chose a Schottky diode because its lack of minority charge carriers supports a fast response with a wide-bandgap material. The active layer contains a depletion region and a diffusion region. The active thickness equals the depletion depth plus the diffusion length. This defines the depth in which absorbed photons can contribute to photocurrent.

The paper uses a relative dielectric constant of 5.7 and a built-in potential of 2.1 V. It also uses a measured defect concentration of 3.1 × 10^14 cm^-3 and a diffusion length of 2.60 μm. At zero bias, the depletion layer is 2.0 μm thick.

The electric field reaches its maximum next to the Schottky contact, so junction breakdown sets the operating limit. The estimated breakdown field is about 10 V μm^-1 for these defect-containing diodes. This is far below diamond’s theoretical 1 kV μm^-1. The resulting maximum reverse bias is about 50 V, corresponding to a 10 μm depletion length.

Measured 25 μm active-layer photodiodes showed no charge collection efficiency decrease at 2.0 × 10^14 n cm^-2. Their charge collection efficiency fell to 47% at 2.0 × 10^16 n cm^-2. The paper places the failure fluence of conventional silicon diodes at 1 × 10^13 n cm^-2, giving the diamond measurements a clear radiation-tolerance advantage.

The authors modeled charge collection with the Hecht equation. Their SPECTER calculation gave a displacement cross section of 4.22 × 10^-22 cm² for 14 MeV neutrons. The model also used an initial mean drift distance of 400 μm and an equivalent damage factor of 3 × 10^4 μm^-1. A separate experimental fit gave 1.4 × 10^4 μm^-1, but it used only two measurements.

Under the fitted model, a 25 μm active layer maintains more than 50% charge collection efficiency to 1.8 × 10^16 n cm^-2. A 5 μm active layer remains above 50% to 8.8 × 10^16 n cm^-2. Thicker active layers give higher initial responsivity but degrade faster as neutron-induced traps reduce the carrier drift distance.

The model treats responsivity as the unirradiated responsivity multiplied by charge collection efficiency. It does not include neutron-induced reductions in breakdown field. The safe reverse bias could therefore fall below 50 V, and the 50 V design could lose performance faster than the model shows.

Earlier diamond work retained more than 50% of initial responsivity at 1 × 10^16 n cm^-2. SPARC’s expected fluence exceeds that level, so the camera also serves as an in-place test of diamond failure.

A 2 μm graphene filter performs nearly like a 10 μm beryllium filter in the modeled detector response. Its narrow responsivity spike near 250 keV does not affect rejection of the low-energy plasma-edge spike. Thin graphene needs a support mesh, which adds about 15% attenuation. The analysis includes this loss.

Graphene also avoids the beryllium filter’s toxicity, tritium retention, and 855 K melting temperature. Its sublimation temperature is about 4000 K. Early research suggests graphene may self-heal from radiation damage under some conditions, but the paper does not treat self-healing as a guaranteed camera behavior.

Filters reduce signal, and reverse bias cannot recover the attenuation. The modeled graphene filter therefore retains low-energy rejection while accepting the lower signal available from the SPARC plasma profile.

The readout model includes shot noise, Johnson noise, downstream amplifier noise, and digitizer quantization noise. Electromagnetic interference is excluded because it depends on cable geometry. The authors limit cable runs and use shielded cabling. Transimpedance amplifiers sit in the tokamak basement, which shortens the cable run and gives the amplifiers neutron shielding.

At SPARC’s 200 °C operating temperature, dark current remains in the picoampere range at every modeled reverse bias. It can therefore be neglected. The bandwidth study holds overall gain at 1 × 10^5 V A^-1.

Reverse bias raises responsivity and lowers diode capacitance, which increases bandwidth. The paper also treats higher reverse bias as reducing neutron hardness, so the performance gain shortens useful life.

The modeled operating points are:

Reverse bias (V)Feedback resistance (kV A^-1)3 dB bandwidth (kHz)SNR range (× 10^3)Fluence for half initial SNR (n cm^-2)
0116002 to 151.1 × 10^17
10157002 to 166.9 × 10^16
50207502 to 184.2 × 10^16

Resolving MHD activity at about 100 kHz requires at least twice that bandwidth. A value near five times the desired frequency is recommended for better reconstruction. The modeled bandwidths exceed both targets at all three operating points.

ASDEX Upgrade work identified an SNR of 10 as the lower usable limit for tokamak x-ray tomography. The unirradiated SPARC diodes are expected to have at least 100 times higher SNR. Because neutron-model and EMI uncertainties prevent a sound extrapolation to SNR 10, the authors compare the fluence at which SNR reaches half its initial value. SPARC will use unbiased photodiodes because their longer lifetime outweighs the modest losses in bandwidth and SNR.

All camera materials must withstand a large number of 24 h bake-ins at 350 °C, although the camera does not operate during them. Oven experiments found that cycling the diamond photodiodes to 350 °C did not damage the diodes or their microwire-bonded connections. The cycles also caused no performance loss.

The design must also resist heat from the plasma and avoid alignment shifts caused by thermal expansion. That requirement ruled out the previously considered Kapton filter. Finite element analysis indicates that the graphene foil can survive a disruption if it sits directly in front of the photodiodes. Placing it at the aperture would not meet that condition.

The main unresolved issue is the sparse basis for the neutron-damage model. Additional SPARC operation will provide data on photodiode degradation, allowing the model to be checked and tuned. The authors also state that no known detector material can survive the neutron flux expected at power-plant scale. They therefore treat the camera as an early in-vessel diagnostic and a source of training data for models based on sensors farther from the plasma.

The camera requires CVD single-crystal diamond photodiode material with Pt Schottky contacts for SPARC’s in-vessel soft x-ray tomography. The diamond materials catalog lists MPCVD-grown single-crystal diamond and Pt metallization, which correspond to those material requirements. 6CCVD’s engineering support can review the custom photodiode geometry and camera integration requirements.

View Original Abstract

Soft x-ray tomography has been used in tokamaks for decades to interrogate a wide variety of phenomena; however, the standard silicon diodes coupled with beryllium filters will not be able to survive the high heat and neutron fluxes present in burning plasma devices like SPARC. Presented here is the design for a set of soft x-ray tomography cameras that use chemical vapor deposition (CVD) single-crystal diamonds and graphene filters for photodetection. Four of these cameras will be placed in the port plugs of SPARC with a large enough view of the poloidal cross section to enable tomographic inversion and sufficient time resolution to resolve the dynamics of low-m and -n magnetohydrodynamic modes. CVD diamonds have been used successfully for the detection of both vacuum ultraviolet and soft x-ray photons at the Frascati Tokamak Upgrade and have demonstrated the ability to maintain greater than 50% of their initial responsivity at a neutron fluence of 1016 cm-2. The neutron dose in SPARC will exceed this limit, providing a demonstration of how CVD diamonds fail in place while adding value to the early operation of a new tokamak. Numerical simulations indicate that graphene spectral filters will filter nearly as effectively as traditional beryllium ones and are able to resist significantly higher heat fluxes due to graphene’s unique thermal properties. Simulated plasma emissivity profiles have been used to optimize both the camera placement and detector properties.