General Fusion Plasma Injector PI3

• PI3 is a spherical tokamak that uses Ohmic heating to produce well-characterized deuterium plasmas for fusion diagnostic research.
• It accelerates plasma via a coaxial Marshall gun into a water-cooled flux conserver, achieving up to 50 kA current and generating up to 10^8 D–D fusion neutrons per shot.
• Advanced neutron detection combined with Monte Carlo sampling enables time-resolved ion temperature inference, validated against ion Doppler spectroscopy readings.

General Fusion’s Plasma Injector 3 (PI3) is a spherical tokamak specifically engineered as a purely Ohmically-heated, low-density experimental platform to advance diagnostic techniques for fusion plasmas. Operating from 2017 to 2024 at General Fusion’s Richmond facility, PI3 was designed both as a test-bed for magnetized target fusion configurations and as a source of well-characterized deuterium plasmas for time-resolved neutron-based diagnostics, with core applications including the inference of fuel ion temperature from neutron counting (Radich et al., 4 Jan 2026).

1. Device Architecture and Operational Parameters

PI3 consists of a 1 m-diameter, 1.5 m-long vacuum chamber, into which a magnetized deuterium plasma slug is injected and accelerated via a coaxial "Marshall" plasma gun. The plasma is confined within a stationary, water-cooled aluminum flux conserver. Upon entry, plasma completes its own poloidal magnetic circuit by generating a toroidal plasma current $I_p$ (up to 50 kA), inducing a poloidal field $B_θ$ of approximately 0.05 T. An auxiliary static toroidal field $B_φ$ up to 0.1 T is maintained by an axial current passing down the central shaft for MHD stabilization.

Typical plasma shots achieve a line-averaged electron density $\langle n_e \rangle ≈ 3 \times 10^{19}$ m$^{-3}$, with core electron temperatures $T_e > 400$ eV and lifetimes up to 30 ms under favorable "catch" conditions. Deuterium fueling is achieved by a prefill gas puff at a few Pascal, supporting the formation and acceleration of a $\sim$1 m long, $\sim$0.1 m wide plasma slug via a 5–10 kV discharge. Plasma propagation into the flux conserver is driven by $J \times B$ thrust. Martialing these parameters, PI3 routinely produces up to $10^8$ D–D fusion neutrons per shot.

2. Neutron Detection Array and Absolute Efficiency Calibration

Neutron yield measurements in PI3 utilize a surrounding array of four uncollimated organic liquid scintillator detectors—two 0.83 L EJ-309 and two 3.49 L EJ-301 units—each shielded within a 25 mm-thick lead-lined steel enclosure and coupled to dedicated PMTs. Both high- and low-gain anode outputs are digitized at 1 GS/s for time-resolved analysis.

Detection thresholds for proton-recoil energies are 1.0 MeV (for SC9, SC10) or 0.5 MeV (for SC12, SC13), which are selected in software to suppress low-energy backgrounds. The absolute, position-dependent detection efficiency $\epsilon_i(E_\text{thr})$ for each detector is established via MCNP6.2 neutron transport simulations. These simulations model 2.45 MeV neutrons emitted isotropically from a 2D $n_e^2$-weighted source distribution reflecting the reconstructed plasma. PTRAC tallies provide event-level proton recoil data in the scintillator volumes, which are post-processed using a nonlinear light response model (with effective electron-equivalent energy $E_\text{ee}$ mapped from proton energy $E_p$ as $E_{ee} = A E_p - B [1-e^{-C E_p^D}]$). The efficiency is defined as the ratio of detected events above threshold to total simulated source neutrons, with statistical uncertainties of 1–3 %.

The time-dependent total neutron yield $Y(t)$ is derived as a weighted combination of pile-up–corrected count rates from each detector:

$$Y(t) = \frac{\sum_i \dot{N}_i(t)}{\sum_i \epsilon_i}$$

where $\dot{N}_i$ is the local detector rate and $\epsilon_i$ its efficiency (Radich et al., 4 Jan 2026).

3. Signal Discrimination and Event Corrections

Discrimination between neutron-induced and gamma-induced pulses in each detector is achieved through pulse-shape discrimination (PSD). PMT pulses exceeding digitization threshold are processed by integrating "tail" and "total" light regions, comparing their ratio to a power-law bifurcation curve established using calibrated Co-60 gamma-ray sources.

Observable pulse pile-up events—characterized by temporal overlap—are excluded from PSD and binned separately. In each time bin $\Delta t$, let $N_{\text{PSD},n}$ and $N_{\text{PSD},\gamma}$ denote the neutron and gamma counts identified by PSD. The neutron fraction is defined as

$$P_n = \frac{N_{\text{PSD},n}}{N_{\text{PSD},n} + N_{\text{PSD},\gamma}}$$

The excluded pile-up count, $N_\text{pu}$, is multiplied by $P_n$ to estimate the number of neutrons lost to pile-up, $N_{\text{pu},n}$, which, divided by $\Delta t$, gives the pile-up neutron rate:

$$\dot{N}_{\text{n}}(t) = \dot{N}_{\text{PSD},n}(t) + \dot{N}_{\text{pu},n}(t)$$

This aggregate, pile-up–corrected neutron signal enhances the accuracy of subsequent neutron yield and plasma parameter inference.

4. Ion Temperature Inference via Maxwellian-Plasma Formalism

Assuming a Maxwellian velocity distribution for deuterium ions, the local fusion reactivity $\langle \sigma v \rangle(T_i)$ for the D(d,n)$^3$He reaction is evaluated using the Bosch–Hale fitting formula. The plasma is partitioned into 20 flux-surface-indexed shells, with reconstructed $n_{d,\text{shell}}$ and $V_\text{shell}$ (from Bayesian equilibrium and density reconstructions). The neutron yield from each shell is

$$Y_{\text{shell}} = \frac{1}{2}\langle \sigma v \rangle_{\text{shell}} n_{d,\text{shell}}^2 V_\text{shell}$$

The total measured yield thus satisfies

$$Y(t) \approx \sum_{\text{shell}} Y_{\text{shell}}(T_i) = \frac{1}{2}\int \langle \sigma v \rangle(T_i(r))n_d^2(r)dV$$

For the analysis, parameterized ion-temperature profiles of the form $T_i(\bar{\psi}) = T_i(0)(1-\bar{\psi}^2)^\alpha$ are considered, with $\alpha=0$ (flat) and $\alpha=1$ (peaked). A forward table of volume-averaged reactivity is precomputed across a range of trial core temperatures. From the experimentally measured $Y(t)$, the experimental average reactivity is computed as:

$$\overline{\langle \sigma v \rangle}_\text{exp}(t) = \frac{2Y(t)}{\sum n_{d,\text{shell}}^2 V_\text{shell}}$$

Interpolation within the forward table yields the core ion temperature $T_i(t)$ at each instant.

5. Uncertainty Quantification by Monte Carlo Sampling

Uncertainty in inferred ion temperature arises from several sources:

• Poisson statistical noise in the discrete neutron counts ($\sigma_\text{rate} \approx \sqrt{N}$)
• 1–3 % propagation of MCNP sampling uncertainty in $\epsilon_i$
• Uncertainties in reconstructed densities $n_{d,\text{shell}}$ and volumes $V_\text{shell}$
• Ambiguity in the assumed $T_i$ profile shape ($\alpha$)

A Monte Carlo procedure (typically $N_\text{MC} \approx 10,000$ trials) independently perturbs each above input using their respective $1\sigma$ errors, reconstructs the neutron yield and reactivity, and inverts for $T_i$ as described. The final estimate adopts the median over samplings, with the 16th and 84th percentiles reported as asymmetric error bounds.

6. Experimental Outcomes and Comparison to Spectroscopy

Evaluation of PI3 shots via this neutron-counting diagnostic demonstrates time-resolved core ion temperature measurement. In a representative example (shot 22714), total neutron yields reached $2.5 \times 10^7$ within the first 6 ms, and neutron-inferred core $T_i$ rose from $\sim 300$ eV at 1 ms to 500 eV at 4 ms for flat ($\alpha=0$) profiles, with systematically 10–20 % higher values for peaked ($\alpha=1$) profiles. Reported uncertainties (±20–50 eV) are dominated by density, efficiency, and profile-shape effects.

Comparative analysis with simultaneous chord-integrated ion Doppler spectroscopy (IDS) using the C V line at 227 nm (acquired across 33 shots at $t=2~\text{ms}$ and $3~\text{ms}$) yields consistently lower temperatures (200–400 eV) than the neutron-based method (400–800 eV). This observation supports an interpretation that IDS, with its off-axis, edge-sensitive viewing geometry, underestimates the hotter core region as captured by neutron emission, which is centrally weighted.

A summary of these experimental comparisons is presented in the following table:

Diagnostic Method	Typical $T_i$ Range (eV)	Spatial Weighting
Neutron counting + inversion	400–800	Weighted toward plasma core
Ion Doppler Spectroscopy	200–400	Chord-integrated, edge-biased

This suggests a peaked-core temperature profile and validates the neutron counting technique as a robust, time-resolved, remote ion temperature diagnostic—particularly valuable for configurations where line-of-sight optical access is obstructed by hardware, as is expected in future magnetized target fusion systems (Radich et al., 4 Jan 2026).