Published: September 23, 2026 | Category: Technical Guide | QSCompute
Nuclear generation, its fuel cycle, its research reactors and a growing decommissioning programme form one of the few industries where a computer can fail for a reason that has nothing to do with heat, shock or power quality: the silicon itself can be rearranged by the environment it sits in. That single fact drives every decision downstream. The specification is written around radiation, and the electronics are placed, shielded, derated and documented to prove they can still be trusted after years in the field.
The vertical is also unusually well defined. Safety-classified instrumentation and control (I&C) is already regulated far more deeply than most industries, so any edge node bolted alongside it inherits a documentation culture that expects traceability, seismic qualification and a stated support life. This guide covers the radiation effects that shape the design, the zone ladder that decides where a box is even allowed to sit, the workloads worth computing locally, and the standards a supplier will be asked to quote.
An ordinary industrial PC is qualified against the environment it will feel: temperature, humidity, vibration, ingress, conducted and radiated emissions. A nuclear-site computer adds a failure mode that no amount of mechanical ruggedising addresses. Ionising particles passing through a die deposit charge, and that charge can flip a memory cell, latch a parasitic thyristor, corrupt an instruction or permanently shift a transistor's threshold. The enclosure can be flawless and the part still fails.
The engineering response is not one technique but four applied together: put distance between the electronics and the source, put mass in the way, choose parts whose process and layout tolerate the dose, and design the system so that the faults that get through are detected and recovered rather than allowed to become wrong answers. Detection matters most, because on a nuclear site a plausible but wrong reading is worse than a missing one.
Radiation assurance is usually written as three separate budgets, because they have different causes, different units and different fixes. A buyer who conflates them will over-spend on shielding while under-specifying the logic that keeps a corrupted register from being acted on.
| Effect | Cause | What it does | Typical design response |
|---|---|---|---|
| Total ionizing dose (TID) | Accumulated gamma and X-ray energy deposition, quoted in krad(Si) or Gy | Slow parametric drift ending in permanent failure | Radiation-hardened-by-design parts, shield mass, planned replacement interval; MIL-STD-883 method 1019 testing |
| Displacement damage (DD) | Neutron and proton knock-on damage, quoted as 1 MeV-equivalent fluence | Loss of gain and efficiency in optoelectronics and analogue parts; camera and sensor degradation | Neutron-tolerant processes, avoid exposed photodetectors, remote optics via fibre |
| Single-event effects (SEE) | A single heavy ion or neutron strike depositing charge in a sensitive node | SEU (bit flip, non-destructive), SEFI (functional interrupt), SEL (latch-up, potentially destructive), SET (transient in analogue or clock paths) | ECC memory with periodic scrubbing, triple modular redundancy, watchdog and power-cycle recovery, latch-up-immune process or current-limited rails |
The practical asymmetry is that TID and displacement damage are predictable and are engineered once, while single-event effects are probabilistic and are engineered for with architecture. That is why memory scrubbing, lockstep or TMR cores, safe-state watchdogs and a defined restart path appear on every serious radiation-environment procurement list — and why "the vendor says it is rad-tolerant" is not an answer to the SEFI question.
The most effective radiation-hardening technique is distance. Dose rate falls with the inverse square of distance from a point source, so moving a node out of containment and onto a cable or fibre often beats every component upgrade available. Sites are therefore designed as a ladder of zones, and the hardware posture changes at each rung.
| Zone | Environment | Hardware posture |
|---|---|---|
| In-core / in-containment, unshielded | High neutron fluence and gamma fields during operation | Radiation-hardened-by-design MCUs and FPGAs only; typically no camera optics; often sensor and transducer only, with signal driven out on fibre |
| In-containment during outage or shutdown | Residual gamma-dominated field, accessible for maintenance | Shielded COTS in a shielded enclosure with deployed dose logging; duty-cycle-limited, replaced on dose budget |
| Spent fuel pool, hot cell, waste characterisation | Gamma-dominated, high field close to the source, strong distance gradient | COTS behind tungsten or lead, tethered or teleoperated, dose-tracked per deployment; electronics retreat with the manipulator |
| Controlled area, plant rooms, cable spreading | Low, well-characterised background | Wide-temperature fanless industrial PC with derated COTS silicon and field-dose monitoring |
| Electrical building, control room, off-site | Design-basis normal environment | Standard industrial or rack hardware; the constraint becomes seismic qualification, separation and cyber security, not dose |
Two consequences follow. First, the specification must state the zone, the expected dose rate and the total integrated dose over the maintenance interval, because those three numbers decide whether a COTS box is acceptable at all. Second, the deployment becomes a consumable: a node in a radiation field with a finite TID budget should be tracked like a filter with a service life, not treated as permanent plant.
What actually gets computed on site divides by sensor type rather than by plant system. Counting, spectroscopy and vibration diagnostics have very different sample rates and very different tolerances for latency, which is why one box rarely fits the whole site.
| Workload | Signal it actually uses | Where it runs | Compute class |
|---|---|---|---|
| Pump, motor and valve diagnostics | Motor current signature analysis, bearing and gearbox vibration at kilohertz sample rates | Controlled area, outside the field | ARM edge AI or fanless x86 |
| Spent fuel pool and dry cask monitoring | Thermocouple arrays, level, neutron and gamma counting rates | Pool deck or cask pad, shielded | Radiation-tolerant MCU plus shielded gateway |
| Dose mapping and survey | Gamma spectroscopy, dose-rate detector, LiDAR or visual SLAM for navigation | On the robot or the surveyor's cart | ARM SoC for SLAM, MCU for the spectrometers |
| Remote handling and teleoperation | Joint encoders, force/torque feedback, camera feeds, sub-100 ms control loop | Manipulator controller outside the cell | Deterministic ARM or x86 controller |
| In-containment inspection | Visual and thermal imaging over fibre, defect detection | Camera in the field, inference outside it | Radiation-tolerant camera head plus Jetson-class inference node |
| Waste characterisation | Gamma spectra, neutron coincidence counting, package weights | Waste handling building | MCU plus ARM gateway |
The recurring pattern is that the sensor must be in the field but the inference need not be. Fibre-optic links are the standard answer, and they come with their own radiation consideration: gamma exposure darkens optical fibre through radiation-induced attenuation, so harsh-environment installations specify a pure-silica-core fibre rather than a standard germanium-doped multimode cable. Copper is worse still, because the noise it couples from a high field corrupts the very signal being measured.
Radiation tolerance is one line in a dossier that a nuclear utility's engineering group will read line by line. The standards below are the ones that come back in every tender, and a supplier who cannot map product documentation onto them will not clear technical evaluation.
| Standard | Scope | What it means for an edge product |
|---|---|---|
| IEC 61513 | I&C systems important to safety | System-level architecture, independence and documented lifecycle evidence |
| IEC 61226 | Categorisation of I&C functions (A, B, C) | Decides how much rigour the node's software and hardware must carry |
| IEC 60880 / IEC 62138 | Software for category A, and for B and C systems | Development process, verification and configuration control evidence |
| IEC 62645 | Cyber security requirements for nuclear I&C | Hardening, patch and vulnerability handling over the support life — sits alongside the EU Cyber Resilience Act duties now live for connected products |
| IEEE 323 / IEC 60780 | Qualification of Class 1E equipment | Ageing, radiation and environmental qualification demonstrated as a programme, not a datasheet claim |
| IEEE 344 / IEC 60980 | Seismic qualification | Shake-table evidence for anything mounted inside a safety-classified building |
| IEC 60709 | Separation between redundant systems | Physical and electrical independence of the new node from the protected system |
QSCompute supplies the hardware half of that specification: fanless wide-temperature and shielded industrial PCs, ARM gateways for out-of-field diagnostics and Jetson-class inference nodes for inspection pipelines, each documented with the environmental and qualification evidence a nuclear engineering group asks for.
Specifying edge hardware for a nuclear, research or decommissioning site?
Send us the zone, the expected dose rate and the workload list — our engineers return a mapped bill of materials with the environmental, seismic and lifecycle documentation for each line.
Contact: +86 137-1464-6179 | info@qscompute.com