The Automation Core

Microreactor Design Explained: Inside MARVEL's NaK Natural Circulation, Control Drums and Stirling Engines

· control drums, dry criticality, HALEU, Idaho National Laboratory, MARVEL microreactor, microreactor, NaK coolant, natural circulation, TRIGA UZrH fuel,

 


Bottom line: AI’s power problem keeps pushing on-site generation into the conversation, from fuel cells to small nuclear. This month one microreactor took a concrete step. Idaho National Laboratory (INL) said on Oct 6 that DOE approved the final documented safety analysis for MARVEL. Its primary coolant system arrived in mid-September, and the target is dry criticality in December. MARVEL is small, at 85 kWt and roughly 10–20 kWe, but its size makes the engineering trade-offs easy to see. Four points follow. Derived figures are my own illustrative arithmetic, not official data.

What happened

  • Oct 6, INL: DOE’s Idaho Operations Office approved MARVEL’s final documented safety analysis, authorizing a dry-criticality configuration. The primary coolant system (PCS) reached the Transient Reactor Test Facility (TREAT) in mid-September, and criticality is anticipated in December.
  • Oct 9, Nuclear Engineering International: MARVEL is designed for 85–100 kWt and about 20 kWe. It stands roughly 15 ft tall and 4 ft wide. Its preliminary safety analysis was approved in March, and full operation with a microgrid connection is expected from late 2027.
  • Oct 17, Radiant: its 1 MWe Kaleidos microreactor at INL’s DOME test bed was heated above 575°F (about 302°C) with an electric heater in the cooling system, wrapping up non-nuclear testing. No criticality had been announced as of Oct 21.
  • Context: The Duck Curve’s Oct 21 tracker counts five DOE-authorized advanced reactors that went critical between June 4 and August 5. None holds an NRC operating license.

What “dry criticality” means: the fuel is loaded but no coolant is present. The chain reaction is brought to exactly self-sustaining at near-zero power to validate core physics, control-drum worth and instrumentation. Wet criticality (with coolant) and full-power operation are separate, later milestones.

Engineering point 1: no pump in the primary loop

MARVEL’s primary system is a four-loop circuit holding about 120 kg of NaK. The core heats the NaK to 500–550°C. Hot metal rises through the core barrel to the intermediate heat exchangers (IHXs), cools, and falls back through four downcomers to the lower plenum. Buoyancy alone drives the whole loop, so there is no primary pump. INL notes that NaK is liquid at room temperature and needs no pressurization. Compare that with roughly 150 bar on a pressurized-water reactor primary (textbook order of magnitude) and the pressure boundary gets much simpler. The vessel is 316H stainless steel designed to ASME Section III, Division 5. The catch: NaK reacts violently with water and air (textbook fact), so the system relies on argon cover gas and a guard vessel. Removing the pump removes a rotating part that can fail, and decay heat still leaves passively if power is lost.

Engineering point 2: the fuel is its own brake

The core holds just 36 fuel rods of TRIGA uranium-zirconium-hydride (UZrH), the fuel used in many university research reactors. Each rod is 30 wt% uranium enriched to 19.75% (HALEU), with five fuel slugs instead of the usual three. The zirconium hydride acts as the moderator, so when fuel temperature rises the neutron spectrum hardens and reactivity drops. INL’s fact sheet says the hydrogen and graphite in the fuel “ensure safety in cases of loss of power or other abnormal conditions.” The first line of defense is materials physics, not electronics. Core life is two effective full-power years.

Engineering point 3: control drums instead of rods

Four rotating control drums surround the core. Each drum is a beryllium-oxide reflector with about one-third of its surface covered by a boron-carbide absorber. Turn the reflector face toward the core and neutrons bounce back, raising reactivity. Turn the B4C face inward and neutrons are absorbed. A central shutdown absorber stays fully withdrawn in operation and drops in for shutdown. At the ANS annual meeting (reported June 8), INL said the only control-drum design with a scram they knew of was the 1960s SNAP-10A space reactor. A terrestrial system must deliver both a fast, reliable scram and slow, precise trimming, and almost none of the bearings, springs, motors or switches existed off the shelf in nuclear grade. Some core parts needed tolerances of 0.0005 in (about 12.7 µm, my conversion).

Engineering point 4: the Stirling lesson and the scale reality

Power conversion was meant to be four off-the-shelf free-piston Stirling engines, each 5–7 kWe per a 2024 OSTI technical overview, mounted directly on the reactor. In testing, the vibration was so bad an engineer worried “the building was going to shake apart.” The fix was to decouple: the engines went back into their original crates with rubber bumpers, and the new design puts two heat exchangers between engines and reactor. Official electric ratings range from about 10 kW to 10–15 kWe to under 20 kWe. Dividing 10–20 kWe by 85–100 kWt gives a system efficiency of roughly 10–24% (my arithmetic), against an INL model figure of 20–25% Stirling efficiency at a 500°C inlet. Scale check: matching one 1 MWe Kaleidos design takes about 50–100 MARVELs, and each Xe-100 module in the AEP–X-energy–Amazon study announced Oct 19 is 80 MWe. MARVEL is not meant to power a data center. Its job is to rehearse safety analysis, drum control, remote monitoring and microgrid integration so commercial designs make fewer mistakes.

Open questions

  • December dry criticality is a target that depends on assembly, fuel and readiness reviews, and it could slip.
  • Electric output differs across INL’s FAQ (~10 kW), ANS (10–15 kWe) and NEI (~20 kWe). Only a full-power run will settle it.
  • Full-power timing reads late 2027 in NEI and 2028 in INL’s older FAQ.
  • HALEU supply: DOE said in July that HALEU “is not currently available from domestic suppliers,” a shared bottleneck for every microreactor.
  • Radiant’s Q3 target for a 150-hour full-power run has passed with no criticality announced as of Oct 21.

Related videos

Your call: where will the first microreactors actually earn their keep: military bases, remote mines, or AI data centers? Tell me why in the comments.

← Newer articles Older articles →