Nuclear Conversion for Starship

There has been much discussion about converting the SpaceX Starship to use nuclear propulsion. It would allow for a great increase in specific impulse and a massive extension of mission capabilities.

But is it actually worthwhile?

The image above is modified from BocaChicaGal’s photo.

Nuclear thermal rockets do indeed have impressive performance. Their specific impulse is up to three times greater than chemical rockets, they produce comparable amounts of thrust and they could be designed to accept a variety of propellants, from CO2 to ammonia. They can reduce travel times in space, push around much larger payloads and get refuelled with whatever fluid is available at their destination.

Full scale mockup of the NERVA engine.

For these reasons, they are lauded as the best way to accelerate human exploration and expansion into space. 

They are not a new technology either. The idea to use nuclear energy to propel spacecraft dates back to 1944. Serious testing has been done on nuclear rockets, with ground tests in 1955 and functional gigawatt-scale rockets firing for several minutes by the 1970s. The Strategic Defense Initiative rekindled studies into nuclear propulsion with Project Timberwind, a program that ran until 1991 and resulted in modern designs that were even more capable. 

We are now in the midst of another revival of this technology. Millions have been awarded to BWXT, General Atomics and Ultra Safe to restart the development of nuclear rockets. 

Naturally, there have been calls to combine the capabilities of nuclear rockets with the other great aerospace development of our time, which are reusable rockets and their champion; SpaceX’s 9m wide Starship and its SuperHeavy booster. Then, we would acquire the ability to send even heavier payloads to orbit and beyond.

The modern nuclear rocket

The BWXT design.

The design and performance targets for nuclear propulsion has shifted considerably over the last few decades. The initial efforts in the 1970s were straightforward in the desire for maximum power and thrust.

The reactors from Project Rover.

Project Rover, for example, resulted in the Phoebus-2A reactor that managed an output of 4000 MW for 12 minutes during a test run. As a fully developed engine, it would have managed 5000 MW and 825s of specific impulse. It would have held 300 kg of enriched uranium and had a relatively cool core temperature of 2300 K (although the goal was 2500 K). Total mass was 9300 kg, meaning it had an impressive power density of 537 kW/kg.

The Pebble Bed nuclear engine.

The ultimate version of this sort of maximum power engine would be the Pebble Bed reactors from the 1980s Strategic Defence Initiative era Project Timberwind. The largest version aimed to operate at 3000K to achieve a specific impulse of 1000s). 12,000 MW of power served to produce 2451 kN of thrust. Power density was a whopping 1450 kW/kg. 

More recent nuclear thermal rockets have taken a different direction. Each rocket is smaller, aiming for a thrust level of around 100 kN. Multiples are used to ensure reliability during operation. Core temperatures are reduced to prolong engine life. With no hope of being used within an atmosphere, they are vacuum-only designs with a reduced focus on power density. Instead, additional capabilities such as the ability to produce electricity continuously are featured. 

A Copernicus Mars vehicle equipped with three SNREs.

An example of this would be the ‘Borowski SNRE’ NERVA-derived design that produces 111 kN of thrust. When operated at 2800K, it achieves a specific impulse of 925s. Power density is about 152 kW/kg (504 MW for 3300 kg).

An even further diminished version of the SNRE is to be expected in the near future. Highly enriched uranium (enrichment level 93%) will be replaced by HALEU fuel, which has an enrichment level of 20% at most. Five times more fuel in total would be needed to complete the same missions as with highly enriched uranium fuel, or a greatly increased amount of neutron reflector is necessary to surround the reactor core to achieve the same power output. Either way, the power density will suffer. 

An additional aspect of nuclear thermal rocket development needs to be addressed: the choice of propellant. Nearly all tests and designs focus on the use of liquid hydrogen as it has the potential to deliver the highest specific impulse. However, other propellants have been considered, especially in the context of ISRU where spacecraft are refueled with whatever is available at their destination.

Most significant for our purposes is methane as propellant. It is six times denser than liquid hydrogen, can be stored at 100K, which is compatible with liquid oxygen, and it can be produced using water and carbon dioxide. At high temperatures, it breaks down into hydrogen and carbon, turning it from a 16 g/mol molecule into a 3.25 g/mol plasma. That is how it achieves a specific impulse only mildly lower than what is achievable using liquid hydrogen. Zubrin lists its specific impulse as 606s when heated to 2800K, or 625s at 3000K.  

Nuclear Starship

A nuclear-powered Starship would not be a complete overhaul of the design.

A depiction by smallstars.

It will still be a 50m tall steel tube that launches atop the SuperHeavy booster, using vacuum-optimized engines fed by large propellant tanks, and a set of smaller gimballed engines optimized for landing, with flaps to handle reentry. Dry mass in the final version will be 120 tons and about 30 tons of propellant is reserved for landing. 

It might be unsurprising to you that we cannot simply bolt on nuclear rockets to the Starship and expect everything to work. Special modifications have to be made to accommodate the new propulsion system, ranging from new attachment points to control software, but we will focus on the most impactful one: radiation shielding.

General configuration of a nuclear Starship.

The shape of the Starship is not well adapted to handling the radiation from a nuclear rocket. There are large flaps extending to the sides that could scatter radiation back into the crew compartment at the top. Retracting them when the nuclear rockets are in use would be a good idea. Designs which were meant to be nuclear from the start also usually place their reactor or nuclear rocket far from the main body of the spaceship, on the end of a long boom or tapered propellant tanks. Radiation released from a fission reaction spreads as a sphere in all directions - if it is placed further away, the main body of the spaceship intercepts a smaller fraction of it. 

The ideal rear end of a nuclear-propelled spacecraft, based on the RNS.

The fraction of radiation that cannot be avoided is handled using radiation shielding, with different layers meant to absorb different types of radiation. It is placed as close as possible to the reactors or engines to create the widest shadow of protection, which is why they are also called shadow shields.

Illustration of the shadow shield concept.

A fission reaction mainly produces fission fragments, gamma rays and neutrons. Fission fragments are heavy ions that do not travel very far. Gamma rays are penetrating photons that are best absorbed by a dense material. Neutrons are high velocity subatomic particles with no charge; they are best dealt with using a material that contains as much hydrogen as possible, like water. 

We want to use as little shielding mass as possible. 

The densest elements are the best protection against gamma rays, with tungsten (W) being an ideal choice (lead would melt too easily and depleted uranium is not practical). 

Lithium Hydride (LiH) is the most mass-efficient protection against neutrons. Boron Carbide (B4C) is 20% heavier than LiH for the same protection, but it melts at 3036K is a very strong ceramic, which is ideal for surfaces exposed to reentry heating. 

It must be noted that some radiation protection is already built-in. The beryllium or graphite reflector within the nuclear reactor prevents some radiation from leaving. The 30 tons or more of landing propellant, especially methane, is effective at absorbing neutrons too and will always be present while in space. A much larger load of propellant will be drained as the nuclear propulsion is used, representing several meters of shielding. Furthermore, there will be a 25 meter separation between the engines and the crew compartment, so only 1/625 of the radiation is actually intercepted. 

Different estimates for the shielding required place the total at about 440 kg/m^2, which corresponds to 2 cm tungsten plus 2 cm of Boron Carbide. The engines themselves are about 1 meter wide, whether they produce 500 MW or 5000 MW of power, so the shielding to be added to each engine is about 345 kg. Perhaps this estimate is optimistic, but we can rely on all the previously mentioned protections to make up for any deficiency. Consider also that the effectiveness of radiation shielding is not linear but improves exponentially with its thickness - it is easy to adjust protection levels. 

Finally, there would be significant changes to be made to supply propellant to nuclear rocket engines. We would extend them to take up the volume currently occupied by liquid oxygen. We won’t be changing the total volume of propellant tanks available to us, for a fair comparison with other versions of the Starship. If we selected liquid hydrogen, we would need specially designed tanks with insulation and active cooling. If we feed the nuclear rockets with liquid methane, we can use the same type of propellant tanks as exists today. 

The propellant tanks dedicated to landing would also have to be changed. A nuclear Starship is expected to have a heavier dry mass, so more propellant is needed to land it, which means larger tanks. 

With all these modifications in mind, let’s dive into the numbers.

Performance analysis

We will calculate the performance of a SpaceX starship equipped with nuclear propulsion.

Two nuclear rockets are considered: a 150 kW/kg near-term design operating at 2800K, and a 1000 kW/kg far-term design operating at 3000K. They will either replace all engines, or just the three vacuum Raptor engines. We will consider the use of traditional hydrogen propellant as well as the methane alternative, and either try to match the accelerations possible with chemical rockets or aim for a lower performance. 

To begin, let’s begin by breaking down the existing Starship design. 

It has three vacuum Raptors massing 1.87 tons each, three sea-level Raptors massing 1.11 tons each. These are based on their TWR figures. This leaves 111 tons of dry mass in the form of propellant tank walls, the thermal protection system, the reentry fins and other structures, for a total of 120 tons.

There are four propellant volumes: two main tanks and two landing tanks.

The main methane tank contains 614.33 m^3 of propellant, containing 268.5 tons of liquid methane at 437 kg/m^3 density.

The main oxygen tank contains 798.4 m^3 of propellant, containing 983.6 tons of liquid oxygen at 1232 kg/m^3 density.

The methane landing tank contains 13.14 m^3 of propellant, containing 5.7 tons of liquid methane at 437 kg/m^3 density.

The oxygen landing tank contains 14.56 m^3 of propellant, containing 17.9 tons of liquid oxygen at 1232 kg/m^3 density.

The total propellant mass is 1252.1 tons. 23.6 tons are reserved for landing. 

This is only possible thanks to subcooled propellants with increased density.

The Starship is meant to carry 100 tons of payload into Low Earth Orbit. With this payload, a full fuel load and its dry mass, it masses 1495.7 tons (a figure very close to the 1500 tons that SpaceX reported to the FAA). It enters orbit by expending all 1252.1 tons of propellant held in its main tanks, leaving it with 243.6 tons. Its mass ratio is 6.13. After releasing its payload, it comes in for reentry massing 143.6 tons. Alternatively, it can be refilled in orbit.

The Starship is first accelerated by the SuperHeavy booster, which has a dry weight of 200 tons and holds 3300 tons of propellant. We assume it holds 300 tons of propellant in reserve (enough for a 3150 m/s deltaV boostback plus landing, similar to the Falcon 9 booster), it provides 3150 m/s of deltaV to the Starship at staging. 

When the Starship stages off the SuperHeavy booster, all engines fire and produce 13,200 kN of thrust. This gives the Starship an initial TWR of 0.9, but by the time it has exhausted its main tanks, it has a TWR of 5.5. 

A Starship with both sea-level and vacuum Raptor engines.

The vacuum Raptors have a specific impulse (Isp) of 380s. The sea-level Raptors have an Isp of 350s. The deltaV for a Starship climbing to orbit with a 100 ton payload is 6492 m/s, assuming an average 365s Isp from all engines being used. With this 3150 m/s boost from the SuperHeavy booster, this is enough to get into orbit. 

It should take about 207 seconds for the Starship to use up all its main tank propellant with six Raptors running at full power.

If the Starship refills back up to its full 1495.7 ton mass in orbit, it will have 6765 m/s of deltaV from only using the main tanks with the vacuum Raptors, plus 380 m/s of deltaV from its landing tanks. That’s a total of 7145 m/s! However, about 800 m/s needs to be reserved to land the Starship and its payload on another celestial body, so only 6345 m/s is available for interplanetary maneuvers. 

150 kW/kg, hydrogen propellant

Using hydrogen-propelled nuclear thermal rocket engines requires the greatest modification of the Starship, but the least engine development.

Our first option is to replace the vacuum Raptor engines with 150 kW/kg nuclear engines that provide 900s of Isp. The three sea-level Raptors with their landing tanks are preserved but not used during the climb to orbit.

The main tanks are replaced by a hydrogen tank of 1412 m^3. Normally, liquid hydrogen has a density of just 70 kg/m^3, so it would contain only 98.9 tons of propellant. We generously assume that subcooled liquid hydrogen at 15 Kelvin with a density of 76 kg/m^3 is available, meaning this tank holds 107.3 tons instead.

Three hydrogen-propelled nuclear rockets are sized to produce 1523 kN of thrust using 6.72 GW of power. With radiation shielding, they mass 45.2 tons each, totalling 135.6 tons and producing a combined 4570 kN of thrust.  

We also need to expand the landing reserves to 64.7 tons to accommodate for the heavier dry mass at landing, mainly due to the hefty engines. 

This nuclear Starship has an initial mass of 100 ton payload + 135.6 ton nuclear engines + 107.3 ton liquid hydrogen + 3.33 ton sea-Level Raptors + 64.7 ton landing reserve + 111 ton other structure, for a total of 522 tons. This is a lot lighter than the original Starship, and manages the same initial TWR of 0.9, but it does us no good.

Its deltaV is 2032 m/s, mainly because it carries very little hydrogen propellant and so its mass ratio is only 1.258.  

The SuperHeavy booster can accelerate this lighter nuclear Starship to 4703 m/s before staging. The deltaV adds up to 6734 m/s, which falls far short of the 9200 m/s typically required to reach a Low Earth Orbit.

Achieving orbit is actually impossible for this vehicle. Worse, all of the additional dry mass due to the heavy nuclear engines means its center of gravity is at the bottom of the vehicle - that means it will flip over backwards during reentry!

If we remove all payload and replace the entire fairing volume of 933 m^3 with an expanded hydrogen tank containing an additional 70.9 tons of hydrogen propellant, we get the following total:

135.6 ton nuclear engines + 178.2 ton liquid hydrogen + 3.33 ton sea-Level Raptors + 64.7 ton landing reserve + 111 ton other structure equalling 492.8 tons.

Onboard deltaV rises to 3985 m/s. The Superheavy booster can add an increased 4777 m/s. It is still short of the 9200 m/s needed to reach orbit.

Removing anything more, such as reducing the landing propellant reserves or using smaller nuclear engines, just means the Starship fails earlier or later. 

Hydrogen propellant with weak nuclear thermal rocket engines is a losing combination.

1000 kW/kg, hydrogen propellant

Different DUMBO designs with 1 to 5 MW/kg.

We now replace the weak engines with the 1000 kW/kg powerhouses from decades past. An improved 1000s Isp is available.  

As before, we replace the three vacuum Raptor engines with three nuclear thermal rockets sized to produce 3568 kN of thrust in total. They add up to 18.6 tons now, including shielding, and have an output of 17.5 GW.

A necessary modification is to reduce the payload volume by 450 m^3 to accommodate more liquid hydrogen. It would bring the total propellant mass up to 141.5 tons, just enough to help it make orbit while carrying the full 100 ton payload. The landing reserve also needs to be increased to 34.3 tons.

The initial mass of the Starship becomes 100 ton payload + 18.6 ton nuclear engines + 141.5 ton hydrogen + 3.33 ton sea-Level Raptors + 34.3 ton landing reserve + 111 ton other structure, for a total of 408.7 tons. Its initial TWR is 0.91 and it has a deltaV of 4186 m/s.

The SuperHeavy booster propels this even lighter nuclear Starship to 5009 m/s at staging, allowing for a total deltaV of 9195 m/s. 

It is still hard to justify the existence of this nuclear Starship. It has less deltaV than the original Starship, and it cannot increase it much by sacrificing payload capacity. 

A trip that starts in Low Earth Orbit and ends with a landing on the lunar surface requires 5.9 km/s of deltaV, to be provided by both the nuclear rockets and then the landing engines. This is only possible if the payload was reduced to 34.5 tons.

This configuration can only land on the lunar surface by sacrificing some payload. This reduction in capability comes on top of halving the fairing volume available. The same is true for reaching Mars: it must either take a slower trajectory or reduce its payload capacity.

Furthermore, it imposes the need for three separate sets of ISRU machinery, for oxygen, methane and hydrogen, if it is to be refuelled on the lunar or martian surface for a return trip. Liquid hydrogen is the most energy-intense propellant to produce, which is an additional complication. 

To add to all these deficiencies, a new problem arises: the SuperHeavy Booster would destroy itself. Having a Starship stage that is too lightweight means the SuperHeavy booster reaches extreme velocities that a boostback burn cannot sufficiently reduce, and without any form of thermal protection, it could become too damaged to land itself. Preventing this means reserving more propellant for the boostback burn, but this in turn means the Starship stage is released at a lower velocity. For hydrogen-propelled nuclear Starships that already struggle to reach orbit, it becomes unworkable.  

In short, a hydrogen propellant nuclear Starship is not saved by better engines. 

150 kW/kg, methane propellant

Internal configuration of the advanced KANUTER design.

Previous calculations using hydrogen propellant revealed how volume-limited the Starship design was. There was no room for the bulky liquid hydrogen, and getting to orbit meant sacrificing the payload mass and volume advantages that the Starship is built around.

These could be solved by using denser liquid methane as propellant for the nuclear propulsion system. The Isp will be lower, but the mass ratios become so much better that more deltaV is available overall.

Now, let’s remove the three vacuum Raptor engines and the main tanks. In their place we add 150 kW/kg methane-propelled nuclear rockets delivering 600s of Isp and a single large propellant tank containing 617 tons of liquid methane. We also need to expand the landing tanks to 79 tons.

The nuclear engines are sized to output 28.5 GW and deliver 9682 kN of thrust each. They mass 191 tons together.

The initial mass of the Starship becomes 100 ton payload + 191 ton nuclear engines + 617 ton methane + 3.33 ton sea-Level Raptors + 79 ton landing reserve + 111 ton other structure, for a total of 1,101.3 tons. Its initial TWR is 0.9, as required, and it has a deltaV of 4834 m/s.

The deltaV is not better due to the ridiculously large engines needed to achieve a sufficient TWR. The Superheavy booster is only able to accelerate this Starship configuration to 3623 m/s, bringing the total to 8457 m/s, which is far short of reaching orbit.

Orbit is only possible by reducing the payload to 20 tons. Alternatively, we can bring 100 tons to orbit by sacrificing 450 cubic meters of payload volume to an expanded methane propellant tank. Of course, this payload will have to be very dense to fit inside the remaining volume… and the TWR will drop to 0.76!

Again, we have an unworkable nuclear Starship. Reduced payload mass or reduced payload volume are the only way to reach orbit. The mass of the engines is overwhelming. In this case, they are 63% of the Starship’s empty mass. 

The only advantage of this configuration is perhaps the high amount of deltaV within the Starship stage. It is comparable to the deltaV of the original chemical design, so it can perform the same missions. But getting barely the same performance by going nuclear is not what we want. 

1000 kW/kg, methane propellant

Time to try the most promising combination. Powerful engines and denser propellant.

625s Isp rockets with an output of 8 GW each yield 2610 kN of thrust. They mass 8.4 tons with their radiation shielding, for a total of 7828 kN of thrust and 24 tons of mass.

The same-sized methane tank holds 617 tons of propellant. The landing reserve is expanded a bit to 36 tons.

This gives the more powerful nuclear Starship an initial mass of 100 ton payload + 24 ton nuclear engines + 617 ton methane + 3.33 ton sea-Level Raptors + 36 ton landing reserve + 111 ton other structure, for a total of 891.3 tons. From this we get the first actually interesting result so far.

The deltaV of the powerful nuclear methane stage is 7209 m/s. It is finally higher than that of the original chemical configuration!

The SuperHeavy booster provides another 3944 m/s, for a total of 11,153 m/s. 

There are two ways to use this improved performance: increase the payload or reach for harder missions. 

This nuclear Starship can carry 245 tons of payload to orbit if it increased its engine thrust to 9128 kN (engine mass would increase to 29 tons). It means that the number of missions to deliver a certain payload amount to any destination is more than halved, and also that the number of refuelling missions needed to get one Starship filled up and ready to go from LEO drops from a dozen to just three.

There’s also the option to only partially fill the Starship. It can perform the orbital mission (9200m/s total deltaV) when loaded with only 323 tons of propellant instead of the full 617 tons.

Lunar missions become much easier.

The original chemical Starship could take up to 215 tons from Low Earth Orbit and land it on the Moon (5930 m/s total, with 800 m/s covered by the landing engines) but it would have to stay there. It cannot return from the lunar surface to Earth’s surface. If it had no payload, it could go to the Moon and insert itself back to Low Earth Orbit, but what’s the point in that?

The methane nuclear Starship can perform a one-way mission with 271 tons of payload, if it could land using its nuclear rockets. That figure is reduced to 234 tons if it landed on the lunar surface using Raptor engines. What’s more interesting is that it could take a reduced payload to the Moon and return on its own to its launchpad on Earth.

The 138 ton dry mass nuclear configuration departs LEO with 25 tons of payload, 617 tons of methane propellant in the main tanks and 36.7 tons of methane-oxygen in the landing reserve. It heads to the moon and lands there, consuming 520 tons of main tank methane (5930 m/s). It then unloads its payload and then heads back to Earth with a 2700 m/s maneuver. After aerobraking, it lands using sea-level Raptor engines. A great win for reusability!

Mars missions benefit as well.

The methane-propelled nuclear Starship has access to 7209 m/s of deltaV it can use for interplanetary maneuvers. The usual 120 day trip is reduced to 88 days or less. Even under the perfect alignment of planets that allow the Starship to perform the shortest possible 65 days trips, the nuclear version can shave off nearly two weeks days and bring it down to 52 days. 

As with the lunar missions, this additional performance opens up more options. For example, the nuclear Starship can load up on 165 tons of payload instead of 100 tons, while performing the same trips. 

The chemical Starship could potentially load up with 465 tons of payload and slow-boat it to Mars on a minimum energy trajectory. This nuclear Starship can do the same with 495 tons of payload, limited mostly by the huge landing propellant reserves it needs. 

Or, it could reduce its payload capacity to aim for even more deltaV and even faster trips. With reduced payload, it could widen the Mars launch window by several months, and often be able to go to Mars and return (with refueling on the surface) before the two planets move too far apart. 

So will going Nuclear be worth it?

The short answer is no.

We’ve gone through and calculated the performance of different nuclear Starship configurations and only found one that has advantageous performance. It is also the one that is least likely to exist in the near future. No large nuclear thermal rocket is being developed today, and no testing of nuclear rockets with methane propellants has ever been performed. The current efforts will revive decades-old hydrogen-propelled nuclear rockets at a scale completely unsuited for the Starship.

Elon Musk is unlikely to fund the development of the necessary technology, especially as it does not his vision for how SpaceX should operate. He wants to build a modest number of launch vehicles that are reused as much as possible. This is how a very low cost per launch is achieved. If there is a deficiency in performance, more launches and not better Starships are the solution. More expensive nuclear designs with a small performance advantage, mainly in the form of fewer launches, go against this philosophy.

Kiwi-A being tested in open air.

This comes on top of the various difficulties of developing nuclear rockets compared to chemical rockets. It wouldn’t be possible to return to work a week later if a test model explodes on the stand, which is completely antithetical to the way SpaceX operates. There’s no ‘move fast and break things’ when the US government swoops in every time things go wrong. That is, if they give Elon Musk access to enriched uranium. Or if they allow large-scale testing outside of close government supervision in the first place.

Another problem is radiation.

Nuclear rockets are safe to handle on the ground without radiation shielding or many precautions, especially when loaded with Low Enrichment Uranium. They only ignite after staging off the Superheavy Booster, far off the ground, so they do not pose a radiation threat to the launch site. If there is an accident upon launch, the uranium could be dispersed, but it is not dangerous - it is safe enough to touch (but don’t eat it)!

The challenges and safety risks of a nuclear payload in the (early) Space Shuttle.

The problem comes after the ignition of the nuclear rockets. The fuel becomes intensely radioactive. After shutdown, up to 1% of the maximum power output keeps getting released. That’s several megawatts in this case. It falls off rapidly, but radiation levels near the engine would remain lethal for days and harmful for weeks. Remember, it is unshielded around the sides and rear, so there is no protection for someone coming from those angles. NASA estimates that a nuclear rocket engine returns to its ‘safe’ state after a month. 

Some nuclear reactor designs needed to be cooled by cold propellant for several hours after use.

Rapid reuse becomes complicated. Discharging the payload to orbit and then reentering means the nuclear engines are still ‘hot’ after landing. Even if the landing itself is performed with chemical Raptor engines rather than with active nuclear engines, the residual radioactivity means that any ground crew would need to be fully protected, the refueling facilities will all have to be shielded and adding a new payload then stacking it back on top of a Superheavy Booster without contaminating them become very difficult tasks. 

Even in space, where we don’t mind irradiating the empty environment, there are issues. Approaching the International Space Station becomes impossible unless the Starship ‘cools down’ for a month in orbit. Docking maneuvers between a Starship and the craft meant to refuel it have to be done along a narrow corridor between each ship’s radiation shadows. Moon landings take place about 3 days after departure from orbit and the use of the main engines. Nuclear rockets would still be ‘hot’ by then and dangerous to any astronaut approaching from the surface. They would have to land far away from any lunar bases, and rely on shielded rovers to transfer payload across the Moon’s surface across a safe distance. The lack of any air to grant free radiation shielding means this safe distance will be very large.

It is less of an issue for Mars missions. Even the shortest missions take more than 2 months and this gives enough time for the nuclear rocket engines to become safe again. Landing is done with chemical rockets, so the Starship is safe to approach once the Martian surface. But this is necessarily a smaller number of missions compared to the Earth-Earth or Earth-Moon missions. 

And finally, there’s the ISRU.

Martian Starships return to Earth after being refilled by propellant produced by CO2 and water found locally. Vast fields of solar panels or fission reactors produce electricity to crack these molecules and reform them into oxygen and methane. A methane nuclear Starship needs nearly three times as much methane than a chemical Starship. It needs no oxygen, but that is a byproduct of the reaction that produces methane anyway - it is not a true saving. Three times more methane means that ISRU facilities have to be three times bigger or refuel three times less Starships, a hefty penalty. 

Can the Nuclear Starship be saved?

It is possible to envision a nuclear Starship in the far future. Someone else decides to develop the necessary methane-propelled propulsion technology. Perhaps the basic Starship is adapted to carry more propellant volume, increasing overall mass ratios and making full use of the increased exhaust velocity. And maybe large launch facilities are constructed for billions of dollars to refuel radioactive Starships on the ground, like those once proposed for handling nuclear-powered bombers during the Cold War.

The GE Beetle, designed to handle radioactive B-36 bombers.

But it is more likely that none of these things happen. Huge performance gains could be had by specializing the chemical Starship for Lunar or Martian missions. These would never land back on Earth’s surface, but they perform their own missions far better than a multi-purpose Starship ever could. And let’s not forget that the final dry mass of the Starship will be lower than 120 tons. SpaceX has released (and tweeted) estimates as low as 60 tons for the final uncrewed version. At the extreme, we have the Starship Lite, stripped of all aerodynamic features, payload fairing and landing systems. It would have a deltaV of 12.7 km/s, thanks to a dry mass of just 40 tons. 

If we need the full performance advantage of nuclear propulsion, we should design a spaceship that is intended for it from the get-go. It never lands, only going from orbit to orbit, so there is no need for heat shielding, flaps, high thrust engines, thick steel structure or aerodynamic shaping requirements. Without these constraints, it can instead utilize huge hydrogen tanks and a lightweight structure made of aluminium or carbon composites. Low pressure rockets with 1300s of Isp would be available, as there is never a need for high thrust. 

A 30 ton craft with 10 tons of nuclear propulsion, 263 tons of hydrogen propellant and 100 tons of payload would have 13,500 m/s of deltaV, enough to get to Mars in 100 days and brake into a low orbit.

It is fast and economical and closer to the current vision for nuclear-powered transportation than a Starship conversion.