You are on page 1of 4

TOKAMAX

Part A

As part of their normal pre-run checklist, Many experts say commercial fusion would be
Tokamax CEO Grace Riley and CTO Ned Jones decades away if we already knew how to build
excused themselves from the buzzy scene of their reactors, but the enormous payday of succeeding at
control suite and reactor floor area to confer about that challenge is attracting ambitious investments in
the final go/no-go decision for the ambitious run a new wave of projects and companies. In 2022, for
they were about to undertake. Video crews were in example, the DOE announced a commercialisation
the final stages of setting up cameras everywhere to plan that includes $50M for investment in a
capture the action, but in the Tokomax ethos, this magnetic fusion plant, with the goal of a practical
decision point is part of a checklist that is more than reactor in only a decade. Whilst governments have a
a formality. Everyone knew the decision could be to history of over-promising on timelines for major
go ahead or postpone. The Tokamax way is to be projects in basic science and energy, this optimism
ambitious, yes, but science-the-you-know-what out about timing is echoed by developments reported in
of the details to achieve great things. a 2021 Nature article on fusion, “The chase for fusion
energy”. According to the newly created Fusion
The video crews and Tokamax investors were
Industry Association (established 2018), more than
hoping Grace and Jo would proceed and succeed
US$5B has been invested in private fusion efforts
with their goal of today’s run. Data from their recent
over just the last two years—twice the investment
successes suggest if they pushed their systems to their
seen in the prior two decades (see also McKinsey).
limits, they could just break the record for the triple-
product of magnetic plasma confinement. Based on Though there are a number of competing
their data and, the go/no-go decision Jo and Ned approaches to reactor design, they all use very high
were about to make would commit to an attempt fuel densities and temperature inspired by the
that, if successful, would blow minds in their physics of solar fusion. Controlling these star-like
community by vaulting to a run time way beyond temperatures and pressures is an extraordinary feat
the current art. Success would mean big things not of engineering.
just for Tokamax and the team, but for the world.
British fusion start-up Tokamax is a leader in
At the same time, such a push risks damages magnetic confinement fusion. Their distinctive skill
that could be serious. Jo and Ned are aware the is using AI-enabled controllers of super-conducting
team have concerns that this “stunt” could damage magnets to achieve superior confinement and
the reactor enough to put the company at risk. They configuration of reactor fuel in place as it reaches
take this seriously, but with everything to play for, temperatures ten times hotter than the sun’s core.
they are feeling it is time to go all-in. The name Tokamax is a riff on tokamak, a toroidal
shaped reactor enclosure.
Background
In the race to build systems that deliver plasma
For decades, fusion has always been 30 years
confinement systems suitable for commercial fusion
away, but a fusion race is now on! Recent advances
reactor designs, Tokamax is part of a rich ecosystem
have cut predictions about when we will see
of different kinds of labs and organisations seeking to
practical fusion. In 2022, the National Ignition
push frontiers in temperatures, fuel densities, and
Facility (NIF) at the Lawrence Livermore National
confinement times to thresholds of this fusion triple
Laboratory (LLNL) in California achieved an
product needed for commercial fusion.
historic milestone with the first instance of fusion
ignition. As LLNL is funded by the US Department Overall, the twin-goal of plasma confinement is
of Energy (DOE), Secretary of Energy Jennifer to avoid both total disruptions of plasmas and edge-
Granholm joined the LLNL team at the press localized modes (ELMs), which are temporary
conference called to announce their achievement. In instabilities in the perimeter regions of plasmas.
her comments about the ignition event, Secretary Whereas fuel leaks are very dangerous in nuclear
Granville said, “It's the first time it has ever been fission reactors, there is little risk that the deuterium-
done in a laboratory anywhere in the world — tritium fuel used in fusion reactors could escape and
simply put, this is one of the most impressive cause significant environmental damage. Tritium is
scientific feats of the 21st century.” LLNL Director radioactive and has a half-life of 12 years, but
Kim Budil also spoke, putting the ignition plasmas that disrupt just disperse inside the reactor
achievement in context by looking to the future: enclosure, which just shuts off the reaction. ELMs
“Ignition is a first step—a truly monumental one usually don’t shut off reactions, but they can interact
that sets the stage for a transformation decade in with plasma-facing components (PFCs) in the wall
high energy density science and fusion research— blankets of reactor enclosures, damaging systems
and I cannot wait to see where it takes us.” that capture energy or breed the tritium used in the

© 2023. Mark Thomas Kennedy.


reactor fuel. Theoretically, confinement failures China: from 100M °C for 10 seconds in 2018 to a
could damage reactor enclosures badly enough to 120M °C for 101 seconds in 2021. This gain is
cause fuel leaks, but again, the dangers of such already greater than the theoretical maximum from
events are low because such damage is very unlikely pushing harder on temperature and density.
given the small quantities of fuel used and the low
To grab the lead in this race, the Tokamax
radioactivity of tritium. That said, confinement
team have broken new ground in applications of
failures can cause damage to sensitive systems
reinforcement learning (RL), a major family of
requiring costly repairs and, even worse, reactor
machine learning in AI, to tokamak operation and
downtime that means falling behind in the fusion
design. Specifically, they have used RL skills to build
race.
an innovative programme for fusion R&D that they
Decision Time call “Fast Forward Mode” (FFM). In essence, FFM
is a system that uses their physical reactor and a
Tokamax CEO Jo Riley and Principal Scientist
high-fidelity digital twin to simulate and learn from
Ned Jones faced a tough decision: should they push
longer and longer runs. FFM is built around two
their system to set a record for the triple-product of
neat tricks. First, a second set of controllers is added
magnetic plasma confinement? Recent runs suggest
to the blanket to induce ELMs and plasma
they just might succeed, and savvy onlookers are
disruptions that match the profiles of similar events
watching closely to see what they can do.
they encounter in test runs. Second, reactor sensors
Setting the record would be a game-changer were modified to achieve ultra-fast identification of
for Tokamax. It would land $500M investment and ELMs, and magnet controllers were optimised for
put them at the top of the table of teams competing super-low latency in controllers that correct
to build commercial fusion reactors. The money and problems in plasma confinement as they arise. With
prominence would put them in a position to learn the speed of these systems, the team is able to do
from a higher rate of simulations and experiments in runs in which events follow the cadence of arrival
systems under development by a range of potential times for real events, but at a rate that makes real
customers—i.e., the programme management teams time seem dilated. Thus, FFM enables Tokamax to
and general contractors leading reactor builds and practise detecting and containing the volume of
testing programmes at both research labs and fusion plasma confinement problems that should occur in
start-ups. Also, their name would become known for much longer runs. Tokamax have now built the
contributing to giving the world an energy source to world’s largest data set of real plasma confinement
replace fossil fuels, thereby offering a way out of the problems and recoveries, and this has enabled them
climate change crisis. The combination of money, to generate even larger scale synthetic data sets of
prominence, and brand would give them a source of confinement problems and recovery events. FFM
significant competitive advantage needed to survive provides data useful for identifying and managing
the rising heat of competition coming as the fusion the wider array of wonky ELMs, like edge cases,
race heats up. There was everything to play for. only seen in much longer runs. Data from actual
reactor runs was used to refine the magnets that
At the same time, going for the record comes
perturb the plasma and related routines in the
with the risk of significant costs and delays. Whilst digital twin of the reactor.
the triple-product includes temperature, density, and
time, Tokamax is focusing on time because of limits Based on data from physical and virtual runs of
on reactor vessel pressure (temperature x density). increasing duration, the team have set an extremely
Increasing density eventually causes a phenomenon ambitious time goal: run for 24 hours without
physicists call bremsstrahlung, a process that radiates interruption. Internally, they refer to this stretch
away a plasma’s power. Based on current theoretical goal as “Le Mans”, and they can often be seen in T-
understanding of magnetic confinement fusion, it is shirts sporting the logo of the 24 hours of Le Mans,
feasible only between 100 and 200 million °C and at the famous endurance race that takes place each
vacuum-like pressures—on the order of a million year at the town of Le Mans, France. For Jo and
times less dense than air. New theoretical work on Ned, the race provides a metaphor that has helped
density suggests it could be increased by a factor of staff to focus on extreme endurance. As Jo puts it,
two over widely used “Greenwald limits” derived
“Practical fusion requires scaling confinement
from empirical results. Again, based on current
times—radically. Current knowledge suggests
understanding, neither temperature nor density can
longer times are the key, so we are lasered in
be increased beyond 2x current benchmarks.
on that. Our goal? Run for 24 hours at
Because power output increases with the square of
benchmark temps and densities. That’s almost
pressure, gains from temperature and density appear
a 100x increase over current bests. Hitting that
to be capped at 8x (23).
time would mark the dawn of a fusion age.
Time, however, is a different story. Consider Everything about power would be different
performance increases at the EAST Reactor in than everything that came before.”

© 2023. Mark Thomas Kennedy.


The Risk-Reward Equation Just before Jo and Ned left the room to confer
on the go/no-go checkpoint, Pat Smythe, Director
Say Tokamax produces a physical and virtual
of the Blanket Engineering Team (BET), asked for a
run that validates their model’s potential: what
quick word.
might that be worth? Consider the California team
that closed a £220M round of funding in 2022 as “I’m not gonna say I doubt your FFM
they achieved a temperature of 75M °C in their modelling, and I’m confident in the ELM counts
reactor, bringing their total investment to £1B, I’ve contributed from our blanket systems,” said Pat
which was enough to finance a next generation Smythe. BET are the group who comb over the
reactor. In this climate, Tokamax investors have Blanket Modules and the code that runs them both
promised a new investment if the team manages a before and after every run. “I really don’t want to
time that is at least “on the order of” 24 hours. The rain on the parade, but my concern is that longer
investors have resisted being more specific, but they run times are interacting with blanket systems in
have hinted this could be any fraction of a day you some way we don’t understand. In our longest
would round up to 1. The investment under virtual runs, we’re seeing degradations to precision
discussion is another£400M of their own money of magnet controller hardware that are not
plus syndicate partner funding likely (“90% correlated to the ELM counts, which are still below
confident”) to total £4B. In private conversations, thresholds for replacement. As a caution, we’ve
the investors are convinced Tokamax can do it, and replaced everything that’s even a little iffy, but I
they want to make some history. gotta be honest, I’m worried about what would
happen if one of the controllers failed in the wrong
Against these benefits of hitting the 24 hour
way. We all know a disruption just shuts down the
target, however, running for longer times raises the
reaction, but a messy one could stress the vessel. In
odds of costly damage from interactions between a
the extreme, we could lose control of the fuel and
reactor’s plasma and its blanket, a lining inside the
therefore the whole reactor hall. The building would
reactor comprised of heavy modules that capture
be a write-off. The business end of this is your area,
emissions for power generation, shielding, and
not mine, but could we really bounce back from
Tritium capture for fuel replenishment, or breeding.
that?”
The hazards associated with fusion reactor blankets
have been understood since the 1970s, with the Ned countered, “You’re right, Pat, as usual,
biggest risks arising from containment problems that about all of that. But then, as you say, the reactor
lead to interactions between wall systems for write-off disaster scenario is extreme. Our data
generation and for shielding and fuel breeding. doesn’t point to that at all.”
Treating permanent staff as fixed costs of the “Yeah,” replied Pat, “but let’s be honest, what
company and ignoring the opportunity cost we are about to try is pretty effing extreme!” Pat had
associated with other projects they might do (the a wonderful way to make points with humour, and
company is solely focused on FFM-driven fusion that got a laugh that lightened the mood.
breakthroughs), the direct variable cost of a run
“Pat, thank you; you’re brilliant, as ever,” said
includes both per hour expenses for computing and
Jo, reaffirming the team bond.
energy and per-run costs of follow-up inspection and
maintenance along with estimated allowances for If they had cash reserves to do it the slow-and-
repairs and refurbishment, as follows: steady-wins-the-race way, that would be tempting.
At current burn rates, however, Tokomax only have
Computing cost: £250K per hour
one year, and the slow-and-steady approach would
Energy cost: £400K per hour likely take something closer to 18 months. Plus, the
whole point (something that Pat tends to miss) is to
Inspection: £400K per run
demonstrate the value of their FFM toolkit. Even if
Module Refurb: £100K per bdi they failed in this attempt, they would get attention
for their work that could well lead to acquisition
Note: The energy costs are high because the
offers, though that was not the dream. They hoped
team is optimising for time over the more marginal
to manage the company as a going concern and
increases in temperature and more efficient energy- make sustained contributions to clean energy.
capture they need to nudge the reaction toward the
longer-term goal of positive net energy output. For Time to make the decision: should they go for
now, these experiments are still very costly. the record-breaking run, or drop back and work up
to this goal?
Concerns

© 2023. Mark Thomas Kennedy.


TOKAMAX
Part B

Normally, scientists are keen to publish their ground-breaking results as soon as possible, but sensing the
chance to make history, the Tokamax team have kept quiet about both the FFM scheme and the impressive
results they have posted. The chart below combines results for virtual runs (i.e., simulated) with the longest
actual run have achieved to data (664 seconds, highlighted in yellow). In addition, the team have conducted
many shorter runs to build the dataset needed for FFM–that is, the data needed to learn and simulate blanket
damage incidents and impact, as measured in the number of modules that need repair or refurbishment. A
blanket module is one section of the enclosure lining.

Runs Times (sec) and #Modules Refurbed or Replaced (mrr)


For Runs with Blanket Damage Incidents (bdi)
5

3
(mrr)

0
- 5,000 10,000 15,000 20,000 25,000 30,000 35,000 40,000 45,000 50,000
(sec)

Obviously, the actual runs are expensive in terms of both energy cost and the time Tokamax scientists and
engineers devote to setting up for a run, inspecting reactor systems afterward, and refurbishing or replacing parts
and equipment as needed. Less obvious is the fact that virtual runs are also expensive in terms of computing
costs, data storage, and post-run analyses by the software team, scientists, and reactor engineers. Data from both
actual and virtual runs is analysed to learn patterns that drive the team’s digital twin model of their reactor. The
table below hardly looks like anyone’s idea of what people once faddishly called “big data”, it summarises
massive amounts of work. Columns show run times (t), the number of runs at each time (N), the number of
blanket damage incidents (bdi), and the number of modules refurbed or replaced. Again, the longest actual run
is 664 seconds, but with FFS, the team regard virtual run data as indicative of results to expect of actual runs.

t (sec) N bdi † #mrr *


1 1 0 0
31 1 0 0
76 1 0 0
153 1 0 0
437 2 0 0
664 2 1 2
1,333 1 0 0
1,795 1 0 0
3,026 4 2 2
3,814 1 0 0
4,732 1 0 0
5,790 2 0 0
6,997 1 0 0
9,905 1 1 1
23,350 1 1 1
29,699 1 1 1
41,278 1 1 3
1 day = 86,400 sec

† bdi = blanket damage incidents


* #mrr = number of modules refurbed or replaced

© 2023. Mark Thomas Kennedy.

You might also like