- Fusion events last only a few microseconds. - All control actions (switching currents, diagnostics) must occur in nanoseconds to keep up. - Fiber‑optic links replace traditional wiring because photons travel at the speed of light, allowing nanosecond‑level signaling.

- Diagnostics are essential to verify that every switch fires correctly and that the plasma behaves as expected. - Rowski coils (electromagnetic coils powered by the same current) emit optical signals indicating conduction status. - All diagnostic data is streamed back to a central repository for immediate analysis.

Inside the Lightning‑Fast World of Pulsed Fusion: Control, Diagnostics, and Simulation

Q: Programming Languages and Hardware

- Early fusion pioneers programmed without computers; today gigahertz‑scale processors enable thousands of operations per microsecond. - Control logic is implemented on programmable logic devices (FPGAs) using low‑level assembly language for speed. - High‑level code is written in Python, Java, and legacy Fortran; many fusion codes still run in Fortran due to their numerical heritage.

Q: Fiber‑Optic Switching Architecture

- Thousands of parallel electrical switches (each limited to ~30 kA) are coordinated to deliver total currents of ~100 MA. - Each switch is driven and monitored via fiber‑optic channels that both send trigger signals and return status data. - Real‑time monitoring ensures any mis‑firing switch shuts the shot down instantly.

Lex Clips

Summary Date: 2025-11-23 03:36:03

•

3 min read

Inside the Lightning‑Fast World of Pulsed Fusion: Control, Diagnostics, and Simulation

Overview

The interview explains how modern pulsed‑fusion devices operate on microsecond (µs) time scales—so fast that the entire fusion event is a flash invisible to the human eye. The system must be started, energy extracted, and shut down before any operator can react.

Timing and Control

Fusion events last only a few microseconds.
All control actions (switching currents, diagnostics) must occur in nanoseconds to keep up.
Fiber‑optic links replace traditional wiring because photons travel at the speed of light, allowing nanosecond‑level signaling.

Programming Languages and Hardware

Early fusion pioneers programmed without computers; today gigahertz‑scale processors enable thousands of operations per microsecond.
Control logic is implemented on programmable logic devices (FPGAs) using low‑level assembly language for speed.
High‑level code is written in Python, Java, and legacy Fortran; many fusion codes still run in Fortran due to their numerical heritage.

Fiber‑Optic Switching Architecture

Thousands of parallel electrical switches (each limited to ~30 kA) are coordinated to deliver total currents of ~100 MA.
Each switch is driven and monitored via fiber‑optic channels that both send trigger signals and return status data.
Real‑time monitoring ensures any mis‑firing switch shuts the shot down instantly.

Diagnostic Systems

Diagnostics are essential to verify that every switch fires correctly and that the plasma behaves as expected.
Rowski coils (electromagnetic coils powered by the same current) emit optical signals indicating conduction status.
All diagnostic data is streamed back to a central repository for immediate analysis.

Numerical Simulation Hierarchy

MHD (Magneto‑Hydrodynamic) Codes – Treat plasma as a fluid with electromagnetic forces; used for overall machine design, capacitor sizing, cable dimensions, and magnetic topology.
Circuit‑Level SPICE Models – Simulate the entire electrical network, feeding results into the plasma model.
Particle‑in‑Cell (Hybrid) Codes – Resolve ion behavior at the particle level; require orders‑of‑magnitude more compute power and now benefit from GPU/AI accelerators.
Simulations are run ahead of experiments; results are compared with measured data to refine future shots.
AI and reinforcement learning are being explored to close the loop, providing near‑real‑time predictions that can guide the next shot.

Magnetic Field Scaling and Fusion Performance

Fusion power scales roughly as B³·⁷⁵ (magnetic field to the 3.75‑3.77 power). Higher magnetic fields dramatically increase density, temperature, and thus reaction rates.
Pulsed magnetic systems have demonstrated >100 Tesla fields, far exceeding the ~20 Tesla steady‑state fields of tokamaks or stellarators.
The goal is to achieve high magnetic pressure (β) while maintaining sufficient confinement time (τ) – the classic n·T·τ product.
In pulse‑magnetized inertial fusion, confinement times range from 100 µs to a few milliseconds, allowing hundreds to thousands of microseconds of useful plasma life.

Self‑Organizing Energy Recovery

As the plasma compresses, its pressure pushes back on the magnetic field, inducing currents that flow back into the system’s capacitors.
This creates a self‑recharging loop where the fusion plasma helps replenish the energy used to start the shot.

Future Directions

Continued migration to GPU‑accelerated and AI‑driven simulations to reduce turnaround from days to minutes.
Development of smarter real‑time control algorithms that can adapt shot parameters on the fly.
Ongoing search for programmers proficient in legacy languages (Fortran, COBOL) to maintain critical scientific codebases.

Pulsed fusion combines ultra‑fast microsecond control, fiber‑optic‑driven switching, and multi‑scale simulations to achieve magnetic fields far beyond steady‑state devices, making it a promising path toward practical fusion energy.

Full Transcript

That's how we think about the systems.
We measure everything in microsconds,
not in seconds. And so when you do
fusion, it's pretty wild. It's literally
a flash. Fusion happens and it's over.
You start it, you do a lot of fusion,
you recover energy from it, and then you
turn it off before the human eye can
really respond even.
>> And there's a computer managing all
this. Like how do you even program these
kinds of systems to do the switching? Is
there some innovation required there?
>> So, I'm continuously amazed by what the
pioneers in Fusion were able to do
before the computer existed cuz they had
to control things at this scale, but
maybe it was pretty hard and and and why
we've been able to be take what they did
and build on it because now we use
modern gigahertz scale computing to be
able to do this. And so, even when I
started my career, we talked about like
megahertz processors. U meghertz is
microsconds. That's great. You're kind
of at the border of fast enough, but you
can't do computation at that speed if if
all it can do is respond in one
microcond. But now gigahertz means I can
do a thousand operations in that one
microcond. So I can do more useful
things. So we use mostly this is way too
fast for any human to respond to. So we
use what's called programmable logic. So
we program in sequences to the fusion
system to be able to do this reversal.
We pre-program it and then we run a
sequence and then fusion happens. Um and
so in this sequence uh programming
language we use a variety of them. Some
of the fusion codes are actually written
in forran still
>> nice and though a lot is now more and
more run in Python and so we do a lot of
Python we do some Java and then we also
have uh because of the speed of this
it's a lot of assembly language
programming. So we go right to the
assembly level of the programmable logic
FPGAAS and we program those. And so to
be able to run one of these systems, we
typically have a series of electrical
switches that turn on this electrical
current. Those are controlled via fiber
optic because the wires are just too
slow. And so fiber optic I can respond I
can send photons at the speed of light.
And so those fiber optics can respond in
nanoseconds. And then I trigger those
fiber optics with programmable logic
that we've programmed in the hard
hardware assembly language.
>> As a small tangent, let me do a uh call
to action out there. I'm still looking
for the best for programmer in the
world. If people uh to talk to them cuz
so many of the essential systems the
world runs on is still programmed in
Forran. I think it's a fascinating
programming language. Cobalt too, but
Forran even more so. It's one of the
great sort of computational numerical
programming languages.
Uh anyway, what uh in terms of the
sensors
that are giving you some kind of
information about the system in terms of
the diagnostics
like what kind of at this time scale
>> what can you collect about the system
such that you can respond at the similar
time scale? So I'm also calling out for
four trend programmers. So [laughter]
for different reasons. So yes, great.
>> The diagnostic systems is really one of
the keys to how we do this effectively
because you need to be able to tell the
system we're going to trigger electrical
current and we're going to do it in a
microcond. And we need to know if it's
working right. And so in one of these F
FRC or these pulsed magnetic systems,
you won't have just one electrical
switch. I mentioned 100 mega amps, 100
million amps of electrical current. Each
even the big transistors we use can only
run at 30,000 amps. So you'll end up
with tens of thousands, in fact the
systems we build now, tens of thousands
of parallel electrical switches all
operating in harmony together. And so
you need to be able to be build a system
and this is what we spend uh a lot of
time with. And I made the joke that in a
lot of ways Helon's an electrical
engineering company
to be able to both program control and
then detect how they're operating
and do it all very fast. Um so in a
typical sequence we will pre-program the
operators will pre-program a sequence um
usually fed from a numerical simulation
of expecting how the fusion system will
perform. We start with a a set of
calculations. We then pre-program all of
these electrical switches to a certain
sequence to be able to inject the fuel,
reverse it, and then compress it up to
fusion conditions
and then we trigger that and [snorts]
then and then let it go and and and
measure fusion happening. Um, but during
that process have to be real time
recording and and measuring all of the
semiconductors and all of the switching
in the system. I talk about measuring
fusion diagnostics. That's a whole
another thing which we can talk about.
This is just on the electrical control
side. Um, and so some of the pioneering
things we've been able to do is that
real time you're monitoring all of these
switches. You're watching who is
triggering correctly, who is not
triggering correctly. And if systems
aren't working, you're shutting down
this because you want to make sure that
all the sequences are are are operating
correctly. So some of the key
diagnostics, it's actually pretty
amazing that even early in my career, we
didn't have a lot of fiber optics built
into the system. And now it's absolutely
essential. And so every one of these
electrical switches has fiber optic
signals going into it and fiber optic
signals coming out understanding how
it's actually operating. Um, and real
time all of these systems are being
monitored by more fiber optics. Um, we
call these Rowski coils, but they're
electromagnetic coils that are powered
by the electrical current themselves.
So, as these switches are conducting,
they broadcast a signal that says, "Yes,
I'm electrically conducting an optical
signal, fiber optics, that come back to
a central repository where we detect
those signals." Um, and so real time
we're monitoring all of this so that we
know that these systems are behaving and
operating at their their optimal
performance.
>> What's the role of numerical simulation
in all of this?
sort of
I guess ahead of time
uh how much numerical simulation are you
doing to understand how the system is
going to behave how the different
parameters all come together the the
electrical system and how that all maps
to the the the fusion that's actually
generated
>> yeah the operation of a fusion system is
is pretty fascinating because all of
this happens on a time scale where human
operators cannot be cannot really be
involved
Um, and so, uh, you have to have
pre-programmed the majority. We call
them shots. You're going to do a shot.
And when you're operating them
repetitively and you're running long
periods of times, you still have all
computers doing both the triggering and
the and the measuring of of how they're
performing real time the whole time. Um
and so um how this typically works at
least in our systems is that we will
design a system with a combination of
with with some numerical simulation
tools that we we've developed based off
of decades and decades of amazing
government programs. National Lab
programs developed these numerical
codes. Um we use a kind of a code called
an MHD magneto hydrodnamic code. Um and
that's uh for people for the engineers
out there who are used to CFD
computational fluid dynamics. This is
very similar. You take the same sets of
equations actually and add the
electromagnetic equations on top of
those and so you get magneto
hydrodnamic.
>> Are you simulating at the level of a
particle? Is there some quantum
mechanical aspects to this also? Does
how low does it go?
>> Yeah, we have multiple codes at
different levels. Um because one of the
the main computational challenges is um
amazingly even given all that we are
have been have built for fusion systems
computers are still not fast enough to
measure to simulate everything. Um and
so we have uh a number of codes that we
use. Um, one we call fluid codes where
you treat the ions, the electrons, all
these fusion particles, you treat them
as as fluids, as gases, ideal gas law
with electromagnetic forces. In those we
can simulate not just the fu fusion
fuel, which is important, but all of the
electrical circuitry. We talked about
capacitors and magnetic coils and the
electrical current and the switches.
Well, we actually simulate the full
thing starting literally with the SPICE
model. uh more of that electrical
engineering. We start with the spice
model and use that to drive the plasma
physics model and that's one level of
simulation. We use that to do design
work and then also to try to understand
how we think the machine will run. But
then we go one level deeper and we start
thinking about particles and we think
about the ions and we treat the ions as
particles and we look at the ion
behavior and for that one the
computational resources are several
orders of magnitude larger. Uh luckily a
lot of the work in GPUs the AI data
center work is directly applicable to
those simulations. It's been able to
speed up our work which is pretty
fascinating. Um that's a whole another
tangent we can go down. Those
hybrid codes we call them particle and
cell codes uh now treat the ions as
particles and that lets us measure and
and simulate the behavior. I mentioned
the stability criteria Sarov the top
behavior. that behavior. We now need
these more advanced codes to be able to
simulate and those are more modern.
Those we've only been able to apply in
practice for the last few years
actually, which is pretty fascinating.
Um that the old stability rules were
built off of testing, empirical tests
where now we can simulate that and we
know why they work and how they work and
we can do some predictions on them. And
so that's really fascinating that we've
been able to push those boundaries. And
what are the different variables you're
playing with? Are you still playing with
like topology? like what are the
different variables in play here?
>> Yeah, each of the different simulations
we analyze and use it to design
different parts of the machine. So at
the MHD level where we have the spike
where we actually have the circuit model
now we uh our design team uses this to
design the circuitry where we're
designing which capacitor to use which
switch to use uh how many cables to use
literally to that level how big of a
cable to use. So as we're doing power
plant designs right now, those are the
tools we're using today. Every day the
team is using
then you can go one level deeper and say
okay let's use these more advanced
computational tools to about stability
to say okay great but I now know the
circuitry but let's look at the magnetic
field topology. How do I design the
magnet the shape of the magnet exactly
the timing of the magnet exactly? I have
to trigger one magnet and the next
magnet next to it and the next magnet
next to it. How do I have that shape and
that that design? And so that's where
you're using those more advanced tools.
Now those unfortunately those are still
too slow and so those simulations may
take a day or two to run. And so a data
an operator right now does a lot of
simulations ahead of time then collects
data uh through their their operations
of the machines making these field
reverse configurations going through
parameter sweeps and then the simulation
team then goes back and looks at that
data and and compares it with
simulations. Um I'm really excited about
some of the things we're seeing in
artificial intelligence and reinforced
learning to be able to speed up that
process. And so I'm I'm we're watching
and starting to work on that now of can
we now rather than using it where we use
it today where we do a simulation to
design a machine or a test run the test
and then over the next couple of days
compare the testing with the simulation
and use that to inform what we're going
to run for the next set of tests. but in
fact do it more real time where you're
now an operator can pull up what the AI
or what the m the machine learning would
have predicted it should have done and
then use that to understand what's
happening in in the actual programs and
the actual generators themselves. All
right. So there's a million questions
there. So first of all,
how much understanding do we have about
how many collisions happen? Can we go to
the fusion?
>> Mhm.
How many collisions are there and how
does that map to the electricity
and maybe can you just even speak to the
directly mapping to the electricity
which is one of the differences between
[snorts] this approach and the the tuck
approach. So how much fusion do you get
out in these systems? And that's really
the right key question. So we already
talked about beta that B^2 the magnetic
pressure is equal to N KT N being the
density T being temperature and then we
talked about fusion where your goal for
fusion is to get particles hot
temperature get enough of them together
density and then you want to get them
together long enough we call that towo
so n t and long enough that fusion
happens and a lot of fusion happens more
than any of the loss rates that are
happening in TTL and in beta with B^2
you know already two of those parameters
N and T are equal and so that tells you
right away the goal is to maximize
magnetic field absolutely maximize
magnetic field and most folks in
magnetic fusion whether it's a tokamac
or it's a theta bench or it's an F fr
FRC are attempting to do that maximize
the magnetic field and so we're all
pushing to that um what's really nice in
pulse systems is that we know how to do
that in fact um in a pulse system. Uh
researchers in pulse magnetic fields
have demonstrated over 100 Tesla
magnetic fields in pulse magnets. That's
much higher than you can get in a steady
magnet or what's been demonstrated so
far.
>> Just a clarification question. Uh so
maximizing magnetic field is about the n
and the t the beta.
>> So we're not talking about towel yet.
>> Not yet. But we need to cuz that's
really important. Um, and so we can even
talk even a little bit further about how
fusion scales. And so in fusion, the
hotter you get the fuel, the more fusion
you get. Um, and we know that by
increasing the magnetic field, B^2 is in
T, you increase density and temperature
together, more density, more temperature
is more fusion. Plus more temperature is
even more fusion. And so what we see is
that in our in the in these types of
systems uh a scaling very clearly of
magnetic field to the 3.75 power or even
uh in a lot of a lot of demonstrations
3.77 that that specific scaling. That's
a very strong scaling of fusion power
output um and fusion reactions. And so
that tells you you want to go to as
maximum magnetic field as you can. Pulse
systems are really powerful. Pulse
systems have showed when you do pulseed
magnetic fields compared to a steady
magnetic field researchers have shown
over a 100 Tesla magnetic fields where
in a steady system people have showed in
the 20 maybe high 20 Tesla systems and
if it's B to the 3.77 power already you
can see massive fusion power outputs by
doing a pulse system.
>> Okay, got it. So we we're maximizing the
magnetic field. So that's going a number
go up super up. How do you get the
duration the towel?
>> But then I said pulseed and pulse
already implies shorter tow.
>> Yes.
>> And so that is in the fusion field the
name of the game.
>> Folks will will have a very uh inertial
fusion will have a nancond towa very
short but then very high pressure. They
don't have magnetic fields but very high
pressure. Um, and then in stellarators
and tokamax, your goal is very long tow,
but you'll have much lower density and
and and you can't really go too much in
temperature, but they'll have much lower
density. And so, where we live in the
pulse magnetic or the magneto inertial
fusion is in the middle. Um, is in
extremely high magnetic fields,
increasing pressure as much as you can
and then keeping them around long
enough. Um and so that gets to the towel
that gets to that energy confinement
lifetime and also it gets to stability.
And so this is the thing that this field
reverse configuration which has showed
that we can um build we that these
plasmas can last for hundreds or
thousands of times. The basic theory has
shown that now you can have long enough
lifetimes. So what that means is in a in
a practical fusion system uh that there
are lifetimes of these high beta pulse
systems between 100 microsconds and a
few milliseconds thousands of a second
and you hold on to it for a few thousand
of a second. You do fusion and then you
exhaust it. And so the whole process in
this is we start with uh a magnetic
field that fills the full chamber. You
then inject fusion fuel. You ionize it,
superheating it now to an ice cold 1
million degrees, but hot enough that you
have charged particles. You have
plasmas.
You can then in start increasing the
magnetic field. You form an F a field
reverse configuration and then rapidly
increase the magnetic field further
>> increasing from one to five to 10 20 to
even higher magnetic fields and as you
do that the plasma heats it. You
compress it increasing the field and
pressure. Fusion is now happening. New
charged particles are being born inside
this system with a tremendous amount of
heat and energy but in charged
particles. And this is where the beta
really really works in in in your
advantage is that just like magnetic
pressure on the outside magnetic
pressure is in KT compresses
the the fuel and increasing pressure and
temperature. When the pressure and
temperature of the plasma increase in KT
increases, it pushes back on the
magnetic field, increasing the magnetic
field on the outside of the plasma. And
what that does is magnetic field is
electromagnetic current and current
running in a wire. And what that does is
pushes current back in the wire. And so
the plasma itself now pushes back on the
magnetic field, pushing electrical
current out of the system and recharging
the capacitors where we started this
whole process. All in a self-organizing
way.

Summary

Inside the Lightning‑Fast World of Pulsed Fusion: Control, Diagnostics, and Simulation

Overview

Timing and Control

Programming Languages and Hardware

Fiber‑Optic Switching Architecture

Diagnostic Systems

Numerical Simulation Hierarchy

Magnetic Field Scaling and Fusion Performance

Self‑Organizing Energy Recovery

Future Directions

Share This Summary

Embed This Summary

Stay Updated!