Behind the dusty hills of Sand Hill Road, across the street from the birthplace of modern venture capital you can see it from the I-280: a three mile long shed on broken slab concrete. Inside is one of the most sophisticated devices on the planet, a technology that is the future of semiconductor lithography.
The Free Electron Laser.
By the end of this article you will know exactly how a free electron laser works, what they are used for in science, and the startups today like @xLight_Inc that have raised hundreds of millions of dollars - in funding and revenue - to commercialize accelerator technology to take up the torch of Moore's Law and carry semiconductor manufacturing into the future.
Team of xLight hanging out in the headquarters of Playground Global, a venture fund that is in my estimation the premier applied physics deep tech VC fund
It's a brand-new era, and the birth of the Beamline Tycoon.
To preface, that shed at the top of Sand Hill Road is the home of the Linear Coherent Light Source, now LCLS-2, the most powerful X-ray laser on the planet, something that produces coherent waves of light at a wavelength so small it can see individual atoms, create movies of molecular reactions, track an electron as it changes orbitals. To do this is produces bunches of electrons of 100 pico-Coloumbs, and fires them down a hall of superconducting radiofrequency cavities one million times a second into an array of magnets. The magnets makes the electrons wiggle, wiggling electrons spit out photons, the photon electric field bunches the electrons even more, making cleaner wiggles and sharper photons.
There is cool physics here we will get into in a second but that is in essence of an X-Ray Laser: extremely short wavelength, coherent light, thanks to the physics of wiggling electrons (to be precise FEL's use undulators, wigglers are the technical term for synchrotron light sources that have broader spectrum and larger wavelength).
By way of background here is a photo of me at the control room of LCLS-2, and another hanging somewhere along the superconducting RF cavity hall where the electrons are taken to 4 GeV energy during my days as a humble RF engineer.
On the screen is a internal report on the event known as the "Switchyard Squirrel Massacre" which I will just comment on by saying that particle beams produce deadly radiation when turned on, and, it is important to secure all ventilation shafts against wildlife.
Now, as of this week apparently TeraFab will construct a Free Electron Laser to achieve its goal of producing one Terawatt of computing power per year. That's 5% of todays global energy production going into chips that come from a single factory. LCLS produces hard X-rays; the but chip production uses EUV light which is longer wavelength.
So then why would TeraFab need a Free Electron Laser, and why wouldn't they just use what ASML does currently, laser-tin droplet ablation? To answer this we should start at the beginning: the fundamental physics of semiconductor photolithography.
Crash Course in Semiconductor Photolithography
Photolithography packs transistors onto silicon by shining short-wavelength light through patterned masks onto a light-sensitive photoresist, which is then chemically developed to define each circuit layer. Multiple masks and multi-patterning steps are often used for a single layer. A typical wafer carries many identical chips patterned by the same mask; Cerebras is the notable exception that turns an entire wafer into one giant circuit.
The scanner—the mechanical core of the system—moves the wafer and optics in opposite directions at 1–4 m/s with accelerations up to 30 g while holding nanometer and nanosecond alignment. Any deviation produces misaligned transistors or interconnects. Feature size is set by wavelength divided by the numerical aperture (NA) of the optics; achieving smaller features therefore requires either shorter wavelengths or higher-NA systems that place the wafer closer to the lens and demand still higher stage speeds.
The history of the field is the progressive use of shorter wavelengths: mercury-vapor lamps in the 1950s, successive generations of excimer lasers from the late 1970s through the 2010s (shrinking features from ~250 nm to 10–20 nm), and EUV sources based on laser-ablated tin droplets from 2019 onward. Immersion lithography further improved resolution by surrounding the optics and wafer with a high-index liquid, effectively shortening the wavelength without changing the light source’s frequency.
One difficulty with trying to get bigger and bigger NA is the requirement for more optical elements, and its hard to get high reflectivity at EUV wavelengths, meaning you lose more and more light with each bounce off a lens or mirror. This is a big reason why 13.5nm is the target wavelength - its basically the intersection of where laser-tin droplet ablation produces lots of usable light and where optics can effectively use it.
It should be noted that the laser-tin droplet ablation is a damned impressive feat of engineering, hitting tiny droplets with a CO2 laser tens of thousands of times a second - actually hitting the same droplet twice, since the first hit flattens it, the second hit vaporizes it forming a plasma.
To sum, to make tons of chips at the cutting edge you need lots of photon power at short wavelengths.
Here are the big bottlenecks with current laser-tin droplet ablation photolithography:
- Wavelength is fixed at 13.5nm, can't go any lower
- Wall plug efficiency is brutally low, 0.05% - 1%. To get 1kw of light power you need 2MW
- Need one light source per scanner, which is expensive
- Yield quality limited by available photon power
The last one is a real doozy - as an example, for intel chips i5, i7 and i9 all come off the same silicon wafers using the same masks, the difference is in the statistical error rates induced by limited or irregular photon power during the masking process.
Free Electron Lasers outclass laser-plasma photolithography on all fronts.
Far more power efficient, huge available power into the multi-kilowatts, improving chip yield, and they have tunable wavelengths that for chip production can extend down into the 8 nm range (it can go lower in to the hard X-ray regime like LCLS-2, but, its impractical for chip production). It should also be noted that right now, the EUV systems and operations are a major cost driver for semiconductor chips overall - about 40% of the cost.
These plots are taken right off of xLight's website (@xLight_inc), one of the coolest companies on the planet right now developing FEL's for chip production:
TeraFab will need dozens of scanners, likely tens of kW of short-wavelength light.
Okay, sold on FELs yet? Ready for the physics on how they work?
More importantly, how can you become a Beamline Tycoon and turn relativistic electron bunch energy into cold, hard cash?
Part One: Beamline Basics 101
We're going to get extremely technical but I'll do my best to convey this as comprehensibly as possible.
First some basics: in the world of particle beam lines and accelerator physics we don't measure particles in kilograms, because they weigh far too little (we also don't measure them in slugs, the Imperial unit for mass), instead they are measured in electron-volts per c², from the familiar Einstein equation of E=mc², where electron-volt is the energy an electron gets by passing through one volt. Conventionally people drop the c² everywhere for mass and just say eV, KeV, MeV, GeV, TeV for powers of 1, 10e3, 10e6, and son.
This is nice because then we can measure the mass of the particles and the energy of the beam in the same units.
Protons are heavy, they weigh ~940 MeV; electrons are extremely light, they only weigh 0.5 MeV or 500 keV. These are the rest masses of the particles, how much they weigh when completely still. If they get accelerated close to the speed of light they start to weigh more, and the total energy of a particle is the sum of both its rest mass and its kinetic energy, simply E = Ɣmc² where gamma is the relativistic correction factor of 1/sqrt(1 - v²/c²), or more simply 1/sqrt(1 - β). Beta is a common term for the velocity relative to the speed of light. A Beta of 1 is physically impossible to reach, if you look at the shape of Energy, the gamma term explodes to infinity as Beta->1.
This is useful - when you collide two particles together, its the kinetic energy that goes into the collision, if they are anti-matter pairs, then its the total rest mass plus kinetic energy. If you pass a 511 keV electron through 511 kV of electrical field, it has equal amounts of rest mass and kinetic energy, so gamma = 2, and you can work out Beta as being around 0.8.
The most relevant first-order intuition for accelerating and handling particles is Beta, its velocity, telling you where the position of the particles are as a function of time. Synchronizing the particle bunches to hit the accelerating electric fields at just the right time, chopping the beams, tuning magnets that bend the beam, are all functions of beta (for bending a particle beam with magnets the mass is also important - heavier particles make for more rigid beams). The kind of physics you can get from the beam depend on its energy, and so as Beta gets bigger, and gamma starts to blow up you can dump more and more energy into the particle bunch while hardly changing its velocity.
This is very counter intuitive but at high beta, you can 10x or 1000x the particle beam's energy while changing the velocity by less than 1%, and the change in velocity asymptotes to zero as energy becomes extremely large relative to the rest mass of the particle.
The second-order intuition is thinking about the beam like ray of light passing through optics - if you want to focus the beam down to a small spot, then the rays have to converge. If you want the rays to not diverge as much, then the spot size will be bigger. The product of spot size and divergence of these rays is importantly a conserved quantity, and a fundamental measure of beam quality called Emittance. For light rays the optical elements that do focusing are lenses, for charged particle beams moving close to the speed of light, the focusing elements are magnetic lenses called quadrupoles.
In the diagram above, the circle describes 'all the places the beam particles might be' in terms of a physical axis and the momentum along that axis - this is the phase space. Trading divergence for spot size means the circle just rotates, to be more spread out in space and tighter in velocity, and vice versa. In ray optics this is called Etendue, and more formally its a result of Liouvilles theorem that information is conserved, meaning the volume of phase space the particle bunch occupies remains constant.
In general the emittance of a particle beam only gets worse as it gets accelerated, since particles that deviate from the idealized particle trajectory, or the path the beam line components are designed for, starts to get knocked around more. Think of a pendulum - it means we can't arbitrarily make the pendulum swing less and less without extracting energy from it. Some particle beam lines in fact do this by injecting a second, colder bunch of particles that mix with the main bunch, absorb some thermal energy, then are extracted to 'cool' the beam, to shrink the emittance.
Part Two: Forming a Particle Beam
The strategic defense initiative launched particle beams into space to shoot down ICBMs in the boost phase, before they separated into a hypersonic warhead cloud spread over tens of kilometers.
There's different kinds of particles we can accelerate - electrons, protons, heavy ions, anti-protons, and so on. Each needs a different means of extracting the particles and forming it into a beam. Electrons are the most popular for most research machines because they are the best at acting as light sources, emitting high energy photons through synchrotron radiation. Synchrotron radiation is light emitted by charged particles when they undergo centripetal acceleration, shooting out like the headlights of a car as it goes around a corner. The amount of synchrotron radiation emitted goes up as the mass of the particles go down which makes electrons a favorite since they weigh so little.
Particles blasting out synchrotron radiation is great if you want a light source, and bad if you're trying to get high-energy collisions since it means energy is leaking out of the particle. If you want really high energy collisions you want to keep adding energy to the same bunch of particles, so you make it go in a circle, which means you use heavier particles like protons so they leak less. Electrons are called leptons in particle physics, and the rule of thumb is leptons go in linear accelerators and protons, or hadrons, go in circular accelerators. Since heavier particle beams are more rigid and harder to bend, at higher energies the hadron colliders get larger radiuses.
Now you know why the Large Hadron Collider has a 27km circumference and why short-wavelength FELs are linear accelerators.
But before we get in the business of designing accelerator facilities from first principles lets start with the beginning - creating a beam of electrons. There are a couple most-popular ways of getting an electron beam:
- Thermionic: Heat up a block of metal to a couple thousand kelvin and electrons start boiling off. You can then draw them off and shape them with electric fields. They come off as a continuous gas, and so have to be bunched, and also come off with a high thermal random velocity which means poor emittance or being spread around a lot, so larger spot size or more diverging rays.
- Photoelectrode: Shoot a block of metal with lasers, so the photons excite electrons to jump off via the photoelectric effect. Tightly controlled and low emittance. Also known as an RF Photoinjector, these are the more modern upgrades to FELs, have less thermal energy and so better emittance, smaller spot size, tighter focused beams.
Whether you are stripping electrons off a piece of metal with a laser, or heating them up so they jump off, or making a plasma to extract something like protons, the first thing you need to do is start accelerating the charged particles to shape them into a beam. For a continuous beam source like thermionic emission from a cathode, this is done with a DC electric field that pulls charged particles and starts accelerating them.
Now here is our first subtle issue - the charged particles in the beam have their own electric field and will 'pile up' until their own field cancels, in their frame of reference, the applied electric field. The amount of current you can extract therefore depends on the strength of the electric field, but, if you see the beam as a cylinder, the particles in the middle see their fellow charges from all sides so experience net zero radial field, but the charges on the edge only see other charges towards the inside center of the beam. This radial electric fields makes the beam want to defocus, meaning the emittance gets terrible quickly.
Fortunately there is a solution and its one of the few exact analytic solutions in practical RF engineering, called the Pierce angle, where a correctly angled electrode applies a voltage that has an opposing radial component which exactly cancels out the radial field from a space-charge limited charged particle beam. In other words, if you max out the amount of particles you can scoop up into a beam with a high voltage electrode, all the charges start to repel each other, and so the field that scoops them up has to be curved oppositely so they get drawn out in a smooth cylinder and not a diverging cone.
I used to work at a little SBIR shop off the side of the highway in Redwood city building beam-driven neutron generators that works by taking a plasma of deuterium gas and accelerating it across 110kV so it hits a titanium target, here's a photo where you can see at the bottom the plasma source and the beam coming out towards an electrode - the electrode has a hole cut in it, and the bottom surface of it has an angled surface designed to cancel the space charge of the beam itself. Can you guess what that angle is?
Building beam-driven fusion devices was the plan B after my 2021 startup attempt to use LLMs for knowledge management in early stage drug discovery fell apart. Investors thought the premise was retarded and would never work. The cofounders went to start an NFT company, and me? I went back to hustling on the beamline, and now, write about how software investors are dumb.
Next time you're at a party and someone mentions particle beam sources, you'll be able to chime in that if they want a zero divergence beam for a space-charge limited extraction optics they should use an angle of 62.5 degrees from the beam axis for the first electrode.
After spurting that out you can safely retreat to the corner to peel the label off your beer.
Part Three: You Must Bunch Additional Electrons
While most modern FEL's use a photo-injector electron source, which produces pulses of electrons because the laser hitting it is pulsed, understanding how you would take a continuous beam and shape it into particle bunches develops good intuition for understanding one of the fundamental principles of Free Electron Lasers which is the Self-Amplified Spontaneous Emission - the process by which electron bunches spit out photons, and those photons then help bunch the electrons even further, getting more concentrated photons.
Why do you want bunches in the first place?
The earliest particle accelerators like Van de Graff and Crockoff-Walton generators were DC, but to make these work all the accelerating electrodes need to be at higher and higher voltages to keep the electric field pointing in the right way. An analogy here is putting energy into someone on a swing by having them start at higher and higher heights - eventually they drop from such a high height there is so much force at the bottom of the first swing the chain breaks, and you get electrical breakdown or a short circuit. If you have an AC acceleration, then you can just push the person on a swing a little bit each pass, using the same amount of force each time such that it keeps adding up, and you never load the chain so much at once that it breaks.
Now suppose you applied a sinusoidally shaped electric field to a continuous stream of electrons, and the wave travels at the same speed of the electrons. One half of the sinusoid has an electric field pointing in the direction of travel, and the other half points backwards, which means that as the beam travels electrons get pushed to the point where the electric field is zero - any further up, they;'re getting slowed down, and behind, they're getting sped up. Since the sine-wave has two zero-crossings there is one stable zero-point and one unstable, at the unstable zero-crossing the electrons are getting pushed away - in other words, there is a spot where the E-field is zero that on either side the E-field points to, and a zero-spot where on either side the E-field points away from.
If you make an electromagnetic wave travel down a tube with a stream of electrons, the electrons get bunched up with the same frequency as the wave. If the beam is slightly faster than the wave, it will push on the field and amplify the waves strength. This is how you make a traveling wave tube amplifier, the RF source used for the Phalanx ship-board defense the US Navy uses to shoot down incoming missiles.
The stable zero-crossing point is what creates the electron bunch, and at the very center is the 'idealized' bunch trajectory, the point around which all further downstream beam line elements are designed. The overall region of electrons you've bunched up into a group is called the "RF Bucket" - its the bucket of electrons that will be effectively accelerated so long as they stay within that small region, outside of that they are on the wrong side of the accelerating field and so get further away from the zero-crossing and become lost.
A particle right at the center of the RF bucket is called synchronous, its well-synced, while particles outside that will trace a circle as they pass through successive RF accelerating cavities so long as they are still inside the RF bucket - at first they are too early to the RF field and get pushed back, drift past the synchronous point to lag behind, then at the next cycle they get kicked forward more strongly because they're running behind, and drift back ahead of the synchronous point, sort of like a dog running ahead of you while walking, stopping to sniff something, getting yanked forward again, and so on.
There's a dividing line around the RF bucket called the separatrix which is a common term in physics which basically means "things roll either way either side of this line" and is like the highpoint or ridge in an energy landscape, so another mental model is that the RF bucket is a bowl with marbles rolling around it in nice orbits - at the center of the bowl the marble isn't orbiting at all, its at the lowest point, the synchronous marble. I mention the word separatrix because its just a cool word that everyone should know.
Part Four: Superconducting RF Cavities
Now for the good stuff - superconducting radio-frequency cavities, or SRF for short, something I worked on throughout most of my engineering physics degree. SRF is one of the most mature and impressive feats of engineering, material science, and physics on the planet and something few people have ever heard about.
The standard workhorse in SRF cavity design is the 9-cell Tesla cavity, which is the standard workhorse of modern particle beam lines, originally developed for the TeV-Energy Superconducting Linear Accelerator and is used in all the latest and greats FELs - European XFEL in Hamburg, LCLS-2 at Stanford, and so on. Why 9 cells and not 1000 all welded together? It's a practical tradeoff between controlling the fields inside nicely and stably, and operating the liquid helium cooled cryostats that fit around them. If you wanted to save money on cryostats you'd make it one super long cavity, thats impossible to control. If you wanted a perfectly controlled SRF cavity it would have just one cell, and you'd go bankrupt operating 1000 separate cryostats.
The Tesla SRF accelerating structure is made of high purity niobium shells that are welded together with electron beams (guess how those beams are made?! thats right, by tycoons), it acts essentially like a wind instrument for electromagnetic radiation - you blow into it with 1.3 GHz EM wave, and because its superconducting the currents induced in the conductive walls can travel without any (or essentially any) losses, meaning it is an extremely high quality resonator.
Here's me hanging out at LCLS-2 with something like $800 million dollars of SRF cavities and required cryogenics and RF power sources behind me, real tycoon stuff:
If you tell people in the bay area you work at SLAC they usually think you build workplace messaging apps, and its even worse if you say you work at an accelerator.
There's some subtleties in exciting an SRF cavity with an EM wave to make it resonate, which is to get as strong as possible electric field inside to give the particles passing by as much energy as possible, the conductive shell needs to support larger and larger surface currents. As we all know, currents produce magnetic fields, and magnetic fields can induce currents; in a normal conductor a magnetically induced current exponentially decays as energy is lost to resistance.
In a superconductor the magnetic field induces a current that stays circulating forever. It seems to defy physical intuition like a perpetual motion machine, but its true.
Why this matters is because the Earth has a magnetic field, and even the Earth's field is enough to induce eddy currents inside an SRF cavity as it cools down from room temperature to cryogenic temperatures and superconducting 'turns on' - any magnetic field that was in the cavity to start with becomes trapped.
This trapped flux becomes pinned at grain boundaries, defect, imperfections, and then when you apply an RF field it starts getting shaken back and forth by the field. Along with the inertia of electrons themselves, the force it takes to move these trapped magnetic flux centers back and forth constitutes the AC resistance of the SRF cavity, usually in the nano-Ohm per centimeter squared which is absurdly low, but which still matters a lot.
The Q-factor for an SRF cavity represents how much stored energy divided by how much energy it loses per cycle, and Q factors are regularly in the billions. If Galileo set a pendulum swinging in 1640 with a Q-factor of an SRF cavity it would be swinging today with about half its original amplitude.
The stored energy of a field goes like E², so if Q drops by a factor of 10 then the E field is down by a factor of ~3.2 - all this to say, in the world of maximizing the performance of SRF cavity accelerator gradients, to get as high a beam energy as possible in as short a distance as possible, you are battling in the weeds of cancelling stray magnetic fields from the Earth or other equipment, even rebar in the concrete floors, surface polishing the cavities, baking them out to remove any impurities. All the current travels in the first few nanometers of material anyway, so surface finish is make or beak.
Here is a device I designed and built during my undergrad years to test SRF cavities by canceling out the Earth's magnetic field, or, producing arbitrary internal fluxes to help out with some SRF cavity design at TRIUMF, Canada's national accelerator lab. It's called a hemholtz coil, which is a fairly simple device, but what was tough was getting it to fit into a cryostat with just a few mm of clearance on either side, and then, getting the coil geometry so as to have an even-field cancellation across the primary current carrying surface of the SRF cavity. Note that this is a single-spoke cavity, not a tesla cell, designed for low-beta heavy ions.
I made all the parts during an absurd 24-hr waterjet marathon over the christmas break when no one else was using the machine shop. Perhaps one of my fondest memories of university.
Like I said SRF technology is extremely mature and reaches acceleration gradients of 35-45 Megavolts per meter, and thats running at continuous-wave operation, with ~145 kiloamps per meter circulating in the surface, dumping 10-200 kilowatts of power into the particle beam while losing perhaps 1 - 10 watts of power via the remnant surface resistence, e.g. the inertia of electrons, the jiggling of trapped flux. All of that happening in the first few nanometers of pure niobium, running at 5 degrees Kelvin.
As we mentioned earlier, if you take an electron of 511 keV mass and put it through 511 kilovolts of electric field, its going 80% the speed of light.
Now take an electron and run it through almost a kilometer of SRF cavities that give it 20 Mega-volts per meter of acceleration, like the LCLS-2 does, and you get a beam of electrons that is 4 Giga-electron volts in energy.
With just a tiny current of 100 micro-amps that's still 370 kilowatts of particle beam power, and we are going to squeeze it for every Joule we can to produce the worlds most powerful x-ray lasers.
Part Five: Jigglers, Wigglers, Undulators and X-Rays
We have our 370 kilowatts of electron beam-power traveling at eight 9's of the speed of light, or a gamma of 7828, noting that going from 100 MeV to 4 GeV only chanced the true speed by going from four 9's to eight 9's of the speed of light (99.999....% of c) - so 97.5% of the energy gained only changed the velocity less than 0.01% - time to make photons.
Light sources from electron beams are in fact not that new, synchrotron light were first developed in the 1960s and FEL's not long after, in the mid 1970s at Stanford. But, the quality and quantity of the light coming out varies by how the electrons themselves are jiggled - gently, roughly, with a fine touch, and so on.
Reasonably enough with a single magnetic lens the electron beam intensity is proportional to the beam current, if you add alternating directions of magnetic fields the electron beam slaloms back and forth dumping more and more light, so you get current * turns intensity, thats called a wiggler.
Where this gets non-linear is if the size of the oscillations and periodicity of the magnet poles is small enough such that at each turn, the electrons keep spitting out photons that add up on top of each other.
This is called an undulator and thats what separates a FEL from a synchrotron light source. There's a subltely here:
The electrons are traveling essentially at the speed of light, so when they emit a photon while bending, the photon travels right alongside them, and the path difference between the small-curvature bending lines the electron takes and the straight line of the photon is negligible, meaning at each turn the electron is still right next to the photon it just emitted, so photons are stacking up on top of each other. Each new photon contributes linearly to the electric field strength, but the intensity of the beam is quadratic in the E-field - if the waves were incoherent, the E-field wouldn't be stacking up like that and the intensity is linear like for the wiggler.
The electrons can keep dumping energy into the photon field while hardly slowing down, in fact, they could lose 97.5% of their energy while slowing down less than 0.01% (realistic power transfer is more like 0.5-1% from electron beam to photon beam).
Now we have all the pieces we need to understand the state-of-the-art in Free Electron Lasers - Self Amplified Stimulated Emission.
Remember the RF buncher that we used to take a continuous beam of electrons and form it into bunches that can get accelerated by a well-defined RF bucket formed by an accelerator cavity? If we make the undulator long enough, there's now enough time for the more-spaced out electron bunch to feel the local electric field of its own emitted photons.
The emitted photon electric field then starts to squeeze the electron bunches tighter and tighter until they form 'micro bunches', now all the electrons in a micro-bunch start radiating photons in the same place at the same time, amplifying the photon electric field, bunching the electrons more and more, in an exponential positive feedback loop.
The emitted light beam now has an intensity that is quadratic both in the beam current, and the number of magnet poles along the undulator length. This intensity grows until saturation, at which point the electrons stop dumping net power into the photon beam and start picking up energy from it instead, just like they would in a normal RF bucket - rotating around the synchronous particle point, getting kicked forward and back by the field strength, and so on.
The maximum photon power you can get from a beam is P*Beam Power, where P is the FEL Pierce Parameter - yep, the very same Pierce who figured out the exact solution to extracting a space-charge limited beam with zero divergence. The physics of making a SASE FEL is closely related to the theory behind traveling wave tubes, something Pierce made foundational contributions in.
The klystron is a kind of compact particle beam thats used to power larger particle beams, and if you wiggle the particle bunches just right, they also act as RF amplifiers to make photon beams brighter.
Traveling wave tubes are a kind of particle beam used as an RF amplifier, just like the Klystron, something invented by the Varian brothers at Stanford in 1937.
That dusty shed stretched out across 3 miles of broken pavement slabs underneath the I-280, just off Sand Hill road exit, is called the Klystron Gallery: 245 high-powered Klystrons thrumming along, each pumping out 65 MW peak power for a peak RF power of 15.9 GW.
Those klystrons are still working, though they power LCLS-1 and FACET, SLAC's plasma wakefield accelerator which pushes accelerating gradients far past what an SRF cavity can support using plasma as a focusing and accelerating medium - see my thread here, and another thread here on RF sources
Now, the RF pulse itself is only a few microseconds, so total power is around 10 MW, about a small city worth...
Closing - The Future of Semiconductor Production
LCLS-2 produces hard X-rays, something that would be tough to use for making semiconductors since they are effectively absorbed by everything and so can't really be focused or channeled with conventional optics.
A FEL for semiconductor lithography would differ from LCLS-2 in pretty much every design goal:
- Electron energy of 300 - 500 MeV instead of 4 GeV
- High average power (multi-kW) instead of high peak brightness
Because the beam is a much lower energy its now feasible to recirculate it through a bending section - the Pierce FEL parameter is quite small, so the electron beam still has most of the initial energy in it.
@xLight_inc is building exactly this - an FEL designed for the future of semiconductor production, with a 13.5nm wavelength, and 120kW of photon power - feeding up to 20 scanners per source, each with 1-4 kW of power.
Now you're well-versed in all the basics of FEL design and operation - from RF bunching, acceleration, SRF cavity design, and the physics of self-amplified spontaneous emission (really, just more RF bunching). What's next?
Building beamlines for profit, for business, for trade.
To become... a Beamline Tycoon
as it happens there is a RCT2 style game that teaches you all of beam line engineering I'll release soon
(P.S. You can see this kind of future in person at @deeptechweek where the CEO of xLight presented recently at SF in June 2026)
