Back to top
The pulsed power units behind Advanced Sources and Detectors (ASD) Scorpius accelerator have begun production at Lawrence Livermore National Laboratory, representing a significant achievement for the Laboratory and a key milestone toward advancing the nation’s stockpile stewardship work. Eventually to be installed underground at the Nevada National Security Site (NNSS), the 125-meter-long linear accelerator will produce multiple short, high-energy bursts of x-rays for radiographic imaging of dynamic subcritical experiments with plutonium. These experiments simulate the late stages of a nuclear implosion, when materials are under the most extreme conditions from explosively driven shocks.
Scorpius will provide the power and resolution required to take radiographs of plutonium behavior during these crucial moments. The data will help study the effects of aging and manufacturing methods on nuclear weapons and inform both the assessment of the current stockpile and the design and certification of the future stockpile.
As part of a multilaboratory effort led by Los Alamos National Laboratory, Livermore is delivering 984 pulsed power modules called line replaceable units (LRUs) that generate the accelerator’s power. Pulsed power systems accumulate energy and release it in short, intense pulses that greatly amplify instantaneous power. In Scorpius, each LRU stores energy in 45 circuit board stages that, when triggered, discharge and add their energy to the total output. The process then repeats on the machine scale, as all 984 LRU outputs combine and energize the accelerator’s injector, which produces the electron beams. The electron beams are amplified to 22.4 megaelectron volts (MeV), converted to x-rays after colliding with a metal target, and pass through the experiment, after which they are absorbed by a detector and converted into high-resolution images.
Livermore’s solid-state pulsed power system (see S&TR April 2021, Shining a Bright Light on Plutonium) is the first of its kind. Not only is it more compact than traditional capacitor-based systems, but each LRU can also be turned on and off at will, allowing researchers to adjust how long each pulse lasts and the gap between pulses. “The fully solid-state design is a Livermore innovation,” says Mike Zika, Principal Associate Deputy Director for the Strategic Deterrence Principal Directorate. “This innovation creates huge possibilities for designing experiments because instead of programming pulses based on the requirements of the machine, the team can ask for a pulse and the machine can deliver it.”
Livermore achieved its latest milestone in 2023 by delivering its first complete cluster of 24 LRUs—41 clusters will eventually be deployed on Scorpius. This accomplishment proved the design could be mass-manufactured and moved the project into production. As external vendors begin delivering production unit LRUs, Livermore’s focus has shifted to guaranteeing that the devices will work dependably across Scorpius’s expected 30-year lifetime. This means ensuring that roughly 1,000 LRUs—with 45,000 stages and 100 million parts—can generate optimal pulses for high-resolution radiography every time on an estimated 400,000 shots. “If we don’t build 1,000 LRUs that all perform within one percent of one another, the physics becomes extremely complex. We would have 1,000 things functioning 1,000 different ways and would need 1,000 ways to run the machine,” says Saeed Assadi, National Security Engineering Division (NSED) researcher and project lead.
LRUs for a Lifetime
Livermore sought to experimentally demonstrate 30-year reliability by testing 24 LRUs 200,000 times to collect performance data and learn how the parts behave as they age. “So far, these systems have shown that when we need them, they are ready for operation around 97 percent of the time,” says Assadi. “Compared to other pulsed power systems, which have approximately 80 percent reliability, these devices are the best in the world.” Assadi believes this reliability will scale up, assuming production continues with the same level of quality. “Building something a thousand times means we have to repair it a thousand times, and the cost becomes significant,” he says. “For Scorpius and its 100 million parts, manufacturing, as well as part selection, quality, and lifetime are crucial.”
Taking inspiration from industry, NSED hardware engineer Elaine Chung led a team that developed a series of functional tests to evaluate LRUs, LRU subsystems, and the equipment that controls each cluster. The tests put equipment under similar conditions to what they will experience in the accelerator to collect data and verify that they behave as expected. Chung’s team started with schematics and the National Nuclear Security Adminstration’s (NNSA’s) performance parameters for Scorpius—which describe the environmental conditions and required outputs for high-resolution radiography—learned how to simulate them for each component, and used that information to design and refine the tests. For more common components like LRUs and the LRU subsystems, the team developed seven fast, automated test stations to assure quality while preventing production delays. “Since we engage with multiple vendors, we needed some way to test that each subsystem is functioning the way we believe it should,” says Chung.
Operation involves placing subsystems in a specialized test fixture, in which a series of pins apply voltages, read currents, and take measurements. The stations quickly return a result and record it in a database for Livermore to use. As of December 2024, vendors had tested 8,700 carrier boards, 9,400 carrier control boards, and 4,300 test assemblies—representing over half of the 220 LRUs being produced during this phase of the project. “We went from a very cumbersome testing process to devices that can basically live at a vendor and be operated with minimal intervention,” says Chung. “All they need to do is put a part on the tester and either send it back to us if it fails or send it on to the next vendor if it passes.”
Programming the Perfect Pulse
Radiographic imaging of plutonium faces the same challenge as taking photos in the dark. Like a camera flash, Scorpius “lights up” plutonium to penetrate the extremely dense material with enough x-rays for the detector to generate clear reconstructed images. This process requires pulse shaping. Multiple pulses are generated with waveforms that resemble perfect rectangles to take high-resolution images at the precise time to understand how the material is evolving.
Scorpius’s multipulse sequences make pulse shaping difficult because each pulse causes reflections that interfere with the following pulses and degrades the quality of subsequent images. “We can’t make the second pulse in the same way as the first because the accelerator is resonating, and shaping becomes more difficult with every pulse,” says Zika. “The control systems are not fast enough to do this on the fly, so we have to plan in advance.”
The LRUs can turn on and off on a nanosecond timescale, allowing the team to program and run complex modulation patterns to compensate. “Making changes even over 5 nanoseconds enables us to compensate for any reflections coming back from the unmatched load,” says controls lead Terence Brown. “If a large voltage peak is reflected back, we can lower the output of the LRU to achieve the desired output voltage.”
Pulse shaping needs to account for a significant number of variables that affect LRU performance, including environmental conditions, voltage or current change, the interaction between devices, wear and tear, and whether all 984 pulsers are in sync. To do this, the team is developing machine learning (ML) and AI techniques to ensure the LRUs produce high-quality images with every shot. “Our system is non-linear and time-variant, which is why we started looking at optimization techniques and AI to see patterns that we can’t,” says NSED researcher Sam Brockington.
Brown’s team used a computer model to generate the large datasets required to train an AI system. Working backward from NNSA’s performance parameters, they established how each component should perform and ran simulations to determine which variables impacted performance the most. The team used the results and existing experimental data to train the model to produce synthetic data that looked like the real results. After optimization, the model could produce large enough quantities of good synthetic data to complement experimental results and train an AI system. “With our framework, we can train big machine learning models like neural networks on tons of simulation data and fine-tune them on smaller sets of experimental data to reproduce the facility’s behavior when we’ve taken experimental shots,” said Kelli Humbird, Strategic Deterrence group leader for computational physics. “We can then use that experimentally informed model to drive our optimization.”
In only seconds, the AI system can design a modulation pattern and calibrate LRU performance based on its environment to produce clean waveforms on nearly every pulse. Since first being demonstrated in 2022, the system has grown increasingly accurate as a result of continued optimization and new experimental data. “Using AI is a much faster way to get a pulse train that meets specifications,” says Zika. “AI is a huge enabler of efficiently delivering the data to meet program needs.”
Humbird’s team is also using the model to explore different ML and AI approaches that might be best suited to calibrating the full accelerator. “Scorpius was designed to take advantage of these machine learning and automation tools,” says Humbird. “The facility is so forward thinking that the concept of enabling the accelerator to tune itself in an experiment isn’t far from reality.”
Accelerating into the Future
Livermore has begun receiving LRU production units and plans to deliver them to NNSS for integrated testing in 2026. As the Scorpius project approaches completion, Laboratory researchers will work with collaborators at NNSS and Sandia and Los Alamos national laboratories to install and demonstrate the technology and bring the accelerator online at the end of the decade.
The chance to innovate while working on a high-impact national security project galvanizes the team and unites the four national laboratory partners. Assadi sees this reflected dedication, diversity, and collaborative spirit in the Laboratory’s engineers, physicists, and computational researchers that make their contributions possible. “All of us wear the Livermore jacket and we are proud of it,” says Assadi. “I am indebted to every one of them and their hard work, and what we produced is without a doubt the highest quality product I’ve ever seen.”
—Noah Pflueger-Peters
For further information contact Saeed Assadi (925) 424-6540 (assadi1 [at] llnl.gov (assadi1[at]llnl[dot]gov)).
