Although the technology of 3D printing objects using metal is well established, it tends to be too complex, expensive, or imprecise to be adapted to traditional methods at scale. Freeform, which has raised $14 million from Nvidia and Boeing, aims to change that by building a game-changing new metal additive manufacturing process. And yes, there is also an AI perspective.
Co-founders Eric Palitsch (CEO) and TJ Ronacher (President) both worked at SpaceX, where they served as lead architect and lead analyst for the Merlin engine and other programs, respectively. While there, they realized the potential of 3D printing parts using metal, but also experienced first-hand the drawbacks of this method.
“We realized the potential of metal printing, which has the potential to transform essentially every industry that manufactures metal products. However, adoption has been slow and success has been marginal at best. ,” Palitsch said. “Why isn't it practical to use at scale? Basically, it's due to three things: Poor quality, inconsistent. Speed — Commercial printers are very slow. And the cost – the price of these printers is astronomical.”
They concluded that if they could operate a process that provided printing services rather than selling printers, they could unravel the whole thing in a big way. So they teamed up with Tasso Lappas, former CTO of Velo3D, to start Freeform.
The main mistake companies made was using something like CNC machines commonly used in traditional manufacturing as a model for their metal printing business. In that case, you can sell the machine and its software and make it work in any shape or process you use it. But metal additives are different, Palitsch says.
“The way these work today is 'open loop,' meaning they're basically playing a file,” he explained. “They had to be smarter than that because the process of melting metal powder with a laser is very complex and in some ways infinitely variable.”
Selling people a machine and saying, “Become an expert at running it, good luck'' is not the recipe for success.
“But when you decide you're not going to build a printer and package it in a box, there's a lot you can do if you have the freedom to build an automated factory from a blank sheet of paper,” Palitsch said.
Image credit: Freeform
Their solution provides printing as a service using a closed-loop process on custom machines that monitors printing on a microsecond scale, and combines various elements to achieve the kind of printing expected at workplaces like SpaceX. It's about adjusting.
The company has many technology advancements to be proud of, but two are most directly relevant: feedback loops and the AI that manages them.
“Our system has high-speed computer vision feedback that operates on the microsecond scale, and all of that data is processed on state-of-the-art FPGAs and GPUs that have become available in the last few years. We had to build this entire stack ourselves,” Palitsch said.
A closed-loop system with real-time monitoring reduces quality issues while enabling rapid printing of complex geometries. We also keep our business model simple by operating as a printing service.
But a second technological breakthrough was needed to make that part of the system work. It's a machine learning model that's fast enough and has enough expertise to actually perform that monitoring.
Image credit: Freeform
“Eric and TJ went through this situation and came to the same conclusion: this industry needs a level of computing and sensors that no one has ever deployed before,” Rapace said.
“To properly understand how to control the process, we needed a dataset that worked on timescales that no one else had. So we created a curated, managed, almost self-labeled dataset. We have begun building a state-of-the-art telemetry system, a platform that collects
This data allowed us to bootstrap the model and generate additional data for a better model.
But then they faced the need for speed.
“Generative models have a lot in common, but they also don't have a lot in common. But one critical difference is latency,” LaPaz explained. “Inference needs to run in microseconds so that we can close the loop on these processes.” Since there were no off-the-shelf solutions available for data and compute, GPU/FPGA “AI on Steroids” ” I had to build the combo from scratch.
Resulting side effects: Freeform is “building the largest metal additive dataset in the world, which is why companies like Boeing come to us,” Palitzsch says. “We have this fundamental, core data collection and processing capability that no one else has.”
Add to this the fundamental benefits of print-based manufacturing, such as factory agility and versatility, and you have a very compelling business case.
Boeing's AE Ventures and Nvidia invested a total of $14 million, but did not provide further details. Each company's investment comes with perks. NVIDIA will provide companies with access to the H100 and other computing hardware, while Boeing will guide suppliers through the certification process and potentially buy parts in bulk. (Freeform will also participate in Nvidia's Inception startup program.)
Palitsch said the company has customers in “all nine” sectors: aerospace, automotive, industrial and energy. They declined to go on record, but did note that they make everything from rocket engine parts to exhaust parts for Formula 1 cars. They plan to use the funding to scale up, build the next generation of (much faster) printers, and hire up to about 55 people in total over the next year.
He acknowledged that it took some time for their approach to grow from theory to reality, but that their methodical, technical approach was also what led to their success.
“The transition was slow,” Parich said. “But when we look back… six of us built the fastest laser melting platform on the planet, and the hardware and software for it, from the ground up. has come true.”