By applying machine learning to traditional computational design, Cary-based ExLattice is increasing engineers’ ability to test new designs quicker while simultaneously lowering costs. Led by Co-Founder and CEO Runze Huang, the startup is positioned to be the acceleration launchpad of growth for the additive manufacturing market.
More commonly known as 3D printing, additive manufacturing arguably may be changing the world of supply-chain production, with its growing $18 billion market that is expected to surpass $25 billion by 2032.
However, only 15% of this market is dedicated to mass manufacturing, as the current market focuses mostly on prototyping. The mass production side of additive manufacturing could potentially replace the need for traditional machinery to mass produce finished products at a large scale, but it faces challenges such as high production costs and slow production speed.
ExLattice’s B2B SaaS-based solution incorporates traditional computational physics with machine learning kernels—which are a collection of distinct algorithms for pattern analysis— to speed up the production process without losing accuracy or causing damage to the product.
When Huang was studying computational design theory for his postdoctoral at Carnegie Mellon, he realized that the current software to produce physical products was not only slow, but also didn’t leave enough room to analyze inaccuracy in the product design until after it was finished making a product.
This pain point inspired him to launch ExLattice in 2018. The ExLattice “engine” uses computational math to simulate the physics of manufacturing processes so that it could produce designs that comply with all requirements before doing the physical manufacturing.
Then, Huang realized that if the “engine” ran with the traditional kernel, that iterative process would still take two weeks to generate a solution. That’s when he decided to pursue a hybrid model of taking the traditional physics simulation model and replacing the time-consuming steps with AI and machine learning models to solve problems and identify issues faster.
According to Huang, no matter how accurate a simulation is, there will always be uncertainty. Engineers would need to make lots of assumptions in the inputs they make into the model. ExLattice’s engine is able to tackle this problem with machine learning to significantly speed up the process with only a very small, acceptable sacrifice in accuracy.
By only giving up, for example, two percent of accuracy in the computational model, it will take that “extra room” to enhance the processing speed to show engineers those iterative simulations and problems in real time, so that they can address them much faster. Instead of waiting days or weeks to find the issues to solve, ExLattice will show those potential future problems in a few seconds, allowing engineers to be able to fix issues almost immediately.
“Now we have something that potentially can be disruptive and can lead to a complex software product that never existed before,” Huang said.
ExLattice plans to launch a co-pilot product that will work with popular computer-aided design (CAD) software programs. As changes are made in the CAD platform, ExLattice can diagnose whether they might create problems in the manufacturing process. If they will, the ExLattice “engine” will highlight the issues and make suggestions to improve the designs to avoid those issues.
“3D printer manufacturers generally have their own simulation kernel that takes too long to compute for their iterative designs,” Huang said. “We come in, customize our technology for their machines and then demonstrate simulation jobs that could take, what used to be half an hour, down to 10 milliseconds. So we can do crazy things that they couldn’t imagine before.”
Huang’s vision for the future of additive manufacturing machines is going to be similar to electric vehicle (EV) sensor technology. EVs possess autonomous sensors because of chips and pieces of code that are dedicated to processing vehicle data in real time. Current manufacturing machines have similar sensors but have little to no pieces of software dedicated to processing product creation in real time.
Huang said that the kernel ExLattice is building is going to be the fundamental piece to lead to real time processing of data collected by those sensors to inform the machines on how to manufacture products better and faster.
ExLattice was recently chosen to be a part of the Techstars Industries of the Future Accelerator class of 2023 cohort. In 2022, ExLattice received a Phase 1 $250K award from the National Science Foundation (NSF) Small Business Innovation Research (SBIR) Program to further develop its accelerated simulation engine. In 2021, it joined the NVIDIA Inception program, which is designed to nurture startups with advancements in AI; and in 2018, the startup received a $50K SEED grant from NC IDEA.
ExLattice plans to offer a subscription plan that will have three tiers of licensing in order to target three different groups of customers, depending on their usage and company size.