From Horns To Armor: Biomimicry, Computational Design, And The Future Of Impact Protection
Matthew Shomper – Six Questions – CDFAM DC
Matt Shomper returns to CDFAM in DC this year, having presented at the inaugural event in 2023 and again in 2025 where he has discussed everything from Mantis Shrimp claw impact, bending saplings on hiking trails, building machine learning pipelines from CT scans for surgical guides and his background redefining spinal implants for, humans.
This year those threads converge around impact protection, with the bighorn sheep’s horn (previously mentioned in 2023) as the case study: a structure that survives their questionable decision to subject their brains to repeated high-speed collisions because of how the horn is formed rather than the material it is made of.
In this interview, Matt walks through where the engineering and manufacturing investment in architected materials actually pays off, how data moves from a CT scan to a machine-ready (3MF) build file, and the F13LD tools he is building to design and analyze these structures in the browser.
He also explains why he is keeping the work open and free, and what real-time homogenization reveals about geometry performance that batch analysis tends to hide.
You first presented on the design of biomimetic structures at the inaugural CDFAM in 2023, and then on using ML from CT scans for the design of surgical guides and implants in 2025. What can we expect from your presentation in DC this year?
Those two talks were really two halves of one idea, and this year is where they meet.
In 2023 I was making the case that biological structures are worth translating into manufacturable geometry; that nature’s microstructures aren’t decoration, they’re engineering. In 2025 I showed the pipeline side: how you get from a CT scan to a patient-specific guide or implant without hand-modeling every case.
This year the subject is impact protection, and the example is the bighorn sheep’s horn – a structure that survives repeated high-speed collisions because of how it’s formed rather than what it’s made of.
What’s different is that I’m not just showing slides of finished parts. The internal geometry is the thing I’m designing, and I’ll demonstrate how that forms interactively: browser-based tools where you steer an impact-deflection field through a structure and watch the geometry reorganize as you go.
Same horn-keratin logic, made interactive. So you’ll get the biology, the math that captures it, and a process for forming it inside of any unique geometry!
What are some of the applications you are seeing where the requirement is critical enough to justify the cost of the engineering investment and the manufacturing process for these kinds of structures.
Short answer: wherever just picking a better material has run out of room, and wherever the part is already bespoke. When both are true, the cost of designing the structure is small next to what it buys you.
Three modern areas where it makes sense. First, orthopedic and spinal implants, where the enemy is stress shielding.
Bone resorbs when an implant carries load it should be sharing, and the fix is matching and grading the implant’s stiffness to the patient’s bone. Emerging implant design is keyed to one person’s anatomy, so the digital pipeline pays for itself.
Second, protective equipment: armor, helmets, defense and sport. The spec is energy absorbed per unit mass and volume, and a few percent at the same weight is the difference between a concussion and not, or between a part that survives a hit and one that doesn’t.
Third – aerospace and space, where mass is expensive and energy-absorbing structures, brackets, and vibration control all reward a carefully designed internal geometry.
The thread running through all three is high consequence at low volume. Architected materials make no sense for a million identical injection-molded parts. They make complete sense when failure is expensive or irreversible and every part is already custom.
Tell us about the F13LD tools you are developing, how they relate to this work, and how others can use and/or extend them.
F13LD is a set of browser-based tools for designing and analyzing architected materials (i.e., metamaterials, structures, lattices, etc). Each one does a focused single job and hands off to the next. There are generators for geometry families: TPMS (product of sines and cosines), Grain (spinodoid, kernel fields, reaction diffusion), Beam families, and more. There’s an analysis tool that runs a full elastic homogenization and draws the directional stiffness surface so you can read the anisotropy directly.
There’s a parameter-space explorer, an inverse-design tool that picks a recipe to hit a target stiffness, and a shared database of measured structures the others can draw on.
Under the hood they run in the browser, with WebGL and WebGPU for visualization and a WASM geometry kernel for watertight mesh export (3MF only of course!), so there’s no install and no license server.
They’re meant to be opened, read, and forked. Each tool is one self-contained file, and they share a common recipe format and pass straight to one another, so adding a new geometry family or a new export doesn’t mean taking on a whole platform. That’s on purpose. I’d rather hand someone a thing they can take apart than a black box.
There are many tools out now that focus on design for additive manufacturing. A lot of them are fantastic, and the point of this suite is not to replace them.
Over the years I’ve found that the vast majority of engineers, researchers, technologists, etc still don’t fundamentally understand why you’d pick one structure over the other. And I’d like to change that via education. The quickest way to do it was to build it myself and make it free.
This presentation extends the idea of geometry from function, by focusing on one F13LD tool specifically – Grain – and its ability to form spinodoid fields. The geometry engine is ported into an interactive tool that displays how this type of mathematical field can be developed directly by its inputs.
Walk us through the data flow on a ‘typical’ project. Where does the data begin, say a CT scan, and how does it move through F13LD to a machine-ready build file, such as .3MF?
Take a patient-specific spinal interbody case. It starts as a CT scan: the DICOM stack gets segmented into the patient’s bone geometry and exported as an STL, with the anatomical landmarks pulled out alongside it.
Those landmarks feed a measurement step that computes the per-level numbers that actually define the part: disc height, lordosis, endplate footprint, across a multi-segment construct. That’s where the envelope and the mechanical target come from, how stiff the device needs to be, level-by-level, to share load with the bone instead of shielding it.
With a target in hand, I use the explorer and inverse-design tools to choose an architected-material recipe – smooth TPMS, intricate reaction-diffusion pattern, directional spinodoid, bundled helicoil – that hits the effective stiffness, checked against the homogenization tool and the measured-structure database so it isn’t a guess. I can also use the tools to cross-reference other design targets like porosity, permeability, curvature, and about 50 others! The generator writes that recipe to a compact file, and the patient STL comes along as the clip domain.
The meshing tool brings the two together. It evaluates the implicit field, clips it to the patient geometry, runs a level-set to produce a single watertight body, and exports .3MF. I use 3MF rather than STL on purpose: it’s the open AM standard, it stays watertight, and it carries the metadata downstream instead of flattening everything to a triangle soup. From there it’s machine-ready, and as-built performance can feed back into the database to sharpen the next design.
What is the most surprising thing you have discovered in developing these tools, and what insights about geometry performance have they uncovered that you could not discover with other tools available?
The most surprising result was geometric. The phase-intersection idea behind PI-TPMS, taking the region where two TPMS phases both stay near zero, produces a pipe network whose volume fraction scales with the square of the pipe radius rather than linearly. That separates pipe size from volume fraction in a way that opened a far wider, more controllable design space than I expected going in. The companion fix, correcting the distance formula so the pipes stay truly cylindrical no matter what angle two phases meet at, only mattered because I could see in real time that the naive version wasn’t.
And the real-time part matters more broadly. Because the tools homogenize in the browser as you update the design, you watch the directional stiffness surface change shape continuously instead of comparing the endpoints of a batch of FEA runs. That “real-time” view shows things batch tools hide: connectivity thresholds where the stiffness reorganizes abruptly, non-monotonic transitions, regimes you’d never think to sample on a discrete sweep.
The interesting behavior lives in the transitions, and you only find it by watching the structure move. The horn-keratin tubule pattern in the deflect tool is the same story; it fell out of the math agreeing with the biology, not from me prescribing it.
What do you hope to gain from participating in CDFAM, and what is the one takeaway you hope others leave with?
CDFAM is the one room where the implicit-modeling people, the FEA people, the computational pipeline people, the AI for geometry people, and the ones who actually build the parts are all there at once.
The DC edition adds the government and defense side, which is squarely where protective applications live.
I want to put these tools in front of that audience and see where they break, find people to build with, and get a clearer read on where AI and implicit design are heading. The most useful feedback I get is always from someone who pushes on an assumption I’d stopped questioning.
The one takeaway: architecture is the material now. For impact protection, the thing that matters most isn’t a better alloy or a tougher polymer, it’s the internal geometry, and we finally have both the math to design it deliberately and the manufacturing to make it.
A bighorn sheep has developed this over millions of years. Our job is to stop treating structure as something we pour into a volume and start treating it as the first thing we engineer, and to keep the tools for it open and interactive enough that it isn’t locked behind a six-figure seat.
Matt’s 2023 and 2025 talks are both up on the CDFAM YouTube channel, worth a look ahead of DC if you want the full arc of this work to do your background research after you register to attend DC/DC, and worth subscribing to while you are there.
CDFAM runs in Washington DC this July 15-16. If you are working on architected materials, design automation, or the manufacturing behind them, it is where you will find others working on the same problems. Come join us.
