Latest / Hardware Startups with Fexingo: Devices, Robotics, and Manufacturing Tech Companies / How Hardware Startups Use Generative Design
Transcript
- Lucas: I was looking at some recent case studies on hardware startups, and there is one trend that keeps popping up that feels fundamentally different from what we discussed in episode one hundred and seventy-two about hermetic sealing. It is not just about how you seal or test the device, but how you actually create the geometry of the part itself. We are seeing a massive shift toward generative design, where engineers stop drawing lines and start defining constraints. Luna: Right, so instead of a human sketching a bracket, you feed the software the load paths and let it figure out the shape? That sounds like a huge leap from standard CAD workflows. Lucas: Exactly. The most compelling example I found involves a mid-sized robotics company that makes autonomous mobile robots for warehouse logistics. They were struggling with the weight of their chassis components, which directly impacted battery life and payload capacity. Traditional machining meant they had to leave extra material for structural integrity, adding unnecessary mass. Luna: And they couldn't just machine away the excess without risking the structural integrity, so they needed a way to remove material intelligently. Lucas: Precisely. They switched to a generative design workflow using cloud-based software. They defined the space envelope, the bolt holes for existing mounts, and the maximum stress limits under full load. The algorithm then produced dozens of organic, bone-like structures that looked nothing like anything a human engineer would draw by hand. One single chassis plate went from being made of three separate machined pieces welded together to a single, unified print. Luna: That is a significant consolidation. Did the cost analysis show that printing one complex part was cheaper than machining three simple ones? Lucas: It did, but not just because of labor. The part weight dropped by roughly thirty-five percent. Since these robots run on batteries, every gram saved extends operational time. But more importantly for their bottom line, they reduced assembly time from twenty minutes per unit to zero minutes for that specific component. No welding, no alignment, no fasteners. It was a net win on both capital expenditure and operating expense. Luna: So the value isn't just in the raw material savings, but in the elimination of the assembly step entirely. That changes the supply chain dynamic significantly. Lucas: It does. And here is where it gets interesting for other hardware founders. You might think this technology is only viable for high-volume additive manufacturing, like metal laser sintering. But many startups are now using generative design even when they plan to injection mold or cast later. They use the output as a highly optimized master model, reducing the amount of raw plastic or aluminum they need to purchase upfront. Luna: That is a smart pivot. If the tooling costs are already sunk, optimizing the part geometry to use less material during production still improves margins per unit. Lucas: Correct. However, there is a cultural hurdle that most teams underestimate. Engineers are trained to understand why a part looks the way it does. When an algorithm spits out a lattice structure that looks like a piece of coral, early-stage engineers often reject it because they cannot intuitively verify its safety. Trust becomes the biggest bottleneck, not the software capability. Luna: So validation is the key. You have to prove the algorithm didn't just find a loophole in the physics simulation that would fail in the real world. Lucas: Exactly. Leading firms are now building rigorous physical testing protocols specifically for generative designs. They run finite element analysis to predict failure points, then physically destroy prototypes to compare actual breakage patterns against the simulation. Once that correlation is established within the team, adoption accelerates rapidly. Luna: It reminds me of how automotive companies moved from wind tunnel testing to computational fluid dynamics. It took years before engineers fully trusted the numbers over their gut feel. Lucas: A perfect analogy. And just like CFD, once you trust the tool, you unlock design spaces that were previously impossible. We are seeing heat sinks that look like internal organs, offering vastly better thermal dissipation than traditional finned blocks, all generated by optimizing for airflow volume and surface area. Luna: If you can make a heat sink lighter and more efficient, you also reduce the fan power needed to cool it, which adds another layer of efficiency to the system. Lucas: You hit on a critical point. Systems thinking is essential here. Generative design works best when you define constraints for the whole subsystem, not just individual parts. If you optimize a motor mount in isolation, you might miss an opportunity to integrate the mount into the motor housing itself, saving even more weight and parts count. Luna: That requires a much broader definition of the problem space. Most startup founders are used to solving discrete problems, not holistic system optimizations. Lucas: It is a mindset shift. But the ROI is clear. For hardware startups operating in tight margins, reducing part count is one of the fastest ways to improve reliability. Fewer parts mean fewer failure modes, fewer inventory SKUs, and simpler logistics. It is a defensive strategy as much as an offensive one. Luna: Defensive in the sense that complexity is the enemy of quality. Simplifying the bill of materials reduces the risk of sourcing delays or quality inconsistencies across suppliers. Lucas: Spot on. And speaking of resources, if these conversations about building smarter hardware are useful for what you're running or creating, consider supporting the show. Listener contributions keep our independent research and deep-dive analysis ad-free and sustainable. You can support us at buy me a coffee dot com slash fexingo. It helps us continue bringing you these technical breakdowns without corporate interference. Luna: Yeah, your support literally pays for the data subscriptions and expert interviews that make these episodes possible, so thank you for keeping the lights on. Lucas: Back to the engineering challenge. There is a common misconception that generative design outputs are only printable via additive methods. While true that complex lattices require 3D printing, many successful startups use the software to simplify parts for traditional manufacturing too. They might generate a part that looks strange but can be cast in a single mold, eliminating the need for multi-part assemblies. Luna: So the algorithm finds a shape that is manufacturable in a specific way, rather than just mathematically optimal? That seems like a crucial constraint to include. Lucas: It is mandatory if you want to scale. You have to tell the software whether you are targeting injection molding, die casting, or CNC machining. Each process has draft angles, minimum wall thicknesses, and tool access limitations. A good generative platform respects those boundaries while maximizing performance within them. Luna: Otherwise, you end up with a beautiful digital model that costs millions to tool and produces parts that crack during ejection. Lucas: Exactly. Manufacturing constraints are not obstacles; they are creative parameters. Some of the most elegant designs come from severely restricted tooling options. It forces the algorithm to get clever rather than just adding material wherever it thinks it is needed. Luna: It turns limitation into innovation. That is a powerful framework for any hardware team stuck in traditional design habits. Lucas: I think the next wave will see generative design integrated directly into PLM systems, allowing for continuous iteration based on field data. Imagine a drone company receiving telemetry on which arm joints fail first, automatically updating the design constraints, and pushing a new optimized version to production within weeks. Luna: That closes the loop between product usage and product design completely. It moves hardware development from a linear process to a continuous feedback loop. Lucas: Which brings us back to the robotics company we started with. They are now using field data to refine their next generation of chassis plates. The algorithm doesn't just know the static loads anymore; it knows the vibration profiles and impact events that occur in real warehouses. The result is a part that is not just strong, but resilient to the specific abuse patterns of its environment. Luna: That level of contextual intelligence is something traditional design processes simply cannot capture efficiently. It feels like the difference between designing for a textbook and designing for reality. Lucas: It is. And for hardware startups, that edge between textbook and reality is often where margin lives. Optimizing for the actual use case, down to the gram and the cycle count, compounds quickly over thousands of units. Luna: Do you think this technology will democratize further, or remain locked behind expensive enterprise software licenses for larger players? Lucas: The barriers are coming down. Cloud computing has made the heavy lifting affordable, and open-source tools are beginning to emerge. But the expertise required to frame the problem correctly remains valuable. The software is getting cheaper, but the judgment needed to set the right constraints is becoming more important. Luna: So the skill shifts from manual drafting to strategic problem definition. That is a fascinating evolution for the engineering profession. Lucas: Indeed. It raises the bar for entry but potentially raises the ceiling for output. Startups that master this workflow early could have a significant cost and performance advantage over incumbents stuck in legacy design routines. Luna: It seems like the companies that treat design as a dynamic, data-driven process rather than a static phase will be the ones leading the next decade of hardware innovation. Lucas: I agree. The question is no longer just how to build the product, but how to let the product design itself based on real-world demands. It is a profound shift in how we approach creation. Luna: A shift that rewards those willing to trust the process enough to let go of the pencil, metaphorically speaking. Lucas: Metaphorically speaking, yes. And perhaps literally, given how few designers are picking up styluses compared to those configuring parameter sets in the cloud today.