Latest / Hardware Startups with Fexingo: Devices, Robotics, and Manufacturing Tech Companies / How Hardware Startups Solve Thermal Management for High-Power Devices
Transcript
- Lucas: You spend months perfecting your PCB layout and then, first time you push current through it, you touch a component and burn your finger. Thermal management is probably the most overlooked failure mode for hardware startups building anything that draws more than a few watts. Luna: Right, and it's not just about not getting burned — heat destroys reliability. Every ten-degree rise in junction temperature roughly halves the lifespan of a semiconductor. Lucas: Exactly. And that compound effect is brutal. A startup building a 10-kilowatt motor controller for drones or e-scooters might have a bill of materials of, say, 80 dollars. If they blow a FET because they skimped on a heat sink, they're not just losing the part — they're losing weeks of debugging and a potential customer. Luna: So what's the right approach? Do you start with simulation, or just over-engineer the cooling in the first prototype? Lucas: Honestly, both. But the smartest startups I've seen follow a simple rule: calculate your thermal budget before you lay out the board. You estimate your worst-case power dissipation, then figure out the maximum ambient temperature and acceptable junction temperature. From there, you know how many degrees Celsius you have to drop, and that tells you your cooling strategy. Luna: That sounds straightforward, but I see people skip it all the time because they're in a hurry. They just slap on a fan and hope. Lucas: And fans are actually a great example of a shortcut that can backfire. They give you huge cooling capacity cheaply — a 5-cent fan can move a lot of air — but they introduce moving parts, acoustic noise, and dust accumulation. For a consumer product like a home robot, that's a non-starter. So you're pushed toward passive solutions: heat sinks, heat pipes, vapor chambers, or phase-change materials. Luna: Phase-change materials — those are the materials that absorb heat by melting at a specific temperature, right? Like a wax that turns to liquid inside a pad? Lucas: Yes. They're great for thermal buffering if you have short bursts of high power. A drone that does an aggressive climb for 30 seconds — the phase-change material can soak that transient heat spike and then slowly release it over time. But for sustained power, you need a continuous path to a finned heat sink or a metal enclosure. Luna: And what about the new vapor chambers? I've seen them in high-end smartphones and some laptop coolers. Are they accessible for startups? Lucas: They're getting there. A vapor chamber is essentially a flat heat pipe — a sealed copper plate with a wick structure and a small amount of fluid. They spread heat very efficiently over a large area. Five years ago they cost 10 to 20 dollars each, but now I'm seeing Chinese suppliers offer custom sizes for under 5 dollars at moderate volumes. That's within reach for a startup doing a few thousand units. Luna: So which solution should a hardware startup start with for their first prototype? Simulation tools? Lucas: I'd start with a simplified thermal spreadsheet calculation first — your power, your allowed temperature rise, your thermal resistances. Then you can validate with a cheap thermal camera, like a FLIR One for your phone. That alone catches 80 percent of problems. Once you have a board, you can run it at full load and see where the hot spots are. Then you iterate. Luna: Let's talk about a concrete example. Say I'm building a drone motor controller that draws 50 amps continuous from a 6S battery. That's about a kilowatt. I'm using six MOSFETs in parallel. How do I size the heat sink? Lucas: Okay, let's run the numbers. Each MOSFET might have an Rds of 5 milliohms when hot. At 50 amps, the conduction loss per MOSFET is I-squared R — so 50 squared times 0.005, that's 12.5 watts. But you have six in parallel, so the current splits roughly equally, so each MOSFET carries about 8.3 amps. That gives about 0.35 watts per device. Wait, I realize my math was off: with six parallel, the effective resistance is 0.005 divided by 6, so about 0.83 milliohms total. At 50 amps, the dissipation is 50 squared times 0.00083, which is about 2.1 watts total. Plus switching losses maybe double that. So you have maybe 4 to 5 watts to get rid of. That's not a lot. A small finned heat sink with natural convection can handle that without a fan. Luna: So thermal management is not always a huge problem. But it depends on the efficiency. If your MOSFETs are sub-optimal or you're pushing higher currents, it gets tough. Lucas: Right. And that's what trips up startups. They pick a part based on cost or availability, and then discover its package has a junction to case thermal resistance that's way too high for their application. So the rule is: always look at the thermal resistance numbers in the datasheet, and derate for real-world conditions. Luna: This kind of practical knowledge is exactly why we do this show — no ads, no sponsors, just real engineering conversations that help you avoid expensive mistakes. Lucas: Absolutely. We deliberately keep these episodes ad-free because we think that makes them more useful for people building real things. If you find value in what we do, you can support the show at buy me a coffee dot com slash fexingo. It helps us keep doing this. Luna: And it really does make a difference. So, back to thermal management: what about using the product's own enclosure as a heat sink? A lot of drone frames are metal - can you just bolt the MOSFETs to the frame? Lucas: You can, but you need to be careful about electrical isolation. Most metal frames are aluminum, which is a great thermal conductor but also an electrical conductor. So you need an electrically isolating thermal pad — something like a Bergquist Sil-Pad or a ceramic pad — between the MOSFET tab and the frame. Then the frame becomes one big heat sink. But you also need to ensure the path to ambient air is good; if the frame is enclosed in a shell, it might not get enough airflow. Luna: Another approach I've seen in some robotics startups is using liquid cooling — running coolant through channels in the enclosure. But that seems overkill for most products. Lucas: It is overkill for anything under a few hundred watts. Liquid cooling adds a pump, a radiator, hoses, fittings — the BOM cost jumps, and reliability drops because of potential leaks. I'd only recommend it if you're pushing 300 watts or more of continuous dissipation in a tight space. For example, some autonomous lawn mower startups with heavy-duty blade motors have gone to liquid cooling to keep the electronics compact. Luna: What about thermal simulation software? There are free tools like OpenFOAM, and then commercial tools like Ansys Icepak. When should a startup invest in simulation? Lucas: If you have a complex geometry — multiple PCBs in an enclosure, stacked components, blocked airflow — simulation can save you time and money. But it also requires skill to set up correctly. My rule is: if you can build a physical prototype and measure it, that's usually faster and more accurate for a first design. Simulation is best for optimizing later stages or when you can't easily build a physical mockup. There are also cloud-based tools like SimScale that let you do CFD for free if you're okay with limited compute. Luna: Let's talk about testing. What's the minimum thermal testing a startup should do before shipping? Lucas: You need to run the device at its maximum rated power in the worst-case ambient temperature it will see. For a outdoor device, that might be 50 degrees Celsius in direct sunlight. You monitor all semiconductor case temperatures with thermocouples — cheap type-K thermocouples and a data logger — and ensure you stay below the maximum junction temperature with a safe margin. Most engineers aim for at least 20 percent derating. If you're close to the limit, you need to improve cooling or choose a different part. Luna: And you should test for long enough to reach thermal equilibrium. I've seen startups test for 5 minutes and call it good, but then a device thermal soaks over an hour and fails. Lucas: Exactly. Thermal time constants can be many minutes for large heatsinks. I run tests for at least an hour, sometimes longer. And I also test the thermal cycling — turn it on and off repeatedly to see if expansion and contraction cause cracks in solder joints. That's a common failure in outdoor products. Luna: What about using thermal fuses or active thermal management like throttling back power when it gets hot? Lucas: Throttling is a great safety net. Include a temperature sensor — a cheap NTC thermistor — on the hottest part of the board, and have the microcontroller reduce power or shut down if it exceeds a threshold. That lets you design for average usage and survive peak conditions. Thermal fuses are good for catastrophic failure protection, but they're one-time use, so they shouldn't be your primary defence. Luna: So the key takeaways: calculate your thermal budget early, choose parts with good thermal specs, prototype and measure with a thermal camera, and build in throttling. Anything else? Lucas: One more: don't forget about thermal interface material — the stuff between your component and the heat sink. A poorly applied thermal pad can add 10 degrees or more of temperature rise. A good quality pad or paste is cheap insurance. And don't assume that more paste is better — a thin, even layer is ideal. Luna: Interesting. I've heard that some startups use graphite sheets now, like from Panasonic's EYGC series, for very low thermal resistance without the mess of paste. Lucas: Graphite is great for spreading heat laterally — it's anisotropic, so it conducts much better in-plane than through-plane. It's excellent for heat spreading in thin gaps, like between a PCB and a metal cover. But it's expensive. For most startups, a good silicone-based thermal pad is more cost-effective. Luna: Thanks for the deep dive. This is one of those topics that can make or break a product launch. Lucas: It's a classic case of pay now or pay later — a few hours of thermal design upfront saves weeks of rework. And your customers get a product that doesn't burn them — literally. Luna: Let's hope so. Up next: we're looking at how startups are using embedded ML on microcontrollers to add intelligence without a cloud connection.