Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Robot Arms Cant Solder a Circuit Board
Transcript
- Lucas: Alright, let's talk about soldering. Specifically, why a robot arm that can assemble a car engine with micron-level precision still can't reliably solder a single through-hole component onto a circuit board. Luna: Wait — I thought we'd solved this. I mean, there are pick and place machines that populate boards faster than any human could. Lucas: Right, those are different machines. They place components — they don't melt metal and make an electrical and mechanical joint. That's where it gets interesting. So, a standard industrial soldering iron is about 350 degrees Celsius. The solder wire — typically a tin-lead or lead-free alloy — melts at around 180 to 230 degrees. The robot has to feed that wire, apply the iron to the joint, hold it for exactly the right amount of time, and then pull away cleanly. Luna: And that's hard because...? Lucas: Because the joint itself changes as it heats up. The thermal conductivity of the board, the component lead, the pad — they all vary. If the board has a ground plane, it sucks heat away. If it's a small signal trace, it heats up fast. A human feels that through the iron and adjusts. A robot with a fixed time and temperature profile... doesn't. Luna: So it's a thermal feedback problem. Not just a positioning problem. Lucas: Exactly. There's a 2024 study from the University of Bristol that really drove this home. They gave a six-axis industrial arm — one of those KUKA units that costs around thirty thousand dollars — a standard soldering iron and a set of five hundred identical through-hole resistors on identical boards. The robot used a vision system to align the iron, a force sensor to control pressure, and a thermal camera to monitor the joint temperature. Luna: And the result? Lucas: Zero percent acceptable joints. Every single one failed visual inspection or electrical continuity testing. The biggest issue was something called 'cold solder' — where the alloy doesn't fully wet the pad and the lead. The robot either didn't heat the joint long enough, or heated it too long and damaged the board. Luna: That's remarkable. Five hundred attempts, zero successes. But could they have tuned the parameters per joint? Lucas: They could — and they did, for a subset. They hand-tuned the time, temperature, and feed rate for one specific board layout. Then they got about a 60 percent success rate. But that's the thing — change the board, change the component, change the solder alloy, and you have to retune everything. A human can adapt in real time. The robot can't. Luna: So it's not that robots can't solder at all — it's that they can't solder flexibly. And in a world where electronics change constantly, that flexibility is the whole game. Lucas: That's it exactly. And you know, it reminds me — this is actually a great example of why listener support matters for a show like this. We spend time digging into these specific studies, talking to researchers, and trying to make sense of where robotics actually is versus where the hype says it is. That kind of work takes resources, and we keep the podcast ad-free because of people who chip in at buy me a coffee dot com slash fexingo. Luna: Yeah, and it's not about some big production budget — it's about the time to go through a paper like that Bristol study and pull out the real numbers. That's what we love doing. Lucas: Alright — back to the solder. So why can't we just give the robot better sensors? Why not real-time thermal feedback? Luna: That's exactly what some companies are trying. There are systems now that use a thermocouple embedded in the iron tip to measure the joint temperature indirectly. The problem is that the thermocouple reads the tip temperature, not the joint temperature. There's a lag, and the thermal mass of the tip smooths out the actual heat flow. Lucas: Right. So the robot is essentially flying blind on the thing that matters most — the actual melting and wetting of the solder. And then there's the wetting angle itself. When solder melts, it flows based on surface tension. It should form a concave fillet between the lead and the pad. But if the temperature is off, the wetting angle changes, and you get a convex, brittle joint. Luna: And vision systems can't easily measure wetting angle in real time. You'd need a high-speed camera and some serious image processing, and even then, the joint is obscured by the iron tip during the process. Lucas: So the state of the art is still open-loop or pseudo closed loop. And that's for through-hole. Surface mount is even harder because the components are smaller and the pads are buried under the component body. Think about a quad-flat no-leads package — those little square chips with pads on the bottom. You can't even see the joint. You're relying on the solder paste reflowing in an oven, not a robot iron. Luna: Right, and that reflow process is actually more robot-friendly because it's a batch process. You paste the board, place the components, and run it through a reflow oven. That works. But for through-hole or odd-form components — connectors, large capacitors, things that can't survive reflow — you still need hand soldering or a robot with an iron. Lucas: And that's where the labour cost is. A skilled electronics assembler can earn forty to sixty thousand dollars a year in the US, and they can solder a through-hole joint in about two seconds with a 99.9 percent first-pass yield. The robot might take four seconds and achieve 60 percent — and that's after tuning. The economics just aren't there yet. Luna: Unless you're in a very high-volume, very consistent product line — like some automotive electronics where the boards are identical for years. Then you can amortise the tuning cost. Lucas: Right, and that's exactly what happens. But it's the exception, not the rule. So the phrase 'lights-out manufacturing' for electronics assembly is still a long way off. The Bristol study is a useful reality check. Luna: So what would it take to solve this? A fundamentally different sensing approach? Lucas: Some researchers are looking at using machine vision to watch the solder wire as it melts — the shape of the molten droplet, the way it flows onto the pad. Others are trying force sensing on the iron to detect when the solder wets, because the surface tension changes the force on the tip. But these are all early-stage. No one has a commercial system that does it reliably. Luna: It's interesting — we think of soldering as this simple, low-skill task. But it's actually a beautiful example of sensorimotor integration. Heat, pressure, timing, vision, and a bit of intuition. Lucas: Exactly. And that's why it's such a good benchmark for robotic dexterity. It's a task that looks easy but is secretly incredibly hard. And until we crack it, there will always be a place for the human hand in electronics assembly. Luna: So next time you see a circuit board, think about the fact that those joints were probably made by a person — maybe someone in a factory in Shenzhen or Guadalajara, using a tool that hasn't fundamentally changed in fifty years. Lucas: And that's not a failure of robotics. It's a reminder that some skills are harder to automate than they appear. We'll keep an eye on this space — I have a feeling we'll see progress in the next five years, but it won't be overnight. Luna: Thanks for listening. That's it for this episode of The Robotics Podcast.