Latest / Hardware Startups with Fexingo: Devices, Robotics, and Manufacturing Tech Companies / How Hardware Startups Use Automated Optical Inspection for Solder Joint Quality
Transcript
- Lucas: Luna, that conversation we had last week with the CTO of a medical device startup — the one making a wearable continuous glucose monitor — it got me thinking about something. He mentioned that after they introduced automated optical inspection on their SMT line, their field failure rate dropped by about 40 percent over six months. Luna: Forty percent — that is significant. And field failures are brutal for a medical device. So AOI caught defects before they left the factory? Lucas: Exactly. AOI — automated optical inspection — is basically a camera system that checks solder joints on assembled PCBs. It's been around for a while, but the technology has evolved a lot in the last five years, especially with the shift to 3D AOI. And for startups building complex electronics, it's becoming a must-have step between pick and place and functional test. Luna: So walk me through what a typical AOI system actually does. Is it just taking a picture and comparing it to a golden board? Lucas: That's the basic idea, but it's more nuanced. The system uses multiple cameras and lighting angles — often top-down and angled — to capture several images of each solder joint. There's usually a red, green, and blue LED ring light that can be programmed to highlight different features. The software then compares each joint against a set of algorithms trained to detect specific defect types. Common ones include insufficient solder, excess solder, open joints, bridges, and tombstoning — where a component stands up on one end. Luna: Tombstoning. I've seen that — it looks like a tiny grave marker on the board. Usually happens with small passive components like resistors and capacitors. Lucas: Right. And with miniaturization, those components are getting smaller — 0402 and even 0201 packages. The human eye can't reliably inspect those. That's where AOI shines. But what's interesting is the move from 2D to 3D AOI. In 2D, you're looking at a flat image, so you can't really tell if a joint has enough solder volume. 3D uses structured light or laser triangulation to build a height map of each joint. That gives you actual volume measurements. Luna: So with 3D you can quantify whether a joint has, say, 50 percent of the expected solder volume versus 90 percent. That seems like a much more objective pass-fail criteria. Lucas: Exactly. And that matters because a joint with 60 percent volume might pass electrical test initially but fail after thermal cycling. The startup I mentioned — they were using 2D AOI and still seeing early-life failures. When they switched to a 3D system, they started catching those marginal joints. Their CTO told me they now have a 'solder volume' parameter for every joint type, and if it's below a threshold, the board gets reworked before it ever sees functional test. Luna: That's a concrete improvement. But I imagine AOI systems aren't cheap. How does a hardware startup justify the cost? A decent inline 3D system can be a hundred thousand dollars or more. Lucas: It's a big upfront cost, but the ROI comes from avoiding field failures. For a medical device, one field failure can cost thousands in replacement, not to mention reputation damage. The startup I mentioned — they're producing about 10,000 boards a month. They calculated that each field failure costs them roughly $200 in warranty and logistics. By catching 40 percent more defects, they're avoiding about 80 failures a month — that's $16,000 a month saved. The system paid for itself in under a year. Luna: And that's not even counting the soft costs like customer trust. So the math works, especially for higher-margin products where reliability matters. Lucas: Right. But there's another dimension: false positives. A lot of startups get frustrated with AOI because it flags too many good joints as bad. That slows down the line because someone has to verify each one. Some systems have false positive rates as high as 30 percent on dense boards. Luna: Thirty percent false positives — that's a huge efficiency hit. So how do you balance catch rate with false positives? Lucas: It's a tuning problem. You can set the thresholds aggressively to catch every possible defect, but then you get more false calls. Or you can set them looser and risk missing real defects. The key is to have a good verification workflow — usually a human operator at a review station looks at each flagged joint and decides. Over time, you can use that data to tighten the algorithm. Some modern AOI systems use machine learning to reduce false positives by learning the difference between real defects and cosmetic variations. Luna: Machine learning applied to AOI — that's becoming more common. I've seen startups use edge AI on the camera itself to run inference in real time. Lucas: Yeah, and that's exciting because it allows for adaptive inspection. The system could, for example, relax the threshold for a specific component that consistently shows minor cosmetic issues, but tighten it for critical joints like power supply pins. That level of granularity wasn't possible five years ago. Luna: I want to come back to something you said about inline versus offline AOI. A lot of startups I talk to are confused about which setup makes sense for their volume. Lucas: Sure. Inline AOI means the PCB passes through the inspection station right after the reflow oven, on the conveyor, without anyone handling it. That's the fastest approach — the board is inspected immediately, and if a defect is found, the system can trigger an alarm or even stop the line. But it's expensive and requires floor space. Offline AOI, sometimes called 'batch AOI', means an operator takes a stack of boards to a standalone machine and inspects them manually. It's cheaper and more flexible, but slower. For a startup doing low volumes — say 100 boards a month — offline might be fine. But once you hit a few thousand a month, inline becomes almost necessary. Luna: And inline also gives you real-time feedback to the pick and place machine. If you see a recurring misalignment, you can adjust the placement parameters immediately. Lucas: That's the real power — closing the loop. AOI isn't just about catching defects; it's about process control. The data from AOI can feed a statistical process control system that tracks trends. If the solder volume on a particular joint starts drifting over several boards, you can adjust the solder paste printer or the reflow profile before you produce a whole batch of bad boards. Luna: So it's a quality tool and a process improvement tool in one. That's the kind of thinking that separates hardware startups that scale from ones that get stuck in firefighting mode. Lucas: Exactly. And I think that's part of why we do this show — to have conversations that actually help founders and engineers make better decisions. We deliberately keep this podcast ad-free because we want the content to be useful and trustworthy, not interrupted by commercials. Luna: Yeah, that's something we hear from listeners a lot. If you appreciate that approach and want to support keeping it that way, there's a simple way to do it: buy me a coffee dot com slash fexingo. Lucas: And every bit of support goes right back into making the show better — more research, better guests, maybe even a microphone upgrade someday. So thanks to anyone who chips in. Now back to AOI. Lucas: Where we left off — closing the loop. One startup I've been following, they make a smart lock for commercial buildings. They use a compact inline AOI system from a company called Mirtec, I believe. They run about 5,000 boards a week. Their AOI caught a tombstone issue on a capacitor that was happening because the nozzle on the pick and place head was slightly worn. They replaced the nozzle and the defect rate dropped from 500 parts per million to 20 ppm. Luna: That's a huge improvement. And it shows how AOI data can pinpoint mechanical issues in the assembly process, not just solder defects. Lucas: Right. And that's the kind of insight you can't get from functional test alone. A functional test might pass a board that has a weak joint, but the AOI will flag it. That's why many startups are now putting AOI before functional test. It's a two-step screening: optical inspection catches assembly defects, then functional test catches electrical issues. Luna: So the order matters. If you do functional test first, you might damage the board or waste time troubleshooting a defect that AOI would have caught in seconds. Lucas: Exactly. And conversely, if you run AOI after functional test, you're inspecting boards that have already passed electrical test — so you're only catching cosmetic issues or defects that didn't cause electrical failure. The ideal sequence is: solder paste inspection, then pick and place, then reflow, then AOI, then in-circuit test, then functional test. That way you catch process drift early. Luna: And for startups that can't afford a full inline AOI system, what are the alternatives? I've seen some use benchtop microscopes with manual inspection, or even phone cameras with magnification. Lucas: Sure, but the manual approach doesn't scale. For a prototype run of 10 boards, a good technician with a microscope can inspect every joint. But at 500 boards a week, it becomes impractical. There are lower-cost AOI systems — some desktop models start around $20,000. They're slower, but they can handle a few hundred boards a day. Also, some contract manufacturers offer AOI as a service. You can send them your boards and they'll inspect them for a per-board fee. Luna: That's a good option for pre-revenue startups. You get the data without the capital expenditure. And the CM might even give you the inspection reports so you can analyze the data yourself. Lucas: Exactly. And that data is gold. For example, one startup I know designed a board with a QFN package — no leads visible from the top. They couldn't visually inspect the solder joints under the package. Their CM ran X-ray inspection but that's slow. They used AOI to check for proper solder wetting on the side of the package — the 'fillet'. AOI can measure the side fillet height and width, which correlates with joint quality. Luna: So AOI isn't just for thru-hole or discrete components. It can handle BGAs and QFNs to some extent, though not as well as X-ray. But for side fillets, it's effective. Lucas: Right. And the trend is towards integrating AOI with other inspection methods. Some systems now combine AOI with X-ray in a single station. But that's a big investment — usually only for high-reliability industries like aerospace or automotive. Luna: Let me ask a practical question: for a startup that's about to set up their first SMT line, what's the minimum viable inspection strategy? Lucas: I'd say start with a good solder paste inspection — that's the cheapest and catches stencil issues early. Then for the assembled boards, either a benchtop AOI or a partnership with a CM that offers AOI. And definitely have a high-magnification microscope at the rework station. If you see a defect, you need to understand its root cause. Over time, as volume grows, invest in inline AOI. The key is to not skip inspection entirely. I've seen startups that just do a quick visual check and then functional test. They end up with field failures that destroy their margin. Luna: And field failures can kill a startup. One bad batch can wipe out months of progress. So AOI is really an insurance policy. Lucas: Exactly. And as components get smaller and boards get denser, the insurance becomes more valuable. I think the next big shift will be ai driven AOI that predicts defects before they happen, using real-time data from the entire line. But for now, the smart move is to get a system in place and start collecting data. Luna: Good advice. So for any hardware founder listening: if you're not inspecting your boards optically, you're flying blind. Lucas: Couldn't agree more. That's all for this episode of Hardware Startups with Fexingo. I'm Lucas. Luna: And I'm Luna. Catch you next time.