Latest / Hardware Startups with Fexingo: Devices, Robotics, and Manufacturing Tech Companies / How Hardware Startups Use Solder Paste Inspection to Catch Defects Early
Transcript
- Lucas: So you spend weeks getting your PCB layout right. You pick the perfect components, you ship the Gerbers off to the fab, the boards come back looking gorgeous. Then you send them to the assembly house, they print solder paste, place parts, run them through reflow — and somehow, a quarter of the boards fail functional test. Luna: And you have no idea whether the problem was paste volume, misalignment, or something that happened inside the oven. Because you didn't look at the one thing you could have measured before it became invisible. Lucas: Exactly. That one thing is solder paste inspection — SPI. And today I want to talk about why this is probably the most overlooked quality gate in hardware startups, and why skipping it is almost always a false economy. Luna: I've definitely been in rooms where founders say, 'We'll just catch defects at functional test.' What's wrong with that approach? Lucas: The problem is you're catching the symptom, not the cause. By the time a board fails functional test, the solder joints are already formed. If the issue is insufficient paste, you've got a cold joint or a head-in-pillow defect — and the only fix is desoldering and rework, which costs time, damages pads, and introduces new failure modes. Luna: Right. So SPI is basically a checkpoint before reflow, telling you 'this board has good paste — send it through' or 'this one has a problem — clean it and reprint.' Lucas: Exactly. And the key metric is paste volume. Industry standard for a good print is typically within plus or minus 25 percent of the stencil aperture volume. But a lot of startups don't even know what their target volume is, because they never measured it. Luna: So walk me through the numbers. What does an SPI machine cost, and what's the return? Lucas: Sure. A brand new inline 3D SPI system from a major vendor like Koh Young or CyberOptics runs anywhere from forty to sixty thousand dollars. But you can pick up a used 2D system for under thirty thousand. For a startup doing, say, a thousand boards a month, that's three dollars per board — if you amortise over a year. Luna: And the alternative is catching defects later. What's the cost of a single board-level rework cycle? Lucas: Depends on complexity, but let's be conservative. If you have a tech manually diagnosing and reworking one board — desoldering a QFN, cleaning, reapplying paste, reflowing — you're looking at maybe thirty to fifty dollars in labour, plus the risk of damaging the board. If you have to scrap it entirely because the pads lifted, you've lost maybe a hundred fifty dollars in components and fab cost. Luna: So if SPI catches just one bad board per hundred, it pays for itself in a year. That's a pretty easy case to make. Lucas: It is. And there's a real startup example I want to mention. There's an IoT sensor company — let's call them SensorGrid — that was seeing about a twelve percent failure rate at functional test on one of their flagship products. They brought in an inline 3D SPI system, and within two weeks they found that their stencil was slightly misaligned, causing paste to deposit off-centre on fine-pitch QFN pads. Luna: So just a mechanical alignment issue that was invisible to the operator. Lucas: Exactly. They realigned the stencil, and their failure rate dropped from twelve percent to under two percent. That's an eighty percent reduction in field failures. The SPI machine paid for itself in about three months. Luna: And that kind of data also tells you something about your process capability. You can start tracking SPI metrics over time and see drift before it becomes a defect. Lucas: That's exactly what they did. They started using the SPI data to create control charts for paste volume. They found that their paste viscosity was changing over the course of a shift — the paste was drying out, so volume was decreasing. They adjusted their stencil printing parameters mid-shift based on that data. Luna: That's a really smart use of the tool. But I imagine there's a difference between 2D and 3D SPI. When does a startup really need 3D? Lucas: Good question. 2D SPI measures the area of the paste deposit — the footprint. But it doesn't tell you height. So you could have a deposit that looks fine from above but is too thin, which leads to insufficient solder joint volume. 3D SPI measures volume directly. For most modern designs with fine-pitch components — 0.4 millimeter pitch QFNs, or BGAs — 3D is really the way to go. The extra cost is maybe ten to fifteen thousand, and it's worth it. Luna: What about false calls? I've heard that SPI machines can be overly sensitive and flag boards that are actually fine, slowing down the line. Lucas: That's a real issue, especially with older 2D systems. False calls happen when the machine's algorithm mistakes a reflection or a board warpage for a paste defect. Modern 3D systems are much better — they use multiple cameras and structured light to build a height map, so they're less fooled by surface artifacts. But you still need to tune the inspection thresholds. If you set your pass/fail limits too tight, you'll get false calls. If you set them too loose, you'll miss defects. Luna: So there's a calibration phase. What's a reasonable starting point for a startup? Lucas: Start with the industry standard: plus or minus twenty-five percent of the stencil aperture volume. Then run a hundred boards, track which ones the machine flags, and manually inspect those. If you find that ninety percent of the flags are actually good boards, loosen the threshold to plus or minus thirty percent. The goal is to catch real defects without slowing down throughput. Luna: And what about stencil design? I've heard that SPI data can actually feed back into stencil design decisions. Lucas: Absolutely. That's one of the most powerful things. If you see consistent low volume on a particular pad, you might need to increase the aperture size or change the aspect ratio. Some startups use SPI data to validate their stencil design before going into production. One company I know — a medical device startup — iterated their stencil design three times based on SPI data from their first prototype run. Each iteration took about a week. Without SPI, they would have built a hundred boards, found failures at test, and had no idea the stencil was the root cause. Luna: So they saved weeks of guesswork. That's the kind of acceleration that matters for a startup. Lucas: It is. And it's not just about catching defects. It's about knowing your process. Once you have SPI data, you can correlate it with functional test results. You can say, 'Boards with paste volume below X have a thirty percent higher chance of failing test.' That's a predictive model. Luna: And that becomes a design rule for future products. You can tell your PCB designer, 'Don't put a via this close to a fine-pitch pad because the paste won't deposit evenly.' Lucas: Right. And that's how you build institutional knowledge. The startup that treats SPI as a learning tool, not just a pass-fail gate, gets compounding returns. Luna: You know, it's interesting — we talk a lot on this show about the tools and processes that separate hardware startups that scale from those that stay stuck in prototype hell. And SPI is exactly that kind of tool. If these conversations are useful for what you're building, listener support is what keeps the show ad-free and going deep on topics like this. Lucas: Yeah, it really does make a difference. If you find value in episodes like this, you can support the show at buy me a coffee dot com slash fexingo. No pressure, just helps us keep the lights on and the content independent. Luna: Alright — back to SPI. One thing we haven't touched on is whether a startup should buy its own SPI machine or rely on the contract manufacturer's machine. Lucas: Great question. If you're using a turnkey CM, they almost certainly have SPI on their line — most reputable SMT houses do. But the question is whether they're actually using it effectively. Some CMs set the thresholds very loose to avoid slowing down the line, so they might be passing boards that have marginal paste. Luna: So you need to ask your CM: what are your SPI pass/fail criteria? Can we get the data? Lucas: Exactly. And if they won't share the data, that's a red flag. Some CMs will give you a report per panel. Others will only give you a summary. Ideally, you want the raw data so you can do your own analysis. If you're doing high volume, it might make sense to buy your own machine and put it at the CM's facility. That's actually a common model for mid-size hardware companies. Luna: But for a very early stage startup doing just a few hundred boards, it probably makes more sense to use the CM's machine and negotiate data access. Lucas: Exactly. The breakeven on owning a machine is somewhere around five hundred to a thousand boards per month, depending on board complexity. Below that, you're better off paying the CM's per-board SPI fee, which is typically a few cents per board. Luna: And what about board complexity? Does SPI matter more for certain types of boards? Lucas: It matters most for boards with fine-pitch components — anything under 0.5 millimeter pitch — and for boards with BGAs or QFNs. Those are the joints where paste volume is critical. For through-hole or large-pitch SMDs, you have more margin. But even then, SPI can catch issues like paste bridging that would cause shorts. Luna: So it's not just about volume — it's about shape and position too. Lucas: Right. Modern SPI systems also measure paste height and area, and they can detect smearing or slumping. Smearing happens when the paste spreads too thin after printing, which can cause shorts. Slumping is when the paste collapses after sitting for a while — that's a paste viscosity issue. Both are detectable with 3D SPI. Luna: And you can catch those before reflow, so you can reprint the board instead of scrapping it. Lucas: Exactly. And that's the whole point: catching defects at the cheapest possible point in the process. The cost of quality goes up exponentially as you move downstream. Catching a defect at SPI costs pennies. Catching it at functional test costs dollars. Catching it in the field costs thousands. Luna: So SPI is really the first line of defence in the SMT process. And it's one of those things that, once you start using it, you wonder how you ever lived without it. Lucas: That's exactly the reaction I've seen from every startup that's adopted it. They start with a mindset of 'we can't afford it' and end with 'we can't afford not to have it.' The data is just too valuable. Luna: Alright, so for a founder listening who's about to do their first production run: what's the one action item? Lucas: Call your contract manufacturer and ask them three questions. One: do you have an inline SPI machine on the line that will run my boards? Two: what are the pass/fail criteria, and can I set them? Three: can I get the SPI data for every board in a report? If they say no to any of those, consider finding a different CM — or at least plan to buy your own machine if your volume justifies it. Luna: That's a solid checklist. And if the volume isn't there yet, you can still do offline SPI — you manually take boards off the line, measure them with a benchtop system, and adjust the printer. It's slower, but better than nothing. Lucas: Absolutely. There are benchtop 3D SPI systems for under ten thousand now. They're not as fast as inline, but for a startup doing low volume, they're perfectly adequate. And they give you the same data. Luna: So the barrier to entry is actually pretty low. It's more about awareness than cost. Lucas: I think that's right. A lot of hardware founders come from software or mechanical engineering backgrounds, and they don't know what they don't know about SMT process control. SPI is one of those things that seems like a nice to have until you have a disaster. But if you put it in place early, you avoid the disaster altogether. Luna: And that's the goal, right? Ship reliable hardware, build customer trust, and scale without quality issues. Lucas: Exactly. And SPI is a big part of how you get there.