Latest / Hardware Startups with Fexingo: Devices, Robotics, and Manufacturing Tech Companies / How Hardware Startups Use In-Circuit Testing to Catch Assembly Defects
Transcript
- Lucas: So you’ve designed this beautiful PCB, you’ve got it assembled by a contract manufacturer, and the first batch comes back. You plug one in, nothing happens. No smoke, no LED, just — nothing. You check the power rail and it’s shorted. The question is: how did that short get past quality control? Luna: And the answer is often: they didn’t do an in-circuit test. Or if they did, the fixture wasn’t catching it. Lucas: Exactly. In-circuit testing, or ICT, is the unsung hero of high-volume hardware. It’s the test that happens before you ever apply power to the board. You literally press the board onto a bed of nails — a custom fixture with spring-loaded pins that contact every net on the board — and the tester measures resistance, capacitance, diode drops, all at once, in seconds. Luna: It’s like a full-body scan for the PCB. It can find a cold solder joint, a missing resistor, a bridge between two pins, even a capacitor that’s the wrong value — before you risk blowing something up. Lucas: Right. And the key word there is ‘before.’ Most hardware startups I talk to jump straight to functional test — plug it in, run some firmware, see if it works. But functional test can miss a lot. If a 10 microfarad decoupling cap is actually a 1 microfarad because the reel got swapped, the board might still run — just with more noise. That’s the kind of thing that shows up in the field as intermittent failures, and those are a nightmare to debug. Luna: So ICT catches the latent defects that functional test might not see until the product is already in a customer’s hands. But there’s a catch: the fixture itself can cost anywhere from a few thousand dollars up to twenty, thirty thousand, and every board revision means a new fixture. Lucas: That’s the trade-off. ICT is fantastic for mature designs with high volumes — think tens of thousands of units a year. For a startup that’s still iterating, the upfront cost can feel prohibitive. But there’s a middle ground: flying probe testers. Those use two or four moving probes that touch each net sequentially, no custom fixture needed. Slower — a flying probe might take a minute per board versus ICT’s ten seconds — but the setup cost is near zero. Luna: And that makes flying probe the go-to for prototypes and low-volume runs. But once you hit production volumes, the per-board test time adds up. I’ve seen startups do the math: at fifty thousand boards a year, ICT pays for itself in labor savings alone. Lucas: Let’s talk about a real example. I spoke with a medical device startup that was building a wearable glucose monitor. Their first functional test pass rate was around seventy percent — meaning three out of ten boards either didn’t power on or failed some firmware check. They’d debug each failure by hand, sometimes taking an hour per board. After they added an ICT step, their pass rate jumped to ninety-five percent. The remaining five percent were mostly functional issues that ICT can’t catch — firmware bugs, calibration offsets — so those still needed a functional test. But the bottleneck disappeared. Luna: And the ICT caught things like a missing 0402 resistor that was barely visible to the naked eye, or a solder bridge under a QFN package that optical inspection missed. Those are exactly the defects that cause field failures. Lucas: Yeah. And the thing is, ICT isn’t just about catching defects — it’s about data. Every test result gets logged. You can track which assembly line or which shift is producing more shorts, or which component vendor has a higher failure rate. That’s information you can feed back into your design for manufacturing process. Luna: Before we go further, I want to say — if these kinds of nuts and bolts conversations are useful for what you’re building or running, and you’d like to help keep the show ad-free and independent, a couple of dollars a month is genuinely what keeps these going. Buy me a coffee dot com slash fexingo, if you’ve gotten something out of them. Lucas: Absolutely. It really does make a difference. And we’re grateful for every bit of support that lets us keep diving into these details. Luna: So — back to ICT. One thing that often surprises founders is that ICT isn’t just for high-volume production. Even at a few thousand units, if the board is complex enough, the debugging time saved can justify the fixture cost. Lucas: Exactly. I’ll give you a rule of thumb: if your board has more than a few hundred nets, and you’re planning at least five thousand units, ICT is worth a serious look. The fixture cost is typically between five and fifteen thousand dollars, and the per-board test time is under thirty seconds. Compare that to manual visual inspection or even automated optical inspection — AOI can find some defects, but it can’t measure resistance or capacitance. Luna: And AOI relies on good lighting and camera angles. A tombstoned component — where one end lifts off the pad during reflow — AOI can catch that. But a cracked resistor that still looks okay on the outside? ICT will catch it because the resistance will be off. Lucas: Right. So the ideal test strategy for a hardware startup is usually: flying probe for prototypes and first articles, then transition to ICT once the design is stable and volumes ramp. And always follow with a functional test that exercises the actual firmware and product features. Luna: But there’s a gotcha: ICT fixtures are specific to the board layout. If you change the PCB — even moving a via or adding a test pad — you might need a new fixture. That’s why you don’t want to commit to ICT until the design is frozen. Lucas: And that’s another reason to design for test from the start. You need test points — small pads or vias that the bed of nails pins can contact. A common mistake startups make is routing all the signals and leaving no room for test points. Then your only option is flying probe or functional test, and you lose the speed and coverage of ICT. Luna: Yeah, I’ve seen boards where the engineer had to add a whole separate test header just to get access to certain nets. That adds cost and complexity. Better to plan for it upfront. Lucas: A good rule is to add a test point for every net that isn’t already accessible via a through-hole component or connector. And keep them on one side of the board if possible, so you only need one fixture instead of a clamshell that contacts both sides. Luna: Speaking of fixtures — there’s also the option of using a ‘flying probe with fixture’ hybrid. Some advanced flying probe systems can use a simple fixture that holds the board but doesn’t have individual pins — just locates it precisely. That can speed up the flying probe process without the full ICT fixture cost. Lucas: That’s a nice middle ground. But for really high volume — like consumer electronics, automotive, medical devices that ship millions — ICT is the standard. The big contract manufacturers like Foxconn or Jabil have entire lines of ICT fixtures running 24/7. Luna: And the testers themselves — the hardware from companies like Keysight or Teradyne — can be expensive, but many contract manufacturers already have them. You just pay for the fixture and the programming. Lucas: That brings up another point: the programming. The ICT test program is essentially a list of measurements — for each net, you specify what resistance or capacitance you expect, within some tolerance. Creating that program takes engineering time, but once it’s done, it’s repeatable and fast. Luna: And you can use the same program to do board bring-up in development. When you get the first prototype back from the fab, you can run it through ICT — if you have a fixture early — and it’ll tell you if there are any assembly issues before you even solder the power connector. Lucas: That’s actually a smart workflow: build a low-cost bed of nails fixture early, even if it’s just a prototype one made of machined plastic and pogo pins. Use it to verify the first boards. Then later, when you go to production, you invest in a more robust fixture made of fiberglass or aluminum. Luna: I’ve seen startups do that with a 3D-printed fixture and spring-loaded pins from a company like QA Technology. Cost maybe a few hundred dollars and a weekend of design work. Totally doable. Lucas: Let’s talk about the downsides, because ICT isn’t perfect. It can’t test things like high-speed signal integrity — it’s purely DC measurements. It also can’t verify that firmware loads or that the device communicates over I2C correctly. That’s what functional test is for. Luna: And there’s a physical limit: very fine-pitch components, like 0.4 millimeter pitch BGAs, might not have pads on the outer rows that are accessible to a test pin. You can sometimes use boundary scan — like JTAG — to test those nets indirectly, but that adds complexity. Lucas: Boundary scan is a whole other topic, but yes. The point is: ICT is one tool in the toolbox. The best hardware startups combine it with automated optical inspection, X-ray for hidden solder joints, and functional test. They don’t rely on any single method. Luna: So if you’re a hardware founder listening, what’s the one takeaway? I’d say: start thinking about test strategy early. Decide where you’ll be in two years — volume-wise — and design your board to support the test method you’ll need then. Don’t paint yourself into a corner. Lucas: Yeah. The cost of adding test points is negligible during layout. The cost of redesigning the board later because you can’t test it — that’s huge. And if you’re working with a contract manufacturer, ask them what test options they support. Some CMs specialize in flying probe, others have full ICT capabilities. Knowing that upfront can save months of headaches.