Latest / Hardware Startups with Fexingo: Devices, Robotics, and Manufacturing Tech Companies / How Hardware Startups Use Modular Design for Mixed Signal Test
Transcript
- Lucas: So we're a hundred and fifty episodes in, and I think we've done our due diligence on things like functional test fixtures, vibration testing, even conformal coating. But there's one topic that keeps coming up in conversations with founders that we haven't really drilled into, and that's the mixed-signal testing problem. Luna: Mixed-signal, meaning chips or devices that handle both analog and digital signals? Because that sounds like nearly every connected product out there these days. Lucas: Exactly. Think of any wearable health sensor, a smart thermostat, even a motor controller in a robot. They all have a sensor, which is analog, a microcontroller, which is digital, and some interface in between. The challenge is that the analog world is continuous and noisy, while the digital world is discrete and clocked. Testing a device that straddles both is a whole different beast from just checking digital logic or a simple analog output. Luna: And that's exactly the kind of problem that keeps hardware founders up at night, because it can bite you in production. Lucas: Right. And before we go deep on this, I want to mention something that keeps this show going. We deliberately don't run ads on these episodes, and that's a choice we've made because we think you should get straight-up engineering conversations without interruption. If that's valuable to you, and you want to support that, there's a link in the show notes at buy me a coffee dot com slash fexingo. It's a small way to keep this ad-free, and we appreciate it. Luna: Yeah, and it really does make a difference. We're not asking for anything else, just that if you get something out of these, you can toss a coffee our way. Now, back to mixed-signal testing. I'm guessing the core issue is that analog and digital don't play nicely on the same test bench. Lucas: You hit it. Analog signals are sensitive to noise, timing jitter, and grounding, while digital signals are fast and can introduce noise that corrupts the very analog measurements you're trying to take. So you can't just slap a digital multimeter on an output and call it a day. You need a test strategy that isolates the analog measurement from the digital switching activity. Luna: Isolation. That's a good word. So how do you actually do that in practice for a small team that doesn't have a custom test rack budget? Lucas: The key is to think in layers. First, you partition the device under test into functional blocks, so you can test the analog section separately from the digital section. Second, you design the test fixture to allow for separate power domains and grounding strategies. And third, you use timing to separate the measurements in the time domain, like letting the digital settle before you sample the analog. Luna: So it's almost like you're running a small experiment that mirrors the product's actual behavior, but you control the environment to get clean data. Lucas: Precisely. And the best way to do that without breaking the bank is to use modular test instruments, like a PXI system. Instead of buying a dedicated box for every function, you buy one chassis and plug in cards for digitizing, signal generation, switching, and digital I/O. That way, you can reconfigure the test setup as your product evolves. Luna: That sounds like a good strategy for a startup that might pivot or iterate on the design. But doesn't that require a lot of upfront engineering to set up the modular system? Lucas: It does take some investment, but the payoff can be huge. I was talking to a founder of a wearable health sensor startup, and they had exactly this challenge. They were building a device that measures heart rate and skin conductance, which are both analog, but it also has Bluetooth and a microcontroller, which are digital. Their first attempt at a test fixture was a mess, because they were trying to test the whole thing in one go, and the digital noise was swamping the microvolt-level sensor signals. Luna: Microvolt-level, that's tough. So what did they change? Lucas: They moved to a modular PXI approach. They created separate test layers: one for the analog front-end, one for the digital core, and one for the wireless interface. They also designed a custom fixture that allowed them to connect to the board at specific test points, bypassing the parts they weren't testing at that moment. That way, they could run a clean analog test without the digital noise, and then switch to a digital test without worrying about the analog signals drifting. Luna: So they were effectively doing a divide and conquer strategy on the test problem. Lucas: Exactly. And the numbers were impressive. They cut their test development time by about 40 percent, because they could reuse the same test code and hardware for different product variants. Their test cycle time dropped by 30 percent, because they could parallelize the analog and digital tests. And they saved about 25 percent on test equipment costs, because they didn't have to buy a separate box for each function. Luna: Those are concrete numbers. But what about the test coverage? Are you sacrificing coverage when you partition the tests like that? Lucas: That's a fair question. You do lose some system-level coverage, because you're not testing the full interaction between analog and digital at every test step. But you can compensate by adding a few end to end tests, like running a full sensor readout through the microcontroller and out over Bluetooth, and checking the digital output against a known analog input. That gives you confidence that the interfaces work together, even if you don't test every possible combination. Luna: So it's a trade-off between exhaustive coverage and practicality, which is a constant theme in hardware. Lucas: Absolutely. And you have to design for test from the start. If you don't expose the key test points on the circuit board, or if you don't think about how you're going to isolate signals, you're going to have a nightmare later. That's why many engineers are embracing design for test principles, like adding test pads, using multiplexed channels, and even building in self-test routines that can run on the microcontroller itself. Luna: Speaking of the microcontroller, how do you handle the ADC timing? Because that seems like a classic place where analog and digital can interfere. Lucas: Great point. The ADC, or analog to digital converter, is where the two worlds meet. If you sample too early, you might catch a transient from the digital switching. If you sample too late, the signal might have drifted. The trick is to synchronize the ADC sampling with the digital clock, using a trigger signal that tells the ADC when the analog signal is stable. Many microcontrollers have a built-in feature for that, but you have to configure it correctly. Luna: So it's not just about hardware, it's also about firmware and software configuration. Lucas: That's right. And that's why mixed-signal testing is so multidisciplinary. You need to understand the analog characteristics, the digital timing, and the firmware that runs on the device. It's not something you can just hand off to a test engineer who only knows one domain. Luna: What about using golden units? I've heard that term thrown around, but never really understood how they help in this context. Lucas: A golden unit is a device that you've characterized thoroughly and know its exact performance. You use it to calibrate your test setup and to verify that your test is actually measuring what you think it's measuring. For mixed-signal testing, having a golden unit is almost essential, because it gives you a reference point for the analog measurements. If your golden unit reads a certain value, then you know your test setup is working correctly, and you can use that to set pass-fail limits for production units. Luna: That makes sense. So you're not just testing against a spec, you're testing against a known-good sample. Lucas: Exactly. And you have to be careful about maintaining your golden units, because they can drift over time, just like any other electronic device. So you might want to have a few of them, and you calibrate them against each other periodically. Luna: I also imagine that temperature and humidity can play a role in analog measurements. How do you account for that in a production test environment? Lucas: That's a really important point. Analog components are sensitive to temperature, so if your test lab is at 25 degrees Celsius, but your factory floor is at 30 degrees, you might see false failures or false passes. The best practice is to control the temperature, or at least to characterize how the device behaves across the expected temperature range. You might also want to use a thermal chamber for some tests, but that's expensive and slow, so you usually reserve it for a small sample of units, not for every one. Luna: So what about speed? In a production environment, you want to test as fast as possible. How do you balance speed with accuracy in mixed-signal testing? Lucas: That's the eternal struggle. You can't test everything at the highest accuracy and still keep the line moving. So you have to prioritize. For mixed-signal devices, you often do a quick functional test to catch gross failures, then a more detailed characterization test on a sample basis. Or you use parallel testing, where you test multiple devices at once, even if each test is slower. The modular architecture helps here too, because you can add more test cards to increase parallelism without redesigning the whole system. Luna: Parallel testing seems like a no-brainer for high volume. But I guess it only works if you have enough test points and the right switching matrix. Lucas: Right. And that's another advantage of modular systems, because you can scale up switching and measurement channels as needed. The key is to design your test fixture with the right connectors and routing from the start, so you don't have to rework it when you want to add parallel testing. Luna: So, for a startup that's just getting ready for production, what's the one thing you'd tell them to focus on first? Lucas: I'd say start with a test plan, not a test fixture. Write down what you need to measure, what the tolerances are, and what failure modes you're trying to catch. Then think about how you're going to isolate the analog and digital parts. Only then do you start picking instrumentation. If you go out and buy a bunch of test equipment without a plan, you're going to waste money and time. Luna: That's solid advice. And it's nice to know that you can start simple and grow into a more sophisticated setup as you scale. Lucas: Exactly. And that's the beauty of modular test systems. You can start with just a few cards and add more as your product volumes increase. You don't have to buy a huge capex system on day one. Luna: So whether you're building a wearable, a robot, or a smart home device, mixed-signal testing is one of those things that can make or break your production ramp. Lucas: It really is. And the earlier you think about it, the better. We've seen too many startups get burned by not considering test until they're ready to ship, and then they have to do a costly redesign or delay their launch. So my advice is, start the test conversation on day one, not the month before you go to manufacturing. Luna: That's a great note to end on. Next time we'll look at something else, but for now, if you're working on a hardware product, think about your mixed-signal test strategy.