Hardware & Manufacturing

Choosing an electronics manufacturing services (EMS) partner is one of the biggest decisions in a hardware product's life. A low quote per board can hide long lead times, rework and field failures. This guide lists what to check, and what to ask, before you commit. What to evaluate DFM feedback. A good partner reviews your design before quoting and returns specific comments on pad geometry, panelisation, test access and component choices. If a quote arrives without a single q

Most IoT projects do not fail because the idea was wrong. They fail in the gap between a working prototype and a device that can be built, certified, installed and supported by the hundred or the thousand. This article walks through how that gap is crossed in a product design project, and calls out the decisions that are expensive to reverse. The stages of an IoT product design project Requirements and constraints. Start with the environment and the economics: where the devic

Every hardware team with more projects than schematic engineers eventually asks the same question: can an AI tool actually take real work off a PCB designer's plate, or does it just create a new category of work — reviewing what the AI got wrong? The honest answer, after running a handful of these tools against real board work rather than a vendor's demo project, is: it depends entirely on which part of the workflow you point it at. Some parts speed up. Others get worse, not

Almost every IoT product looks successful at the prototype stage. The sensor works.The dashboard loads.The demo impresses investors, internal stakeholders, or the first customer. And then, quietly, somewhere between pilot deployment and real-world scale, the product starts to crack. Battery life collapses.Devices behave differently in the field.Firmware updates become risky.Manufacturing costs balloon.Certification timelines slip.Support tickets multiply. By the time these pr

Traceability Is No Longer Optional — How Data Lineage Turns Devices into Business Evidence Most organisations do not think about traceability when they begin building connected products. The early focus is almost always on functionality: whether the device works, whether data flows to the cloud, and whether dashboards show something meaningful. Traceability usually enters the conversation only when the system is already in use and someone asks a difficult question that cannot

Most connected products do not fail in the lab. They fail somewhere between the moment engineering signs off and the moment manufacturing tries to repeat that success hundreds or thousands of times. This gap—often labelled EVT, DVT, or PVT depending on maturity—is where optimism collides with reality. On the engineering bench, the device works reliably.In small pilot runs, everything appears manageable.On paper, the BOM looks acceptable. And then manufacturing begins. Suddenl

"It's all in the datasheet" is the reassurance every junior firmware engineer hears the first time they struggle with a serial integration, and it's true just often enough to be dangerous. For a real, undocumented, in-market device — especially anything reverse-engineered from an older instrument or a legacy sensor without a full technical reference — the datasheet is frequently a starting point, not a complete specification, and treating it as complete is exactly what turns

The pitch for AI-assisted PCB design tools is built almost entirely around boards of a certain size — enough real estate for the tool's training data to have seen comparable layouts, enough component spacing for its constraint-solving to have real room to work with. Most of the marketing demos reinforce this: a mid-sized development board, a reasonably generous four-layer stack, components with normal spacing. Push the same tools toward a small board — the kind where every sq

A load cell's raw output is a tiny differential voltage off a Wheatstone bridge — typically a few millivolts per volt of excitation — and getting that signal into a microcontroller as a clean, precise digital reading is truly the hardest analog design problem in an otherwise straightforward weighing instrument. Two Texas Instruments parts come up constantly in this exact design: the INA333, a standalone precision instrumentation amplifier, and the ADS1232, a fully integrated

Here's the rule: a PCB fab quote is only as accurate as the manufacturing detail behind it, and every detail left implicit in a quote request gets filled in by the fab's default assumptions — which are frequently more conservative, and more expensive, than what the actual design needs. A quote that comes back higher than expected is very often not the fab overcharging; it's the fab correctly pricing a worst-case interpretation of a spec that was never fully specified. The Rul

Here's the rule: bandwidth and sample rate answer two different questions, and a scope that's generous on one while thin on the other will misrepresent a real signal in a specific, predictable way. Most bench engineers can recite that "more bandwidth is better" and "more samples per second is better," without necessarily being able to say what either number protects against — which makes it easy to overspend on the wrong spec, or underspend on the one that actually matters fo

A connector is the single most common point of field failure in an otherwise well-engineered rugged electronic enclosure — not the PCB inside, not the main housing seal, but the specific interface where a cable meets the box. A mil-spec-class connector, the kind Harwin and similar manufacturers build for defence and harsh-environment applications, costs meaningfully more than a commercial equivalent, and understanding exactly what that cost buys, feature by feature, is what s

A hardware team sources a precision wire EDM part from a locally-qualified vendor, receives a batch that looks visually correct, integrates it into an assembly — and discovers weeks later, once the assembly is under real operating stress, that a meaningful fraction of the batch fails prematurely at exactly the machined feature the EDM process was supposed to produce reliably. The root cause, traced back, is almost never that wire EDM as a process was the wrong choice — it's t

The question arrives early in nearly every precision instrument design project involving a compliant mechanism: "can we 3D print this instead of machining it?" For a monolithic flexure-based weighing cell — the precision mechanism at the heart of an analytical balance, where a single block of metal is machined to form the levers, linkages, and flexure hinges of a force-restoration weighing system — the honest, physics-grounded answer is no, not today, and it's worth understan

A prototype board works. It works on the bench, it works through thermal cycling in a lab freezer-oven, it works after the founder personally demos it to a client. Then a hundred-unit production batch comes back from assembly and eight of them fail within the first week in the field — not catastrophically, not obviously, but with intermittent connectivity dropouts and readings that occasionally don't make sense. Nothing changed in the schematic. Nothing changed in the firmwar

The Number on the Box Is Not a Deployment Specification Every LoRa module datasheet has a range figure. 10 km. Sometimes 15 km. The number is real — under the right conditions, LoRa can achieve that range. Those conditions are: line of sight between antennas, antennas at height, correctly matched 50-ohm antenna systems at both ends, a spreading factor of SF12 (at the cost of very low data rate), and no significant interference on the 865–867 MHz (India) band. Real deployments

The Prototype Is a Sample of One A prototype that works is a necessary condition for product launch. It is not a sufficient one. A prototype is assembled by hand, with careful component selection, on a PCB that was reviewed multiple times, by an engineer who understands the circuit and can compensate for marginal conditions through judgement and adjustment. A production batch of 100 units is assembled by a process, with components from a BOM that was sourced at the time of pr

The Algorithm Has Arrived. The Hardware Has Not. Precision medicine — the discipline of tailoring clinical decisions, treatments, and preventive interventions to the individual characteristics of each patient — has been one of the most heavily funded and intensely discussed areas of healthcare innovation over the past decade. The computational infrastructure is maturing rapidly. Genomic sequencing costs have fallen by orders of magnitude. Biomedical knowledge graphs integrati

The Most Expensive Blind Spot in Embedded Engineering There is a skill gap in embedded engineering that nobody talks about loudly, but everybody who has shipped real hardware products has encountered. It sits at the boundary between firmware and hardware. It is not about writing better drivers, not about mastering RTOS scheduling, and not about optimising power consumption in software. It is about understanding what happens to a signal before it reaches the microcontroller's

Reading Time: 25 mins Topic: Product Management & Execution Bridging the gap between academic projects and industry reality. Visualizing Frustrated engineer looking at a circuit board. Photo via Unsplash. The Hook: The 9-Month Illusion "We'll have prototypes in three months, beta in six, and we'll be shipping volume by Q4." If you've been in the hardware industry for more than a week, you just chuckled. Or maybe you winced. The reality of hardware product development is that