From Prototype to Product: A Hardware Story
First-person hardware case study: breadboard to PCB, component sourcing, enclosure, manufacturing, certifications, and the mistakes that cost time and money.
By Velloreth
The jump from a working breadboard to a product you can ship is wider than it looks. A prototype proves the idea. A product proves the idea can be built ten times, a hundred times, without failing in the hands of a stranger.
I built one hardware device — an environmental monitor logging temperature, humidity, and CO₂ to a local display and a cloud endpoint — and took it from a breadboard to fifty assembled units. Here is what broke, what cost money, and what I would do differently.
The prototype
I started on a breadboard with a microcontroller development board, a CO₂ sensor, a temperature/humidity sensor, and an OLED display. Jumper wires everywhere, powered by USB. It read sensors, displayed values, and posted readings over Wi-Fi. It worked on my desk for three days.
That prototype taught me the sensing was feasible. It taught me nothing about power draw in an enclosure, about what happens when the sensor warms up, or about how the device survives being plugged into a random wall adapter.
Prototypes lie about power. The development board has a regulator that wastes power, LEDs that stay on, and a USB-to-serial chip drawing current even when the device sleeps. My breadboard prototype ran for days on USB; the first battery-powered version died in eight hours.
From breadboard to schematic
I drew the schematic in KiCad: the microcontroller, the two sensors on I²C, the display on SPI, a voltage regulator, and a charging circuit for a Li-ion battery. I kept the design to a two-layer board to keep fab costs low.
The first mistake was the regulator. I picked a part rated for the peak current but not for the heat it would generate inside a sealed enclosure. After twenty minutes the board was warm to the touch. I swapped in a lower-dropout regulator with better thermal performance and added a note to measure case temperature in the next revision.
The second mistake was the CO₂ sensor. It needs a heater cycle and draws 80 mA while heating, once a minute. I sized the battery for the sleep current and forgot the heater. The first prototype drained a full charge in a day. I added a larger battery and changed the heating interval to once every ten minutes, trading some freshness for battery life.
Ordering the first PCB
I sent the gerbers to a low-cost prototype fab and ordered five boards. They arrived in two weeks. Assembly was by hand: solder paste, a hot air rework station, and a magnifying lamp.
The first board did not boot. The microcontroller’s decoupling capacitor was on the wrong pin — I had swapped two pins in the schematic and the layout followed. I cut a trace, ran a bodge wire, and it booted. That board became the test fixture, not the product.
Lesson: order one assembled board before you order fifty. A $20 board finder is cheaper than a $2,000 reel of boards that need rework. I now assemble the first panel by hand even when I plan to use an assembly house, just to verify the design works as built.
Component availability derailed the second revision. A display I had designed around went from “in stock” to “20-week lead time” between the first order and the second. I respinned the board for a pin-compatible alternate, but the footprint differed by half a millimeter and the connector did not match. I learned to check stock across multiple distributors before freezing a design, and to avoid single-source parts.
The enclosure
I designed a two-part ABS enclosure in Fusion 360: top half holds the PCB and display window, bottom half holds the battery and mounting posts. Mounting holes matched the PCB standoffs. A USB-C port cutout on the side.
The first 3D-printed version did not close. The display bezel was 0.3 mm too wide and the PCB sat 1 mm too high. I printed four iterations before it snapped together. 3D printing is forgiving; injection molding is not. I used the printed enclosure for the pilot run of twenty units and only ordered a mold after the design was frozen.
Ventilation was an afterthought. The CO₂ sensor needs airflow to read accurately, and the sealed enclosure trapped heat, shifting readings by 40 ppm. I added four 3 mm vents in the bottom half and a small gap around the display. It is not perfect, but it is measurable now.
Manufacturing
For fifty units I used a local assembly house: they populated the boards from my BOM and gerbers, ran reflow, and did automated optical inspection. I supplied the PCB panels and a bill of materials with Digikey and Mouser part numbers.
The assembly house flagged two parts as out of stock and asked for substitutes. I approved them without checking the datasheets. One substitute had a different tolerance on the voltage divider feeding the ADC, and temperature readings drifted by 2 °C. I replaced them in the next batch and now lock critical components in the BOM with “do not substitute” notes.
Testing was manual: a jig with pogo pins contacting test points, a script checking each sensor and the radio. Every unit got a 24-hour burn-in before shipping. Two units failed — one bad solder joint under the microcontroller (reflow issue), one sensor out of calibration. Both caught by the burn-in.
Certifications
Because the device has Wi-Fi and connects to mains power through a USB adapter, I needed to consider FCC (United States), CE (Europe), and RoHS (substance restrictions). I used a pre-certified Wi-Fi module so the radio certification transferred to my product, which saved months and thousands of dollars. I still did RoHS documentation for the BOM and a basic emissions test with a local lab.
If your design transmits radio or plugs into a wall, do not guess about certifications. Find a lab early, even if you only do a pre-scan. A failed certification test after you have inventory is an expensive lesson.
Costs
| Stage | Approximate cost |
|---|---|
| Prototype (dev boards, sensors) | $120 |
| First PCB panel (5 boards) | $60 |
| First enclosure iterations (filament, prints) | $40 |
| Pilot run (20 assembled units) | $900 |
| Certifications (pre-scan + docs) | $1,100 |
| Full run (50 units) | $1,600 |
The prototype was cheap. The product was not. Most of the money went to assembly, certifications, and the enclosure mold — none of which appeared in the breadboard phase.
What I would do differently
- Measure power on the prototype. I would put a current probe on the breadboard before designing the PCB. The battery sizing error cost one redesign.
- Lock the BOM early. Component availability changes between design and order. I would buy long-lead parts before freezing the design.
- Test the enclosure with the PCB inside before printing. I modeled them separately and discovered interferences late.
- Start certification research at schematic stage. Knowing the radio strategy up front shaped the whole design; I got lucky with the pre-certified module.
The device works. Fifty units are in the field, and the readings match a reference instrument within tolerance. It is not a commercial success, but it is a real product, and the difference between the breadboard and the box is everything I learned in between.