The Route Runs Through Stations

Founders often want to go straight from idea to a finished custom board, on the theory that intermediate steps are wasted money. In practice the shortcut costs more, because a first custom board carries every unproven assumption at once and there is no way to tell which one failed. Each stage exists to retire a specific class of risk cheaply, so the expensive stage only has to answer the questions that are left.

Four stations, roughly. Not every project needs all four, but skipping one should be a decision with a reason behind it, not an oversight.

Station One: An Off-the-Shelf Development Board

Buy a vendor evaluation board for the microcontroller family you are considering, plug breakout modules into it, and prove the single riskiest thing about the product. Not the whole product. The one thing that, if it does not work, ends the project.

If the product is a sensor that has to detect something subtle, prove the detection. If it is a device that must transmit through a wall, prove the link. If it is a motor controller, prove the torque and thermal behavior. Everything else can wait.

Cost is $150 to $600 in hardware and one to four weeks of engineering. The output is a bench setup and a decision. Choose the microcontroller family here rather than later, because switching families after firmware exists costs weeks; the criteria are in how to choose a microcontroller.

What this stage cannot tell you: power consumption, size, cost, or EMC behavior. A dev board draws ten to fifty times what a designed product will, and measuring current on it produces a number so wrong it is worse than no number.

Station Two: A System of Modules

Now build the whole functional system out of separate boards, wired together, sprawling across a bench or screwed to a plywood panel. Every subsystem present, nothing integrated. This is the stage that looks embarrassing in photographs and does the most useful work.

Here you write the real firmware architecture, discover that two peripherals want the same timer, find out that the display refresh interferes with the ADC reading, and learn that the sensor needs to sit eight inches from the motor. You also start measuring power properly, subsystem by subsystem, because that budget determines battery size and therefore product size.

Cost is $800 to $3,000 in modules and cables plus four to eight weeks. If the product is connected, this is also where the cloud and app scaffolding should come up alongside the hardware rather than after it, as described in how to prototype an IoT product.

Station Three: The First Custom Board

Everything moves onto one PCB, sized roughly for the intended enclosure. Expect four layers for most designs, six if there is a fast bus or a radio. Fabrication and assembly of five boards runs $600 to $2,200 with a two to three week turn, on top of $12,000 to $40,000 of schematic capture and layout work for a moderately complex design. Detailed pricing is in PCB prototype cost.

Plan the first board to be debugged, not to be perfect. That means test points on every rail and every bus, series resistors on lines you might need to cut, footprints for parts you are unsure about, a debug header, and generous silkscreen labeling. Nobody has ever regretted adding test points.

Expect blue wires. A first custom board with three to eight cut traces and patch wires is a normal, successful outcome. Bring it up carefully rather than plugging it in and hoping; the sequence is laid out in PCB bring-up. Most of the errors at this stage are the recurring ones catalogued in first PCB design mistakes.

Station Four: The Mature Board

Revision B or C folds in every fix, drops the debug scaffolding that production does not need, and adds what production does: fiducials, panelization, manufacturable footprints, and a firmware test mode. Mechanical integration becomes real, with mounting holes matched to the enclosure, connectors positioned against the housing openings, and the board profile checked against the actual molded part.

This is also when compliance stops being theoretical. Run an EMC pre-scan on this revision, because finding a radiated emissions problem now costs one more spin, and finding it after tooling costs the tooling. Formal testing follows on a later build. The build sequence from here is conventional and worth naming properly, since factories and investors both speak in those terms; see EVT, DVT, and PVT explained.

A Legitimate Shortcut: A Compute Module

If the product needs Linux-class compute, a system-on-module collapses stations three and four considerably. The hard routing and the software base come pre-solved, and your custom board becomes a carrier with connectors and conditioning. The tradeoff is a higher unit cost, which stops mattering below roughly five thousand units a year.

The related shortcut is going the other way: many products that were scoped for an application processor turn out to need only a microcontroller, which removes an entire class of expense. Question the compute requirement before accepting it.

How Many Rounds and How Long

A realistic count for a moderately complex connected product is three board spins to a production-ready design, occasionally two if the design is conservative and the team is experienced, four or five if the requirements keep moving. Each full loop, meaning design change, fabrication, assembly, bring-up, and test, runs six to ten weeks. That is the number to plan around, and it is the single largest driver of the overall prototype timeline.

Two habits shorten it. Batch changes into one revision rather than sending a fix every few days, because each spin carries the same fixed overhead regardless of how many changes it contains. And keep the mechanical design moving in parallel from station two onward, so the board and the enclosure converge instead of discovering each other at the end.

Plan the Stages Before You Spend on Them

Projects House takes electronic products through all four stations: risk-first bench proving, integrated module builds, custom board design and bring-up, and a production-intent revision with compliance built in. Send a description of what your device has to do through our contact form and we will map out the stages and what each one will cost.