Oct 08, 2026 / Avik Arefin

Custom PCB Prototype or Dev Board? When to Switch

Build the first prototype on a dev board, move to a custom carrier board that reuses pre-certified radio modules when one of five triggers applies, and leave chip-down production boards until volume pays for their own radio certification.

Author

Avik Arefin

Published

October 08, 2026

Custom PCB Prototype or Dev Board? When to Switch

Build your first prototype on a dev board. Move to a custom PCB prototype only when one of five triggers applies: size, battery life, unit cost, wiring reliability, or a demo others handle. When you do switch, the first custom PCB should be a carrier board that holds pre-certified modules. A pre-certified radio module lets your product reuse the module's FCC grant and EU radio test reports, within the module maker's conditions. A radio chip placed directly on your board needs its own certification. The production board, with chips placed directly and the layout reworked for factory assembly, comes after the design is proven.

Custom PCB prototype stages: dev board, carrier board, production board

A dev board (development board) is a ready-made circuit board that puts one chip on a board with USB, a power regulator, and pin headers. You plug in sensors with jumper wires and write firmware on day one. Common examples are Espressif's ESP32 boards, ST's STM32 Nucleo boards, the Raspberry Pi, and NVIDIA's Jetson developer kits.

A module is a small, finished sub-board that you solder onto your own board like one large part. Examples are the ESP32-S3-WROOM-1 and STM32WB5MMG radio modules, Raspberry Pi's Compute Module 5 (CM5), and NVIDIA's Jetson Orin Nano module.

A carrier board is your own printed circuit board (PCB) that holds one or more modules plus your sensors, connectors, and battery circuit. Because a microcontroller module already contains its radio, memory, and antenna, its carrier board mostly routes power and the module's input and output pins.

A production board is the version built for volume. It may place the bare chips directly on the board, which engineers call a chip-down design. It may also be reworked for DFM (design for manufacturing: changes that make a board cheaper and more reliable to build in a factory). For most founders, the first custom PCB is a carrier board, and the production board is a later project.

flowchart TD
    A{Does the core feature work<br/>on a dev board?} -->|No| B[Stay on the dev board]
    A -->|Yes| C{Does a trigger apply?<br/>size, battery, unit cost,<br/>wiring, demo handled by others}
    C -->|No| B
    C -->|Yes| D{What does the board carry?}
    D -->|Microcontroller module<br/>ESP32, STM32WB| E[2-layer carrier board<br/>with pre-certified modules]
    D -->|Linux or AI module<br/>Raspberry Pi CM5, Jetson| F[Off-the-shelf carrier, or a custom<br/>4+ layer carrier: specialist job]
    E --> G{Annual volume in the<br/>hundreds of thousands?}
    G -->|No| H[Production board keeps the modules]
    G -->|Yes| I[Chip-down radio with<br/>its own certification]
    F --> J[Production carrier keeps the module:<br/>NVIDIA sells Jetson Orin only as modules]

When to switch: five triggers for a custom PCB

Stay on the dev board until the core feature works end to end, then switch when one of these triggers applies. NVIDIA describes its developer kits as tools "to develop and test software in a pre-production environment."

Size and form factor

A dev board adds USB connectors, buttons, and pin headers around the chip it carries. A CM5 module measures 55 × 40 mm, less than half the area of the 85 × 56 mm Raspberry Pi 5. If the enclosure your product needs cannot hold the dev board plus its wires, a carrier board is the fix.

Battery life

A dev board keeps drawing power while its main chip sleeps. Espressif does not recommend measuring a module's current on a dev board, "as some circuits still consume power on the board" in deep sleep. The ESP32-S3-WROOM-1 module itself draws 7–8 µA in deep sleep and up to 355 mA while Wi-Fi transmits. On a carrier board, you choose the voltage regulator, the part that turns battery voltage into the chip's 3.3 V. You can pick one with low standby current.

Linux and AI boards draw far more. Jeff Geerling measured a Raspberry Pi 5 drawing 1.2–1.6 W while shut down, until a firmware setting cut it to 0.01 W. At 1.2 W, a 10 Wh battery, roughly a 2,700 mAh phone cell at 3.7 V, empties in about eight hours with the board off. NVIDIA rates the Jetson Orin Nano at 7–25 W, so these products run on wall power or large batteries. For a device that must last weeks on a small battery, these numbers point to a microcontroller module on a carrier board.

Cost per unit

For microcontrollers, a dev board costs more than twice the module it carries. In October 2026, DigiKey listed the ESP32-S3-DevKitC-1-N8R8 at $15.00 and the matching module at $6.32, or $4.05 each on a 1,950-piece reel. For Linux and AI compute, switching saves less. A CM5 starts at $67.50 against $77.50 for a 2GB Raspberry Pi 5, and the CM5 still needs a carrier board. NVIDIA lists the Jetson Orin Nano 8GB module at $399 at 1,000 units, the same price as the developer kit that includes it. Unit cost is a reason to switch for an ESP32 or STM32 device, not for a Jetson device.

Stock is a second cost risk. On October 8, 2026, DigiKey showed that module out of stock until June 2027. The N8 version, without PSRAM (extra working memory), was in stock. Confirm stock of the exact module variant before a board is designed around it.

Wiring reliability

Solderless breadboards, the plug-in boards that hold jumper wires, have three documented limits. They are usually limited to signals below 10 MHz, and their contact resistance grows over time. Their plug-in connections are also easy to disturb by accident. The ESP32-S3 camera interface can run at up to 40 MHz, four times that limit. If you spend more time reseating wires than testing features, move the wiring onto a PCB.

Investor demo

An investor demo of a works-like prototype is a wiring and size test under worse conditions. The prototype travels, gets handled by people who did not build it, and has to work the first time. A carrier board in an enclosure replaces plug-in connections with soldered ones. If the demo happens on your own table or on video, a tidy dev board build can be enough.

If none of the five triggers applies, keep the dev board and spend the budget on the features.

Custom PCB cost: 2-layer fabrication and assembly prices

A 2-layer board has copper on its top and bottom faces only. It is the cheapest board type to make and suits a carrier board for a microcontroller module.

  • JLCPCB (China): boards from $2.00 for five pieces with a 24-hour build time, a price it offers only through its JLCONE desktop app (October 2026). Its Economic assembly service charges an $8.18 setup fee, $1.53 for the stencil, and $0.0016 per solder joint. Each "extended" part, one that needs a machine feeder loaded, adds $3.07 (price page updated September 9, 2026).
  • OSH Park (US): $5 per square inch for a set of three 2-layer boards, made in the United States, delivered in 9–12 calendar days.

For example, take five JLCPCB boards with five extended parts and 200 solder joints each. The fees are $2.00 + $8.18 + $1.53 + $15.35 + $1.60 = $28.66. That total excludes components, X-ray inspection of leadless parts, shipping, and duty. For US buyers, JLCPCB pre-collects a 35% duty on PCB and assembly orders, per its tariff update of March 17, 2026. OSH Park's US-made boards carry no import duty.

The board itself is the smallest cost in a custom PCB. The larger costs are design hours and the wait for each revision. Each revision repeats fabrication, assembly, and shipping, so the number of revisions sets the prototype schedule more than the board price does.

Radio certification: pre-certified modules vs chip-down radios

Any product that sends radio signals needs radio approval before it is marketed, and your board design decides how much of that work you inherit. In the US, the FCC requires intentional radiators to be certified by a Telecommunication Certification Body, with narrow exemptions. An intentional radiator is a device that transmits radio on purpose.

FCC modular approval in the US

The FCC lets a radio module get its own grant, called modular approval. The module maker tests once; your product reuses that grant if you follow its installation rules and carries a "Contains FCC ID" label.

Modular approval covers the radio, not your whole product. The FCC told test labs in 2020 that the host product "needs an equipment authorization as an unintentional radiator." That category covers electronics that give off radio noise as a side effect. Espressif's FCC grant for the ESP32-S3-WROOM-1 requires "separate approval" for different antenna configurations, so a carrier board should keep the module's certified antenna.

The same grant requires separate approval for portable use, which the FCC defines as a transmitter within 20 cm of the user's body. So a device worn or held close to the body needs its own RF exposure approval on top of the module grant. That approval checks how much radio energy the body absorbs. The FCC exempts only transmitters below a power limit that shrinks as the antenna gets closer to the body.

CE marking and radio modules in the EU

Under the EU's Radio Equipment Directive (RED), responsibility sits with the finished product. ETSI guide EG 203 367 states that the maker of the combined product "is responsible to ensure the conformity" of that product. The maker can reuse the module's radio assessment if the module sits in equivalent conditions and follows its installation instructions. ETSI calls for a new radio assessment after a change of antenna or of the module's layout.

The module's reports cover its radio, not the rest of your electronics. The RED also requires the finished product to meet safety and EMC (electromagnetic compatibility: not disturbing, or being disturbed by, other electronics) requirements. ETSI's guide says the product's non-radio part still needs its own EMC assessment. If the radio and the rest of the device operate at the same time, the whole product needs an EMC re-assessment.

Since August 2025, many internet-connected radio products in the EU also have to meet cybersecurity requirements under Delegated Regulation (EU) 2022/30. The Commission has repealed it from December 11, 2027, when the Cyber Resilience Act fully applies. Both duties fall on the finished product, whatever module it uses.

What a chip-down radio costs

A chip-down radio puts the bare radio chip on your board, so your board design becomes the device under radio test. Compliance Testing, a US lab, listed FCC prices in April 2025: $6,500–$10,000 for products using FCC-certified modules and $9,000–$12,000 for Bluetooth or Wi-Fi transmitters. On that list, the module path saves $2,000–$2,500 in US lab fees.

The larger cost is redesign risk. Silicon Labs, which sells both modules and chips, compared a Bluetooth module and chip from its 2019 Series 2 launch in an undated whitepaper. It calls it "very typical" for chip-down designs to fail transmit-power tests, forcing a lower output power or a board change. It estimates six extra months of development for a chip-down design. It assumes a $2.99 module against a $1.11 chip plus $0.55 of extra parts and testing, at 300,000-unit pricing. On those inputs, chip-down pays off at 100,000–200,000 units a year. Counting sales lost to the delay raises that to 500,000–1,300,000 units. Below those volumes, keep the pre-certified module on every board you build.

If your prototype has a radio, send us three sentences through cortextech.dev on what it does and where it will be sold. We'll reply with the trigger that applies first and a plan to test it.

AIoT prototypes: Jetson modules always sit on a carrier board

AIoT (AI on the device) products run machine-learning models on the device instead of in the cloud. Espressif's ESP32-S3-WROOM-1 datasheet (v1.8, March 2026) lists wake-word detection, speech commands, and face detection and recognition among the module's uses. That module costs $6.32 at DigiKey, against $349 for the cheapest Jetson Orin Nano module at 1,000 units. Test whether your AI feature fits a microcontroller before you plan around a Jetson.

For camera-based AI that needs more than a microcontroller, one option is an NVIDIA Jetson module. NVIDIA sells Jetson Orin only as modules, so chip-down is not an option. An NVIDIA forum moderator answered "Only modules are sold" when asked about buying the chip alone in December 2023. The Orin Nano developer kit pairs a module with a reference carrier board, and NVIDIA's FAQ says developer kits "are not for production use." Kits carry a one-year, development-only warranty and no availability guarantee. Production modules carry a three-year warranty and are available for at least five years. So a Jetson product ships a production module on a carrier board, not the developer kit.

A Jetson carrier board is harder to design than a microcontroller carrier. It carries fast signals: PCIe for storage, USB 3, and MIPI CSI camera lanes. NVIDIA's design guide fixes these lines' electrical values within ±15%, which board makers call controlled impedance. It also asks for USB 3.2 lines with ground on the layers above and below, which a 2-layer board cannot provide. JLCPCB offers controlled impedance only on boards with four or more layers. That makes a custom Jetson carrier a 4+ layer board and a specialist designer's job. The same applies to a CM5 carrier that uses its USB or PCIe lanes, which the CM5 datasheet also specifies by impedance.

At prototype stage, an off-the-shelf carrier board avoids that design work. NVIDIA's FAQ says its ecosystem partners "offer many production-ready carrier boards." Before any carrier, prove the model fits the module you plan to buy. NVIDIA states that the Jetson AGX Orin Developer Kit can emulate any Jetson Orin module. The Orin Nano 4GB module costs $349 at 1,000 units, $50 less than the 8GB.

CortexTech designs 2-layer carrier boards that integrate existing modules for prototypes. For a 4+ layer controlled-impedance carrier, such as a custom Jetson carrier, we bring a specialist board designer into the project. For an AIoT prototype, choose the Jetson module on a dev kit first, and decide on the carrier board after the model runs on it.

Battery risk: how CortexTech tests it before a custom PCB

We build works-like prototypes of IoT and AIoT devices for non-technical founders. Our past work includes a non-technical founder's connected home-entry device with video calling, on-device AI, and a companion app.

The method below is for a different kind of product, a battery-powered sensor. It starts with one measurement.

The problem: a dev board hides where the battery goes

Consider a founder whose battery-powered sensor runs on an ESP32-S3 dev board and drains its battery too soon. A custom PCB looks like the fix. From the outside, the founder cannot tell where the current goes. Part of it feeds the dev board's own circuits, which a carrier board removes. Part of it feeds the radio each time the device wakes, which a carrier board does not change. Espressif's own measurement guide makes the same point: on a dev board, other circuits keep drawing power while the module sleeps.

The solution: measure the module alone in the first days of the sprint

When a founder brings us a device like this, the first thing we test, early in a 1–2 week prototype sprint, is the module alone. We flash the founder's firmware onto a bare ESP32-S3-WROOM-1 module, the same part the carrier board would carry. We keep the founder's sensors wired and powered through the profiler, so each wake includes real sensor reads. Then we record full sleep-and-wake cycles with a power profiler. Espressif recommends an ESP-Prog programmer plus a Joulescope or a Nordic Power Profiler Kit II for this. The datasheet gives the floor, 7–8 µA in deep sleep, and the peak, up to 355 mA while Wi-Fi transmits. The recording adds what the datasheet cannot: how long each wake lasts and how often it happens.

Espressif's alternative is to cut the dev board's power path to the module. We use the bare module because it is the part the carrier board will carry. From the measured cycle and the founder's battery capacity, we calculate a ceiling on battery life, not a forecast. On a carrier board, the regulator's standby current and a battery-voltage monitor each lower it. The regulator also stops holding 3.3 V once the battery voltage falls too close to it, a limit TI calls dropout. So a Go decision needs the ceiling well above the target.

The result points the founder to one of four decisions, made before anyone pays for a board:

  • Go: the ceiling clears the target with margin, so a 2-layer carrier board with that module and a low-standby regulator is worth designing.
  • Wait: the existing dev board already lasts long enough for the demo, so the carrier board can wait, and we say so.
  • Change the design: the wake cycle alone misses the target, so a carrier board will not fix it. The firmware has to wake less often or send less, or the product needs a different radio.
  • Change the power source: no change closes the gap, so the product needs a larger battery or wall power.

For any device, find the part that decides go or no-go, and measure it on the cheapest hardware that gives a true number.

Send us your device idea in three sentences through cortextech.dev. We'll reply with its riskiest part and a 1–2 week sprint plan to test it before you pay for a custom board.

Sources

Want to read more?

More engineering deep dives from the Cortex R&D team.

All Articles