If you do care about cost (you said you didn't, but now that's all you seem to be considering?
) then by far the cheapest is to just place the NRF51 directly on the board. Takes very few additional components and not hard to do if you just follow the reference layout (and you said you don't care about certification).
"As for the certification, I don't care about it that much."
Love that statement. Especially taken in conjunction with another thread here, where someone has a problem finding a rogue 433 MHz transmitter. And in my own neighbourhood, where cheap Chinese babyphones have killed all the garage openers as well as basic BT functionality.
"Nah, don't care about that. It's the neighbours' problem."
Until the Polish equivalent of the FCC or the "Bundesnetzagentur" are at your door with a 50k Euro fine.
Using the reference layout from an SoC vendor is not very different from using a module, if you think about it. Think of the reference layout as a module that you plug into the rest of your PCB layout. It's possible to make a module misbehave if you don't follow the guidelines from the module vendor (wrong groundplanes, wrong antenna, poor power supply). Same is true with the SoC (wrong crystal load capacitors, say). In fact, it would be easy to get higher unintentional emissions from a module if you're not careful. Most modules try to break out as many of the pins from the SoC as possible, and consequently have compromised groundplanes and a lot of short traces dangling off into space, perfect little unintentional antennas if not properly terminated. Consider this module for a particularly egregious example:
https://www.aliexpress.com/store/product/PTR5528-Fingertip-size-nRF51822-Module-Ultra-Low-Power-Bluetooth-4-0-Low-Energy-RF-Module-Free/130096_1705429590.htmlSome IC pads broken out helpfully to completely unterminated traces that go nowhere. Massive interruption in the groundplane making a perfect slot antenna and terrible ground return paths. Laughable PCB trace antenna and placement. Ludicrous crystal placement with decoupling capacitor bridging over one of the crystal traces. You'd really have to try hard to make a layout that bad on your own boards. Of course most modules don't look like that, but they typically try to be as small as possible while breaking out as many pads as possible which will always be a compromise.
The rules from the FCC made sense when they were written, but for most intents and purposes the NRF51 and similar chips (even more so for say NRF52 with built-in balun) are "modules on a chip". If you're using the softdevice for all the radio handling, you will not be able to exceed the Bluetooth limits on channel occupancy. And exceeding transmit power limits would also be pretty hard to do as the chip doesn't have a PA powerful enough for that. That pretty much leaves harmonics, which you could get if you're using the wrong balun components with NRF51 (so best to stick with the ones from the DK).
In contrast, some of those Chinese babyphones and similar equipment are built from the cheapest of hand-soldered discrete transmitters with hand-rolled coils and little regard to harmonic suppression. It's kind of amazing they work at all. Much easier to get that wrong especially when the drive for cost reduction is taken into account.
Making stuff carries risks, and if you try to protect against all of them, you'll never really get anywhere. If you're Apple, it's appropriate to spend a few million $ or so to comprehensively test the iPhone and production processes against every possible eventuality. And it's a good idea because phones catching on fire doesn't make for great marketing.
If you're making 100 units of something, and your overall budget is say $10000, you're going to have to think carefully about what testing is appropriate, or the project can not happen. Some people take the attitude that those projects just shouldn't happen, but I don't personally agree. HP started with $538 in working capital (including the drill press). If they had decided that they couldn't sell their first product without full FCC certification, the company would have most likely never have happened at all. Same for Apple, for that matter. And Dell too.