How do electronics engineers pick manufacturer/chipset?

[Tuppe]

Hello,
I'm EE hobbyist/student and I'm again facing this problem; how do I pick the most suitable solution of all the manufacturers?
Choosing a product from a single manufacturer catalog is easier, but choosing similiar solutions between manufacturers is way harder.

For example now I'd need ultra-low power 433Mhz solution, compatible with encryption and long range features. Preferably SoC solution.
I have seen similiar features for rfPIC, Semtech SX1230, Silicon Labs Si4010/Si10XX, TI CC430, Freescale MC12311 etc.
These solutions seem all so equal to me and fill my criteria. I'm pretty sure that I can't ever find one answer to this.

But I'm not looking for an answer to this problem("teach a man to fish and he will eat for a lifetime"), but I want to hear some tips how the professional guys deal with these extremely common situations, as I will encounter these a lot. Do you just randomly go with one? Do you rely on past experiences? Do you email the manufacturers? What is the tipping point to choose X over Y?

For some reason I get stuck on making these choices. Sometimes I have picked the wrong part and wasted weeks worth of work just to find out that it isn't suitable/compatible.
I don't know is this a stupid question, do professional EEs just to flip a coin and be done with it?

Obviously I don't expect to hear 100% full answer to this issue, but I'd like to hear tips and methods how experienced guys pick the best solution fast.

Thank you for reading!

[hamdi.tn]

the way i do it and get lost between parts references some time 
Criteria by order of importance to me

- it do the job... obviously
- Availability of the part
    i usually check distributor stock such farnell and mouser and RS to get an idea about the availability of the component am going to choose
- How well is documented
   if i have in mind some references to choose between, i choose the one better documented and the manufacturer give more data to work with, make my life easier
- Price
   well have to keep an eye on overall price of the design so the cheaper the better, again distributor can give an idea about large order quantity pricing.
- Sampling policy of the manufacturer
 
for all of those ( except price  ), i generally start by choosing anything from TI.

[tggzzz]

This is a very good question. There is no simple answer.

Obviously you see if the part meetsall the technical spec, including environmental, production lifetime, and anything else you can think of.

Then you look for good datasheets and, for the more complex parts, a thriving "community" (i.e. people asking questions about it and the questions being answered). If there is a community then the part is both available and likely to be bought for longer - and therefore to to be manufactured for longer.

Then you try to find if it is actually obtainable at the moment. Some components remain "unobtanium". Some components are made in one big batch (think RF) and never made again. Some people will claim that being listed as available in the large distributors' catalogues is necessary and sufficient.

And then you just keep your eyes open to learn which manufacturers have a reputation for leaving designers in the lurch w.r.t. unbotanium, optimistic specs, suddenly dropping parts - or not.

[Howardlong]

Absolutely agreed with previous posters, and I will add one further thing. Never believe marketing mantra, including expected shipping dates, or pre-release datasheets. Even early "production" silicon can have serious flaws.

I had my fingers burned by Microchip recently when they launched their PIC32MZ series of MCUs. It boasted a 12 bit 28MSPS ADC with 5 or 6 sample and holds, perfect for one of my applications. I worked with the chip for four months (after waiting another three months for silicon) before giving up. The ADC just wouldn't work even closely to the spec. More recently they reduced the spec to 16MSPS, one S&H and 10 bit. Then about a month ago, over a year since its announcement and ten months since parts became available, it was reduced to 8 bit, one S&H and 1MSPS. Clearly they hadn't even tried characterising the part.

Use parts with at least some heritage of you can. It's not always easy, we all want to use the latest and greatest, but it has risk.

[Dave Turner]

Also think about your bill of materials. The more you can source from one manufacturer, all else being equal, the better discount might be possible. Of course the downside is that if the manufacturer has problems you might be screwed.

[theatrus]

I had my fingers burned by Microchip recently when they launched their PIC32MZ series of MCUs. It boasted a 12 bit 28MSPS ADC with 5 or 6 sample and holds, perfect for one of my applications. I worked with the chip for four months (after waiting another three months for silicon) before giving up. The ADC just wouldn't work even closely to the spec. More recently they reduced the spec to 16MSPS, one S&H and 10 bit. Then about a month ago, over a year since its announcement and ten months since parts became available, it was reduced to 8 bit, one S&H and 1MSPS. Clearly they hadn't wven tried characterising the part.

Use parts with at least some heritage of you can. It's not always easy, we all want to use the latest and greatest, but it has risk.

Analog components always seem to be the worst part of any micro controller package - very few work.

Until you have a working production part in your hand, its foolish to put any roadmap in place needing that part.

I've been burned by a lot of things, including the Energy Micro 2.4GHz radio (Draco) that never ended up shipping. TI canceled their entire Luminary Micro LM3 line, even though some (very bad) silicon was shipping as samples (I have a collection of devkits for parts that TI doesn't even list anymore). Atmel had major flaws in the ADC on the XMega line of processors (so much so they were nowhere like spec).

Re-validating parts is time-consuming, and debugging why the part fails validation is annoying. However, its part of any product life cycle.

[SL4P]

If it's a programmable part - also consider support, community, eas of use, and PRICE of the tool-chain needed to work with them !

They range from free, free + add-ons, to very expensive!

[theatrus]

Agreed, supported toolchains are a big deal.

For example, the supported toolchain for TI's radio SoC (BLE) is IAR, for the 8051. Never mind the 8051 is so incredibly old, but the toolchain license remains about a $2k expenditure per seat. (There have been efforts to get it working with SDCC, but I'm not aware of any results).

Contrast that with the newer Nordic parts, which are Cortex-M0 based and GCC is supported, however a SWD JTAG unit can run several hundred $ (Segger).

Microchip has bastardized compilers which require paying $1k per architecture to turn on -Os (small code) support. Even for GCC on the PIC32 (MIPS 4k).

[tggzzz]

I'll second the point about looking very carefully about the toolchain and programming/debug mechanisms in terms of:

• upfront cost
• purchased or rented. No I'm not thinking about licencing/buying, I'm thinking will you be able to return to the project in 2 years time without having to "re-buy" it again
• upgrades/updates, which can be independent of purchasing/renting, e.g. updates for 1 year but they can be used forever
• alternatives for programming hardware, preferably non single-sourced
• operating system, especially the ability to function in a virtual machine (a toolchain for a WinXP machine)

Most importantly for anything that is complex enough to need a toolchain, do I want to invest part of my remaining life learning how to use this product/toolchain. Consider whether the knowledge will be profitable for you in 1/2/5/10 years. Of course, "merely" being fun is also a sufficient reason for doing something!

[jimon]

(A bit of offtopic, sorry)

I've only started with electronics and I have huge background in software engineering. And what I can say - vendor lock-in is total and absolute bullsh*t in pure form 2k$ for compiler per seat ? Software engineering world doesn't have such prices anymore, they all switched to subscription models (you can buy 1k$ products for 20-40$\per month). You can hire a programmer for 5k$ to fork llvm to compile to ANYTHING for free emscripten project for example - llvm-to-javascript compiler, and all optimization algorithms from llvm and C++ compiler are free, you only need to write llvm IR to target platform translator, which is like 2-4 weeks of work if you have asm compiler or hex listing of opcodes.

And guess what ? PIC16 was supported in llvm, but then they dropped it I hope MCU compiler world will change in 5-10 years, and llvm will become defacto standard for everyone. So no more vendor lock-in.

So I have some idea about what is good or bad in MCU worlds, but no strong opinion about other types of IC's. Really curious how to select proper chips from 74 series, or ADC, or OpAmps, or other IC's.

[Tuppe]

Thank you everyone for the informative comments, I learned many new viewpoints to this issue.
My applications are yet pretty simple(basic sensors/inputs) so I don't think that sets any limits.

Most importantly for anything that is complex enough to need a toolchain, do I want to invest part of my remaining life learning how to use this product/toolchain. Consider whether the knowledge will be profitable for you in 1/2/5/10 years. Of course, "merely" being fun is also a sufficient reason for doing something!

I was thinking to get familiar with different developement environments, as I have only done only Arduino IDE based projects with bootloaders, which is laughable in the industry.

I need to look into this toolchain issue more closely(I didn't understand all of the information with my current knowledge). As I beginner I'd need a very well supported setup and free software. I think this will limit my choices a lot.

Such fast and active community here! I didn't expect so vast amount of responses in a day! I need to bookmark this topic and come back few more times to understand all the information...

[tggzzz]

I was thinking to get familiar with different developement environments, as I have only done only Arduino IDE based projects with bootloaders, which is laughable in the industry.

Everyone has to start somewhere. The first computer I built (6800-based, inspired by the Altair 8080) was dire in many ways - but I learned a heck of a lot through my many mistakes. And at the end I knew a lot more than my peers that hadn't tried to build anything - which impressed employers.

Remember, the most difficult part of any journey is the first step.

More concretely, one possible next step would be to program an atmega328 without a bootloader, starting from bare silicon. Perhaps you could duplicate some functionality you have already developed, by writing the configuration setup and device drivers yourself. For that you would need Windows, the Atmel development environment (free), a programmer/debugger (e.g the AVRDragon), a USB cable, and the desire to read the atmega328 data sheet. That setup made progamming and symbolically debugging (i.e. single stepping the processor through source/assembly code) pleasantly straightforward.

[T3sl4co1l]

Good question.  There's no good answer.

The solution is usually to dig through as many choices as possible and find something probably mediocre, and stick with it.

Things that are fungible and easily substituted and searched for, are pretty easy.  Resistors and capacitors, say.  That's pretty well a solved problem, these days.

More complicated things, you are generally stuck with the same old fashioned process that you've always been stuck with.  Call up some distributors and representatives, have a guy come out in person and talk about your application, and have them suggest something.  Get some idea of pricing for your target quantity and schedule.

Then repeat until you're fed up or out of time, pick the best one and go with it.

Often times, engineers will disregard all of this shit, because it is, and just go with what they're already familiar with, even if it's a sub-standard solution.

For chipsets and stuff like that, specifically... you might be lucky if you can even tell that a manufacturer makes the thing you're looking for!  Many of these SoCs and stuff are only available under tight NDA and $10,000 organization membership!

Tim

[Laertes]

Good question.  There's no good answer.

The solution is usually to dig through as many choices as possible and find something probably mediocre, and stick with it.

Things that are fungible and easily substituted and searched for, are pretty easy.  Resistors and capacitors, say.  That's pretty well a solved problem, these days.

More complicated things, you are generally stuck with the same old fashioned process that you've always been stuck with.  Call up some distributors and representatives, have a guy come out in person and talk about your application, and have them suggest something.  Get some idea of pricing for your target quantity and schedule.

Then repeat until you're fed up or out of time, pick the best one and go with it.

Often times, engineers will disregard all of this shit, because it is, and just go with what they're already familiar with, even if it's a sub-standard solution.

For chipsets and stuff like that, specifically... you might be lucky if you can even tell that a manufacturer makes the thing you're looking for!  Many of these SoCs and stuff are only available under tight NDA and $10,000 organization membership!

Tim

Have to agree here - especially because for many many parts, going with something you know well already will mean reduced development cost because you will work faster - which will easily mitigate significant price or feature differences between two parts. Also, paying an engineer to compare parts for a week will only pay back if the difference between two solutions is massive. If you only build small series up to say hundreds or even thousands of boards, it's probably not worth your while to compare too much.

Personally, for almost all of my parts, I go to Digikey and give it my parameters, then choose the first component I find that does what I need, is available and matches my price expectation. When it comes to processors, FPGAs and the like, I look at Digikey and then at my available toolchains. For example, we have a lot of Xilinx programmers and the software licenses around and I am very familiar with the ISE/with Vivado, so I always go with their stuff if I have a choice at all(which I have, until now, always had).
I use Digikey because I find their part selection and their search engine to be the best of all the big distributors like Farnell and Mouser and RS, and because our procurement likes to buy from them. You can however use any distributor you like.