Selecting Your Next Project's MCU
Selecting the best chip can be tedious work but the best chip can save you a lot of time and money, and might even be faster! So should you spend time finding the best? I have some words on the topic.
If a primary goal of your next project is to learn a new MCU, you want to create something easily reproducible, or if there will be only one machine building the project, then I recommend you to go with the chip you want to learn, the chip most readily available, or the easiest one to work with.
If you want to create a professional product that might be built thousands of times with reasonable effort to be sold at a reasonable price point, then read on.
Table of Contents
The Best MCU for an EE
I assume you, the reader, are the person responsible for the firmware of your project and that you work with an electronics engineer (EE). If the EE presents you with the finished PCB: Yeah, you don’t have to decide on the chip to use. However, you might run into issues that the EE did not take into account.
For an EE the best chip has no pins at all, is the size of an SMD resistor, has no special needs to power rails, runs at 0 Hz clock, and costs nearly nothing. That would make it easy for the EE and getting it to work is within the firmware’s responsibility.
Often the EE will use the parametric search to find the microcontroller to use. The parametric search is provided by the big distributors (Digikey, Farnell, Mouser to name a few). It allows you to filter all available chips according to their features. The EE can filter out all chips with the wrong voltage requirements, unpractical packages (DIL, SMD, BGA), or improper operating temperature ranges that the product needs to endure. These and many other filters are provided. In the end, the EE sorts by price and selects the cheapest. Maybe they select a handful and let the firmware engineer choose.
Easy right? Not exactly. The cheapest chip that meets the hardware criteria is usually one of the most painful chips to bring up and write firmware on. We’ve all been there.
EE and Firmware Engineers & MCU Features
The requirements of an MCU for a firmware engineer look quite different. They care about significantly different things! Does the MCU have the peripherals needed, does it have good performance, and enough RAM and flash memory. As for the software requirements, firmware engineers are also looking for a solid set of drivers, supported OS’s, and tools. Researching all of this is tedious work.
One way around this is to stick to the processors already in use in other projects. The argument would be that the people involved already know the chip, the tools are already available, and buying more of it might make it even cheaper also for the other projects. If the current standard chip can not be used then you can probably find one that can be used in the same family or at least from the same vendor that fits the requirements. That might not be the cheapest chip, it might not even be on the EE’s list, but we can select it anyway. We convince the manager that it is easier to use this chip as we already have the knowledge and tools and that new tools would come in more expensive than the cheaper chip the EE has found. And this might be the right way to go. It all depends.
If you are true to yourself, you probably have to admit that the effort planned for “learning a new chip” is a combination of not wanting to learn something new and the fear of running into issues with the new chip. There is the risk of something going catastrophically wrong with the new chip, such as a missing required feature or an unforeseen errata. But if the benefits outweigh the risks, we should jump into the adventure.
The differences between chips are bigger than one would expect. For example, the Raspberry Pi RP2040 is a 32bit dual-core processor that runs at a maximum of 130 MHz and costs about one dollar. The ATMEGA649-16MUR is an 8-bit single-core processor running at a maximum of 16 MHz and costs $3.69. Pricing is a dark art of the vendors and many factors influence it. If the vendor provides a widely used framework for an established chip, the prices may go up (or not go down) as people will want to stick with it. Then producing more of a chip that is being used in a high turnaround product (Mobile phone,..) results in less effort than switching the machines to something only a few customers need.
The Best MCU for a Firmware Engineer
What if we turn the tables and come up with the firmware list of chips that we then present to the EE? What should we filter the available chips for?
Peripherals and Software Stack
We should probably start with the available peripherals. If the product needs one or more UART, SPI, I2C, CAN, USB, or Ethernet interfaces, then the chip should better have them. Be cautious not to filter out too many chips at this stage. Especially if you are designing a product that will be mass-produced. A small difference in price can become a huge factor. Therefore, if the required peripherals limit the selection too much, think about bit-banging an interface. If we only need to read a few bytes from an external memory using I2C then doing that with two GPIO pins and some code will probably be OK. Also don’t forget about the “internal” peripherals, DMA, crypto accelerators, (watchdog) timers, etc.
We also need to check the compatibility of the chip. Interfaces like USB, CAN, and Ethernet are usually used in combination with a so-called “stack”. That is a software library that implements all the complicated features of the interfaces. If we need or want to use a certain stack then the peripheral in the selected chip needs to be compatible. Otherwise, we might need to implement a driver or switch to a different stack. Both can be a lot of work. Also, other code that we want to reuse like libraries or (real-time) operation systems need to be checked for compatibility. We need to weigh the price difference of the chip against the amount of work needed to get the incompatible chip to work.
Another thing is tooling. We can not use a chip if we do not have a compiler to generate code for it. We will need some way to flash the chip. (flash = write the compiled code into the chip) A development board or simulator for the chip will speed up firmware development before the project’s PCB is available. A debugger that lets you step through the code and check the values of variables speeds up the finding of issues. If it can also provide trace information then that helps to find the hard-to-find bugs. An IDE with syntax highlighting, that allows for debugging and integrates with the version control system, will help make software development more efficient.
The freely provided vendor tools might often be enough to get the firmware done. But software development is often like driving in the dark. One has the map, so what’s the problem. But switching on the headlights helps with making the turns. Good tools provide light. More light leads to fewer bugs. And therefore it leads to new features faster. Even expensive tools can pay for themselves due to less wasted development hours rather fast. The internet is full of war stories of long bug hunts that would not have been necessary if the right tool would have shone some light on the issue early on. Also, the best tools are usually not chip-specific. Heck. Even the ST-provided ST-link debugger can work on Microchip (and other) devices if used with OpenOCD! So don’t hesitate to ask your manager for the expensive tools even if you could get it done with what you already have. Less development time spent is a good argument.
Community & Popularity
Another plus side for a chip can be if it is part of a bigger chip family. (The base for the “standard chip” argument) Don’t value this too high. But in case of an emergency, it can help. If at the end of firmware development the release date of the product is close but the firmware is just a little bit too big, then switching to the pin-compatible variant with just a bit more flash or a bit more RAM might save the day. It is always good to have a plan B even if you never expect to use it. Don’t plan to fail!
Community support and documentation quality might also be decision-leading factors. Popularity plays a role. If you select a chip that nobody uses but you, if the documentation is brief but incorrect and even the vendor suggests using a different chip, then better listen. Writing firmware for a widely used chip might help if you run into issues as someone else might have had the same issue and already found the solution. It will also help on the compatibility side. The chances that a new library you want to use has support for the chip will be higher. But again just being a popular chip alone doesn’t mean much. But if you need to decide between two otherwise equal chips, the more popular will probably be the better choice. It also has a lower risk of production being stopped. So unless you are in a chip shortage where the most popular chips are sold out the fastest, the more popular chip is the better choice.
If you work in a company that buys a lot of chips then you can and should also ask the distributors and their Field Application Engineers (FAE) to suggest a chip that meets your requirements. Don’t expect them to recommend a competitor’s chip though. It is also a good confirmation if the FAE recommends the same chip you selected.
After all these considerations you now have filtered out the chips that won’t work for this project. The ones left can be sorted by price and then let the EE select the one he likes the most 😉.
If you are now asking where you can find the information to all chips to get your selection process started, then that is the reason why I started ChipSelect. It has just launched and will need a lot of work. It’s open-source and I welcome any improvements.
It intends to help firmware engineers, like myself, find information regarding the plethora of available microcontrollers and make it easier to find, decide on, and get started with a new microcontroller.
See anything you'd like to change? Submit a pull request or open an issue at GitHub