Entry cost is a significant concern for me. With PIC, you can get a decent programmer (the PicKit2) for $35 direct from Microchip, and there’s several cheaper programmers available, including several do-it-yourself designs that can be assembled for under $10 if you have a serial port. When looking into Atmel, I initially couldn’t find anything under $100 that would get me a working programmer and software, though now I’ve found a good, cheap in-circuit debugger for their 8-bit products that only costs $34, so there do seem to be options there as well. However, I have never found a way to figure out how much an AVR MCU will cost without specifically asking Atmel for a quote (unlike Microchip, Atmel’s site doesn’t list prices), and their samples process is much more involved and restricted than Microchip’s.

Another thing that PIC does very well on is providing high-end devices in DIP packages. I recently looked at Atmel when trying to put together a parts list for a custom keyboard controller I’m building, and was disappointed to find that Atmel doesn’t actually sell any MCUs with USB support in a DIP package; anything they have that supports USB is surface-mount, which makes it impossible to use with a standard prototyping board without first attaching it to some sort of adapter. Microchip, on the other hand, sells almost 150 different models of PICs which include USB support and come in a DIP package. There are some third-party solutions available to adapt AVRs with USB support to a DIP format, such as the Teensy and Teensy++ provided by PJRC, but they’re much more expensive… the Teensy++ (which is the only model I found with enough pins to work for this project) is $24 each; the PIC I ended up choosing instead (the PIC18F4550) only costs $4.47 each.

Atmel has interested me for a long time, but given that this is only a hobby for me, I haven’t seen enough reason to spend the time and money required to switch. If I were to get the opportunity to use microcontrollers in a project at work, I would choose PIC without a second thought, since I’m more familiar with it and already own all of the tools required.

Finally thank you for such a thoughtful comment. One of the things I really like about those who comment on this blog is that the comments are well reasoned and devoid of religious fervour. It’s quite refreshing compared to some of the things that get posted on other forums.

More seriously, IMO the main reason PICs are so popular (beside being popular, which is a big factor) is that Microchip has a very good reputation for delivering the chips, and continuing to deliver them. If you make a product now and still want to sell it (without the hassle of re-design, re-testing, re-certification, etc) I think Microchip is the only choice. If OTOH you want a the best bang for the buck right now (and don’t care much what happens in 5 years, or don’t mind sitching to a different chip) Microchip it is definitely the wrong choice.

I use PIC because when i was a student, i got an old programmer. then i had many sample of their circuits, free. it was a criteras for me, having them free cause it’s a little bit expensive for a student without money.

I meet your point, i agree. architecture is awfull, the two interrupts vectors are not sufficient for effective programming, sometimes it’s a nigthmare… small stack, really small.
some bugs in the compiler also..

But the computing power is sufficient for me, and when i need more power, i have my own DSP board.

Maybe i’ll change for ARM one day..

An then you hear:
- “Use PIC. I’ve dealt with them.. in fact i have a programmer in my house”
- “Hey! but I have a programmer in my office.. give me just a minute..”
- “Well, I have half your code written with test cases and everything!”
- “You know what? let me just do that with you and by the end of the day will have it ready!”

and so it goes…

(Anyone happen to know the first microcontroller from any other company with on-chip electrically erasable program memory — EEPROM program memory or flash program memory?)

Since the firmware was on-chip in electrically erasable memory, that made it vastly superior for rapid prototyping and debugging firmware compared to every other CPU of the time.
All the other CPUs around that time period either
(a) had a quartz “erase window” for erasing EPROM, such as earlier some versions of the PIC and Motorola 68HC11. The chips were expensive, and it was annoying waiting for long erase times in the UV eraser every time we wanted to make some small, quick change.
(b) identically the same chip packaged without the quarts erase window, making it one-time programmable (OTP).
Alas, if you made even the tiniest error in the firmware you had to throw away the whole chip and start over.
These chips were relatively inexpensive, so if you could somehow fix all the bugs in only one or two iterations it’s still cheaper than the quartz window chips.
Alas, I don’t know anyone that can fix all the bugs in only one or two iterations.
Many times a piece of firmware would *still* have bugs even after dozens of iterations, so you have dozens of chips in the trash, and you slowly realize that a couple of re-usable quartz window chips would have been better than dozens of OTP chips.
(c) had mask-ROM internal program memory, with huge lead times and NRE costs.
(d) no internal program memory, so it requires external program memory — so lots of pins on the package were eaten up by the memory interface. Therefore, to get the same number of useful general-purpose I/O pins as the 16C84 required a package with more pins, therefore the packaged chip was larger and more expensive. Or else external bus decoder chips, therefore the board as whole was larger and more expensive. Also a bit of a hassle hooking up the address bus and data bus when you just wanted a quick prototype. Also much more difficult to pass FCC testing with a external memory bus than when all the memory is internal.

Since this was such an huge advantage of the 16C84 over all other CPUs, lots of people switched to it.
When they needed to pick a chip for some new project, they *started* with the PIC because rapid development was much faster and cheaper, even if they thought in the back of their mind that once the feature set had “stabilized”, they would switch to some other processor — perhaps mask-ROM.
When it came time to teach some new guy how to program microcontrollers, those people would use the chip that was cheap to reprogram over and over — because we know the new guy is going to make the same sorts of common program errors that all new guys make.
But once development was “finished”, they were too busy with the next project to port it,
so they stuck with they knew was already working.
Or they made an easy switch to some other Microchip PIC processor that used basically the same instruction set and chip burner.

Switching to some other company’s microcontrollers was much more expensive.
That required buying a new chip burner, and either
(a) buying a new C compiler that targeted that chip, which at the time was typically over $1000, which allowed you to start programming right away in the familiar C programming language. Or
(b) learning yet another assembly language, which took a long time to become fluent.

Today there are lots of competing chips — AVR, ARM, and Freescale — with on-board flash program memory.
While they have many small advantages over the Microchip PIC series, I don’t see such an overwhelming advantage that makes it worthwhile to spend thousands of dollars in up-front switching costs.

Because lots of people were using Microchip PICs, we had an reinforcing economy-of-scale feedback loop that caused prices of PICs to fall and prices of other chips to well, not exactly rise, but fall less rapidly.
Causing more people to switch to PIC because of its lower cost, which led to even lower cost PICs.

However, since the cost of C compilers such as SDCC and gcc has bottomed out at “free”, and the cost of chip burners has plummeted, the switching cost is far lower than it used to be. So even small advantages of other chips are enough to make it worthwhile to switch.

I suspect most if not all of Microchip’s current popularity come from a combination of this high switching cost which has only recently disappeared, and the related reasons Doug pointed out — strength of installed base, engineer familiarity and sheer inertia.