Fujitsu 32-bit MB91F364G flash microcontroller

[vdaniel]

Does anybody body used the Fujitsu 32-bit MB91F364G flash microcontroller?
It is nice and rather cheap part but with very difficult to read documentation and with a very passive support from the manufacturer. I have used it in two projects, but now, starting anothere one I hesitate, maybe it will die before my project will be ready.

Varuzhan

 

[Andrea]

__wait_nop();

In my company we are using it in many projects, even in some under development at the moment. I do not know anything about support, since we rarely had to ask for it. My only concern is about the price. I see very powerful 32 bits microcontrollers (for example LPC ARM by Philips) at a very low price now that can make F364 obsolete soon if Fujitsu does not decide to lower its price. By the way, how much are you paying fo it?

Bye,
Andrea

 

[vdaniel]

__wait_nop

Thank you for reply. At last I know somebody, who use this part. We bought a smal quantity, about 5 parts two yers ago. As I remember, they cost in range from $15 to $20. I know that in large orders they cost $15.
As for the Philips LP2100 series, yes this is very nice chip, but the external interface is not so good. For example, you cannot make design, using all 6 PWM outputs and UARTs.
Besides in Europe the very good design sofware is free.
That is why I am still looking in the Fujitsu direction.

Varuzhan

 

[Andrea]

datasheet mb91f364g

I agree with you. The price here is around 10€ for medium quantities. The micro is very good, and the development software is ok. I have been using Fujitsu 16 and 32 bits since a long time and I have many boards in production, but at the moment I'm a little concerned about Fujitsu policy, mainly because it seems they are not going to lower the price, as the market requires, and they are not supporting all part numbers equally (some p/n promoted in the past are now hard to find or sold at very high prices). That's why I'm starting to use LPC for new designs. I hope that technically it can have the same performance, but thanks to the very low cost I can even accept some compromise...

Bye,
Andrea

 

[vdaniel]

Re: MB91F364G

Ok, thank you very much,

this small discussion convinced me that it is really not a reliable solution
to choose the Fujitsu part for a new design.

Varuzhan

 

[Andrea]

Re: MB91F364G

I hope Fujitsu is not going to kill me for this....

Another bad thing is that it seems that only two people are interested in this p/n in our forum. Is there anybody else that can share his/her experience with Fujitsu microcontrollers?

 

[vdaniel]

Re: MB91F364G

Yes,

And the extremely passiveness of the dedicated forum in www.accemic.com is very
suspicious too.

Varuzhan

 

[KrunCH]

Re: MB91F364G

Hi

I have just started working with a Fujitsu microcontroller, too. Actually it is a MB91F362, but it's the same family. Our faculty received some development boards from Fujitsu (maybe they are getting obsolete and now they are getting rid of the stock :cunning: )

Actually I've got a question for you, I couldn't find the answer anywhere
what is the actual MIPS throughput of this processor?

I've done some experiments measuring the execution time with a timer and I've got some really strange results:
I already concluded that running the program from tha flash is about 4 times slower than running from IRAM. That makes sense as there are some waitstates when reading from the flash. But running the program from IRAM doesn't reaches the 32MIPS at 32MHz.
Furthermore: for the following sequence of code

	for (i=0; i<10000; i++) 
	{
		__wait_nop();
		__wait_nop();
		__wait_nop();
		__wait_nop();
		__wait_nop();
		__wait_nop();
		__wait_nop();
		__wait_nop();
	}

I measure 3000 us. When there is only one NOP inside the loop (with the same loop counter) the elapsed time is 825us and for 4 NOPs is 1750us.
Can you explain this to me?

 

[Andrea]

Re: MB91F364G

Your conclusion is right.

To achieve full speed you need to execute code from IRAM. Flash has wait states so it's slower.

The reason why you cannot measure declared MIPS is probably related to how pipeline works in instruction fetch. So the MIPS value that Fujitsu declares is a theoretical value based on the shortest execution times of the instruction set.

 

[KrunCH]

Re: MB91F364G

Yes, I figured that I cannot achieve theoretical MIPS value. In the loop I have a branch instruction which consumes more than 1 cycle (for clearing the pipeline).
What troubles me that it seems that the number of cycles necessary for that loop is not constant (2 as in the datasheet) but varies for the different loops (the time consumed by the loop does not increase linearly with the number of nops inside)
Maybe there are some penalties accessing the IRAM not mentioned in the datasheet?

 

[Andrea]

Re: MB91F364G

Maybe you are right. I've never been able to compute exactly the execution times too and I concluded that there is something not completely explained and/or understandable in the datasheet. When I need to evaluate the execution time of some routine, I use average estimated times, and this is enough for my applications.

You can try to ask more about this issue in Accemic forum, and see if they can give more information.