STM32 at 128 MHz.

[seriouserg]

Posted on November 02, 2009 at 06:06

STM32 at 128 MHz.

[seriouserg]

Posted on May 17, 2011 at 13:26

Hi!

Today I experimented with fclk of STM32F103ZE...

I use FSMC (mem & LCD320x240), DMA, NVIC, Timers (TIM1 & Systick)...

128 MHz - fly is normal :o

Everything work fine, LCD benchmark shows FPS improvement, MIPS, MMACS, MFLOPS benchmarks too... Chip's temperature is a little more than normal mode. Current consumption is about 30mA higher than normal.

Tomorrow i try ADC and USARTs, but I think, that it will work too.

Somebody can explain this fact? ST Engineers, bravo! But I want to know something about, please ;)

[seriouserg]

Posted on May 17, 2011 at 13:26

Danish, I agree.

I thought that limiting the maximum clock frequency associated with the technology and guaranteed design. I tested the same way Atmel - the results were much worse - at 70 MHz is already a big part of the periphery was not working.

That is, we can say that STM32 design has some potential to increase the clock frequency and would like very much (and it is not excluded) in the future to see the chip, similar to F103, with a frequency of> = 150 MHz. It would be very good.

P.S. SPI, USART, ADC is working :D

[ This message was edited by: seriouserg on 12-10-2009 18:54 ]

[jj_it]

Posted on May 17, 2011 at 13:26

@danish-

Very thoughtful - clever - well detailed post - Bravo!

[danish]

Posted on May 17, 2011 at 13:26

ST never claimed the chip _wouldn't_ work at 128 MHz.

All they claim is that it will work at 72 MHz. Over a wide range of supply-voltages and temperatures. For many years. Or your money back.

Some chips might work beyond that frequency. Others might not. And of those that do today, you might find that tomorrow or next year they don't. Or they might ''almost'' work in that most things work but some instruction-sequences, peripherals or whatever will not give the expected result. Or running them this hard for a couple of years will burn them out.

You would be amazed how good the measured performance of a chip has to be in order that you can guarantee that every chip will work to a lesser performance.

What you see is interesting, but not surprising.

Regards,

Danish

[seriouserg]

Posted on May 17, 2011 at 13:26

Quote:

We don't believe that you should draw any ''real'' conclusions...

And I do not pretend to do so. I just stating a fact. By the way, the MCU of the early lots. It was received at June of last year. (220H0 VC, MLT22820, Rev code Z).

Please consider my findings no more than experiments. I'm not going to disperse and deliver the STM32 serial products with an overclocked MCU, I was curious to establish the fact - if he could earn a higher frequency. I understand that the parameters may degrade over time, that he may fail, and that a single MCU is not an indication of actual results.

Just fact: one STM32 - one result.

[jj_it]

Posted on May 17, 2011 at 13:26

Please keep in mind danish's guidance that your tests may be ''too comforting'' as they involve only 1 mcu, over limited test conditions, from a single lot, single fab, single production run etc, etc...

We don't believe that you should draw any ''real'' conclusions...

[ccowdery9]

Posted on May 17, 2011 at 13:26

http://www.arm.com/products/CPUs/ARM_Cortex-M3.html

So if the STM32 was on a 90nm process, it would run at 191MHz.

The limitation is the bus & peripherals attached to it.

Chris.

[st3]

Posted on May 17, 2011 at 13:26

When you operate any device outside its specified limits, the behaviour is undefined; that means that absolutely anything can happen - it doesn't have to be predictable or repeatable.

[tomas23]

Posted on May 17, 2011 at 13:26

Have got pictures of exploded STM32s? ;)

[andy239955_stm1_st]

Posted on May 17, 2011 at 13:26

FYI:

Due to failing to set the ADC prescaler I was running the ADC at 64MHz instead of below 14MHz. At that rate about 10% (2 of 20 units) had problems with the sequencer occasionally skipping a programmed conversion. An actual power cycle was required to get a device to function properly again.

My first fix still used in incorrect prescale value and at 32MHz 1 of 40 devices still did the skip.

Temperature mattered, none failed at 25'C, the numbers above were at close to 40'C.