Does hardware I2C work on RP2040 and SAMD21?

[charkster]

I have Pico and Xiao boards for RP2040 and a Xiao board for SAMD21. I am trying to get hardware I2C working at 400kHz and 1MHz, but the SCL frequency is much slower than expected (half the requested frequency in some cases). I suspect that software I2C is being used, as I see the same SCL frequencies on all 3 boards (even across different MCUs).

Here is my Pico instance:
(Bus 1 is for pins 2 and 3)
https://learn.adafruit.com/assets/99339

import machine
i2c = machine.I2C(1,sda=machine.Pin(2),scl=machine.Pin(3),freq=400000)
i2c.scan()

Here is my Xiao RP2040 instance:
(using Pico uf2 file and mapping the pins with the schematic, pins 6 and 7 are Bus 1)
https://files.seeedstudio.com/wiki/XIAO-RP2040/res/Seeed-Studio-XIAO-RP2040-v1.3.pdf

import machine
i2c = machine.I2C(1,scl=machine.Pin(7), sda=machine.Pin(6),freq=400000)
i2c.scan()

Here is my Xiao SAMD21 instance:
(pins 8 and 9 are for Bus 0)
https://wiki.seeedstudio.com/XIAO-SAMD21-MicroPython/#software-development

import machine
i2c = machine.I2C(0,sda=machine.Pin(8),scl=machine.Pin(9),freq=400000)
i2c.scan()

On all three instances I measure a SCL frequency of 200kHz when requesting 400kHz. When I request 100kHz I see 82kHz on all three instances. To get close to 400kHz I have to request 800kHz on all three instances. When I use mem32 to read the peripheral hardware registers I see expected values for 400kHz settings, but I always measure 200kHz. I have used both my oscilloscope and logic analyzer and I measure the same frequencies (I don't think this is an instrument issue). I can measure expected SPI frequencies using the same instruments (no issues there).

I suspect that hardware peripheral writes are being done, but at the pin level software I2C is being used. I was under the impression that software I2C would only be used if machine.SoftI2C is called.

Any ideas about what could be happening would be appreciated. I am using MicroPython version 1.22.2 on all three boards. I have 1k ohm pull-ups to 3.3V on all three boards with no slaves attached.
Thanks.

[robert-hh]

I'm pretty sure that both the RP2040 and SAMD21 port use hardware I2C when requested so.
But I can confirm some deviation of the frequencies. Generally that does not matter, since I2C is a synchronous protocol. But one may expect that it is in the order of the requested frequency.

When set to 400kHz, I see 380kHz at the SAMD21 and about 250kHz at the RP2040 for freq=400000. The frequencies may deviate slightly from the requested to get proper duty rates at I2C and may vary with the rise time of SCL. I made my test for SAMD21 with a Adafruit ItsyBitsy M0, but that does not make a difference. The I2C part is identical. The firmware only differs in the Pin mapping and the support of external flash.

For RP2040 one must look into the details, which most likely are part of the RP2040 SDK. But it seems that the granularity of frequencies are limited. When requesting 400kHz I get 250kHz, when requesting 600kHz and up, I get 500kHz, when requesting 200kHz, I get 165kHz. Looks like a frequency of 500k can only be divided by integers. But interestingly, when setting up SoftI2C I get 500kHz when 400kHz are requested.

Note: According to the documentation, the value freq=n at the I2C constructor sets the maximum frequency.

[robert-hh]

I repeated the test SAMD21 test with a XIAO SAM21 and get 408kHz when 400KHz is requested, using a 3k3 pull-up resistor. With a 10k Pull-up resistor I see 390kHz with that board. As said, the effective frequency varies with the actual rise time, which depend on many factors.

[charkster]

Thank you very much @robert-hh! I cycled power on my Xiao SAMD21 and am seeing the expected frequencies now. I can get very close to 400MHz and 1MHz now (with those frequencies requested). I must have created a problem when changing frequencies without complete power cycling the part. I am also seeing the 250kHz when requesting 400kHz on the RP2040 (Pico and Xiao boards).

[charkster]

@robert-hh I am going to close this issue as I will just use my SAMD21 board for precision I2C work. Thank you again for your time.

[robert-hh]

Just of interest. For which application do you need precise I2C clock frequencies?

[charkster]

@robert-hh I am wanting to test an IC that my company is making for a customer. It is specified for both 400kHz and 1MHz operation. My current I2C master (FT4222) has a lot of latency between bus cycles and I wanted to perform many random bus cycles as quickly as possible to see if I ever missed a NACK.