Excessive power usage when using i2c/twi

RE: Excessive power usage when using i2c/twi

Simonr — Mon, 27 Sep 2021 12:03:05 GMT

According to our developers, this issue should also be fixed in NCS v1.6.0, where a new config option is available for the nRF GPIO driver in Zephyr. It allows avoiding the use of GPIOTE for generation of the edge triggered interrupts, please see GPIO_NRF_INT_EDGE_USING_SENSE for details. Using this option should help keep the current consumption low in your case.

Best regards,

Simon

RE: Excessive power usage when using i2c/twi

Simonr — Tue, 21 Sep 2021 13:21:21 GMT

It took a while and some discussion with my colleagues, but it seems like we've gotten to the bottom of it. There seems to be some strange bug wherethe I2C driver doesn't handle the errata 89 workaround after your accReadRegister() function when the GPIO edge trigger is set. When reproducing this, we saw that commenting out the accReadRegister() function from the main loop also put the current consumption back to normal, and implementing the errata 89 workaround after the accReadRegister() function put the device back down to ~20uA. I have reported this bug internally, so the developers will take a look at it and fix this for a future release of the NCS.

So the workaround here will be to, immediately after the accReadRegister() function in your application, add a short "wait" before implementing the errata 89 workaround manually.

Best regards,

Simon

RE: Excessive power usage when using i2c/twi

Poorp — Fri, 17 Sep 2021 12:46:02 GMT

Thanks for taking the time to look into this!

I realized that I can trigger the same problem with the nrf52dk (PCA10040, v1.2.4 2019.19) so I created a minimal project to that replicates it. Hopefully it'll make it easier for you to figure out.

I've attached it below, and included pre-compiled hex files. Incase it only triggers with certain versions of the nrf connect sdk.
I compiled it with "cmake -DBOARD=nrf52dk_nrf52832 -B./build_output && make -j9 -C ./build_output"

The below graph shows the problem on as measured with the nrf52dk
Green is with is using i2c (with nothing physicall connected to dk) and an edge trigger
Red is with using i2c (again nothing connected) and a level_low trigger

Yellow is with the edge trigger and no i2c

devzone.nordicsemi.com/.../bug.zip

RE: Excessive power usage when using i2c/twi

Simonr — Fri, 17 Sep 2021 06:11:37 GMT

Thanks for specifying.

I'll need some more time to look into what might be causing this, as I don't see why the combination of TWI and EDGE_FALLING interrupt would result in a higher current consumption than using just one of the two.

I'll get back to you on Monday when I've done some more digging on my end.

Best regards,

Simon

RE: Excessive power usage when using i2c/twi

Poorp — Wed, 15 Sep 2021 12:12:19 GMT

Hi

What I meant is that the excess current consumption only happens if we set the interrupt to GPIO_INT_EDGE_FALLING _and_ read to or write to the i2c bus.

Using any of the following uses a normal amount of current:
* only I2c
*only an interrupt on GPIO_INT_EDGE_FALLING

* i2c and interrupt on GPIO_INT_LEVEL_LOW

I can't remember off the top of my head what the normal power consumption is, and it depends on if we run the i2c bus or not. But I believe around 20 microamps or so. Please let me know if exact numbers would help and I'll get them for you.

Could using the i2c/twi cause the device to switch what event type it uses from a Port to IN_EVENT for GPIO_INT_EDGE_FALLING?

RE: Excessive power usage when using i2c/twi

Simonr — Wed, 15 Sep 2021 11:51:48 GMT

By "the problem" I assume you mean the excess current consumption you're seeing only occurs when using the GPIO_INT_EDGE_FALLING instead of I.E. GPIO_INT_LEVEL_LOW, correct? What current does the device draw when the GPIO trigger is set to LEVEL_LOW instead for instance?

I think this is due to the GPIOTE is using an IN event rather than a PORT event. An IN_EVENT will require the device to use the HF clock, explaining the increased current consumption.

Best regards,

Simon

RE: Excessive power usage when using i2c/twi

Poorp — Mon, 13 Sep 2021 15:01:59 GMT

Edit: I was mistaken with my previous information, updated the post with the correct information

We've managed to narrow down the problem further. But I'm not sure why we're seeing the behavior we're seeing,

The problem only occurs if we configure i2c, write or read from it, AND setup a pin interrupt on one of the gpioe to trigger on an edge.

That is if we set the trigger to GPIO_INT_LEVEL_LOW, instead of e.g GPIO_INT_EDGE_FALLING, the problem does not occur.
I don't believe that it is caused by interference triggering the interrupt, because I added a print statement to the example code interrupt (see "triggered" below) and that doesn't get triggered except e.g when the button is pushed.

What particular pin the interrupt is configured for doesn't seem to matter, we've tried pin 25, 29 and 6 all with the same effect. Pin 25 is connected to a button, and 29 and 6 are not connected on our board.

static void trigger_reset_polling(const struct device *dev, struct gpio_callback *cb,
		    uint32_t pins)
{
	LOG_INF("triggered");
}


void reset_init(void)
{
	reset_button_dev = device_get_binding(RESET_LABEL);
	__ASSERT(reset_button_dev != NULL, "error device not found");

	int const conf_ret = gpio_pin_configure(reset_button_dev, RESET_PIN, GPIO_INPUT | RESET_FLAGS);
	__ASSERT(conf_ret == 0, "Failed to configure pin, error: %d", conf_ret);

	int const inter_ret = gpio_pin_interrupt_configure(reset_button_dev,
					   RESET_PIN,
					   GPIO_INT_EDGE_FALLING);
	__ASSERT(inter_ret == 0, "Failed to configure interrupt");

	gpio_init_callback(&reset_cb_data, trigger_reset_polling, BIT(RESET_PIN));
	gpio_add_callback(reset_button_dev, &reset_cb_data);
	LOG_DBG("Set up button at %s pin %d\n", RESET_LABEL, RESET_PIN);

	k_work_init(&detect_reset_work, detect_reset);
}

RE: Excessive power usage when using i2c/twi

Poorp — Mon, 13 Sep 2021 10:34:13 GMT

Hi

Thanks for taking the time to look at this.

The excess current consumption is about ~270microamps

As you asked I have now checked and the above errata code is executed and returns true. It had/has no effect on the excess current.

My understanding is that the i2c bus will use about ~200microamps when active. Could it be that it never goes in-active after having transmitted the data? If so what is the correct way to tell it to do so?

Best Regards
Poorp

RE: Excessive power usage when using i2c/twi

Simonr — Thu, 09 Sep 2021 11:35:33 GMT

Can you specify just how large the excess current consumption you see is? If it's ~400uA it does indeed seem like the erratum 89 is what's causing this, but it should be implemented for nRF52832 projects in the nRFConnect SDK from C:\...\NCS_FOLDER\vx.x.x\modules\hal\nordic\nrfx\mdk\nrf52_erratas.h. If it seems like this is it, please try implementing the fix as it is done in nrf52_erratas.h instead of how it's described in the Infocenter, as that's the workaround how it should be used in the nRF5 SDK.

Best regards,

Simon