DRV8837: Improper charge-discharge cycle on the load side

Naveen,

Can you please share the top-side markings of a "OK" DRV8837 and that of an "NOK" DRV8837?  I want to verify we are dealing with TI components.  Units were all purchased thru an approved distributor or TI directly?

Regards,

Ryan

Hi Ryan,

Both the "OK" and "NOK" sensor PCBs have the following H-Bridge markings - 837 11I CIV7, 837 17I C72E and 837 17I CS4Q.

The "OK" H-Bridges that I tested had the following marking -  837 268 A812.

The units were all purchased through an approved distributor - ARROW.

Note - Though the H-Bridge from same production lot (If I could call it), whenever I make a swap between OK and NOK sensor, the H-Bridge from NOK sensor creates an issue on the OK sensor as well.

Regards,

Naveen.

Naveen,

OK, so looks like we can rule out issues with non-TI parts.

Can you get some scope captures of the input vs. output waveforms and measure propagation delays on both edges?  

Regards,

Ryan

Hi Ryan,

Here are the input vs output waveforms along with the propagation delays.

The propagation delay varies between 39 ns and 46 ns (for both OK and NOK cases). In addition, the spikes on the generator coil side are seen under both OK and NOK cases. Hence, it is not always guaranteed that an NOK event will accompany a spike in the voltage.

Regards,

Naveen.

Naveen,

I have to admit that I am a little confused on what exactly the issue is when it appears that both DRV8837 devices (OK and NOK) are behaving within the specifications of the device. 

Please help me understand what I am missing...I think I am definitely missing something and not able to understand clearly the behavior of a NOK DRV8837 and a OK DRV8837 in your circuit.

Regards,

Ryan

Hi Ryan,

I would like to refer to the description that I gave at first.

"If one notices the highlighted section, there is a clear deviation on how long the charge is help up, before getting discharged from the coil (load). This "charging area" should ideally be <500ns but it is moving up to 650 ns, leading to noisy behaviour from the sensor side."

I have highlighted (in all the four images) the "charging area" which I will again highlight below (orange area inside red square).

This area directly corresponds to the current flowing to the generator coil, after which, the coil starts discharging slowly, until the switch happens (GEN_IN will go from HIGH to LOW while GEN_OUT will go from LOW to HIGH).

For all the OK sensors, this highlighted area stays constant (500 ns) but for the NOK, this highlighted section has a deviation of more than 30% (>650 ns). This is the main issue here, since this directly influences the sensor side, increasing the overall noise up to factor 5.

Coming to the additional information I provided, this issue with the deviation in the "charging area" did not occur on the newer variants of the H-Bridge (DRV8212P). If I am correct, the newer one has the same Rds but has a slower rise time than the DRV8837. Does this factor influence the issue I am facing here? If so, why the DRV8837 functioned well for the same setup for more than a year but failing at a higher rate in the recent past (6 months).?

Please let me know if additional information is required. Thanks!!

Regards,

Naveen.

Naveen,

Thank you...makes sense now, but not sure what factors are influencing.  

The rise/fall times of DRV8212P are given only as typical values of 150ns.  DRV8837 can be anywhere between min of 30ns and max of 188ns.  So, technically it is possible to have the same rise/fall time within the spec.

Can you check rise time of the NOK DRV8837 vs. OK DRV8837 and see delta.  Maybe also check DRV8212P as that seems to work OK as well.

Regards,

Ryan

Hi Ryan,

I measured the following parameters on the four different variants - 8837, 8837C, 8212P, 8210P.

1. Rise/Fall time of Inputs (PWM1, PWM2)
2. Rise/Fall time of outputs (GEN_IN, GEN_OUT)
3. Propagation delay between Input and Output (PWM1 and GEN_IN, PWM2 and GEN_OUT).

All units are in nanoseconds.

Test Parameters -> VM = 3.3V, VCC = 3.3V, L < 50 µH, RL < 5 ohms (DC resistance).

Regards,

Naveen.

Naveen,

I am not sure what to make of this.  The NOK sample definitely has higher edge rates vs. OK sample, but I am surprised the DRV8837C works OK as those are similar to NOK.  

It appears the edges of the DRV8210/2P are much slower.  

Have you tried hanging like 10nF or some other smaller capacitor off each output and see if that turns a NOK sample into OK?

Regards,

Ryan

Hi Ryan,

With regards to the DRV8837C, I think the Rds has an additional role to play as well. The issue here is not only with the fast rise time but the fast charging of the coil, and I believe, the 8837C does not charge the coil as fast as the 8837. But I am surprised by the results as well.

The 8210/2P on the other hand, have much better EMC performance, with smooth edges. This could be the reason why we never face issue with 8212P, though the Rds is similar to 8837.

Regarding the filter capacitors, the measurement is currently planned and will be performed within the next days.

In addition to the above, I have an interesting update. Measurements were performed on two separate evaluation modules (DRV8837EVM) - one with OK H-Bridge and the other with NOK H-Bridge. Both of them have similar results. The rise time is constantly below 30ns but the fall time is always within specs (~100 ns).

The measurements were done with 90 kHz PWM frequency (IN1 and IN2 have complementary channels of PWM signals as input). The VM was supplied as 5V from an external supply, with the bypass capacitor values mentioned in the table.

Now I am even more surprised, because by this measurement, both the H-Bridges are NOK. Is there a specific 20 ohm load that one should give? I took a standard 20 ohm resistor with 0.5W dissipation. Here is a screenshot of the whole measurement (PWM2 is not recorded but present as IN2, with PWM1 as IN1).

Regards,

Naveen.

Naveen,

As you mentioned, Rdson also plays a role here.  If so, it is possible that Rdson is lower on the NOK sample vs. OK sample.  If timing is similar, that is the only thing I can think of.

Regards,

Ryan

Hi Ryan,

Thanks for the response. But the main question here is, why the H-Bridges are outside specs? According to the datasheet, if a load of 20 ohms is connected to the H-Bridge with VM = 5V and VCC = 3V (I used 3.3V here in my test setup), the rise time should be minimum of 30ns. 

However, I am getting value between 17ns and 22 ns, and not more than that. I checked with multiple configurations (higher/lower capacitance between VM and GND, lower load) but the rise time hardly changes more than 2 ns.

Does this mean the H-Bridge is already out of specs and because of the "tolerance of Rds" we get OK ones? Is there a way to measure the Rds?

We got thousands of H-Bridges (> 100k) directly from TI and the production is on stand-still because of this very issue.

It will be of great help if you can guide me in solving the main issue at play here (rise time and not Rds).

Thanks!

Regards,

Naveen.

Additional Information on tests performed - 

The test with bypass capacitors were performed (1nF and 10nF) and the results are as follows.

1. The overshoot reduced considerably with the introduction of bypass capacitors (>1V without caps, <0.3V with 1nF and <0.15V with 10nF). This is sort of expected, since similar configuration is already used in another project of ours.

2. The rise/fall time was considerably reduced (~100ns for 10nF configuration and <20 ns for the rest).

3. The sensor is still NOK, irrespective of the changes.

In addition, the test on the eval board was repeated for DRV8212P, and the rise time seems to follow the specs.

There is an additional parameter called "Output Dead Time" in the new variants, which is not there in the old.

This makes the switch from OUT1 to OUT2, slower than the previous variants (8837). Here is the one from 8212P (~ 60ns).

Here is the one from 8837 (~ 20ns). 

The range is between 15-25 ns for both OK and NOK H-Bridges. Hence, I think this might not be a "decisive" parameter but nevertheless, can you let me know why this parameter was included in the new H-Bridge?

Naveen,

All of our integrated bridges have dead-time built in to avoid shoot-thru.  It was just an oversight to not include this in the specification.  Anyway, it appears that NOK and OK have a similar dead time.  So, I guess that is also not the culprit here.

Regards,

Ryan

Hi Ryan,

Thanks for your response on the dead-time. Even I described it as not a decisive parameter, and hence, we are on the same page.

However, it will be great to have your response on a more crucial- question that I described earlier? I am quoting the same below.

"Does this mean the H-Bridge is already out of specs and because of the "tolerance of Rds" we get OK ones? Is there a way to measure the Rds?"

The reason for my question is, we wish to have a parameter that could be tested on a H-Bridge, before it is soldered on to the PCB. Until the H-Bridge is soldered, and the sensor goes through the end of line testing, one never know if the H-Bridge is OK or NOK.

And do you agree that the H-Bridges (8837) are out of specs, with respect to rise time?

It will be great if you can provide assistance here on these two questions.

Thanks!

Regards,

Naveen.

Naveen,

Please see this e2e post for advice on measuring Rdson.  You would want to fully enhance high-side FET by applying DC inputs to input.  We specify based on a particular current in the datasheet.  Read the last part of the post below...first part can be ignored.

https://e2e.ti.com/support/power-management-group/power-management/f/power-management-forum/233577/measured-rds-on-is-greater-than-that-specified-in-datasheet

It does appear the device is switching hotter than expected.  

Regards,

Ryan

Hi Naveen,

While "charging area" is getting longer on GEN IN is it getting shorter on GEN OUT at the same time?

If so, I suspect there may be some DC current (average current) flowing through coil. I guess such current may flow if both half-bridges are not identical (rise/fall/delay times, Rds etc.). I would check coil current with current probe if its average value is 0. I would also make some comparison test of good and not good drivers.

Instead of coil I would connect between GEN IN / GEN OUT and GND RC filters (lets say 5 Ohm + 100uF, resistor to GEN X and capacitor to GND). Then I would compare voltages on both caps if they are similar.

Regards,

Grzegorz

Hi Ryan,

Thanks for the reference. I will discuss this internally to see the possibilities of measuring Rds accurately, and have a comparison between OK and NOK parts.

And thanks for the response regarding the rise time. I will discuss internally regarding how we can proceed with the parts we have here (> 60k).

Regards,
Naveen.

Hi Grzegorz,

The change in "charging area" doesn't happen simultaneously i.e., its either on GEN_IN (mostly) or GEN_OUT, but not both. And the change is always making the area to get bigger and not shorter.

And related to your suggestion, I actually performed a test with capacitor values of 1nF and 10nF (Cap between GEN_IN || GND and GEN_OUT || GND). And the results are as follows.

Note - I measured the rise time from 10% - 90% of the signal.

1. The overshoot of the GEN_IN and GEN_OUT reduced considerably with the introduction of bypass capacitors (>1V without caps, <0.3V with 1nF and <0.15V with 10nF). This is sort of expected, since similar configuration is already used in another project of ours.

2. The rise/fall time of GEN_IN and GEN_OUT was considerably reduced (~100ns for 10nF configuration and <20 ns for the rest).

3. The sensor is still NOK, irrespective of the changes i.e., the change in "charging area" still occurs.

I also build a C filter on the input (IN1 and IN2 -> to GND), and the rise time of incoming signal was close to 75 ns, with the output signal rising at ~120 ns. The problem exists nevertheless.

In a separate test, I built a serial resistor between the OUT1 of H-Bridge and GEN_IN (1 ohm/2.2 ohm/4.7 ohm/10 ohm), but it did not result in getting any better results.

Regards,

Naveen.

Hi Naveen,

If coil is close to 50uH and 5 Ohm then its time constant L/R is around 10us and current will not reach stable value which is VM / (Rcoil + Rdson) at 90kHz. Coil will be filtering output current and if both half-bridges are not identical some DC current may flow in the coil what would extend "charging area" on one output and shorten on another one at the same time.

Filter 5 Ohm + 100uF I mentioned was just to compare balancing of half-bridges in OK drivers and NOK ones.

Anyway, if change in "charging area" on one output does not cause opposite change on another output then my theory probably is wrong.

I think that measurement of coil current with AC/DC current probe (BW of 0,5 MHz or more) might help to understand what is causing the problem.

Regards,

Grzegorz

Naveen,

Just another thought, in case of NOK drivers I would try to modify PWM wave from 50/50 to lets say 45/55 and 55/45 and see if it changes anything.

Regards,

Grzegorz

Hi Grzegorz,

A correction to my earlier statement. After going through all the measurements again, I noticed that the "charging area" could change simultaneously for both GEN_IN and GEN_OUT, however, in a similar way only i.e., it might extend on both not the case where one will extend and one will contract. 

However, I should make a remark that this event of a change happening on both the outputs are not often.

I actually performed the coil measurement, but it was inconclusive (on the sensor setup vs evaluation board setup). Hence, I will review the setup once again and perform the measurement.

Regarding the duty cycle change, I could do this to figure out the issue, but the application requires a constant 50% duty cycle. But, I will take this into consideration and update the results once it is available.

Thanks!

Regards,
Naveen.

Hi Ryan,

The Rds measurement was conducted from my side and the results are the following.

The same measurement was repeated for the coil configuration used on our sensor, and the results are as follows.

Though a slight increase in Rds is observed between OK and NOK H-Bridges, it is only in terms of few mOhms. Could this still influence the behaviour that I see on the sensors?

Regards,

Naveen.

Naveen,

And it also appears that the total Rdson on the "NOK" bridge is higher than the "OK".  Am I seeing that correctly?

Regards,

Ryan

Hi Ryan,

Yes. That's true.

Regards,

Naveen.

Naveen,

Then that doesn't match any previous theories as the other drivers with much higher Rdson work fine.  So, it appears we can rule out Rdson as the culprit.

Have you tried any of the other experiments recommended above with altering PWM duty cycle to try and turn a NOK sample into an OK one?

Regards,

Ryan