• TI thinks resolved

TCA9406: TCA9406

Part Number: TCA9406

Hi Team,

For a new design I using TI TCA9406 I2C Level Shifter – to translate from 1.8V to 3.3V voltage levels.

Up to now I have used TI PCA9306, but since the new design needs to support 800KHz mode, I changed to TCA9406

 

However I facing difficulties with the new part, mainly signal integrity issues.

For the same PCB and same I2C master and slave the ne part does not work – and the old one does (at 100KHz/400KHz).

 

I ask for your advice on how to remedy this issue.

 

Thanks,

Shlomi

 

  • Hello Shlomi,

    How much capacitance is on your bus?  Did you measure it?  What are your pull up resistor values.  I would like to see your schematic to verify connections and component selection.

    Can you take scope shots of both SCL_A, SDA_A, SCL_B, and SCL_B.  Ideally you will need a 4 channel o-scope. 

    Depending on parasitics, this parts can sometime have issues due to one-shot accelerators.  NOTE, this is a pass fet architecture and doesn't isolate bus capacitance between the A and B side.

    If you are having problems with this device we have the TCA9617B that is designed to buffer and translate for fast mode plus applications.

    -Francis Houde

  • In reply to fhoude:

    Hi Francis,

    I advanced on solving the issue myself, I did try adding capacitance on the SCL/SDA lines and it seems to improve the signal integrity.

    I am out of office today so I cannot attach scope shots of the signals, but what I saw are spikes on both SCL and SCL when the signal what at "0" – that happened when the other signal what rising from "0" to "1". That is if whenever that was a clock rise – SDA had a significant spike.

     

    This is original schema:

     

    There are 2Kohm pull up resistors to 1.8V on the I2CS_SCL/SDA_S_H_2 lines that are not visible here.

    What I did is to add a 400pF capacitor from both SCL and SDA to GND (close to J36).

     

    Do you have more input?

     Thanks,

    Shlomi

     

  • In reply to Shlomi Yehezkia:

    Hello Shlomi,
    That sounds like too much capacitance. You will be violating the I2C specifications. I will wait and see the waveforms before I give more feedback.
    -Francis Houde
  • In reply to fhoude:

    Hi Francis,

    Please see waveforms attached.

    "I2C_A2" is the with added 400pF capacitors.

    Thanks,

    Shlomi

  • In reply to Shlomi Yehezkia:

    Hello Shlomi,
    It looks like you have cross coupling from SDA to SCL and vice a versa. Is there are harness with twisted pair wiring or is this routed on a board?
    -Francs Houde
  • In reply to fhoude:

    Hi Francs,

    No cross coupling on the PCB - the signal is perfect with TCA9306 wired on the board.

    Thanks,

    Shlomi

  • In reply to Shlomi Yehezkia:

    Hello Shlomi,
    Did you mean PCA9306 or TCA9406?

    If you meant PCA9306 gave you perfect waveforms than that makes sense. It doesn't have the rise time accelerators. The waveforms that you published before either show waveforms that have a lot of noise and what seems to be cross coupling. The noisy waveforms could be an artifact of how you are measuring the waveforms with long ground leads on the scope.

    Have you looked at your local decoupling caps for the buffer? Are they close to the device and have you looked at adding capacitance to the VCCA and VCCB?

    Also, is it possible for me to look at the layout. We generally use Altium here but I have other viewers that I can use for other applications.

    You may also want to order the TCA9802 EVM along with TCA9803 devices and TCA9617B devices. The TCA9802EVM allows you to populate two devices as once for comparison. The TCA9803 would probably be your best bet because it has the strongest pull up strength (it has constant current sources on the B side instead of pull up resistors which save on component count and has better signal integrity. It also has slew rate control for falling edges which is better for signal integrity.)

    -Francis Houde
  • In reply to fhoude:

    Hi Francis,

    I’m considering  using the TCA9406 device for new design  (starting layout) and I’m very warry about the below SI issue.

    I talked with the guy that use the TCA9406 before and It looks that adding capacitors on the lines improved the performance but not  completely  solved the problem. 

    Can you please advise (that suspect that the problem is cross coupling – see below) if you can provide detailed layout guiding and if you can recommend for decoupling capacitor or other thinks  that can prevent the problem because I still in the Schematic/Layout phase of the new design

    Thanks,

    Shlomi

  • In reply to Shlomi Yehezkia:

    Hey Shlomi,

    In terms of PCB design, I recommend you keep the SDA/SCL lines as short as possible and keep them from being too close to each other. As the closer they are to each other, the more likely they will be able to couple onto each other. Keeping the traces short will help keep parasitic capacitance down.

    Any other signals which switch, have low impedance should not cross over the SDA/SCL traces if possible because those signals can also couple to each other.

    You may also want to be careful with the power supplies, if you are using a switching regulator you may see the switching frequency couple onto the SDA/SCL lines. It would be best to use a linear regulator to prevent this. The SDA/SCL lines should not be super close to the power supply traces if possible. The switching frequency of SDA/SCL can couple on to the power supply traces and cause your voltages to ripple into your device. This could also cause signal integrity issues.

    Lastly, keep your decoupling capacitor as close to your VccA and VccB pins as possible to help filter any ripples the power supplies might see. You have also want to add a large tantalum capacitor near the source of your power supplies on your board.

    Thanks,
    -Bobby
  • In reply to Bobby Nguyen:

    Hi Bobby,

    Thanks, I’m already familiar with those requirements. 

    I used 100nF decupling capacitors is (for each of the two power rails) it ok if we take in count the fast switching time?

    Thanks,

    Shlomi