TCA9800: I2C buffering

It is possible that the TCA980x's B side does not work with the level translator. And there must not be any pull-ups on the B side.

Anyway, it's possible that the translators/bus switches make an additional buffer unnecessary. What devices are you using?

Hi Clemens,

Both section A and B use the 74CBTLVD3245PW for the bus switch.

The TCA980x must not have pull-ups on the B side, but the bus requires some pull-up when it is disconnected. The TCA980x would not be a good idea.

What is the purpose of the TCA980x on the electrical level? If the two supplies are nearly the same, there should be no problem.

Why are there two bus switches? You should need only one to disconnect the two boards.

The bus switches are designed on COTS boards for the FPGAs so they're already part of the design. 

I really just need to keep the power rails isolated between section A and section B, so my intuition was that I'd need to buffer the signals going between the two circuits as well, which led me on a search for i2c buffers.

I did just discover i2c isolators for the first time, so is that the terminology for the kinds of parts that I should be looking for instead? I'm new to this kind of design so I'm not very familiar with the terminology yet. On my initial search I came across the ISO1541, which seems like it'll achieve what I want.

Hi Tyler, first of all, thanks for the really thorough response. 

I really only looked at the TCA980x family because a tech support specialist recommended it. I was originally asking whether a PCA9306 would provide the buffering that I need, and was told that it wouldn't and that I should look at TCA980x.

I believe Section A and B shouldn't be getting disconnected during operation.

The core goal of what I'm looking for is to make sure that the I2C signals will communicate properly given the power supply isolation requirement. Since power is supposed to be isolated, I believe I also want to make sure that signals going between A and B should also be isolated/buffered (not sure if I have the terminology correct here). I know that technically since the grounds are tied together, they aren't completely isolated, so the goal is to separate them as much as possible other than where the grounds need to tie together.

All this is to say that I don't have any particular attachment to using a TCA980x, so I'm happy to throw out that idea and use something else.

I think that the block diagram I provided is actually incorrect. I'm suspecting that the bus switches aren't translating up to 3.3V; the switches and FPGA are generating 1.8V signals, with the switches apparently just there to enable the I2C signals. I'll need to double-check whether that's the case or not.

Assuming that the signals that I have to work with are 1.8V, an older design we have used the PCA9306 I2C level translator between A's switch and B's switch (this was before the power isolation was required). Since that worked, I was thinking of chaining a PCA9306 to translate from 1.8V to 3.3V, then a digital isolator (I've been looking at the MAX14850 since I have some other digital signals to buffer, and I glanced at the ISO1450, but again, no particular attachment if you have a recommendation for something else), then another PCA9306 to translate from 3.3V to 1.8V. Does this sound like it'll work?

Looking at the TCA9517, it looks like I'd have to use 3 of them (1.8V <-> 3.3V <-> 3.3V <->1.8V) since I can't connect the B-sides together, right? If so, I think it'll come down to which option saves me more space since my board is getting crowded.

Thanks again for your help.

Galvanic isolation does not make sense if the grounds are the same. And it is still not clear what kind of isolation you need. What is the actual problem that you are trying to prevent with isolation?

I²C uses open-drain signals, i.e., current flows only for low-level signals. So it is not necessary to completely buffer the signals. To keep the high-level voltages separated, a passive switch like the PCA9306 would suffice.

Hi Tyler, I talked to some engineers here about how to describe what I'm trying to do. The aim is to control the return currents, and therefore I was told to use isolators to keep the grounds separate all the way up until they get tied together. 

As I mentioned in my last response, I realized that the I2C signals I'm working with are going to be 1.8V, not 3.3V. I looked at the available I2C isolators, and didn't see any that would work for 1.8V signals, but I may have missed it. I'm hoping I'll be able to save space without needing to add a level translator. Are there any I2C isolators that would work, or maybe any bidirectional standard digital isolators that might work at 1.8V?

Should I ask a new question on the Isolation forum instead?

With 2× P82B96, you could use a higer voltage and a plain 2+2 isolator in the middle.

(And I suspect you could then just leave out the isolator.)

Hi Gillian,

Buffering is mainly for re-driving a highly loaded bus in order to drive I2C signals where the parasitic bus capacitance reaches near the 400pF limit. I2C busses with this amount of parasitic capacitance tend to run into rise-time issues especially at faster data rates, and need to be re-driven, thus implementing a buffer/repeater solution. 

Your response here makes it seem like you are desiring some type of electrical isolation (isolating GND A and GND B). 

Our team supports transceiver devices such as I2C switches/muxes/IO expanders/buffers etc. I will move this thread to the isolation team to see if they have an I2C solution for your needs. 

Regards,

Tyler

Hi Koteshwar, agreed on the grounds not being fully isolated. However, we're controlling return currents in section A and B, and want to keep the grounds separated as much as possible. Understood on the buffer vs. isolator distinction, so let's just assume that an isolator is actually what I want, and GNDA and GNDB aren't tied together.

I'm pretty sure the I2C is 1.8V, not 3.3V though, which is why I was wondering if TI has an isolator that will work at 1.8V as all of the ones I've seen are 3.3V or higher, but I wanted to check in case I missed one that would work perfectly. 

Otherwise, will two PCA9306 to translate 1.8V to 3.3V work with the ISO1640?