Bidirectional level shifter (TXS0108E) randomly changing outputs

Question

Circuit and video of weird behaviour.

I am using a TXS0108E 8-channel logic level shifter to convert a 3V3 input (VA) to a 5V output (VB).

According to the datasheet, when there are no inputs driving either side of the shifter, the pins are all supposed to be 5V and 3V3.

However, the output is not always 5V on the corresponding side. On some pins (random pins, changes every power cycle) the voltage is ~2.3 V. When these pins are measured using a multimeter, tapping the pin with the multimeter probe repeatedly (measuring the same pin over and over again) changes the output to the correct output after a random number of taps.

Even when pins on the input side (3V3) are driven, the output is not guaranteed to be 5V.

I have wired the Output Enable pin (OE) to 3V3 with a 1k pulldown resistor. The 2 pin supply voltages are 5V (generated by LM7805) and 3V3 (generated by AMS1117), both with 0.1 uf bypass capacitors.

Is this expected behaviour? How to fix this?

Could it be related to the auto direction detection feature? If the chip is getting confused by what side is the output and input, is it a good idea to dedicate a pin to always be driven by 3V3 input and the corresponding pin on the output side be tied to ground via a large resistance?

Answer 1

Well you are not using the chip under the rated operating conditions.

Both sides are completely floating, on a breadboard with high stray capacitance, and getting poked randomly with a multimeter probe for surges.

The chip won't see transitions that are within specs of guaranteed operation.

So having said that, most questions seen here about TSX0108 and it's siblings is that why they don't work correctly, and the answer is almost always that the chip is put into am environment that is out of spec for proper operation. Depending what you want to level shift, these may not be the chips you really want to use.

Answer 2

If you're measuring what should be a logic signal, with a multimeter, and it doesn't seem to be a valid logic value (or reasonably pulled away from idle due to load resistance), it's probably because it's toggling.

So the meter isn't very useful here. You need an oscilloscope.

These chips aren't very useful anyway, outside of their very narrow design application: translating well-defined (driven) logic values from one side to the other. Long wires and high impedance/open pins are common pitfalls.

Most likely, what exactly is happening, is the probe cable's capacitance pulls the pin below threshold, the pin toggles suddenly (it has a negative resistance characteristic, hence the amplification), the wave propagates down the wire, reflects off the meter (some nanoseconds later), which the translator dutifully reads as the opposite state, and so on, the cycle repeats. This is much too high a frequency (10s of MHz) for the meter to read directly, and in DC mode it reads the average anyway.

And sometimes pins start in the low state and the meter does nothing, and sometimes pins start in the high state and the meter triggers this behavior. Still others start high but the meter does not trigger this, perhaps the transient (on touching the pin) simply doesn't cross the threshold for various reasons; conversely, low pins might be disturbed by random static and the process starts. Or due to the capacitance between pins on the breadboard (ballpark 4pF between adjacent strips), the chip is oscillating on its own.

On the upside, you discovered this (poor) behavior before trying to put it into any project! Use directional level translators where possible.

Answer 3

Read the details posted on adafruit website (link below). It suggests that this chip has weak output meaning it won't hold at TTL like its counterpart 4LVC245.

https://www.adafruit.com/product/395

Answer 4

Yes, you are correct that both sides have internal weak pull-up resistors on each I/O pin: 4 kΩ when driven high and 40 kΩ when driven low. This ensures that the I/O pins are driven high when left open.

The issue isn't with the TXS0108E. Rather, the problem lies in the multimeter generating a low voltage, high-frequency signal on the probe in voltmeter mode, which is sufficient to trigger the one-shot accelerator.

The ~2.5V that you measured was the average voltage of the voltage swinging between 0V to 5V at high frequency.