How to interface 3.3V part to 2.8V part over SPI?

[Michael J. Noone]

Hi - I'm trying to interface a 3.3V ARM (Atmel AT91SAM7X256) to a 2.8V
accelerometer (Kionix KXP-74) over SPI. The ARM is the master. So -
tell me how much I'm off here: I was thinking I could use a simple
2-resistor voltage divider for the clock, master out slave in, and
slave select pins. For the master in slave out I don't think I need to
do anything, as I believe a 2.8V signal will register as high. (though
I haven't verified that yet).

Is this the best way to interface these parts? Any way to make it
smaller? (6 resistors is kinda chunky - maybe two 3-resistor resistor
arrays, if those exist) What sorts of values would I use for the
resistors? Thanks,

-M. Noone

 

[Tim Wescott]

Hi - I'm trying to interface a 3.3V ARM (Atmel AT91SAM7X256) to a 2.8V
accelerometer (Kionix KXP-74) over SPI. The ARM is the master. So -
tell me how much I'm off here: I was thinking I could use a simple
2-resistor voltage divider for the clock, master out slave in, and
slave select pins. For the master in slave out I don't think I need to
do anything, as I believe a 2.8V signal will register as high. (though
I haven't verified that yet).

Is this the best way to interface these parts? Any way to make it
smaller? (6 resistors is kinda chunky - maybe two 3-resistor resistor
arrays, if those exist) What sorts of values would I use for the
resistors? Thanks,

-M. Noone

First double check to see if the accelerometer isn't 3.3V tolerant --
who knows, maybe it is.

Then consider if you can use some of those handy micro-packaged gates; I
think they put a pair of buffers into a SOT-23; check & see if those are
available in voltage tolerant versions.

Finally, if you have to use resistors you can get 1208-sized 4-resistor
arrays. Use a low enough resistance to drive your expected line
capacitance at your expected frequency with plenty of margin (I'd go for
an RC time constant about 1/10th of the clock).

 

[Michael J. Noone]

I already e-mailed my contact at Kionix - apparently it will work, but
with decreased accuracy. What a nuisance.

I'll see if I can find some buffers like that. Any idea of who makes
them?

I don't understand what you're saying about using the capacitance of
the line. That would be a better method than a V divider? Also - I
don't actually know how to figure out line capicatance at this point,
though that is not a big deal as I am just now drawing up the schematic
- layout won't begin for another week or two. But anyways - once I do
figure out what the line capicatance is - I want a RC value that is
equal to 1/10 of the spi clock period (5Mhz - so 200 ns period and thus
20 ns rc value)

Thanks,

-M. Noone

 

[Tim Wescott]

I already e-mailed my contact at Kionix - apparently it will work, but
with decreased accuracy. What a nuisance.

I'll see if I can find some buffers like that. Any idea of who makes
them?

Um, everybody? Look for "tiny logic" from anyone who makes general
purpose silicon. I just don't know about its voltage tolerance --
looking at everyone's web site is your job.

I don't understand what you're saying about using the capacitance of
the line. That would be a better method than a V divider? Also - I
don't actually know how to figure out line capicatance at this point,
though that is not a big deal as I am just now drawing up the schematic
- layout won't begin for another week or two. But anyways - once I do
figure out what the line capicatance is - I want a RC value that is
equal to 1/10 of the spi clock period (5Mhz - so 200 ns period and thus
20 ns rc value)

You don't want to _use_ the capacitance, you have to size your resistors
low enough so the capacitance doesn't screw you up.

 

[Michael J. Noone]

I'm still not following what I'd be doing with these buffers... Is the
idea that I'd be buffering the signal coming from the 2.8v device to
bring it up to 3.3v? I can't imagine doing it the other way as it seems
like a pretty uncommon characteristic to be able to accept inputs that
are nearly 20% greater than the supply voltage.

Also, regarding resistors - are you just saying that I should use a
voltage divider and just be aware of the rc constant? Or is there some
way you're thinking of where I can get away with just one resistor?

-M. Noone

 

[Rich, Under the Affluence]

I'm still not following what I'd be doing with these buffers... Is the
idea that I'd be buffering the signal coming from the 2.8v device to
bring it up to 3.3v? I can't imagine doing it the other way as it seems
like a pretty uncommon characteristic to be able to accept inputs that
are nearly 20% greater than the supply voltage.

Well, no, it's the other way around - ISTR you've already mentioned that
a 2.8V high output should be interpreted as a high by a 3.3V input, so
that's not at issue.

For the output of the 3.3V logic going to the input of the 2.8V logic,
one, if the data sheet doesn't say that the 2.8 has 3.3 compatible
inputs, then you make a voltage divider.

And I'd say that the starting place would be to look up how much
current this 3.3V device will source when it's high, and use a
resistance that just fits r = e / i. The voltage divider, of
course, is the standard voltage divider formula. I can never
remember the equation, but whatever current you got from the
data sheet, calculate the resistance it takes to drop 2.8V.
Also, take into account the input current of the 2.8V device -
this figures in as an additional resistor, to Vcc or to 0,
depending on which way the leakage current is flowing. If
it's CMOS, then it'll be negligible.

So, say, if your 3.3V device can source, say, 0.1 mA, then
the resistor would be, uh, lessee ... R = E/I? I forget
simple shit like that, but my SSN is etched somewhere in
the Akashic Record... Anyway, 3.3/0.0001 = oh, Dog, will
somebody please do the decimal points for me?

OK, say 33K or 330K or 3.3K or whatever - then, with that
same 0.0001A, what's the value of the resistor that would
drop 2.8V? 28K, of course, and the next closest 1% value
is ... heh! Imagine my surprise! 28K!
http://www.rfcafe.com/references/electrical/resistor_values.htm
So, you subtract that from 33K, and get, um ... three minus two
is one, three minus eight is five, borrow the one... 5K! I
guess you'd use 4.99K in this case.

And don't worry too much about the time constant unless it has
to run at high speeds - somebody else will be much better than
me with advice in that department!

Cheers!
Rich

 

[Tim Wescott]

I'm still not following what I'd be doing with these buffers... Is the
idea that I'd be buffering the signal coming from the 2.8v device to
bring it up to 3.3v? I can't imagine doing it the other way as it seems
like a pretty uncommon characteristic to be able to accept inputs that
are nearly 20% greater than the supply voltage.

In these days of multiple logic levels it _is_ fairly common to have
parts that accept input that are greater than the supply -- 5V tolerant
3.3V parts accept input at 5V, 3.3V tolerant 2.8V parts accept input at
3.3V, etc.

Now, I don't _know_ that there are tiny logic parts that will translate
from 3.3V to 2.8, but it would be a damned good market to be in, so I
_suspect_ that there would be a good selection. It's certainly worth a
look.

Also, regarding resistors - are you just saying that I should use a
voltage divider and just be aware of the rc constant?
Yes

Or is there some
way you're thinking of where I can get away with just one resistor?

No

 

[PeteS]

You could go across the higher to lower voltage boundary using open
drain buffers. The low voltage to high voltage (2.8V to 3.3V) should
require no translation at all (2.8V is a valid high for all standard
3.3V logic families).

The 74xx07 is an open drain non-inverting buffer. Power it from the
3.3V and put a pullup to the 2.8V rail. I use precisely this technique
in an existing design where I have to keep the speed up (otherwise I'd
use resistor dividers).

This part is available from Philips and TI in single gate packages.

Datasheet, hex unit :
http://focus.ti.com/docs/prod/folders/print/sn74lv07a.html

Datasheet for all variants at Philips:
http://www.standardics.philips.com/products/buffers/07/

The device comes in 1 gate, 2 gate, 3 gate and hex gate versions.

Cheers

PeteS

 

[Michael J. Noone]

Hmm - seems like a very reasonable way to do it. But couldn't I instead
use a triple buffer with voltage tolerant inputs, like this:
http://www.fairchildsemi.com/ds/NC/NC7NZ34.pdf and give it a vcc of
2.8V? To me that would make more sense - as that part is very small but
still has leads (this is for a one-off board and I'm soldering it, and
one leadless part (the accelerometer in a DFN package) is enough for
me!) That seems to me the smallest and simplest method - as I'm trying
to save as much space as possible on this board. Thanks,

M. Noone

 

[Uwe Bonnes]

Hi - I'm trying to interface a 3.3V ARM (Atmel AT91SAM7X256) to a 2.8V
accelerometer (Kionix KXP-74) over SPI. The ARM is the master. So -
tell me how much I'm off here: I was thinking I could use a simple
2-resistor voltage divider for the clock, master out slave in, and
slave select pins. For the master in slave out I don't think I need to
do anything, as I believe a 2.8V signal will register as high. (though
I haven't verified that yet).

Is this the best way to interface these parts? Any way to make it
smaller? (6 resistors is kinda chunky - maybe two 3-resistor resistor
arrays, if those exist) What sorts of values would I use for the
resistors? Thanks,

What about Voltage Level translators like the Analog Devices ADG324X series?

 

[PeteS]

Certainly using parts with high input voltage tolerance is a reasonable
way to do this - indeed, that's one of the reasons such parts were made
originally (when the industry started moving en masse to 3.3V from 5V).

Cheers

PeteS