Maker Pro
Maker Pro

HCT4051 leakage

J

Joerg

Jan 1, 1970
0
Hello Bill,
The manufacturer's specified leakage current specification used to be
set by the limitiations of their measuring gear and the time available
to take the measurement.

If you look at the HCT405x specs you'll see that they specified very
different leakage values for the 4051 versus the 4052 and 4053. They've
got to have a reason for that.

I don't know how leaky modern CMOS has become.

My impression is that CMOS processes for the logic families are a whole
lot better in variations than they were 20 years ago. And I have heard
quite some horror stories about the situation 30 years ago.

Regards, Joerg
 
J

John Larkin

Jan 1, 1970
0
I did measure an old CD4053 about two years ago (sorry did not have 74HC).
It had +/-5V supplies and was arranged for 0V common mode at both switch
ends.
IIRC the leakages were about 1pA at room temperature.

One thing that surpised me was that, while static leakages were that low,
dynamic leakages (100kHz switching) were not negligeable at all.

Quizz: how do you pump fraction of a uA through the resistors with the
switch wired as below?

4053
.------------.
| | ___
| |--|___|- GND
___ | o---/| 50K
GND -|___|--|---o--__ |
50K | o--- | ___
| \|--|___|- GND
'------------' 50K


Clock it?

I don't think I have a charge injection problem. After I configure the
switches to put the two resistors in series, I wait 4 milliseconds
before I digitize Rx (which takes 125 millisec) and then digitize
Rref, another 125. Since the worst node has an impedance near 100
ohms, and the capacitances are small, the RC tau is way under 1 usec.
My calculator won't do e^-4000 but I'm guessing it's pretty small.

Hmmm... connecting the resistors does apply a load, about 6 mA, to my
2.5 volt reference. The same ref drives the ADC, so it's all
ratiometric, *unless* the ref has a transient bounce from the 6 mA
load step. Now I must need one of those super-low-noise multi-fet amps
to analyze my voltage reference bounce!

John
 
C

Chris Jones

Jan 1, 1970
0
John said:
Clock it?

I don't think I have a charge injection problem. After I configure the
switches to put the two resistors in series, I wait 4 milliseconds
before I digitize Rx (which takes 125 millisec) and then digitize
Rref, another 125. Since the worst node has an impedance near 100
ohms, and the capacitances are small, the RC tau is way under 1 usec.
My calculator won't do e^-4000 but I'm guessing it's pretty small.

Hmmm... connecting the resistors does apply a load, about 6 mA, to my
2.5 volt reference. The same ref drives the ADC, so it's all
ratiometric, *unless* the ref has a transient bounce from the 6 mA
load step. Now I must need one of those super-low-noise multi-fet amps
to analyze my voltage reference bounce!

John

If you do an experiment where you greatly slow down the cycling of the MUX,
is it still bad? I thought that Sigma-delta ADCs aren't generally supposed
to work well when muxed, their digital filters contain history about old
input signals.

Chris
 
J

John Larkin

Jan 1, 1970
0
If you do an experiment where you greatly slow down the cycling of the MUX,
is it still bad? I thought that Sigma-delta ADCs aren't generally supposed
to work well when muxed, their digital filters contain history about old
input signals.

Chris


(Hmmm, my last post got lost somewhere in usenet limbo for a week or
two)

This is an AD7793 24-bit delta-sigma. It has a mode where you can
trigger it to digitize. It takes twice as long to deliver data as in
the continuous mode, but has no memory of the previous input.

We're not seeing crosstalk between channels, just nonlinearity. We'll
get a few more boards nest week and, if they all have the same trend,
I'll just stick a polynomial into the firmware to uncurve the data and
never understand it.

John
 
C

Chris Jones

Jan 1, 1970
0
John said:
(Hmmm, my last post got lost somewhere in usenet limbo for a week or
two)

This is an AD7793 24-bit delta-sigma. It has a mode where you can
trigger it to digitize. It takes twice as long to deliver data as in
the continuous mode, but has no memory of the previous input.

We're not seeing crosstalk between channels, just nonlinearity. We'll
get a few more boards nest week and, if they all have the same trend,
I'll just stick a polynomial into the firmware to uncurve the data and
never understand it.

John

Fair enough it sounds like the fact that it is sigma-delta is nothing to do
with the problems you are seeing.

By the way, are you able to measure the voltage of the RTD and precision
resistor with no current flowing also? Perhaps there is a thermal EMF or
something like that. If it is leakage in the ESD diodes, then if the
voltages were similar leading to similar leakage with and without the
current flowing (hard to arrange?) then it might be possible to compensate
for the leakage that way.

Chris
 
J

John Larkin

Jan 1, 1970
0
Hi,

Has anybody measured typical leakage currents for an HCT4051 analog
mux? I'm wondering about both ESD diode leakage (ie, to rails) and
leakage through open switches. I'm running 0 and +5V rails.

Indications are that everybody's 0.1 uA max spec is wildly
pessimistic, but I was wondering if anybody knows more, before I have
to drag my butt into the lab and make actual measurements. It's a lot
easier to sit here and type and eat bon-bons.

We're scanning eight RTDs. A +2.5 volt reference goes through a
precision 270 ohm resistor and gets mux'd to a selected RTD. The
voltage drop across the RTD gets differentially mux'd, too. A 24-bit
delta-sigma ADC digitizes the voltage drop across the 270, then the
voltage across the RTD, and does the math. We're getting errors in the
tens of PPM, tolerable, but we're curious where they're coming from.
The sensitivity analysis math here is a nuisance.


John


Followup: we never found the cause of the nonlinearity in the
resistance measurement. It's not leakage, because it's not very
temperature dependent (our resistance measurement tc is 3 PPM/K!)

So we just fixed it:


MOVE.L RTCAL.W, D7 ; CONTEMPLATE FINAL CAL FACTOR...
MULU.Q D7, D3:D4 ; D3 IS RESISTANCE IN OUR FORMAT!

; OK, WE HAVE THE RESISTANCE IN D3 H:L AS OHMS:FRACTIONAL OHMS.
; BUT THERE'S AN AS-YET UNEXPLAINED ERROR CURVE: HIGHER
; RESISTANCES READ TOO LOW, BY ABOUT 120 PPM AT 1500 OHMS.
; SO LET'S DO AN R^2 KLUGE, WHICH SEEMS TO FIT PRETTY WELL.

; THIS IS A SMALL TWEAK, SO WE'LL JUST NAB THE INTEGER OHMS,
; SQUARE, SCALE, AND ADD THAT IN.

; 120 PPM OF 1500 OHMS IS 0.18 OHMS, WHICH LOOKS LIKE
; 11,796 IN OUR FORMAT. SO IF X IS THE FUDGEFACTOR,

; 1500^2 * X = 11796
;
; X = 0.005243 AS AN UNSIGNED FRACTIONAL,

X = 22517998 ; AS A LONG

MOVE.L D3, D5 ; COPY RESISTANCE
SWAP.W D5 ; MOVE 'OHMS' PART INTO D5.W
MULU.W D5, D5 ; AND SQUARE THAT... LONGWORD NOW.
MOVE.L # X, D7 ; NAME THE FUDGE FACTOR
MULU.Q D7, D4:D5 ; DO A FRACTIONAL MULTIPLY INTO D4
ADD.L D4, D3 ; AND BLEND IN THE CORRECTION.

; NOW CONVERT RESISTANCE TO TEMPERATURE




That's the great thing about being an engineer: you don't have to
understand it, you just have to make it work.

John
 
J

Jim Thompson

Jan 1, 1970
0
On Mon, 12 Dec 2005 20:56:49 -0800, John Larkin

[snip]
That's the great thing about being an engineer: you don't have to
understand it, you just have to make it work.

John

Gaaack! Until the next production run doesn't work ?:)

...Jim Thompson
 
J

John Larkin

Jan 1, 1970
0
On Mon, 12 Dec 2005 20:56:49 -0800, John Larkin

[snip]
That's the great thing about being an engineer: you don't have to
understand it, you just have to make it work.

John

Gaaack! Until the next production run doesn't work ?:)

...Jim Thompson


We've tried a large number of boards (three, actually) and they are
all close. Seems to me that, even if parts change later on, doing this
is better than not doing it. My error budget is 250 PPM, so buying
back 120 is pretty good.

We test every board, so if things change we'll know it.

Spoilsport!

John
 
J

Jim Thompson

Jan 1, 1970
0
On Mon, 12 Dec 2005 20:56:49 -0800, John Larkin

[snip]
That's the great thing about being an engineer: you don't have to
understand it, you just have to make it work.

John

Gaaack! Until the next production run doesn't work ?:)

...Jim Thompson


We've tried a large number of boards (three, actually) and they are
all close. Seems to me that, even if parts change later on, doing this
is better than not doing it. My error budget is 250 PPM, so buying
back 120 is pretty good.

We test every board, so if things change we'll know it.

Spoilsport!

John

Any "patch" that works, but I don't understand why, always scares the
hell out of me. I don't make three at a time, I make thousands :-(

...Jim Thompson
 
J

John Larkin

Jan 1, 1970
0
On Mon, 12 Dec 2005 20:56:49 -0800, John Larkin

[snip]

That's the great thing about being an engineer: you don't have to
understand it, you just have to make it work.

John



Gaaack! Until the next production run doesn't work ?:)

...Jim Thompson


We've tried a large number of boards (three, actually) and they are
all close. Seems to me that, even if parts change later on, doing this
is better than not doing it. My error budget is 250 PPM, so buying
back 120 is pretty good.

We test every board, so if things change we'll know it.

Spoilsport!

John

Any "patch" that works, but I don't understand why, always scares the
hell out of me. I don't make three at a time, I make thousands :-(

...Jim Thompson


I could have put that X factor into the module cal table, which is
poked during production test and saved in eeprom. But a certain lazy
youth, who's writing the cal software, talked me out of it. Hell, I
could have put a whole mess of polynomial terms into e-square, and
tuned the hell out of it.

A gimmick like this can be handy for, say, halving a residual error.
Trying for a factor of 10 improvement would be pushing our luck.

I think some fancy chips like delta-sigma adcs have similar hidden cal
tables. I don't know if anybody goes for higher-order terms. Don't you
do zener-zaps and equivalent things?

John
 
Top