AD7793 again

How often is that destructive? Several times I've had a PC or peripheral malfunction and only removing power would fix it, even if there was a hard reset button.

Last night I plugged my secondary monitor into my laptop and it had a smeared picture as if the resolution or frequency was invalid. I could change settings on the main screen and lower resolutions worked. Only removing the battery fixed it.

Chip Scale Package, generally a piece of silicon with solder bumps on the bottom and little or no epoxy case.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

Several terahertz.

Windowed eproms had that issue, too.

Some opamps have optical DC offset.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

There are so many things with PCs that are designed poorly that you don't need to invoke an ESD caused latchup to explain it. My Lenovo is terrible in that regard with the battery being reported as missing only getting straight by pulling it out and reinserting it and many other problems.

--

Rick

One school of thought is that nothing should ever go wrong in a digital system, so one shouldn't expect a gadget to survive missing even one clock, once. After all, the program counter of a computer doesn't count wrong, and the system crashes if it ever does.

Inside an FPGA, we often assume that nothing will ever be done wrong, in things like state machines or sinc filters. An ADC chip, with a shared SPI interface, and flakey ESD diodes, and flakey temperature behavior, is another matter. No harm would have been done if de-asserting CS were to reset the SPI machine.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

I remember several decades ago when LCCs were a popular option, a colleague was working with a quite brilliant scientist to explore new packaging methods. They were already using solder bumps which require "large" pads for the solder. The chip makers didn't like the "waste" of silicon area, so they didn't get used in anything but military designs for another decade or two. I guess there was no push for such levels of miniaturization until cell phones came out. Even in DoD stuff the contractors preferred hybrids because they were already geared up for that.

--

Rick

I believe you said there is a data related reset function, right? So clearly they did give this consideration. The only difference is whether you reset at the end of every transaction (which does not fix the problem completely because you still may have one corrupt transaction) or only when you detect the problem and send the reset sequence.

In most cases there won't be significant difference in disruption of system operation whether it is one transaction or 100 that is corrupted.

You should talk to ADI about the temperature thing and forget the idea of ESD diodes being useful in circuit unless you stay within spec. To deal with your transient problem, you can always add your own limit diodes externally along with a small series resistor to the IO pin which should prevent the possibility of the ESD diodes being a problem.

Consider the millions of AD7793 parts they ship and ask yourself if the problem is you or the chip? Better, ask ADI.

--

Rick

If the SPI state machine does not reliably "state machine", you can no longer trust the part and I'd just as soon not go beyond that.

I suppose you could put Vcc on an I/O pin or as an output from an FPGA, but that's indecent.

--
Les Cargill

BSD is exactly how a software system should behave when presented with unwinnable state.

It's "Ah canna change th' laws of physics, cap'n." done loudly.

--
Les Cargill

If there's enough power supply amps available, SCR latchup is generally destructive. The 'a' series 4000 CMOS logic was notorious for latchup. Modern HC-type logic is a lot less sensitive, but few semi processes claim to be latchup-free. If you're lucky, a data sheet will specify some allowable ESD diode current that won't trigger latchup.

Mixed-signal parts seem to be more sensitive to latchup than pure analog or digital parts, seems to me.

Oh, I think we fixed the product with the AD7793: we added a 75 pF cap from the ADC SPI clock pin to ground. We're going to temperature cycle it all weekend and verify the fix. The longer term fix is to use a TI part.

Some guy rang the doorbell this afternoon. It was a TI FAE sort of knocking on doors at random. He was too late to buy us lunch, so he'll have to come back.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

You can send it a stream of 31 1's, which resets everything in, depending on the paragraph you believe, 500us or 1 ms.

I think we'll do the capacitor kluge, and also read the status and chip ID registers after every conversion. If they don't look right, we'll ignore that ADC data and reset the chip.

We can expect the ID register to conveniently display the hex value

0xXB.

All ugly.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

that the capacitor works could suggest that you are using the wrong clock edge

-Lasse

Yes, but the waveforms look exactly like the data sheet.

It was once observed that bad data was coming out on the wrong clock edge, after it was run cold.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

Thank you for this post. Yesterday, when I noticed that my smart(?) phone wasn't receiving, even though it was right next to a cell phone tower, it occurred to me that rather than just resetting the processor, I should power the thing down and back up.

And voila -- it worked.

Maybe you could make "chip select" for the ADC power it down entirely?

--
www.wescottdesign.com

All sorts of gadgets hang up and need to be power cycled to fix them.

There is a serial command to the ADC to reset everything; send it 32

1's. I guess we should do that now and then, to unfreeze it if it messes up. But then we'll have occasional transient errors instead of permanent errors.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

Particularly gadgets heavily relying on software.

We had to do that in the widget that used the TI 5K DSP. We used a spare CODEC wrapped on itself to detect when the channels were out of sync and then reset the DSP rather than glitch the system regularly. It's not a good solution, to be sure, but that's the solution management wanted (my suggestion was to do the I2S/TDM interface the right way, in a CPLD or FPGA).

We had exactly this problem with another AD part, a DDS chip. There were 3 of them on the PCB, all driven by separate SPI controllers in an FPGA. The problem was worst on the one furthest from the FPGA, trace length was a few inches. Adding the cap to ground on the SCLK pin fixed it, we source terminated all the outputs from the FPGA (used 33R IIRC) for good measure.

And yes, the SPI-registers side of the DDS chip was a nightmare to figure out. It took multiple tech support requests to discover basic stuff which should have been in the data sheet, like how to cleanly re- initialise the chip in software. Sending the commands in the wrong sequence would lock the chip up and require a power-cycle. Our latest spin of this board has designed this part out. I wonder if AD know or care that this stuff loses them business?

They do the analog parts of their chips very well, and the digital parts are horrors.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

Hence the name. ;)

I used to have an old AD databook with a FAQ. One entry read something like:

"Digital: You've got to be joking. See the name."

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC 
Optics, Electro-optics, Photonics, Analog Electronics 

160 North State Road #203 
Briarcliff Manor NY 10510 

hobbs at electrooptical dot net 
http://electrooptical.net