Identification of modules

Retention *time*? What sort of product lifespan are you assuming?

Do you want slots to be wired (nearly) identical? Or, are you willing to have a "controller slot", etc.?

Sole source?

What's wrong with using a tiny MCU to determine which slot *it* resides in and then convey its slot number and "module type" to the controller, regardless of where the controller happens to be located?

Apply Vref to the "slotID" pin of the leftmost slot and GND to the same pin on the rightmost slot. Put a fixed resistance between the slotID pin on each slot and the slot to its right. MCU on the card *in* the slot can then read the voltage on the slotID pin to identify itself.

Alternately, put a unique R value between Vref and each slot's slotID pin then sense this (with a current sink, RC oscillator, etc.).

Another pin can act as a bidirectional bus that the controller assumes ownership of and polls the individual MCU's on each slot.

Of course, this assumes you are taking precautions to protect all inputs that are "exposed" on the backplane...

Mostly identical.

Not a problem, can buy more in advance.

Best regards, Piotr

So, you're planning on discarding the EEPROM in the DS2438? I.e., you can't rely on any "persistent" features in your design that can't be reloaded at POST. (Make sure the configuration isn't implemented in EEPROM if you mistrust it!)

Why not just opt for a one-wire ADC? Pair? Addressable switch?

Or, some discrete logic that you clock, remotely?

(presumably, you have some time to query the system's configuration -- you're not trying to hot swap cards and bring them on-line "instantly"?)

So, you're prepared to do an end-of-life buy as you head into initial production?

What about masked parts?

I'm not clear on what you are using the resistive divider for. Is that to identify the module type or the slot?

The lowest priced 1 wire part I found on Mouser is the DS28E05P+T. It combines the unique 64 bit serial number each 1 wire device has with 112 bytes of user EEPROM for under $0.50 in quantity. The EEPROM can be programmed in manufacturing to provide a module type identification number.

I'm not clear on what your limitations are for identifying the slot number. The easy way is to isolate the 1 wire buses to each slot so you are only talking to one at a time. If you don't want to use that many pins on your MCU, an analog switch multiplexer will allow the MCU to talk to one slot at a time and only need log2(N) I/Os.

Oh, I misunderstood you. Yes, the wiring will be composed of 18 parallel, identical lines + 2 dedicated per-slot (already used) pins on the edges of the connector. All single-sided and reinforced with wire, I don't trust vias that much.

A fair chance for the MCU to develop Alzheimer's disease after 30 years of being kept on the shelf with other spare modules. :-) No stored charge devices here. Which brings the question about DS2438 internals. I've read somewhere that some voltage reference chips are based on a cap holding some charge injected in the factory, which will make them drift due to leakage in the larger time scales. Will need to ask Maxim about that, perhaps.

That's another story, but yes, of course.

Best regards, Piotr

Yes, not going to use it at all.

The problem is not in EEPROM, but the lack of power supply/communitation in the case of spare modules. The powered ones can be automatically reprogrammed once a year, just in case.

Where, on the module? It's exactly how 2438 is going to be used.

Too big.

Yes, no hot-swap capability required.

I can add a spare DS2438 in a bag for every produced module.

The scale is far to small, can't afford.

Best regards, Piotr

There are two pieces of information the controller needs to know, both safely in the range of 5 bits:

1. Where the module is (in which slot).
2. What type of module it is, which says what waveforming is required to drive it. Say, "I am a 230V module stuffed with TRIACS" or "the switches are MOSFETs, so use reverse phase control to decrease the level of EMI" or "mechanical relays, avoid fidgeting", etc.

The more advanced modules, e.g. carrying an MCU, may want to use the location info to configure their I2C/CAN address etc.

Just 10 years of retention time, far to small.

Would require a complex backplane routing.

Would need to be located on the backplane, which is hard to replace.

Best regards, Piotr

Is that five bits each or five bits total? Why do I need to know how many bits this uses? I need to know how you plan to distinguish the board slots.

You didn't answer my question.

"Complex"? How do you define "complex"?

I guess I still don't understand your system. I don't see why any parts would need to be on the backplane.

Where are all the pieces? Where is the CPU? You still have not explained the limitations of your slot identification. How will the board know which slot it is in? Are you trying to identify the boards and positions to a central CPU or to the MCUs on each board?

So, small scale, can't trust stored charge over decades.

Consider manually set switches or manually inserted zero ohm resistors, and a shift register.

Five bits each.

Because that determines the space of possible solutions. One bit would could be solved purely by digital means,

One option is that a designated pin carries a different voltage level. The level is generated by an appropriate resistor divider on the backplane, based on 1% parts.

The type of the board is identified exactly the same way, but the divider is on the board.

Anything that wouldn't fit on a board without vias and the traces cannot be reinforced with wire.

A board composed of 18 parallel wide traces + 2 per-slot dedicated lines is considered to be simple.

Then how?

To a central controller composed of 2 redundant CPUs. An MCU on board is optional and preferably avoided if possible.

Best regards, Piotr

We use those (with trepidation, maxim etc). We use it to store the serial number and calibration data on a tiny sensor module, to make them interchangeable.

You don't need to do that AIUI, the one-wire bus has a discovery algorithm you can use to address any device when all the "one-wires" are connected together. Never had to use it but it's there.

The configurability isn't required, each slot "knows" its number and each module knows it type, which is determined by its physical construction, so the config can be etched on the PCB.

I also thought about CD40103, with the count inputs preset to encode 8-bit information and count how many times do I need to pulse it till it blinks, but it would use many slot pins and is bigger than the SO8 Maxim parts.

But you pointed out an interresting solution, each module contains a shift register (it is the way the desired content is presented to the power devices, but from the controller's POV it is write-only). I can skip the generation of the latching pulse and abuse the value of the shift-out chaining pin in a filthy way.

Since each module is either in normal operation or identification mode, this idea can lead somewhere...

Best regards, Piotr

You haven't explained your system.

You haven't said *where* the CPUs are. Are they on two of the 322 plugin cards? There are lots of things you haven't explained about your system. It's hard to suggest solutions when the system is not defined fully.

That is pretty much exactly what they are intended for. I see now that part is barking up the wrong tree. He doesn't care about the serial number or the flash, he only cares about the ADC.

I've used it before, how else do you learn what devices are on your bus? Addressing the devices is not the same as knowing what slot they are in. That is the purpose of the multiplexers, to isolate the ID pin on each card slot so the slot is known. Instead he wants to supply to the slot an analog voltage through a resistive divider which is read by an ADC on the card. Then he doesn't like the cost of the ADC.

There are other ways to get an ADC such as an MCU, but I can't think of any that won't use flash memory... except for the TI MSP430FR devices that use ferroelectric storage, FRAM. But I doubt that has enough retention time for this requirement either.

The only other ADC that might suit him is one that works over SMBUS. That will use two pins, but it is all a bus. The ADC needs two inputs, one for the slot voltage and one for the card type voltage. Best would be one that can do ratiometric measurements to eliminate the effect of variations in Vcc.

Similar to the shift register approach maybe use something like a SN74ABT8245 to setup a JTAG chain. With that octal part you would have 16 bits of config data/temp data you could clock out. Not a cheap solution though.

So, you're assuming the EEPROM contents will remain intact between times that you get a chance to "examine" it. Note that there are no CRC's on the configuration so if a bit toggles, you'll have no way of knowing that the configuration has changed and will gleefully "refresh" it to its erroneous state.

Same way as yours is used. I'm just eliding the "charge accumulator", voltage/current monitoring, etc. that makes the 2438 so "special".

Masked FPGA/PAL/etc.?

That seems like you're planning for failure -- instead of planning for *success*!

Can't argue with that! :>

From above, through holes drilled in an appropriate place to connect two THT resistors.

This is what I consider simple. 2 busses on the edges of the connectors, common signals in the middle, the slot ID resistor dividers connected to pins 5.

Technically, on a special module, they have their dedicated connectors on the backplane and drive the lines from the image above.

What would you like to be better defined, Rick? The backplane is a row of straight 20-gold pin connectors, the modules are to be installed vertically.

Best regards, Piotr

And the integrated temperature sensor, which is there for free.

Exactly, and the reason to dislike the dedicated ADC (i.e. 2450) is its 3x higher price and the lack of the thermometer.

"The devices are recognized automatically and assigned unique addresses."

Sounds good!

Best regards, Piotr

How limited a flash retention time? Flash is usually guaranteed for 10 years, but I've seen anecdotal evidence for a much longer time.

How much board space can you use up? There used to be some 7400 series parallel-in serial out registers that should be at least semi-compatible with SPI. If they still exist, two or three chips should let you clock out a 16-bit binary number.

There's less and less of that sort of thing out there because there's more and more microprocessors.