Supervisor ICs, doubts

[legacy]

For practical projects, I am willing to use commercial solutions provided by Maxim/Dalsemi, Texas Instruments, or Analog Devices. These solutions are ready to work and well documented.

But about CPU supervisor ICs, I have to say they are intriguing applications of analog electronics, besides doubts in my head.

In this topic, I'd like to ask for some clarification.

doubt number one: the supervisor chip is power supplied by the same voltage used to supply the CPU, it monitors the voltage provided to the CPU and issues a reset if the voltage goes below the safe working condition (Vgood, say for a 5V CPU, Vgood is 4.5V). How can its internal voltage reference work if it's related to Vcc, and Vcc drops?

Assuming that a Zenner diode can provide a good voltage reference even when Vcc drops (I am considering the case that Vcc drops, but it's still below to the Zenner's voltage), the Zenner diode is a component that is notoriously affected by a lot of noise, with a dispersion of +/- 0.2V around the value of the voltage at which the supervisor has to trigger for a reaction. Thus, how can you precise trigger at 4.5V if Vcc is 5V?

doubt number two: for the ICs used to realize non-volatile ram, there is an OA inside the chip. This OA is power supplied by an external battery, usually at 3V. But the OA needs to compare the voltage provided to the CPU (usually 5V) with Vgood (usually 4.5V), thus can you feed in a voltage higher than your supply to the input of an opamp? How does it work? The common-mode input range in of an OA is quite often less than the supply voltage rails, less than Vcc, greater than Vee. Rail-to-rail-OAs have inputs that can accept signals at the same potential as the supply lines, but usually, there are clamp diodes to drop above or below the rail and no guarantee the rest of the opamp will behave nicely in this condition. In short, their inputs must be in the same voltage range of Vcc-Vee.

Thanks 

[MT]

doubt number one: the supervisor chip is power supplied by the same voltage used to supply the CPU, it monitors the voltage provided to the CPU and issues a reset if the voltage goes below the safe working condition (Vgood, say for a 5V CPU, Vgood is 4.5V). How can its internal voltage reference work if it's related to Vcc, and Vcc drops?

Charge pump fills a bucket with current and voltage and a threshold detector intime to shut things down before bucket empty?

Assuming that a Zenner diode can provide a good voltage reference even when Vcc drops (I am considering the case that Vcc drops, but it's still below to the Zenner's voltage), the Zenner diode is a component that is notoriously affected by a lot of noise, with a dispersion of +/- 0.2V around the value of the voltage at which the supervisor has to trigger for a reaction. Thus, how can you precise trigger at 4.5V if Vcc is 5V?

By not using a zener but a comparator?

doubt number two: for the ICs used to realize non-volatile ram, there is an OA inside the chip. This OA is power supplied by an external battery, usually at 3V. But the OA needs to compare the voltage provided to the CPU (usually 5V) with Vgood (usually 4.5V), thus can you feed in a voltage higher than your supply to the input of an opamp? How does it work?

Charge pump?

[helius]

It works the same way that the built-in brown out detection on a MCU does. There is a band-gap reference that is fed into a comparator on one side, with the Vcc on the other side. This should be described in the component's data sheet.
A Zener diode could also be used. The noise isn't a problem if current through the diode is controlled using a current source.

I don't see any opamps inside an SRAM (it's a digital chip!). The chip and output enables are NAND gates. Unlike a DRAM that has sense amplifiers to recover data from the charge on a capacitor, SRAM cells pull themselves to the output level with the 6T structure.

[legacy]

By not using a zener but a comparator?

the Zenner is used to provide the voltage reference for the OA. I haven't understood your answer.

Designing a Voltage pump is an intriguing idea. I have to consider it, pros and cons, advantages and disadvantages

[legacy]

It works the same way that the built-in brown out detection on a MCU does. There is a band-gap reference that is fed into a comparator on one side, with the Vcc on the other side.

the voltage comparator should be a simple OA.

A Zener diode could also be used. The noise isn't a problem if current through the diode is controlled using a current source.

umm, thus a precise current generator, say by using a MOS transistor in saturation when the current on Id is directly controlled by Vgs. Something like that?

I don't see any opamps inside an SRAM

There should be one OA, used as a voltage comparator, inside NVRAM (=SRAM+battery+supervisor+battery). Judging by block-scheme posted by Maxim.

[legacy]

NE555 has two "voltage comparators" (aka "threshold detector", "threshold comparator") inside the chip, and the schematic has been published 



Looks like I can observe how a voltage comparator is implemented by transistors

[MT]

the Zenner is used to provide the voltage reference for the OA. I haven't understood your answer.

I more or less replied on this line: but i might have misinterpreted things.
"Thus, how can you precise trigger at 4.5V if Vcc is 5V?"

Are there precise/exact 4.5V zeners around, i doubt. 5.1 V used to be the most precise. There are overvoltage input OA's around you could use, they take in more voltage then supply without going berserk.

[MT]

Perhaps you could use 7555?

[iMo]

@legacy: you still looking for an off-the-shelf schematics for your Dallas NVRAMs replacement, don't you?

A professional solution which will guarantee your data will not be lost when randomly switching the power on/off at your 68k system is not that simple. You have to provide a lot of effort to get something 100% reliable.
You need for example, a 1.25V ref voltage, 3.3V opamp/comparator, 3.3V logic, to be well below 5V and to have enough headroom for the activities, and create a simple FSM which will monitor the rise/fall of the 5V rail, and manipulate the /CS and /WE, /WR signals of your SRAMs accordingly.

All must be fast enough to provide the all necessary stuff during the power-off.
The power off with your 68k system may last <1ms easily (based on the shot your board takes 1.5A @5V).
Have a look with your o'scope how long does it take the Vcc dropping from 5V to 4.75V when you switch the stuff off. During this time you have to do "something" with the SRAM's signals.
The 3V backup is easy, 2 diodes are ok. You cannot power the above logic from the 3V backup, however, as it takes several mAmps usually (3V CR2032 is 150mAH, you want 5 years).

PS: to be more constructive - I would start wit a small "dev board" with 1.25V ref (easy, ie LM385-1.2), small 8pin microcontroller, something like PIC12F or 16F which run from 2V Vcc up (optionally with built-in comparators), a low power dual comparator (you may find some with 2-3V Vcc sure), and 2-3 logic level fets (not sure if N or Pmosfets at this stage) for gating the 2-3 SRAM's signals (like the /CS, /WR, /WE..).
Add some decoupling., few resistors. With this setup you may start to elaborate something reliable. You would definitely need an DSO handy.

[westfw]

it monitors the voltage provided to the CPU and issues a reset if the voltage goes below the safe working condition (Vgood, say for a 5V CPU, Vgood is 4.5V). How can its internal voltage reference work if it's related to Vcc, and Vcc drops?

 needs to compare the voltage provided to the CPU (usually 5V) with Vgood (usually 4.5V), thus can you feed in a voltage higher than your supply to the input of an opamp? How does it work?

various voltage reference circuits are well-known analog technology.  See wikipedia's entry on "bandgap voltage reference", for example.
Why are you assuming that the monitor circuits have to compare Vcc to anything, when it's perfectly adequate to compare, say, (Vcc/4) to the ~1.2V typical of bandgaps - all you need is for the rest of the circuit to operate adequately at voltages well under the "turn if off" voltage of the target.   With today's 1.8V microcontrollers, you could do it in software...

[legacy]

With today's 1.8V microcontrollers, you could do it in software...

you can't you replace everything with an MPU; in avionics, it's mandatory to avoid to use a CPU/MPU whatever unless you are willing to cover the DO178B's stuff (that requires a lot of effort)

this project is for hobby, but! an as good practice I am trying to avoid to use an MPU for things that I can handle with a pure analogic circuit, and this is the reason of the topic, I'd like to "play" with that, rather than with MPU's

[legacy]

Vcc/4

dividing a voltage through resistors injects up to the 5% of error around the working value, and by reducing the threshold by 1/4 we are talking about 0.5V / 4 ~= 0.1V for the threshold, that is too subjected to false-positives

it'd better be handled by an OA, used to divide Vcc by 4, and reinforcing the precision through the feedback loop, but even a rail-to-rail (SE voltage) OA needs to be powered at Vcc.

[westfw]

dividing a voltage through resistors injects up to the 5% of error

Why do the say that?  1% discreet resistors are readily available, and if they're on-chip you can probably get as precise as you want (especially since the main concern is the ratio), using the same sort of processes (laser-trimming, perhaps) used for other precision analog components. (although it looks like a "typical" supervisor is only about 2%...)A quick look shows that Maxim has some "precision voltage dividers" (MAX5490 series) that claim 0.035% accuracy.And ... a quick look at several "supervisory" ICs shows... voltage dividers before comparison with a bandgap reference...