a 100y lifetime project : incredibly difficult ?

[PA0PBZ]

An interesting problem.
I can't think of a display that will last 100 years and is low power, exept maybe this:



The interesting part is that it is both the display and the memory, but not very attractive.
To make it work you need 1 pulse per day, that should be possible to build with long life components.
To power it maybe a combination of solar and different kinds of energy harvesting.
But a power storage device that will last 100 years... no idea.

[ker2x]

An interesting problem.
I can't think of a display that will last 100 years and is low power, exept maybe this:



The interesting part is that it is both the display and the memory, but not very attractive.

Interesting.
It's not really ugly. It look low tech.
Even the shinyest awesome hitech whatever we could build today will look (and will be) low-tech in a century.
So... who care ? Our today awesome stuff will have the same effect as http://www.rewindmuseum.com/vintagetv.htm in a century 

PS : I'm reading this : http://www.sitime.com/support2/documents/AN10025-SiTime-Reliability-Calculations.pdf

[DanielS]

For setting the DoB and current date, one could use NFC with a mask-programmable (ROM) microcontroller. Since the NFC interface is powered by the master device, the wireless connectivity comes at no cost for the sealed device's battery life.

Will NFC still be around 100 years from now? Probably not. But neither will most of what we use today 40 years from now anyway: the last remaining PC connection from 30+ years ago is the serial port but most modern PCs only expose it as a motherboard header if at all. If the unit fails and requires replacement before then, you can upgrade to whatever is current at that point. In principle, the only time you will ever need to interface with it is soon after purchase or replacement; not a century later.

Since there is no guarantee any external timing network in existence today will still be around even 50 years from now, you will need an internal clock and with a typical 1MHz watch crystal, you will need an extra 37 bits to count microseconds per day... so that would be 52bits total for 100 years.

Achieving 1day/century error could be difficult since crystals could have that much of an error simply from their orientation relative to gravity and temperature. The only way around that would be to use an atomic clock like others suggested but those usually consume too much power for something battery-operated intended to last for years.

Just for fun, I went searching for a low-power atomic clocks and found this:
http://www.microsemi.com/products/timing-synchronization-systems/embedded-timing-solutions/components/sa-45s-chip-scale-atomic-clock
Sub-120mW chip-scale atomic clock, sounds interesting. Only rated for 100k hours MTBF though. Sounds like a great fit/upgrade for those Agilent frequency counters with lousy stock oscillator.

[ker2x]

For setting the DoB and current date, one could use NFC with a mask-programmable (ROM) microcontroller. Since the NFC interface is powered by the master device, the wireless connectivity comes at no cost for the sealed device's battery life.
Will NFC still be around 100 years from now? Probably not.

Not a problem, DoB need to be set only once, and in easily predictable future.

Achieving 1day/century error could be difficult since crystals could have that much of an error simply from their orientation relative to gravity and temperature. The only way around that would be to use an atomic clock like others suggested but those usually consume too much power for something battery-operated intended to last for years.

Is it difficult ? I'm really against any external timekeeping device. 

[SeanB]

EPROM is only to store the MCU program. You would have to store the time likely in SRAM and use a supercap or a silver oxide cell ( rechargeable) that will store enough charge to run the clock oscillator chip and keep the memory alive. You store the start time multiply in the RAM and read them to get the best result of start then use the RTC to count from that. That way you only have the RTC drift to contend with, and this will be within your error margin for most RTC chips.

You will have to have redundant clocks and ram though to be able to recover from single point errors and single failures of the RAM or the clock. You probably will also have to have redundant batteries and switching to enable them to feed the RTC units and the RAM. Batteries would probably be best in separate sealed subcompartments to keep corroding cells from damaging other parts.

[ker2x]

From atmel AVR 8bit FAQ :

Mean Time Between Failures is an indication of the number of hours to pass between failures.

Please find below the predicted MTBF (Mean Time Between Failure) numbers of
our micro's at different temperatures. The statistical calculations are based
on current reliability qualification data in the "microcontroller
reliability data package".

Atmel has been shipping NVM parts for 20 years and Flash Micro's for 13
years. We have never experienced a long term reliability problem.

Here are the MTBF numbers calculated from life test and data retention
results:

65ºC 1.69x10e7 hours. => 1929 years
85ºC 4.46x10e6 hours. => 509 years
105ºC 1.34x10e6 hours. => 153 years

It's not a product supposed to have a very high lifetime, isn't it ?

- But if atmel can do it on their AVR... well, i guess using a microcontroller in this project is a viable option 
- firmware storage look possible too (according to previous post here)
- oscillator lifetime doesn't look like a major problem too.
- we have some options for the "user interface"

Lot of option to explore for the power, which is still the first and main problem 

Energy harvesting (solar/RF/themic,kinetic/...) , if possible, would be best but it can't be the only source of power.
I can't find a supercapacitor with more than 20y lifetime.

[Jeroen3]

The battery with the longest lifespan I could find were NiFe batteries. http://en.wikipedia.org/wiki/Nickel%E2%80%93iron_battery

What would the lifespan of a CR2032 battery be? Vendors don't seem to specify this. I've never seen one leak, as opposed to Alkalines.

[ker2x]

The battery with the longest lifespan I could find were NiFe batteries. http://en.wikipedia.org/wiki/Nickel%E2%80%93iron_battery

What would the lifespan of a CR2032 battery be? Vendors don't seem to specify this. I've never seen one leak, as opposed to Alkalines.

The highest i could find is http://en.wikipedia.org/wiki/Zamboni_pile but it's not an viable option 
I'm just checked some datasheet from energizer. Not much information about lifetime but the few i could find was below 20y.

Nickel-iron battery won't work too.
Energy density is very, very, VERY bad. 
Also : Self-discharge rate   20% – 30%/month 

About the CR2032, the datasheet from enegizer (i checked this one already) say "Self Discharge: ~1% / year"

From TI about rechargeable battery :

Code: [Select]

SELF-DISCHARGE
CELL TYPE        NI-MH NI-CD LI-ION
@ 20°C (%/MONTH) 20-30 15-20 5-10


[Corporate666]

What about a combined electrical and mechanical solution?

You would have to do some calculations to figure out the power involved... but I wonder how much energy a wound-up spring contains and how that compares to a battery? 

You could have a clock-like mechanism but beefier that is wound up when the unit is made... then some sort of latching (or breakable) device that keeps the unit in an inactive state.

When the unit is activated (perhaps by hitting it on a surface, breaking a tiny glass latch that then allows the mechanism to start running), the device starts counting from a set point - like 10 days or 30 days.  This gives the proud parents an opportunity to buy it and have it engraved.  If you can store enough energy in the mechanical mechanism, you will never need a battery and will never have to replace anything.  You could put the whole thing in some sort of glass or plastic container which has a vacuum (or nitrogen) inside so it never corrodes or wears out.

[ker2x]

What about a combined electrical and mechanical solution?

You would have to do some calculations to figure out the power involved... but I wonder how much energy a wound-up spring contains and how that compares to a battery? 

You could have a clock-like mechanism but beefier that is wound up when the unit is made... then some sort of latching (or breakable) device that keeps the unit in an inactive state.

When the unit is activated (perhaps by hitting it on a surface, breaking a tiny glass latch that then allows the mechanism to start running), the device starts counting from a set point - like 10 days or 30 days.  This gives the proud parents an opportunity to buy it and have it engraved.  If you can store enough energy in the mechanical mechanism, you will never need a battery and will never have to replace anything.  You could put the whole thing in some sort of glass or plastic container which has a vacuum (or nitrogen) inside so it never corrodes or wears out.

I love the idea about the breakable mechanism.   

[tszaboo]

100 years contains very low frequencies, so you have very big pink noise. This can include lightnings, nuclear wars, collisions with other space objects... So I would put as much protection devices on board as possible.
I would use an antifuse FPGA for this. The crystal has to be in a super package.
Einstein said, that human stupidity is infinite, so I would power it from that.

[ker2x]

100 years contains very low frequencies, so you have very big pink noise.

I don't understand this part

This can include lightnings, nuclear wars, collisions with other space objects... So I would put as much protection devices on board as possible.
I would use an antifuse FPGA for this. The crystal has to be in a super package.
Einstein said, that human stupidity is infinite, so I would power it from that.

I'll write a memo "warranty void if nuclear winter seal is broken" 
The problem with stupidity : even if it potentially hold an infinite amount of energy, its power efficiency sux 

[max_torque]

if you want to simply count "days" as your smallest unit, could you use some other stimuli to correct the slow drift of a conventional realtime clock?  ie it generally gets dark once a day, so could you use a light sensor with a very very slow "averaging" algorithm of some type to keep the RTO from drifting ??


(obviously, if the device is kept in a drawer for any length of time, that approach will fail!)

[T3sl4co1l]

Well...

Let's make a list of things that absolutely cannot possibly be used:


Energy Storage, Power Generation

- Battery: All types

(Arguably, a Zamboni pile not only could last extremely long, but may even be the only proven multi-century technology!  Even so, note the power density is extraordinarily minuscule.)

- Supercapacitors, electrolytics, polymers (all age limited, susceptible to dry-out or moisture ingress).
- Solar cells (PV) -- dubious; good enough for aerospace, but most are known to degrade over time; also, dust accumulation in dry climates.
- RTGs -- that is, if made with conventional fuels and materials.  With suitable choices, they could last very long indeed, however, and the added weight and space of lower density fuel, extra shielding and poor efficiency converters would be tolerable in a terrestrial application (like the millennium clock thing).


Active Components

- Flash memory, EEPROM (and the vast array of programmable devices which integrate these): charge decays over time, and programming causes incremental damage.

- Fine pitch, low voltage ICs.  Besides being more sensitive to ESD / overvoltage, they're more sensitive, thermodynamically speaking.  (One of the reasons a PC CPU shouldn't go over say 80C, or a GPU over 100C, versus a mil spec transistor over 200C.)  I would be fine with, say, CD4000 series logic at 5V (well within ratings), despite the ESD sensitivity.  74/LS logic should be okay too, but will take a whole lot more power.  Of course, you can't go wrong with discrete transistors.

- ESD, EMC, life in general: everything must be several times below ratings, all circuits well shielded, filtered and protected, with current limiting (of the "big dumb resistor" kind) and multi-level transient protection everywhere.  Most protection devices (MOVs, GDTs, etc.) are wear components, but with current limiting, sufficiently large TVSs should be okay I think?

- What else?


Things that are okay include a surprising variety of mechanical things, but they all have to be magnificently engineered with top-notch materials.  Oxidation, moisture, and lubrication (migration / degradation) are the greatest challenges here.  Examples of energy storage / power generation might include springs (well below ratings -- nearly infinite fatigue limit), weights (gravity isn't going away any time soon, but mind if it moves), and motors (induction motors are essentially indefinite; permanent magnet motors may be susceptible to demagnetization either over time or due to very strong ambient fields).  Electrical energy reserves would be best implemented as massive stacks of film or ceramic capacitors -- very bulky, but won't dry out.

Tim

[FrankBuss]

Let's make a list of things that absolutely cannot possibly be used:
[...]
Active Components

- Flash memory, EEPROM (and the vast array of programmable devices which integrate these): charge decays over time, and programming causes incremental damage.

Both Atmel (for AVR) and TI (for MSP430) claims at least 100 years data retention for their flash, so I don't see a problem. Of course, single bit failures could still happen, like from cosmic rays or if one bit is not as reliable. Best would be to use many microcontrollers and an external majority circuit, like used in safety critical systems like nuclear reactors and spacecrafts, and self testing and error correcting firmware.

LEDs might be a good idea. I found this report with MTBF of 2,000 years for green LEDs. Interestingly red LEDs have a MTBF of only 175 years. But I guess this would be much longer if you don't use the maximum rated current. But it requires more power, or you need a button to show the number of days.

I couldn't find any MTBF rating for epapers.

Good finding for the SiTime spec. So one of those SiTime oscillators, an AVR microcontroller, some kind of sealed mechanical display, solar cells for the power supply and some big conventional film or ceramic capacitors for the backup should last 100 years. You could seal the whole circuit, maybe add some LEDs for status display and some capacitive buttons for setup.

[EEVblog]

I'll write a memo "warranty void if nuclear winter seal is broken" 

Include one of these in your BOM:
http://www.maxwell.com/products/microelectronics/docs/hsn1000_rev3.pdf
When your /NED pin goes low, activate protection circuitry.

[EEVblog]

Just for fun, I went searching for a low-power atomic clocks and found this:
http://www.microsemi.com/products/timing-synchronization-systems/embedded-timing-solutions/components/sa-45s-chip-scale-atomic-clock
Sub-120mW chip-scale atomic clock, sounds interesting. Only rated for 100k hours MTBF though. Sounds like a great fit/upgrade for those Agilent frequency counters with lousy stock oscillator.

From the website:

Underwater sensors for seismic research or gas and oil exploration

Yep! I worked in that field for 10 years, and yes, our biggest problem was always precisely that (pun intended), a precision autonomous low power clock.
When I left in 2007, the problem still hadn't been solved. Now solutions like this make me want to cry 

[coppice]

Just for fun, I went searching for a low-power atomic clocks and found this:
http://www.microsemi.com/products/timing-synchronization-systems/embedded-timing-solutions/components/sa-45s-chip-scale-atomic-clock
Sub-120mW chip-scale atomic clock, sounds interesting. Only rated for 100k hours MTBF though. Sounds like a great fit/upgrade for those Agilent frequency counters with lousy stock oscillator.

Does anyone know how a device like this can be atomic, and yet have such a poor aging characteristic? Atomic clocks are supposed to use very fundamental qualities of matter. How can that age?

[ElektroQuark]

Make a clepsidra.

[mikerj]

Just for fun, I went searching for a low-power atomic clocks and found this:
http://www.microsemi.com/products/timing-synchronization-systems/embedded-timing-solutions/components/sa-45s-chip-scale-atomic-clock
Sub-120mW chip-scale atomic clock, sounds interesting. Only rated for 100k hours MTBF though. Sounds like a great fit/upgrade for those Agilent frequency counters with lousy stock oscillator.

Does anyone know how a device like this can be atomic, and yet have such a poor aging characteristic? Atomic clocks are supposed to use very fundamental qualities of matter. How can that age?

It's aging is specified at 3.0E-10/month, doesn't seem too bad to me!

[ker2x]

but 120mW is HUGE 
We already have a major problem with battery without even considering any power consumption (the aging itself is a problem).

[mikeselectricstuff]

Power is clearly going to be the biggest issue, so if you can decouple the time count from any processing & display, that would probably simplify things a lot.

So for example find the lowest power RTC chip you can find  to keep time, and an MCU/display that's only powered when there is enough power available - I think Solar is probably the only option if you don't want a radioactive battery.
Maybe multiple RTCs with multiple power sources for backup.

Any storage capacitance probably doesn't want to have anything liquid in it, so maybe (a lot of) tantalum, polymer  or ceramic.

[Jeroen3]

A mechanical system (spring) that hits a piezo every once in a while to recharge your energy harvesting solution.
They've made switches that send wireless signals when switched. All powered from a piezo energy of the click itself.

[ker2x]

edited first post

Power is clearly going to be the biggest issue, so if you can decouple the time count from any processing & display, that would probably simplify things a lot.

So for example find the lowest power RTC chip you can find  to keep time, and an MCU/display that's only powered when there is enough power available - I think Solar is probably the only option if you don't want a radioactive battery.
Maybe multiple RTCs with multiple power sources for backup.

Any storage capacitance probably doesn't want to have anything liquid in it, so maybe (a lot of) tantalum, polymer  or ceramic.

Yes, power is the bigest issue. no solution found yet.
I wrote to energizer with a simple description of the project and a simple a question : "is there any battery form factor guaranteed to still exist in 100 years ?"
Perhaps some kind of industrial form factor have a guaranteed minimum EOL.
I'm not expecting a reply but... who knows ? 


Also... why a RTC ? It seems to be much more complex than it need to be.
Why not a simple counter ? It's all about counting days. (also, i didn't find RTC that will still work after 2099, which is a problem)

A mechanical system (spring) that hits a piezo every once in a while to recharge your energy harvesting solution.
They've made switches that send wireless signals when switched. All powered from a piezo energy of the click itself.

We still need to store this energy, it may solve the self-discharge problem, but not the mechanical/chemical lifespan.

Also there will be a problem with isolation from exterior. It would be nice of all the electronic part were completely isolated from air.
Perhaps something with magnet and antenna ?

[mikeselectricstuff]

Also... why a RTC ? It seems to be much more complex than it need to be.
Why not a simple counter ? It's all about counting days. (also, i didn't find RTC that will still work after 2099, which is a problem)

Because an RTC is specifically engineered for the job - the counter isn't a huge issue but the oscillator is, and is probably the largest power consumer. Date range is not an issue - just keep any higher-order count seperately - many RTCs have some extra registers for user data, so have a bit for year rollover