Do EEPROMs lose their data over time?

Yes, they are based on a stupidly tiny charge on a floating gate, so eventually they will lose their memory.

The time is generally pretty long under benign conditions (seldom, if ever, re-written, cool temperatures, no significant ionizing radiation).

If any of those things are not true, the life can be significantly foreshortened. For example, at high temperatures rule of thumb is half for every 10°C (not sure this is 100% valid for EEPROM cells, but let's use it here).. so that 100 years becomes 50 years at 35°C, 25 years at 45°C, 12.5 years at 55°C and so on.. so at the maximum storage temperature of 150°C the retention might be less than a week.

Similarly, a badly written or errant program can quickly exceed the lifetime 10^6 erase/write cycles in a matter of hours.

I believe there is some interaction between the two- many write cycles negatively impact the retention time.

You can find a datasheet for the EEPROM you want to use. I looked at one for a 24LC64 (from Microchip) and it said data retention was greater than 200 years. So, yes, they don't hold data forever, but may be good enough for your use.

There are several methods to manufacture an EEPROM, and I mention this only because most of the time when people see "EEPROM", it can be just about anything. Generally speaking, the architecture depends greatly on what they will be used for. For all cases, it's a variation on the floating-gate MOSFET You have a very little amount of charge on NOR and NAND-based storage for SSDs, but you will have larger capacitors for industrial applications where you care about data integrity. Fundamentally, this is a function of the sense amplifier structure and how much charge it takes to flip the bits. I would put my ICs in an oven at 200C to simulate aging and look at charge loss. Generally, we saw no charge loss due to the gates, but possible offsets due to shallow charge trapped in the oxides.

Here's a summary of how you make floating gates:

• Analog-based that are programmed with injection and tunneling
• Differential. This uses two different floating gates to store a value, so one is high and one is low. TI uses this to store information on non-FLASH processes.
• SONOS The charge is trapped in the oxide. This works well for space.
• NOR/NAND. This has excellent charge density, but is a special manufacturing process.

For all of these, you test them by putting them in an oven and watching the drift over time. The NOR/NAND is the worst because it's driven by density so that there is very little charge. Everything else pretty much will hold charge effectively forever. One needs to also keep in mind that changing the state can be destructive, depending on how you do it.

physics warning: If you are "tunneling", this is a quantum process and this is effectively safe for infinite cycles as you do not have enough energy to move a hole. If you are using hot electron injection, there's a probability that you will eventually have an energy state that high enough to move a hole, and you'll throw an atom into oxide. If you are careful, ie: not in a hurry, you can effectively have infinite programming cycles. The oxides will fill with shallow traps of charge to make a bias, but if you have a large enough capacitor, you'll not have any issues with offsets.