short version: The assumption "What I want to know is, why does it seem like none of these limitations apply to the floating gate layer?" is incorrect, but with some caveats.
You can make a floating gate on any process if you have your oxide clean enough. Your "grown oxide" between the channel and the first layer is much better because it doesn't have traps caused by impurities that you see by the deposited oxides; however, it's actually a bit more complicated than that because with cycles of injection, you do see leakage back from the gate to the drain edge.
For analog processes, I build floating-gates that look something like this:
They look different when I'm just making digital storage, but for this demonstration, the analog floating gate layout makes the most sense. I make the node negative by putting charge on the floating poly through hot electron injection, which is classical physics. I make the node positive through the tunneling junction.
When you inject, you are heating carriers and if you are lucky (not really, the work by Hasler outlines this explicitly), some make it on to the gate, such as:
Injection is a function of field, so you have a high field at the drain edge. Some of the electrons can get stuck in the oxide and you start seeing band-to-band tunneling after a bunch of cycles. The grown oxide is clean, but not perfect. How many cycles? It depends on how fast you tried to program things. If I put a device in subthreshold and put a large Vds (larger than process) across it and hold it for a day, it basically throws enough charge in the oxide that you can watch the leakage. For the process voltage, I usually get about 100k writes at the process voltage before things go bad. This is the same issue that causes FLASH drives to get stuck sectors.
Consider a short to ground on the high side, before the breaker.
If the low side is grounded (not floating) the short to ground is a dead short, and either a breaker will trip, a fuse will blow or something will melt. Either way, it'll get fixed.
If the system is floating, a short to ground on the high side doesn't affect the performance of the system. It continues to work just fine. Some time later (maybe even years later) someone wants to do some work on the system, so they flip off the breaker. But if the breaker only disconnects the high side, then the low side is left with a large negative voltage. The person gets a shock.
Best Answer
Floating signals are usually not a good thing, but can be acceptable in some cases. In all those cases, the value of the signal does not matter.
Not every signal is relevant all the time. A common example is the MISO line of a SPI bus. This is only actively driven when a slave device is selected (enabled). It's value is only relevant for a short time around one of the SCK clock edges. At all other times, the signal can be any state without affect on the system since the system ignores it.
So what happens to MISO when no slave is selected, as is the case whenever the SPI bus is not in use? The answer is you don't care. It doesn't matter what its value is since nobody is looking at it. Since MISO is driven only by the single selected slave, all unselected slaves and anything else on that line must be high impedance. That means when no slave is selected, the line is left floating as described in the passage you quoted above. This causes no bad data, since again, the system is ignoring the line at that time.
While a floating line is OK logically when nothing is looking at its value, it can be a problem electrically. Many logic inputs are intended for the voltage to be either solidly low or solidly high. In-between values can cause higher than specified currents in the input circuit, and in some cases can even cause this circuit to oscillate.
For this reason, there is often a weak pulldown or pullup resistor on lines that could float. I usually use a 100 kΩ pulldown on MISO, for example. When a slave is selected, it drives the line regardless of the small extra current it takes to hold it in the high state. However, when nothing is driving the line, it will go low, preventing the unwanted extra current and oscillations in anything receiving the signal.
There are also types of digital inputs that can handle any voltage within the valid range without undesirable characteristics, like extra current or oscillations. Schmitt triggers are one example. These have hysteresis so that after flipping one way, it takes a different voltage to flip the other way. A floating line may cause the digital signal to be interpreted randomly as its voltage floats around, but the digital input is designed to handle that. Of course the rest of the system still needs to be designed to not care what the value of that digital signal is during the time its floating.
In general, truly floating signals are bad, but can be easily addressed with a weak pulldown or pullup resistor.