In this post I want to expand upon the discussion of the role of quantum tunneling in everyday flash memory as discussed in the [latest video] on the YouTube channel. I’m going to assume you’ve seen that video here so if you haven’t I’d check it out first.
Now, the basic element of all modern microchips and computers and arguably one of the most important technologies in human history is the Metal-Oxide-Semiconductor Field-Effect-Transistor or MOSFET for short. If you crack open a microchip you’ll find a single slab of silicon into which literally billions and billions of these devices have been embedded to make an integrated circuit. (See [this previous video] for a much greater discussion on these guys)
Those little guys are what makes the digital world turn.
In a nutshell, each of these devices is like a little electric switch with a “control gate” and a “channel”. The basic idea behind their operation is that depending on whether an electrical voltage is applied to this top “control gate” electrode or not, the “channel” of the device will be closed or open. So it’s like a valve for electrical current which can be opened or closed depending on the gate voltage. And it’s this open or closed-ness that gives the binary “0” or “1”ness we need to make digital circuits.
On a physical level, to say that I’ve applied a voltage at the control gate is to say I’ve caused charged electrons to accumulate at the gate surface and the electric field due to these charges can be felt in the channel. Without going into too much depth, the effect of this electric field is to alter, in real-time, the conductivity of the semiconducting channel making it either very conductive, like a metal, or very resistive like an insulator on-the-fly. It’s this “field effect” that gives you your on or off-ness of the channel and also is what the FET in MOSFET stands for (“Field Effect Transistor”). So applying a voltage at the gate, bring in electrons and the field due to those electrons alters, from afar, the conductivity of the channel allowing it to be “open” (i.e. highly conductive) or “closed” (very resistive).
However, if the metal gate electrode was literally sitting directly on top of the semiconductor channel and I applied a voltage, I wouldn’t get this accumulation of charge and this field-effect. Instead, any charges that accumulated would just flow through the channel. So I want the charges to be close to the channel but I don’t want them to be able to actually move into the channel. In order to accomplish this I put a very thin layer of insulating material between the two. Historically, this material was basically just glass which is actually silicon dioxide which is what you get when you heat silicon in the presence of oxygen. In other words, it’s kinda just silicon rust. But the point is that silicon’s a semiconductor but silicon dioxide is an insulator and having that thin layer stops unwanted flow of charges from gate to channel.
But what does this have to do with flash memory? Wasn’t that the topic here?
Well, the basic element – the basic object that can hold a binary “0” or “1” state- of flash memories is what is called a floating-gate transistor (sometimes called FG-FETs). These floating-gate transistors are very similar to our MOSFET with the only difference being that in the middle of this insulating layer there’s been inserted a chunk of metal or silicon called a floating-gate. What we mean by saying it is floating is that it’s not actually attached to anything. The top contacts are part of a circuit and connected to different things, but the little island in the middle is just a completely insulating chunk of conductive material.
Now, remember that it was whether charges existed at the gate or not that determined whether a MOSFET switch was open or not and gave us our “0” or “1”. So what then would happen if there were actually electrons trapped somehow, on this floating-island? Well, if they’re right and truly trapped then even when I turn the power off to the circuit, the effect they have on the channel would persist. The channel would be open or closed and it’d stay that way even when I disconnect the power. In that case I don’t have a switching device, but I have a memory; a device that can be set to “0” or “1” and stay that way and I can read its state by seeing if current will flow through the channel or not. More specifically I have a non-volatile memory, which means that the data remains even if you turn the power off. This is in contrast to something like computer RAM, which is called “volatile memory”, that only holds data temporarily while the computer is on.
That’s how flash and solid-state memory works. But the question is, if there little floating-island is completely electrically insulated from any electrical contacts, how do you get charges onto it or take charge off of it in the first place? Well, in the video only quantum tunneling was discussed but in reality there are actually two ways you can do it.
As was discussed in the video, for reasons I didn’t really explain (fodder for another video!) the alleged “most important” electron states are those in a thin band near the edge of a band. Again, I provided no explanation *why* that is the case, but simply said that it was. However, even though that’s where *most* electrons are there actually is still the occasional, rare hotter-than-average electron with energies much higher than that range (“hot” meaning how much energy it has). There are much less of those, but they exist and one way to get electrons on the island is to use the electrons in these states because for them there actually are states in the insulating layer that they can conduct through (see the above figure).
This is called “hot-carrier injection” or “thermionic emission” (an odd term that was inherited from the old days of vacuum tubes).
The down side to charging or discharging the floating-gate with “hot-carrier injection”, however, is two-fold. The first is that in order to have a large amount of this hot-carrier flow the voltages actually need to be much higher. High voltages means the device consumes more power which means it generates more heat which means you can’t stuff as many of them into the same surface area. So you basically get less storage space for the same chip, less gigabytes of data.
The other down-side is that these hot-carriers, by definition, have quite a bit more energy than our tunneling electrons and as a result they can literally cause damage to the insulator as they flow. They do this by, for example, scattering and knocking out atoms in the lattice creating what are called trapping centers that can hold electrons. This allows electrons to tunnel through the insulator even at low voltages by not doing the whole thing in one trip but instead doing multiple shorter tunneling events and kinda hopping from trap to trap through what is called “trap-assisted-tunneling”. Furthermore, these traps can also simply hold onto electrons even if you discharge the floating-gate. So the device is supposed to register like there’s no charge on the floating-island but because of these hot-carriers that got trapped in the insulator the device will still read like they’re there and thus your data is corrupted.
So, in a nutshell, what that means is that memories that work through this kind of hot-carrier charging and discharging have short device lifetimes and cannot be written over (i.e. their data changed) nearly as many times before something breaks.
So tunneling is not the only way to charge and discharge flash memory but a hot-carrier approach typically requires higher voltages (thus less data per chip because of heating) and so-called “hot carrier effects” degrade the quality of the insulating layer with use leading to a device with a shorter lifetime. However, that being said, of the two major types of flash memory that exist (what are called “NAND” and “NOR” flash memories) one is “tunneling charging – tunneling discharging” but the other is still “hot-carrier charging – tunneling discharging”. So it’s not strictly true to say that flash exclusively always works through tunneling (it’s either 100% or 50% of the operation depending on NOR or NAND).
But hopefully that gives a little bit of a fuller discussion of how quantum tunneling relates to flash memory than there was time for in the video.