talked about this multi-level cell or MLC technology.

And I described to you how cool of a technology it was.

That it essentially allowed me to store twice or thrice or

sometimes even four times the amount of data in the same cell.

So it was clearly a technology meant for

tough economic times or you know times of recession,

but I, what I want to do in this video is to talk a little bit about

what is the trade off?

What is the trade off if I store two or three or four bits per cell.

So how does the performance change as compared to,

you know, storing a single level in the cell?

And to do that I need to introduce two new terms in your vocabulary.

One is called the endurance or or also called as cycling, and the other

term I want to introduced is stress induced leakage current, also called SILC.

So let's look at what these terms mean first.

So the first term I want to introduce is called endurance.

it's also sometimes called cycling, and cycling here

has

you know little relationship with what Lance Armstrong used to do.

But, what cycling a cell or endurance of a cell, is, means is that, it's,

each time you program or erase the cell,

each of that operation is called, cycling operation.

And typically for these single-level cells you have up to 10 to power five.

They're meant to, you know,

last up to 10 to power five cycle.

Or you can program or erase them 10 to power five times.

So what do I mean by this endurance or these, cycling?

So what happens when you a programmer, erase these cells

And you inject these electrons out of the cell when you want to erase it.

So you are flowing these electrons in and out, using,

you know, by doing this program/erase operation in the cell.

And what happens is that it degrades the quality of this tunnel oxide.

So remember

these tunnel oxide as well as these inter-poly dielectric or IPD oxide.

These are insulator materials.

Silicon oxide, or, you know, similar dielectric materials.

And what we're doing here is we are essentially

flowing current in and out of these tunnel oxide.

So we are treating them, as conductors when they are really insulated.

So these material don't tend to like that, and

what happens is that over a period of time when

you keep on programming and erasing the cell when you.

you are essentially flowing electrons in and you are flowing electrons out.

Some of these electrons get trapped in this tunnel oxide.

And what that does is that it it changes your

programmer/erase window so now you have some electrons already trapped in

your tunnel oxide.

So your VT or your threshold voltage of your erased state is essentially

goes up because there's already several of trap charge stored in your cell.

At the same time these electrons, they prevent these, these electrons which are

trapped in your oxide, they prevent other

electrons from flowing into your floating gates.

So, you are, you are able to store less number of electron in your program state.

So your program VT degrades, so your

program VT essentially degrades or it, decreases.

And your erase VT increases.

So in turn what happens, is your program/erase window which is the determined by

this difference between my program. And erase state that window starts to close.

Or this window is essentially, it's decreasing.

And so these typically these, cells are build such that they can

survive this program, erase cycling to up to 10 to power five cycles.

Or they are meant to endure this program erase operation.

To up to 10 to the power of five times. The other term or

you know the other phenomena which happens when you program and erase these cells is

a stress induced leakage current, which starts to happen.

It's also known as SILC, standing for stress induced leakage, current.

So what it means is that when I keep

on programming and erasing my cells, I'm stressing these tunnel

oxide materials because, you know, they are not really meant to be conductors.

And when you are flowing electrons in and

out, you are essentially stressing them and you,

they don't like it so what happens is

that these trapped states are developed in my dielectric.

So it creates these trap states in my insulator material.

And it's you know, it's not as

bad if there's, there's only one of these trapped states.

But these trapped states when they line up in this manner as shown over here.

So they can what they can do is they can allow tunneling of electrons

through these trap states, so these trap states can assist the tunneling

of electrons from my silicon into my floating gate.

Also

from my floating gate back to my silicon. And this especially

plays a lot of havoc at low voltage state because ideally if I was in

a, if I was applying a low voltage between my silicon and my floating gate.

I would expect a little or no tunneling but

because of these trap states which build up due

to cycling of my device and the stress which comes with my cycling.

Instead of getting zero current, so I now get a

substantial current at these low voltage at across my dielectric.

And this phenomenon is known Stress Induced Leakage Current.

So this was the ideal case where you know, from a theoretical, from our

or direct tunneling model you would expect, you

know negligible current less than a Picoamp of current

but because of these stress induced leakage.

All these traps states, you get this high

amount of or you get this tunneling current.

And this tends to again also degrade degrade the performance of your cells.

So suppose you had a retention requirement of 10 years and you

had a lot of these trap states created during programming or

erasing your cell.

Your charge can now leak out, you know in less than a year.

So now that we know about endurance and, cell

, let's look at what happens to the

distribution of, threshold voltages as we cycle our cells.

So in one of the previous videos that I talked, you know, that our threshold

voltage, it has this Gaussian kind of distribution for each of

this state, so this is my state zero, this is maybe my state one, two, and three.

So it has, when I initially program

this cell, it has this tight distribution around a narrow,

set of threshold voltages, but what happens as I cycle my cells?

So let's say this was my state you know, at my very first time I program or

cell.

so when I cycle this cell, after, let's

say, 1000 cycles, this distribution becomes more wider.

So it becomes maybe let's say after 1,000

number of cycle it becomes something like this.

When I have a larger range of my threshold voltages because as I

talked about you know, I'll be building up some charge so it will

degrade my erase VT and you know increase you know, degrade, increase

my erase VT and degrade my program VT so as a result.

On a statistically speaking, I'll get a larger threshold voltage distribution.

And as I keep on going even further, maybe you know, I'll get an even

larger distribution when I'm programming it 100k times 10 to power of

five times.

So or, as I keep on cycling myself there's a distribution of threshold voltages.

For one particular state it keeps on increasing.

And this was, this have been fine if I had, you know, just two states and had a

very large separation between these states but as you can see if if I have four

of these states.

When I when I, you know, keep on cycling them,

each of these states will, you know, expand in its distribution.

And if these distributions starts you know, coinciding

with each other, then I have a problem.

And if this was a problem, if you think this was a problem in a two

bit per cell, let's look at, you know, if I had three bits per cell.

So if I was storing three

of these bits file.

We be having, two to power three of these are eight of these states.

And now each of these states, as I keep

on, cycling them, they will set each of them will, will

have, you know, this is very first time

I program or erase them and then this is maybe after.

thousand times and this is after

ten to power five time.

So you know, you can can clearly see that if I if I program three bits

per cell or a TLC cell for ten to power five times, I'll

be getting these overlapping regions which I don't know which state my cell is.

Whether it's in slate three or whether it's in state four.

So one of the major limitation you get

when you store multiple levels in one single cell is that you can.

Only program and erase them for let's say 10 to power

four, or 10k cycles, or 10 to power three times, or 1k cycles.

And this is summarized in this table over here.

So as you can see here, if I had a single level cell at a you know at 50

nanometer technology or even at a 30 nanometer technology,

I could cycle it 10 to the power five or 100 k times.

And I could still distinguish between these two, to my program.

And it is state which had very large seperation between them.

But now, let's I have MLC cell with two bits

stored per cell. So I have these states very close to each other, and I can

only program and erase it in this case you know 10K times.

And again this degrades as I keep on scaling myself, if I have if I go

to a 30 nano-meter technology that leaves me only with even

fewer electrons and then I can program and erase this cell only 5k times.

And if I go to a 20 nanometer technology this number is 3k times.

So and if I have, even if I store three bits per cell or if you know, if I have

if I have that means you know, two to power three or eight levels in my cell.

I can, you can see, I can only program

and erase it over 2500 times and you know, it's less

than a 1000 times if I go to a 20 nanometer technology.

So this is you know one of the main main Bottlenecks or one of the main

trade-offs that if you have a multi-level cell, if you

can see over hear that, if you go from a single-level cell You could program it

a 100k times.

versus if you go to 50 nanometer MLC cell, you can program it only 10 k times.

And if you scale your scale, if it's a 20

nanometer technology, you can program and erase it only 3,000 times.

And to you know, to, to kind of, you know mitigate the impact of this.

What you do is you increase the number of error correction, code.

So you store these extra bits which, correct for

the state of the cell in case something goes wrong.

And you really didn't want to lose data, so you.

introduce this ECC error correction a, you know, a bits.

And these are the number of bits that you have to introduce, which is essentially

a penalty on your overall chip performance that keeps on increasing.

So your ECC requirement it again keeps on increasing if you store multiple

bits per cell. There are other trade-offs as well

So, if you store a multiple of these levels

in one cell, you have to read, or you

have to, you know, essentially distinguish between this multiple

of these levels here or even to read one

of these cells, you essentially need to distinguish, earlier you just had to

distinguish between two levels, now you have

to distinguish between four or eight levels.

So if you're doing a random read, you know

your controller takes some time to figure out, you know,

what state your cell was, so your, so if you're

doing a random read, your read, time increases as well.

So, as we start to

summarize, you know, we, to compare my MLC versus single level cell technology.

my MLC certainly has the advantage that you know,

it can it can, it's very economically viable cell

it can let me store you know twice or

thrice or four times the data on the same chip.

But, on the other hand it comes with these

trade offs that you know, I, my number of cycles that I can program or erase

it degrades from, you know, 10,ah, 100k, to, as we saw,

less than 10k for, you know, a two bits per cell, and, you know, even less

than 100, less than 1k for, for a triple level cell.

at the same time it's slower to read

so that's another trade-off.

So all in all it comes, you know it's a mixed bag of results, you