Tip:
Highlight text to annotate it
X
[MUSIC]
>> So here's how the
band diagram would look like. So, this is the
substrate And remember I told you that two dielectric are the stunner dielectric.
And this is your thicker dielectric or your interpoly dielectric.
Sometimes it's also called as a blocking dielectric.
So when you apply a program, when you program a
cell, you apply a very high-voltage on your control gate.
So Applying high volt when it comes down on this band diagram, and you get very
large amount of tunnel current in and you get a less amount of tunnel current out.
The reverse happens when you will erase the cells.
When you
erase the cell you typically already have electrons stored,
and you apply a negative voltage on your control gate
so it pushes that up and that essentially kicks
in and kicks out these electrons from your floating gate.
The other way, there's also that holes
come from your substrate into your floating gate.
Both are the same effect. They essentially remove negate
the electrons present in your control gate.
So, this is, you know, the simple equation that your charge is essentially
is build up integration of the different currents or tunneling in and out.
And this how a particular transient looks like.
So if you apply a fixed scale voltage, your charge
builds up over time, and either you can achieve a high
vT depending whether you're programming, or you
can achieve Negative vt if you're trying to
erase it and this difference between your program
and erase vt is your total program window
[NOISE]
.
I think I would recommend you to go home and try you know, take your flash drive
your USB drive and take a one gigabyte file and Just copy it into it.
And read the rate at which it's writing into it.
And then copy back to your hard drive and then read the rate at which
its account either the time or, many
a time your operating system also displays the
rate at which transferring.
And you'll find that it's much, faster to
read as compared to write into this flash drive.
Also, you know when you, it
is
[UNKNOWN]
because it's just changing it from the file system.
But usually it's really time consuming to, erase your flash card, as well.
Let's see.
So this, this top dielectric that is ipd, could
either be just a single dielectric that'll adjust to oxide.
Or it could be multiple layers. And typically what people use is ONO.
And the reason they use ono, if you want to.
Higher dielectric constant, and nitrite has
a higher dielectric constant than, than oxide.
So that's why it's ONO. So when you when
you are when you have programmed your cell, what you want to do is
you remove the voltage you apply on your gate, so this term goes to 0.
And this is negative charge, so that the lapse of negative voltage
[UNKNOWN].
Negative voltage
[UNKNOWN]
up in this Venn Diagram. And you want to at this
particular state, you want to retain your retain your
state of the cell and typical specs range between ten years to one year.
But, if you have an iPod, and you
downloaded music, you don't want it to disappear the
next year you turn your device on, so you want at least a few years of retention.
And so, what is needed is that at this particular voltage,
which is a couple of volts, you want your tunneling current to be very high.
So this turning current can be brought across here at
the analog sign and brought across the inter-pole dielectric .
And that inter-pole dielectric could just be either be
one oxide or it be this or and or.
And silicon nitrate has a
buying gap as compared to oxide.
So if I just use nitride I get a higher turning current out.
So I use a composite dialectic to, to essentially give me a better retention
or even to, when I'm programming it, to block these electrons from leaking out.
So normally we will put a lot of these band structures in your
problem set to, so you, so you can practice with these, or however,
[UNKNOWN]
program it is.
The other thing which is important is for
your flash memory endurance or your cycling so
what it means is that how many times you can keep on reading and writing your cell.
So what happens is that initially you have this large program it
is right now between your program
threshold voltage and erase threshold voltage.
But over time if you keep on, cycling it, that is
if you keep on programming and erasing it, it starts to close.
And the reason why it happened is because you get
between in your tunnel dielectric, you tend to trap these electrons.
They were trying to go from your channel to your floating gate,
but over the period of time, there's built up some trapped charge in
your tunnel dielectric, and that pushes your arrays to be
higher, because you have trapped charged in your tunnel dielectric.
At the same time, they prevent electrons from tunneling.
To your floating gate.
So the total charge your floating gate can store goes down.
So your program state goes down, your erase state goes up.
And that's also called as a window closure.
And remember this, you know, these
materials were not meant for conducting current.
They were God made them to be insulator but we are you
know, playing with the property even flow, flow current in and out.
They don't seem to like it and they degrade over time, right?
So, another thing which happens is that if you keep on turning, if you keep on
turning current In and out of your device.
You get this phenomena which is know as Stress Induced Leakage Current.
So you get multiple of these, traps placed, very closed to each other.
And now since, you know, there's not just one of them but, they have formed a gang
and, Many of them, they can combine together and give you a part for current.
So what
happen in your low voltage state you will suppose to get really
this dash line low current but you tend to get this higher current.
And that's really bad for retention because when we are retaining the state as
I describe you, we are floating it their lapse of potential you know, somewhere
in this range and this stress and use leakage, really tries
to limit your retention time.
[COUGH]
So, this is described over here. When, from a retention state,
at any warrent a low floor for my turning current, but as I
get more and more of the stress and use leakage, this current in the lower
[INAUDIBLE]
goes up and that tends to limit my retention requirement.
And that tends to limit my retention requirement.
So I am no longer able to meet that ten years of retention requirement.
[NOISE]
Alright, so this is another band diagram showing that.
So if you, initially if you were, you haven't done any cycling,
you had a very good for sell, you'll have only limited number of turning parts.
Worse is if you cycle your seller if you program any,
let's say 1000 times, you get these trap states in your dielectric.
And that and that tends to
give you this additional component which is
trap-assisted tunneling or so silicon then that
tends to limit your limit your the amount of time you can store your charge.
[COUGH]
So the last concept I want to cover is the, Multi-level cell operation.
So remember your threshold voltage is linearly related to your charge.
So, why not store more than one state? So, this is my window that
[UNKNOWN]
this is my 0 state, this is my program state.
And I think, you know.
I have this large window, why can't I fit my
[UNKNOWN]
cells between them. And that's exactly what people did.
So instead of storing one bit, if you want to
store bit, two bits, you essentially need four of these levels.
So peoples developed you know, you can store four different amount of charge.
And based upon that you get four of these levels
and you can essentially store two bits in the same cell.
And that, that's what's called as multi levels and
people they're still available in the market which are three
levels per cells or to call this TLC, so
for TLC you need to depart three or eight levels.
And believe it or not, there are cells available will store four
bits per cellular, so for four bits in one cell you need.
16 of these levels. And these
levels have to be separated and this is not just one level, it's a distribution.
So these distribution needs to be well separated from each
other. They don't need to close when you cycle
the cell, and that's, but still it's, it's doable and people do it.
So, this really started in 1992, and 1997 was when the first multi-level cell
was shipped.
And I think most of your iPods and your iPhone contain an MLC device.
It's not it's not a one-level cell.
And the downside is that, since you are multiple of your, these levels, your
signal-to-noise ratio, increase, decreases, and also it
becomes, very difficult to separate these cells,
especially when you cycle your device. But nonetheless, this is the
first chip that was, which shipped with the multi-level cell.
And, you can see that this was 1997, and this window between your zero state
and your program state, your array state
and your program state, was around three volts,
and the, this corresponded to around 30,000 electrons stored.
between your program and erase
[INAUDIBLE]
.
This was back in 1997, when the first MLC product shipped.
And people were able to store four levels
in it, and each of them corresponded to around
3,000 electrons, and if you have 3,000 electrons
then you where to retain them for ten years.
It means you need to lose less than 300
electrons a year or less than a electron per day.
So, as you can see it's already a very stringent requirement.
But, none the less I am not really making this up, this is a actual
paper from Samsung, it's shown the recent of these memory and the way
this was done was you apply different amount of voltages on your control gate.
So if you apply 15 volt you reach one state.
If you apply 16 volt, you need another threshold voltage.
And that's the way people used to achieve this multilevel cell operation.
So you're a controller which is controlling this flash
[INAUDIBLE]
now has to generate multiple of these thresh-, multiple of these voltages.
And that's usually not a very easy thing to do.
But look at what's happening in the current technology, so,
this is a total number of electrons that you store between your program and the
[INAUDIBLE]
other function of the fin technology node. So at 20 nano-meter nodes,
you only have around 500 electrons in total between your
lowest, the lowest ratio state and your highest ratio state.
And if you only store ten levels or 16 levels into is,
what you get is essentially ten electrons for each of these states.
And, you are, if you to
retain them for ten years, that means that you will lose one electron over a year.
So, it has gone from losing one electron per day, to losing one electron per year.
But still, you know, I find it hard to
believe it that actually people are able to do that.
But, this is important that you know for a 20 nanometer cell, you only store
around 500 electrons.
And first state you only have ten electrons.
But this, you know it comes with certain trade off
and since you have these many levels you significantly degrade
the number of times you can Cycle your cell the
number of times you can program an area of it, so.
Typically, a 9 flash sale, you can program it tend to par five times.
Or you can program it either 100k times. So,
if you have a might 11 cell you can program
it only, when you're at 10,000 times many cases if its like a
[INAUDIBLE]
cell, you can only read and write into it less than 1000 times.
Also retention is much poorer.
It's only one year instead of ten years.
It's also slow.
If you are doing Since you have very bad retention, and programming is cycling,
you have to add more error correction bits to compensate for it.
So, this is showing your, some actual data showing that if you have a single
layer cell you can program it 100 K times, if you have
multilevel cell, two bit per cell you can program it only 10K times.
If it's actually three level per cell it's less than a thousand
times, and you also have to add more and more error correction bits.
And also it's slow.
So all those companies that I showed you earlier, which were working on flash
memory storage, what they are trying to do is they are all recipe companies.
They,
they take these different amount of some storage here, but allows a
number of storage here, and even large number of storage in in
[UNKNOWN]
cell.
And they come up with algorithms, how to achieve, or our system,
which is faster, but at the same time, it low in price.
[MUSIC]