Subtitles section Play video Print subtitles [MUSIC] As that news article said it has been 25 years since NAND flash memory has been around. So, what I wanted do is give you a 5 minute you know, plug-in of what has happened in this 25 year period. So, these memory cells were available before 1984 as well. You, you can program a non-volatile memory. But to erase them, you had to place them under UV light. So these were you had to simply. I don't know if any of you had used these chips. But you had take them under a UV light if you wanted to erase them. Then in 1984 this gentleman Masuoka san, from Toshiba, he invented this flash memory cell, and it was of the NOR architecture. He presented in the IEDM conference that year. Earlier that year, he applied for a patent for the same. And this was you know, the first cell which you could program using electrical signal and also erase it using electrical signal. But this was not a kind of memory. same set, same bunch of characters, they again target you know this weakened increase the density. And they they came up with using this [UNKNOWN] fashion, this was presented in 1988, [UNKNOWN] in the same year. And you can see that [UNKNOWN] it's was, it just the number size of it has gone up but it essentially that same architecture. And Mr. Masuoka's son was actually, instead of being rewarded for his invention, he was he was demoted from his position and that you know, so this is something very particular to Japanese culture. I showed you that the election on [INAUDIBLE] he somewhat resigned from Hitachi. And he too you know, didn't get any benefit for his innovation. And so there was [INAUDIBLE] awarded him a very big award in 2002. But inspite of inventing this multi-billion dollar market. His parent company did not give him any rewarded, [UNKNOWN] a nice article about him. Now, he is a professor at Tohoku. So, I find this you know, very I, I will tell you another story when we get to LED, that was also invented by a Japanese inventor Nakamura San. And he too, was actually demoted because of that invention, and he, he left that company and sued them later. So this is this is something you know, you find very contradictory to the culture in the valley where you are if you are a nail sticking out you get hammered in. If you're in that culture as compared to here, where you, you get rewarded and you can raise money to pursue your idea. But, anyway, the, the next thing which happened was in 1997, people realized that you know, we have this cell, why not start storing more bits into it? So [INAUDIBLE] cell Not really invented, but came in to production in 1997. And most of the cells, that you see in your iPads and in your iPhones are multilevel cell. And then from 2002 to 2010, which is really the post PC era. 2002 it started to just the beginning of it, and since then, the flash memory capacity has doubled every year. So, it's also known as Hwang's Law, and it's a more aggressive law than Moore's law. Moore's law says the capacity should double every 18 months. This law is at your total capacity that you bu, can buy in one chip, not one die, in one chip, should double every year. And in fact, it has been progressing at, at, at least progressed at that rate from 2002 to 2010. It was named after this guy Hwang Chang Gyu, he was a president of Samsung electronics. And Samsung is one of the dominant players in flash memory. Both flash and DRAM memory. Back in 2003, it also also became cheaper than DRAM and it. Nowadays, in most of the memory systems you very limited amount of DRAM. But most of your storage happens in in your nine, and that's because it's, it's an order of magnitude cheaper than DRAM. And 2000, in this year, you can buy, one single die, which has 128 gigabit or 16 gigabyte of storage in one single die. You can also buy this tinny chip which stores 120 giga, gigabytes. So most people you know, their hard disk, most of the storage requirement can be met by this one single chip, which has almost the capacity of your hard drive. But still remains much more expensive than hard drive. That's why hard drives have been so hard to displace. And it's still an order of magnitude expensive than hard drives. And that's because hard drives prices also keep on falling. So, I described these numbers, right, that nine flash cost a dollar a giga, a gigabyte, or 70 cents a gigabyte. So. But, I was not describing those numbers for a, a single level cell. But those are for a two bit two bit [UNKNOWN] that's most commonly used in most of your iPhones and iPads. Alright. So, any, any questions on history. Alright. So, let's look at how do we makes these make these chips. Right? So let's look at the process technology. It's of course about manufacturing, so. So this is how, if you open one of these chips up, it would look like you see these these break lines. And they'll be separated by these very high aspect ratio trenches. So a quick question to answer that you know, we have learned about lithography. So looking at this picture. what do you think, which lithography was used? If you look a, a little deeper and you think about it, you see these two different trench depths. So clearly you know it looks like the use double patterning technique. And this is not a chip from 2012. It's actually a chip from 2007. And for flash memory has been using double patterning for a while, before even logic started using it. So, if you open up your iPhone as we did in our first task, you see that this big chip that is your nine memory chip. And whether you buy 16 gig phone or you buy 32 or 64 gig phone, you always get that one single chip. So why, how is that possible? Right? So the reason why it's possible is that, if you open up that chip, it's not just one die but it's multiple of these dies stacked on top of each other. So, you can have up to 32 of these dies stacked on top of each other. Each of these would be eight gigabyte or somewhere in that order. And there you know, they are still connected using these wire bonds. And we'll talk about, when we talk about when we discuss packaging in in lecture number six that how these are connected together. And it is quite amazing, that you know, that 32 of these chips and you have these wire bonds and none of them sharp to each other. And if you look at each one of this die, it's essentially nothing but a large area of memory. And often memory are divide, is defined by this term, which is called array efficiency, that is, how much die area is covered with the actual memory cell. So, a DRAM, for example, has an area efficiency of 60%. nine flash has area efficiency of 90 to 80%, that is more than 80% of this dye is just this die is just this nine [INAUDIBLE]. The rest, the 20% of this peripheral circuit circuitry, your charge pumps that would be required to generate that high voltage. And your row and page decoders, but still 80% of the area, it's nothing but these banks of memory. Right? So, how do we make these things? I [UNKNOWN] for that, it for making this a pretty simple device, what you need to do is pattern bit lines, and then pattern word lines. And for each of these intersections, you get a cell below them. Right? So let's look at some of these intersections. So when you pattern these bit lines, you essentially separate them with a shadow trench isolation. Each of them is guiding such a high voltage, you isolate them so that the way deep STI edge to separate them. And when you draw these word lines, these board lines we'll these, they tend to overlap and flow around this bit lines. And I'll show you why that is the case. So, let's look at a few of these little cross sections. This cross section I already showed you. This is a cross section taken along this direction. So, if you have the multiple bit lines and if you take a cross section like this, what do see is these bit lines and they're separated by this very high aspect ratio STI etch. So this is a very sophisticated, not sophisticated but a very high aspect ratio structure that you need to etch. Then you also need to fill it, and that's what separates these different bit lines. And on top of these bit lines, you can see a cell over here that this this is your, where your cell is located. Right? Now, what if we looked a cross section along this site? So if you take a cross section along your, along one of your bit line, that is intersecting multiple of these word lines. What you see in this cartoon is essentially these multiple of these transistors. Right? Or multiples of these apartments, which are connected in series to each other. Right? Or, if you draw if you draw a more device friendly kind of picture, you see these multiple of these gate stacks. And these transistors connected in series to each other. Right? So this is a actual picture which I showed you earlier other side shows. This is a picture taken along one of these word lines, so you see multiple of these transistors. And at the end, you have something connecting, but in between them essentially, it's just a string of transistors and they are just connected in series to each other. And to read anyone of them, I need to turn on all these others so I can read what's in here. Right? So let's look at, a closer look at how it looks like when I look along a vertical word line. So when I look along a vertical word line, this word line is wrapping around my different bit lines. And you can clearly see over here, so this is the [INAUDIBLE] of one of the STI etches. And I have my tunnel oxide over here, and this tall thing is my floating gate. But what I see as my word line, or my control gate, is wrapping around it. So I see this [UNKNOWN] this is my interparty dielectric. And this other whole thing on the top which is wrapping around is my control gate. So the reason why it's wrapped around is because you want to, remember our GCR which I said, we want to keep it as high as possible. So we want to keep this word line capacitance with my cell as high as possible. And one way to increase that capacitance is this wrap around that contact around my gate. Right? So I get more area and I get more capacitance, and that's how people have been doing it. So this word line wrapping around your cell is to essentially achieve a high gate coupling ratio. So now you can see 1, 2, 3, 4 of these STIs standing out. And then you have your tunnel oxide over here. This is your floating gate. And then you can see you this n or n dielectric RER control gate on top of that dielectric wrapping on on top of these cells. Any questions on this one? So actually this [INAUDIBLE] has accelerated this flash memory so much that it, you know, it has developed far beyond what people predicted it to. It has scaled far beyond what people expected to happen. So this a picture from the ITRS road map of 2006. And it says that you know, in 2012, we'll have a memory available with 32 nanometer feature sign. And was circled in red, so people didn't know how it will happen. Right? It was so uncertain back in 2006. But since economy drove it. Since the, iPhone, iPads drove it so much, actually this happened much before. In 2000 12, you can buy a chip which is 20 nanometer in feature size and 64 gigabit in one single die. So it's, you know it's hard to predict the future. And so it's all driven by so this is how aggressive your [UNKNOWN] device that have driven the flash memory. Right? And since it's a very simple layout, I showed you a layout that's essentially nothing but these very periodic bit lines and word lines. So it's a very simple structure to write lithography on. There's no grating, I mean, no, there's no cut required. All that's required is these gratings. so this in fact has become the driver for lithography as well, so NAND flash was the first to use our double patterning. It has been using it for a quite a few generation now, and it's right now they are the first to use quadruple patterning as well.