Hello, I'm Kobi Hanokh, CEO of Weebit Nano.
And what is Weebit Nano about?
Weebit Nano develops non-volatile memory technology called ReRAM. ReRAM is a technology that is replacing flash now. We're focused on the embedded space and we're very happy to show here now our progress, both commercial progress with our customers and also the AI demo that we have here.
ReRAM is actually a back-end-of-line technology. So, unlike flash which sits on the silicon substrate, ReRAM comes between the metal layers. And basically, in a very simplistic way, between two metal layers you have a porous material. When you apply voltage in one direction, ions come in and create a short circuit which is low resistivity or a logical one. When you apply the voltage in the other direction, the ions move apart and that's a logical zero.
Can you explain a bit what we are seeing on the screen?
So, here I guess it's right now it's our name. It's basically talking about the advantages of ReRAM. ReRAM has many advantages over flash technology. It is much faster than flash. It is much lower power than flash, lower power consumption of course, better endurance. I think the two most important things are it is cheaper to manufacture and it can go down to the small geometry. So, unlike flash that gets stuck at 28 nanometer, Weebit's ReRAM can go down to very small geometries.
And is it possible that also what's called, you can maybe you are focusing not so much on the storage capacity but more on the low cost, affordable kind of... does that make sense also?
Weebit is focused on the embedded memory market right now. So basically on the companies that are doing systems on a chip and want to embed a non-volatile memory into their design. So we actually have two kinds of customers. We work on the one hand with the foundries, the fabs and we teach them how to manufacture our technology and we work with them to go through all of the testing and eventually qualify the technology for mass production. On the other hand we work with the fabless companies and we give them a license to embed a module with our technology into their chip so that then they can go to the foundry to manufacture.
Non-volatile memory. So...
That's a memory that doesn't lose its contents even when you unplug it from power.
It doesn't lose its contents. Yeah, so if you think of the SSD drive in your laptop or a USB stick and if the battery in your laptop runs out you don't want all the data to disappear, right? So that's where NVM comes in. It's actually a very important technology because if you think about it basically any device in the world you have that magical moment that you turn it on and you're expecting there to be some software there or firmware or whatever, some data, some security keys, AI coefficients, whatever. All of those things need to be stored in a non-volatile memory that they're there even when there wasn't power before.
RAM is volatile, but RRAM is not. RAM is random access memory, and you have different types of RAM. So, you have SRAM or DRAM that are volatile, and you have like RRAM or flash, which are non-volatile.
And there's a lot of talk in the industry, if we can start talk go over here. There's a little bit of nervousness happening because of the memory problems and flash storage problems and all that stuff. Do you help solve some of that issue?
Yes. So, yes and no. I mean, a lot of the nervousness is around what's called discrete chips, chips that are 100% only memory, and there's a huge demand now for those in data centers and in different places. However, there is also a growing demand for embedded memory because of all of the AI stuff. You also want to have, for example, inference. In AI, you have the learning, which is done at the data center, and you have inference, which is now done mostly on the edge devices. And those edge devices are normally embedded memory. They are systems on a chip that need embedded memory in it. So, we provide that memory for them, and we actually have very good solutions for AI, both what we call near memory compute. So, that's replacing the memory that's existing today in those chips with RRAM because flash cannot scale to small geometries. So, if you want to keep your coefficients, you actually... and since most of these chips are below 28 nanometers, they end up having two chips. They end up having the AI chip and then the flash chip which includes the coefficients and then you load them when you boot, you load them into the AI chip. The thing is two chips obviously are more expensive than one and boot time takes a long time to copy everything and during that copying you have a security issue because it's so easy to eavesdrop. So a lot of issues with that. The simple thing that people are already doing today is replacing the SRAM in the AI chip with ReRAM and now it's a one chip solution. It's instant on when you turn it on. The security issue is obviously gone. And it's non-volatile so that's true because now if the sensor isn't sensing anything you can turn the system off and you save battery power. So that's something that's being done right now. Moving forward, the next big thing in AI and you can hear about it all the time is in-memory compute. ReRAM is a great technology for in-memory compute. ReRAM is actually an analog type memory which is very cheap to manufacture. We have here a demo. What you're seeing here is a demo of using ReRAM in AI and this is a gesture recognition type application that was done by a company called E-MAS. So again, that's where we're heading. Weebit recently set up a systems and AI division and that team now is working on providing the solutions for the AI market. We will be having those solutions, the initial solutions ready already this year so that's another very exciting growth path for us.
So the ReRAM is... am I understanding correctly that it's directly on the die together with the main SOC?
Yes. Well, that's the good thing about it, that it can go down to the small geometry. So, you can actually...
