ReRAM Poised to Surpass Flash

Coby Hanoch, CEO of Weebit Nano Ltd, developer of ReRAM memory technology that’s 1,000X faster, 1,000X more energy-efficient, and 100X greater endurance than existing flash technology was interviewed during IT Press Tour #35.

The non-volatile memory market (NVM) is currently estimated at $60 billion and is expected to grow very rapidly. It’s the fastest-growing segment in semiconductors and expected to grow to $82 billion by 2023 and over $100 billion by 2025.

Why is ReRAM is poised to surpass flash?
Memory is at the center of the digital world. Practically anything we use has semiconductors with memory. The demand for memory is growing exponentially. Semiconductors are the foundational elements in advanced electronics and communications. Chips are used for every electronic device, including computers, cell phones, autonomous vehicles, security cameras, etc. The digital economy is highly reliant on semiconductors with 7 of the 10 largest companies in the world in 2019 were digital/semiconductor companies. Whereas, in 2009, only 1 of the top 10 was a digital/semiconductor company.

These companies are huge and most offer cloud services and have massive NVM server farms. And actually we already have 2 Chinese companies among these.

The Chinese angle is an important one for Weebit. China is, is trying to become the number one company in semiconductors. They decided this a long time ago. They’ve been pouring a lot of money into semiconductors. And nowadays with Trump giving them a few slaps in the face, it just encouraged them even more. So when we come to China it’s really amazing to see just how much they’re putting into semiconductors yet the gap between them and Taiwan and South Korea just growing rapidly despite all of their investments. If you look at what are they actually consuming, practically 60% is from Hynix and Samsung, it’s memory.

When we come to China, they look at us and they say, ‘Wow, you guys have emerging memory technology.’ It’s obvious to them that trying to compete on the existing technology on flash is a big challenge. And they won’t be Hynix or Samsung, if they try to compete on flash. So the Chinese strategy is to look for forward and say eventually the emerging technologies will take over. Let’s just focus on them. Let’s get these things working well. And once we get to using emerging technologies, we’ll be the leader.

It’s similar to what they did with cell phones back in the 90s, they could have invested in landlines but they just skipped and went straight to cell phones, and now they’re clear leaders in this domain.

China is currently building now 30 semiconductor fabrication plants in parallel, it’s just uncomprehensible All of these new fabs need non-volatile memory – for them flash. But flash is the front end of the line technology. It’s a technology that actually sits alongside the design. So when you want to integrate flash into your fab, it’s quite a challenge. And for every technology you want to put in, you need to adapt it to flash. ReRAM is a back end technology, we come in the top in the upper metal layers, and we don’t really disturb the design. So you just need to adapt ReRAM once into your fab or into geometry and you’re done. This is a big advantage when you’re talking about ReRAM vs. flash.

How does ReRAM stack up vs. flash today?
ReRAM is well-positioned to replace flash with faster write time and lower power. ReRAM is faster and more energy-efficient than the existing flash memory It has several major benefits over flash as it is used for both embedded memory and for discrete memory chips Unlike other ReRAM and emerging technologies, Weebit’s is based on standard SiOx, making it easier to manufacture in existing fabs – cheaper and more efficient to produce. Weebit’s ReRAM can be easily integrated with standard CMOS computing devices – critical for embedded modules.

What’s on the horizon for ReRAM?
This is more long term, but it’s what’s called neuromorphic computing. So, it turns out that ReRAM functions in a very similar way to a synapse in your brain. This has triggered a lot of research in many different institutes. Today it’s more of an R&D activity in universities and research institutes. But this enables us to emulate the brain. So, if you look at what’s happening today, with all of the neural network’s activities, we’re basically simulating the brain we’re running on, either standard processors or more special processors. We’re trying to simulate the process that goes into the brain and that’s not a very efficient way to do it.

If you can emulate the brain in hardware with rigor, and you can be probably orders of magnitude more efficient, both in terms of speed In terms of power, etc. So so there’s a lot of work going on into this. There are many institutes, Leti is one and there’s Politecnico di Milano. There are other institutes all over Europe that are looking at this the Technion in Israel. IIT Delhi is the partner for us, we actually started a program where we are cooperating with these institute’s we give them rearm arrays so that they can develop their technology and we have access to what’s going on there.

My goal is that once this becomes a technology that can be productized that will be there will be ready and we’ll be able to lead the way with it. So very exciting for us. This is going to have a significant impact on AI in the future. And what you can see is a little bit of a comparison between the human brain and the big computers today. People expect that with ReRAM, we’ll be able to be much closer to the human brain in terms of the volume, the power and the speed that is needed to perform the same functions.