Beyond Flash: The Future of Non-Volatile Memory

A Conversation with Coby Hanoch, CEO of Weebit Nano

Flash memory has long been the leading trend in non-volatile storage. However, as data demands increase, its limitations are becoming more apparent. Flash memory writes data in large blocks, which slows down performance, and it has a limited lifespan due to wear from repeated write cycles. Manufacturing flash is relatively expensive, and not reasonably possible below 28nm. Additional issues such as power consumption and speed have prompted the search for new memory solutions.

Resistive Random-Access Memory (ReRAM) is emerging as a potential alternative, offering advantages in performance and efficiency. Unlike flash, ReRAM writes data at the bit level, making it faster and much more durable. It also consumes less power than flash, which is critical for battery-operated devices.

Weebit Nano, an Israeli semiconductor company, is developing ReRAM technology to address these challenges. I spoke with Coby Hanoch , CEO of Weebit Nano Ltd , to discuss ReRAM’s potential, its benefits, and the challenges of transitioning to new memory solutions.

Aron: Coby, your journey into semiconductors has led you to focus on memory innovation. What first drew you to the industry, and how did you become interested in developing new memory technologies like ReRAM?

Coby: Back in 1977, when I enrolled in university, I actually originally was planning to learn physics. However, at the time, computers were just starting to emerge into people’s everyday life, and a friend convinced me they were the future and pushed me to learn computer science. But, being a curious person, I felt I couldn’t use a machine without understanding how it works, which pulled me into the semiconductor world. My first job was at National Semiconductor and, while there, I fell in love with this exciting domain. I remember how in those early days the NVM technologies that were widely used were discs and tapes. Not many people know today, but the origin of the term “tape out” was when the design was ready and we would load it on a tape and send it out to the fab.

Discs and tapes were very slow, and when laptops came out, having a moving (rotating) device like a disc in them caused many challenges, which gave flash a huge push in the form of SSD drives and USB sticks. However, very quickly it was clear that flash too had significant limitations, as you mentioned earlier. So, 20 years ago, and even before that, the search for flash replacement began. I followed this from the sidelines, as I was focused more on the architecture of the devices and verifying they were bug-free. I saw how these efforts became even more intense when, on the one hand, Systems-on-Chips became more prevalent and, on the other hand, the market realized embedded flash was not able to scale below 40nm.

When I was approached by Weebit, almost 8 years ago, I felt this was a great opportunity to be part of an important breakthrough in the market. I saw the potential of ReRAM and was very excited about the opportunity to drive it forward. Little did I know that this would actually become the technology of choice to replace flash, and that Weebit would become the leading independent player in this field.

Aron: Given your background in the semiconductor industry, you’ve seen how storage technologies evolve. Flash is now pervasive, but as it reaches its limits, what makes ReRAM a viable alternative?

Coby: As I described previously, I saw first-hand how the market evolved from tape drives to discs, and then to flash. I also remember how,10 years ago, everyone expected flash to die within a few years, but then the 3D NAND came out and enabled discrete flash memory chips to survive. However, 3D NAND is not an option for embedded applications, which are becoming more and more critical. As an embedded technology, flash can’t scale below 28nm, and technologies like ReRAM are now a must have.

Actually, the market was exploring many options to replace flash, like PCM (Phase Change Memory) and its derivatives like Optane or 3D X-Point, MRAM (Magnetic RAM), FeRAM (Ferroelectric RAM), etc. However, most of these dropped along the way as they were too expensive or difficult to manufacture. ReRAM has emerged as the clear winner of this race for most applications since it is cheaper and easier to manufacture.

Some companies tried to manufacture ReRAM with esoteric materials, but they are no longer around, and Weebit, which is focused on fab-friendly materials – those most commonly used in fabs – has a clear advantage. In addition, ReRAM is much faster than flash, consumes significantly less power, has much higher endurance, and is radiation tolerant by nature. It is also a BEOL (Back-End-Of-Line) technology, manufactured between two metal layers, whereas flash is an FEOL (Front-End-Of-Line) technology, sitting on the substrate next to the customer design. For analog designs, being BEOL is a huge advantage. Flash imposes limitations on the design, often forcing the product company to implement a 2-chip solution with a separate chip for the flash.

Aron: While ReRAM presents a promising alternative to flash, adoption has not been immediate. What are the key challenges that have slowed down its path to becoming a viable commercial alternative?

Coby: Semiconductors is a very capital-intensive industry, and therefore also a very conservative one. When you spend tens of billions of dollars to build a fab, you are very cautious to bring in new technologies. The decision to license a new technology is highly strategic. After all, it’s a long-term commitment that can last decades. It includes significant investment in both manpower and money, and it requires involvement from top management. As long as they didn’t see this technology in mass production with good results, they perceive it as risky.

Product companies, on the other hand, can’t commit to a new non volatile memory before they know the fab they use knows to manufacture it and has it qualified. Time and patience are needed, but the potential payoff is significant.

Aron: One of the hurdles in commercializing new memory technologies is manufacturing integration. Given the challenges you've mentioned, how feasible is it for manufacturers to integrate ReRAM into existing processes?

Coby: I don’t believe there is anything from a technical perspective that is holding ReRAM back from being adopted by the fabs. We already have our ReRAM qualified in two fabs, have fully functional samples in 2 others, and successfully simulated it on many different geometries, down to 12nm.

When it comes to integration into existing processes, Weebit’s ReRAM is far easier to integrate than MRAM or other NVM technologies. Weebit has made it a point to use only fab-friendly materials, which are common in almost all fabs. Companies already know how to deal with these materials, and normally already have the tools needed to work with them. This enables them to avoid high cost and potential issues with reliability or cross-contamination with other materials in their flow.

Aron: As ReRAM finds its way into manufacturing, emerging technologies such as AI, IoT, and edge computing are driving new memory demands. What role do you see ReRAM playing in these industries?

Coby: ReRAM will play a key role in all these areas and others. We are particularly excited about “Inference at the edge”, where ReRAM has a key role to play. Since most AI applications are implemented at 22nm and below, embedded flash is not an option, requiring the AI weights to be stored on a separate chip – which is more expensive, increases power consumption, and poses security issues. ReRAM can store these weights on the same chip as the AI engine, providing a low-cost, simple and efficient solution. At the Embedded World conference we will show a new demo with EMASS, an AI chip company. The demo will show the power of Weebit ReRAM for edge AI. EMASS is particularly excited about the possibilities, as they are switching from MRAM to ReRAM.

No matter if you are dealing with AI, automotive, IoT, industrial, or almost any other application, ReRAM offers multiple “must have” advantages. It can be integrated monolithically at advanced processes, it is significantly faster and consumes much less power than flash, is simpler and lower cost to manufacture, etc.

Aron: What advice would you give to entrepreneurs looking to innovate in the semiconductor space?

Coby: The timescales in semiconductors are very long, so the first advice is – you need to combine patience and persistence. Everything is measured in months and years, and you need to be able to maintain your focus and keep pushing without letting go, even when it takes years to reach your goal. This is a very tough challenge as sometimes it seems like there is no progress for months. A side effect of this is that many investors shy away from semiconductors due to the timescales, so you need to look for those who understand the timescales, but also the big potential.

My second advice would be to remember that not everything works the first time and you need to balance the need to show aggressive project plans with some realistic guard bands.

Aron: Given the evolving semiconductor landscape and ReRAM’s growing presence, where do you see the technology in the next five to ten years? Will it fully replace flash, or will they coexist?

Coby: ReRAM usage will grow rapidly in the coming years. The market is now accepting it as the new standard for embedded NVM, and practically all major fabs are looking to adopt ReRAM. So, on the embedded front I see ReRAM taking over significant market share in the next 10 years, although some projects will continue with their legacy flash usage, and some might still use MRAM.

On the discrete side, I believe that in the next 10 years there will be a breakthrough in selector technology for ReRAM which will enable large capacity discrete devices, and their usage will grow. 3D NAND has, and will maintain, a very strong position for many years. We should remember that tape drives and rotating disks are still used widely in data centers as they are low cost solutions for long-term mass storage.

Closing Thoughts

As the semiconductor industry explores alternatives to traditional flash memory, ReRAM presents itself as a technology worth considering. With its potential advantages in speed, durability, and efficiency, it raises important questions about the future of non-volatile memory and its role in next-generation computing systems.

To learn more about the developments in memory technology, visit www.weebit-nano.com.