Resistive Random-Access Memory (ReRAM) is one of the several emerging non-volatile memories (NVMs) that are currently under intense scrutiny as a potential replacement/complement of mainstream technologies (Flash, DRAM, SRAM) in a host of embedded and stand-alone applications, spanning from microcontrollers (MCU) to low-latency drives based on storage-class memory (SCM).
As detailed in Yole Développement’s Emerging NVM Report 2021, ReRAM has attracted a great deal of attention because of various technical advantages, among which stand out its simplicity. ReRAM cells are simple two-terminal devices where a resistive material – commonly a thin metal-oxide film – is sandwiched between two metal electrodes in the back end of the line (BEOL). Unlike Flash, where the information is stored through the retention of electrical charges in an appropriate charge-storage layer, ReRAM relies on the modulation of the resistance-state of the oxide film, which can be easily tuned through the application of electrical voltages.
To make things even more simple, the ReRAM company Weebit Nano is proposing a resistive-switching technology based on materials (e.g., silicon oxide), tools and flows that are common in any CMOS fab. One could argue that simplicity is never a straightforward path, and that hard challenges must be overcome to turn a simple design into reality.
Recently, Weebit has provided important updates on their progress towards commercialization of ReRAM: they demonstrated the integration of an OTS selector, paving the way towards high-density arrays for SCM applications; they completed the design and tape-out of an embedded ReRAM module suitable for new products such as Internet of Things (IoT) devices; they also announced their first commercial agreement to take ReRAM to volume production.
Simone Bertolazzi, Ph.D., Senior Technology & Market analyst at Yole Développement (Yole), had the opportunity to discuss with Weebit’s Eran Briman, VP of Marketing and Business Development, to learn from the source about Weebit’s achievements. Discover the details of their discussion below.
Simone Bertolazzi (SB): Could you briefly introduce yourself and Weebit to our i-Micronews readers?
Eran Briman (EB): Weebit is a ReRAM company. We were founded in 2015 in Israel, and even though we’re a start-up, we’re publicly traded on the Australian Stock Exchange. Working with our research partner CEA-Leti, we’ve created a truly innovative non-volatile memory technology based on over a decade of research.
We have a team that is well known in the semiconductor industry. If you look at our Chairman Dadi Perlmutter, Directors Atiq Raza and Yoav Nissan Cohen, and our CEO Coby Hanoch, just those four people have 150 years of combined industry experience at companies including AMD, Intel, National Semi, Tower Semi, Cadence, and others. We are very fortunate to have their combined technical and executive expertise and benefit from the amazing insights and knowledge they bring to Weebit.
I joined Weebit as VP of Marketing and Business Development in 2020 after spending more than 25 years mixing technology with business roles in semiconductor IP companies such as CEVA and Corephotonics. I am excited to work closely with Coby and the team to drive the company through its next phase as we move out of pure R&D and into the commercialization of the technology.
Along these lines, Weebit recently announced our first commercial agreement – with SkyWater Technology, a U.S.-owned semiconductor manufacturer. We are partnering with SkyWater to take our ReRAM to volume production and, as part of the partnership, SkyWater licensed our ReRAM to use in customer designs.
SB: What are the unique features of Weebit’s ReRAM compared to other established and emerging technologies?
EB: ReRAM combines the advantages of both RAM and Flash. It’s non-volatile and more cost-effective than Flash. It’s got much better memory access time and endurance as well as lower power consumption and operating voltage than embedded Flash. It can also scale well below 28nm – we don’t see a barrier at any point – and so from an embedded perspective, it has clear advantages compared to Flash.
One of the biggest advantages of ReRAM is that it’s a back-end-of-line technology, so unlike Flash, which gets integrated at the front end, you don’t need to make compromises with other components and devices you might be developing at the front end and adapting it for a technology node is good for all its variants.
With Weebit’s ReRAM in particular, we have even more advantages that make it the best NVM in its class when you look at parameters like cost and performance. The big thing here is that we use the most common materials used in fabs today, as well as standard tools and processes. You can easily integrate Weebit’s ReRAM into any standard CMOS flow, with only two additional masks, so the added wafer cost is minimal and lower than that caused by other NVM technologies. There are other advantages too – like our ability to maintain reliability at high temperatures – up to 175oC for over 10 years. This is required by many of the companies we are talking to.
When you consider other emerging NVMs like MRAM and PCM, Weebit’s ReRAM wins on just about every parameter. This is true on tech specs, and more importantly, when you look at the simplicity of our technology and how easy it is to integrate into existing processes, translating to lower cost and lower risk.
SB: Can you provide a short overview of Weebit’s latest technical achievements?
EB: We recently completed the design and verification of our first embedded ReRAM module and taped out a test chip that integrates that module in 130nm. We’re using the test chip for characterization, qualification, and demonstrating to customers a fully functional ReRAM product that they can integrate into their SoCs. We implemented the module with smart digital and analog circuits that lead to better speed, endurance, and retention. It was implemented taking the requirements of our future memory compiler into account. This will allow us to implement the compiler faster, allowing customers to adapt the design to their needs with minimal effort.
Just before taping out the module, we successfully demonstrated an Oxide-based ReRAM cell integrated with a selector. We chose OTS for the selector technology because it enables the smallest ReRAM bit cells, in addition to high endurance, low energy consumption, and high switching speed. The OTS selector helps us achieve the density needed for discrete memory chips. This is a significant step on our discrete memory chip roadmap, which we are working on in parallel with our embedded roadmap.
SB: What are the main technical challenges for integrating a selector with a ReRAM cell? And how did Weebit address them?
EB: The main technical challenges for integrating a selector with a ReRAM cell are developing the two innovative devices (selector + ReRAM) and matching their performance profiles. For example, the selector element must allow the ReRAM cell to switch and be read without interfering with its operation. It must turn on at a specified time and then turn off immediately once switching is complete. The selector should have very low resistivity as it is addressed so that it doesn’t interfere with the ReRAM operation, and it should have very high resistivity when unselected so that it can block any leakage/sneak path current.
The development of the selector itself was a complex undertaking. We relied on the expertise and years of research by Leti, who developed this OTS selector and successfully integrated it with our ReRAM cell. We are now developing and testing larger scale arrays of the ReRAM with OTS, addressing a new set of challenges, which will get us one step closer to commercializing the technology.
SB: Which applications will be enabled by the successful OTS-ReRAM milestone?
EB: A selector is obviously an essential part of achieving the densities needed for a successful commercial ReRAM discrete chip. This is still a few years out on the roadmap, so there are probably a lot of applications that we haven’t even conceived of yet. That said, we do see a broad range of opportunities, all the way from NOR flash replacement to storage class memory. There is also a range of emerging applications that make sense for our OTS-ReRAM integration, such as 5G and AI.
SB: Do you foresee any critical barriers to the continuous advancement of ReRAM towards Storage Class Memory (XPoint-like applications)? What is the next technology milestone to be accomplished in this direction?
EB: Obviously, showing a single memory cell integrated with an OTS selector is just the beginning. The next phase on our roadmap is small arrays, followed by larger arrays. A key challenge is how to achieve production yield with a matching set of parameters/performance specifications.
From a broader industry perspective, there is a perceived challenge in advancing ReRAM – and other new memories – toward SCM, but we see this as more of an ecosystem challenge since SCM for PCs or cloud applications mainly depend on the entire system into which they are integrated, and how developers view storage memory. Developers are used to DRAM and Flash, and SCM is something in between. Once they are comfortable with the idea of SCM, we don’t think it matters to them whether the underlying technology is ReRAM, PCM, MRAM, or something else, as long as it achieves better performance, power consumption, etc., at an acceptable cost. ReRAM in a 3D configuration does all of these things better.
As Weebit is still a few years away from having discrete production chips, we believe that developers will be ready for it when our technology is available. We also think that other SCM-like opportunities would arise in the coming years, beyond PCs and servers.
SB: In the embedded-memory application space, do you see any specific opportunities for Weebit’s ReRAM? What will drive the adoption of ReRAM?
EB: Nearly every electronic device in the world is a target for us. This includes everything from popular consumer electronics applications like smartphones, wearables, and drones, to autonomous vehicles, robotics, and 5G. In embedded, the applications for ReRAM are endless and span different process nodes, memory sizes, and usage profiles.
Some of the places where we see an immediate fit are low-power MCUs, analog and power management chips, and edge AI processors. If you look at IoT requirements like low power, low cost, a high level of integration, and endurance in harsh conditions, ReRAM checks all the boxes. For analog and power management ICs, the fact that ReRAM is integrated in the back end of line is key since these chips must support multiple process nodes and derivatives, and you don’t want to be making compromises between analog components and memory on the front end.
In the area of edge AI, a good deal of on-chip memory is needed for storing the neural net weights. You can use SRAM or Flash, but neither of those is ideal. ReRAM is significantly smaller than SRAM – and is non-volatile. Unlike embedded Flash, ReRAM can scale to the most advanced processes while enabling quick memory access.
ReRAM has been in research for more than 15 years, and over this time, its key challenge has been variability. However, we are nearing the point where ReRAM can be robust even in harsh environments. We recently announced a new agreement with Leti through which we are integrating even more of their IP into our ReRAM – and tests show an order of magnitude improvement in array-level endurance and a 2x increase in data retention versus our previous results. We can also achieve high-temperature reliability up to 175oC.
We believe this, along with the need for alternatives as embedded flash hits the scalability wall, will speed the adoption of ReRAM. The industry will be looking for reliable alternatives that they can easily and cost-effectively integrate into their existing flows and processes, and that’s where Weebit comes in.
As I mentioned earlier, we recently signed our first commercial agreement to bring our ReRAM to volume production. The work with SkyWater is aimed at applications such as analog, power management, automotive, and IoT designs.
SB: Emerging memory technologies are being intensively investigated for neuromorphic computing and sensing. What is the value of Weebit’s technology in this context? How is Weebit addressing neuromorphic applications in its roadmap?
EB: We believe there is a great opportunity in the neuromorphic domain. ReRAM’s attributes make it an excellent fit for emulating neural networks in an analog fashion versus the simulations you see today. Today in the industry, we see partially brain-inspired neural networks, but there are issues like the tremendous amount of power consumed by moving data between computing cores and memory and clocking the entire system. In the future, this is where ReRAM can help through in-memory analog computing.
Together with our partner CEA-Leti, Weebit was the first to demonstrate a ReRAM-based Spiking Neural Network application (at FMS 2019). And we are continuing to collaborate closely (and publish joint papers) on neuromorphic computing with several research bodies and academic institutions to tune better our ReRAM cell architecture to their needs. With ReRAM, the industry can move toward real brain-inspired computing.
SB: How did Weebit cope with the global challenges set by the pandemic?
EB: Like most companies, Weebit had some challenges early in the pandemic. Because of lockdown measures in France, our development partner Leti had to temporarily cease operations in the plant where we are developing our technology, causing a slight delay in our schedules. In addition, the travel restrictions meant we couldn’t travel to meet with partners and customers making it more difficult to achieve our first commercial agreement. I think it also helped that, like many technology companies, our team members were already comfortable working from home before Covid.
Despite the pandemic, we were able to meet, and in some cases exceed our development milestones. It hasn’t been easy, but our teams around the world have managed to work through it.
SB: Do you want to say a few concluding words to our i-Micronews readers?
EB: Flash can’t continue to scale to the most advanced process nodes – it’s hitting a wall as an embedded technology, and the industry is well aware of its needs for a new NVM solution. The key is finding a technology that doesn’t require heavy investments or incur undue risk. We believe that ReRAM is the emerging memory technology that the industry has been waiting for, and Weebit’s ReRAM has unique advantages – like our use of standard materials, processes, and tools – that will make adoption not only painless but extremely valuable to customers. Weebit is positioned to be the first company to successfully commercialize and ship ReRAM in volume.
Eran Briman is Vice President of Marketing and Business Development at Weebit Nano. Hehas more than 25 years’ experience in the semiconductor IP field. Prior to Weebit, Eran was VP of Marketing for Corephotonics, where he helped lead the imaging IP company from start-up through commercialization to its acquisition by Samsung Electronics. Before that, he was VP of Marketing and Corporate Development for CEVA (NASDAQ: CEVA), where he helped lead the company to become the de-facto standard DSP vendor in numerous markets. He was previously Chief Architect at CEVA, and prior to that held engineering and management roles at DSP Group (NASDAQ: DSPG). Eran holds a B.Sc. in Electrical Engineering from Tel Aviv University and an MBA from the Kellogg Business School at Northwestern University.
Simone Bertolazzi, Ph.D., is a Senior Technology & Market analyst at Yole Développement (Yole), working with the Semiconductor & Software division. He is a member of Yole’s memory team and contributes day-by-day to the analysis of memory technologies, their related materials, and fabrication processes. Simone obtained a Ph.D. in physics in 2015 from École Polytechnique Fédérale de Lausanne (Switzerland), where he developed novel flash memory cells based on heterostructures of two-dimensional materials and high-κ dielectrics.
Embedded Non-Volatile Memory (NVM) readies to take off driven by low-power applications. Stand-alone NVM continues its journey toward mass adoption, despite ecosystem slowdowns.
NAND consolidation, China’s bet on two key players, the rise of the CXL interface: as the memory business narrows, the market keeps growing and is poised to exceed $200B in 2026.
After a difficult finish to 2020, the NAND market outlook improves in early 2021.
The DRAM rocket ship is fueled up and ready for blast off to reach 120B$ by 2022 due to limited supply coupled with resurgent demand.