TORONTO — It wasn’t long ago that Weebit Nano CEO Coby Hanoch was focused on building the company’s embedded business with discrete products a few years out. But the company is shifting gears, he said, to respond to customer demand with a program to address the needs of discrete memory components based on its ReRAM technology.

“The plan was to establish ourselves in the embedded market first, get revenue coming in, and then start working on it,” he told EE Times in a recent telephone interview. However, as the company spoke to more and more potential customers, it became clear there was an urgent need on the discrete front. Hanoch said the most pressing demand and realistic opportunity for discrete ReRAM is a replacement for NOR flash, which is running into scaling challenges.

One of those customers is China’s XTX Technology, which in late 2019 announced it had successfully duplicated measurements on Weebit Nano non-volatile memory chips previously achieved at French research institute Leti. This verification is considered an important step to proving the manufacturability of the silicon-oxide based ReRAM. Hanoch said the company wasn’t looking to have its technology verified by a third party such as XTX; the company had a solution it was eager to test. “XTX came to us.”

Weebit’s ReRAM cell consists of two metal layers with a Silicon Oxide (SiOx) layer between them comprised of materials that can used in existing production lines. Source: Weebit

But discrete ReRAM has its own technological challenges, and that’s what Weebit Nano’s program will focus on, said Hanoch. “You need to have a better selector than transistor. In the embedded market a transistor is good enough. It might be big, but people don’t really care that much about the size of the memory array.” For discrete ReRAM to replace something like NOR flash, size does matter, however, and it means having a selector, he said. A selector helps isolate the memory cells so only the specific cells that should be modified are, and all the other cells are disconnected and not affected. “A selector is not something that you do in a couple of days.”

Hanoch said the company’s longstanding partnership with Leti will play a key role in developing the necessary selector, as the research institute has already had one in development for the past five years to fulfill a variety of purposes. This saves Weebit Nano some time, while lowering Leti’s risk. “Now it starts looking more realistic. Eventually, the selector will also be used for larger arrays,” he said, which could open the door to ReRAM competing with NAND flash, although that scenario is years away. “When you have a good opportunity, you need to grab it and use it before it disappears.”

The joint development program between Weebit and Leti goes back to September 2016, and has accelerated progress of Weebit Nano’s technology, which includes use of the company’s unique silicon oxide technology, said Hanoch, who has come to see Leti as extension of his company. “Obviously, a selector is something that is a huge project. It was just natural to work with them.”

Even if the selector challenge is solved, there are other hurdles simply because producing discrete memory is an entirely different ballgame than embedded.

“It takes many years to develop the product and get it to yield at a mass-production scale,” said Michael Yang, director at IHS Markit. How a company’s technology as a standalone chip is judged is different than when it’s embedded with another company’s broader solution. With embedded memories, it’s often an IP arrangement to have someone use the technology as part of their solution, rather than manufacturing and selling the product, he said. “The business model is very different than a discrete product and you don’t have the investment in the manufacturing line.” There’s a lot of upfront capital costs for discrete memory, he said. “It’s much more difficult in terms of scalability, mass-production, the scale, and the cost and all that.”

The embedded business model is simpler and that’s why most of these emerging memory types go that route, at least to start, said Yang, until the technology can be proven and can be taken to the next level in terms of capacity and performance. ReRAM’s path is somewhat similar to MRAM, where companies, such as Everspin, have built up an embedded customer base with revenues that can fuel its discrete memory efforts.

Hanoch said that despite ramping up its discrete ReRAM development, Weebit Nano is certainly not abandoning the embedded market. “That still is clearly the much faster path to revenues. We need to finish developing the memory module to get to the point where we can start having commercial discussions with customers.”

Gary Hilson is a general contributing editor with a focus on memory and flash technologies.