Researchers Extend ReRam capabilities

Not sure if this article or parts of it have been posted.
Apologies if it has.

Weebit Nano of Israel and French research institute CEA-Leti report advances in the development of resistive-RAM (ReRAM) techology, although few are ready for prime time.

CEA-Leti’s advance involves what it calls a “newfangled approach” that enables ReRAM devices to operate as energy-storage elements as well as memory, depending on the applied bias. As part of its roadmap, the institute has been exploring in-memory energy as a supplemental feature for in-memory computing to reduce energy use. ReRAM-based batteries are highly scalable as well as dynamically allocable and can be placed next to memory blocks, near the processor, it said.

Gaël Pillonet, a senior researcher at CEA-Leti, said locating the energy supply close to the processor is especially helpful when the processor requires peak power, typically supplied by an external source. ReRAM has the potential to act as an energy storage device because devices used faradaic processes to store information inside an active volume.

“It’s a unique feature of ReRAM because ReRAM is based on electro-chemical processes,” said Pillonet. This allows for high energy and power densities that far exceed that of electrostatic capacitors, and the approach is far more scalable. “We are comparable to a supercapacitor,” Pillonet claimed.

For Internet of Things applications, a node might include ReRAM for calculations. When idle, it could be used for energy storage. “We can allocate dynamically the memory to be in energy storage mode or in memory storage mode, depending on what you need to do.”

The advance has been achieved in the lab without any radical changes to the ReRAM, but there’s still work to be done to make it commercially viable. “We have to think about a packaging solution” Pillonet acknowledged.

click for full size image

Source: Weebit Nano

While transistors can be scaled down quite easily, the story is different for energy storage, said Pillonet. A button-sized battery cell can’t scale with the same efficiency. “The energy storage scale-down is very critical,” he said. “But if you don’t need too much energy for future IoT nodes, which consume very low power, maybe we can put everything together.”

CEA-Leti and Weebit Nano have also collaborated on memory research, recently announcing an initiative to analyze the environmental impact of Weebit’s ReRAM technology, comparing it to other non-volatile memory technologies. The assessment focuses on total greenhouse gas emissions associated with resource use, energy consumption and the gases and chemicals used in development, manufacture and deployment.

Weebit CEO Cody Hanoch described CEA-Leti as an extension of his company, playing a key role in efforts to develop discrete memory components based on ReRAM technology. To that end, Weebit has broadened its intellectual property cooperation with the French research institute to leverage its memory research.

For example, the partners have jointly developed memory IP, and Weebit has incorporated IP licensed from CEA-Leti into its ReRAM offerings.

The partners recently demonstrated production-level versions of Weebit’s ReRAM technology in a 28-nm process, a key step toward producing embedded non-volatile memory. They jointly tested 1-Mb ReRAM arrays based on 28-nm process technology on 300-mm wafers.

A critical focus has been development of a selector for discrete ReRAM.

Selector

A selector is a critical component for building larger arrays, allowing ReRAM to perhaps compete with NAND. The collaboration advanced with integration of an ovonic threshold switch selector and an oxide-based ReRAM.

Selectors are key elements of a memory chip, helping to access only required cells. In embedded applications, a transistor is typically used as the selector device, but transistors don’t support the densities required for discrete chips.

Hanoch said selectors are a challenging technology requiring time, effort, and money.

Weebit recently announced the partners have demonstrated their first operational crossbar arrays, a key milestone to creating discrete (stand-alone) non-volatile memory chips. Because the one-transistor, one- resistor architecture used in embedded ReRAM arrays is insufficient to support large arrays of memory cells needed in discrete memory chips, a one-selector, one-resistor architecture was instead used to develop Weebit’s crossbar arrays. That enabled the high density needed for discrete chips while allowing for 3D stacking of layers necessary for delivering higher densities.

The joint development program dates to September 2016, and includes Weebit’s silicon oxide technology, said Hanoch. “Obviously, a selector is something that is a huge project. It was just natural to work with” CEA-Leti.

Weebit has also partnered with Minnesota-based SkyWater Technology, where it is qualifying its ReRAM technology to run on SkyWater’s 130-nm CMOS process. Upon qualification, it can be used as embedded, non-volatile memory IP for applications ranging from analog, power management, automotive, IoT and radiation-hardened designs.

Under a transfer and qualification agreement, Weebit’s embedded ReRAM technology will be transferred to SkyWater’s fab for volume production by the end of 2022.

Given ReRAM’s utility in AI applications, Weebit is also exploring neural network and neuromorphic computing applications, said Hanoch. “We have partners in different leading research institutes. As a start up, you can’t just go and work on everything that you want. You have to be focused.”

For now, Weebit’s embedded business is driving revenues as the company develops discrete devices and arrays.