Smaller, faster, more secure: CrossBar builds better ReRAM at NanoLab

Collage of three images. First and third images: Researchers in the Marvell NanoLab at CITRIS, wearing protective attire. Second image: Four members of the CrossBar team. The two people in the center each hold a round silicon wafer.

Sung Hyun Jo, CTO of CrossBar Inc., credits the flexible, accommodating Berkeley Marvell Nanofabrication Laboratory at CITRIS for helping the company discover ‘missing pieces’ during the commercialization process.

CrossBar Inc. intends to revolutionize the nonvolatile memory industry with smaller, faster, more secure resistive random access memory, or ReRAM, technology. According to CrossBar’s chief technology officer, Sung Hyun Jo, the Berkeley Marvell Nanofabrication Laboratory at CITRIS has played a key role in the company’s efforts.

Backed by Kleiner Perkins, one of the biggest venture capital firms in California, CrossBar launched in 2010, spun out of the University of Michigan electrical engineering and computer science research lab where Jo received his Ph.D. Since the company’s founding, it has accrued over 190 issued U.S. patents. 

Improving size and speed with ReRAM

CrossBar is addressing one of the greatest challenges in consumer electronics, as memory determines a product’s performance and price. Their work has profound implications for the future of memory in relation to artificial intelligence and the Internet of Things.

“Everyone’s looking for a smaller device or a faster device,” Jo said. “It’s fair to say most of your electronic devices’ performance is limited by the performance of memory.”

Flash memory, widely used for storing and transferring data, is a type of nonvolatile memory, meaning it can retain data even without power. Flash memory does not scale well below a 22 nanometer technology node (the unit of measurement for semiconductor devices, such as the transistors that make up memory chips). However, central processing units and graphics processing units can be manufactured at or below a 10 nm technology node. Jo said that this difference in size shows that memory technology is around 10 years behind CPU and GPU technology. 

CrossBar improves upon traditional memory by using ReRAM. According to Jo, ReRAM is not only smaller and faster than flash memory, but also more secure. This is because ReRAM stores information through atomic movement, which is harder to reverse-engineer than traditional RAM. ReRAM is also nonvolatile, so, like flash memory, it can retain information even after power is removed.

Finding a home for further development

After CrossBar’s preliminary research in Michigan showed promising results, the company was ready to demonstrate that the technology could be commercially manufactured in a semiconductor fabrication plant, or foundry. They looked for suitable facilities for further research and found the Marvell NanoLab at CITRIS in 2011.

Jo said that they chose the NanoLab because it was one of the few to offer 6-inch wafer processing, and it was also more efficient than the other labs they were considering.

“The Marvell NanoLab at CITRIS is one of the rare labs that has everything you need, so we don’t have to use multiple labs to perform different parts of our experiments,” said Jo. ​“It’s flexible and faster, because in a production lab, everything is slow and takes several months. But in Berkeley, it can take days or weeks.” 

Discovering a collaborative spirit

Jo also found that other facilities were so secretive about other projects that it would slow down CrossBar’s progress and make it difficult for the team to estimate their development timelines. 

“When we start a process, the other labs ask us to prove that we will not affect any other teams, but we don’t know details about their processes,” said Jo. “It’s like a chicken and egg situation.”

Another feature of the NanoLab that Jo appreciates is the clear, accessible documentation and staff, faculty and student support that CrossBar receives. At other facilities, he said, they were fixated on confidentiality, meaning that even if someone working on one project had skills or knowledge that could benefit another team’s research, they could not offer help. The NanoLab’s collaborative spirit allows researchers to concentrate on their own areas of expertise while leveraging commonly available nonproprietary processing platforms.

“At the Marvell NanoLab at CITRIS, you have this extensive knowledge base, so we don’t have to reinvent the wheel,” said Jo. “By using that, we focus on our own results.”

He also noted that when CrossBar found issues with their technology during R&D, they could perform designs of experiments, or DOEs, quickly with staff assistance.

Solving puzzles and seeing results

Through their research at the Berkeley Marvell NanoLab at CITRIS, Jo and his team proved that ReRAM can scale below 10 nm, much smaller than the average 28 nm size of RAM. They also proved that ReRAM can be 10 times faster than traditional nonvolatile RAM.

After working with the NanoLab for 10 years, Jo says that he would recommend the facility to other companies looking for labs in California.

“When transferring technology, we think we know everything, but there are missing pieces in the transfer from technology to commercialization,” Jo said. “Thanks to the NanoLab, we are making good progress in top-tier foundries. All the missing pieces were solved here.”