Resistive RAM (ReRAM) technology has been waiting for its moment in the sun for several years now. Despite its placement on the roadmaps of several chip manufacturers and as part of novel architectures like HPE’s shuttered memristor-based “The Machine” project, it never took off as a revolutionary memory technology, even though the concept makes excellent sense.
Manufacturability, cost, and other factors played a role but for companies like Crossbar, which staked an entire business on the emerging role of ReRAM, the gambit has yet to pay off. That same company today announced a possible new life for ReRAM technology—in security.
CrossBar, Inc. says ReRAM technology can be used in the creation of physical unclonable function (PUF) cryptographic keys since each key is specific to the individual IC.
This idea has some backing, including from Dr. Bertrand Cambou from North Arizona University’s nanotechnology and cybersecurity group. He says that after evaluating other PUF approaches, there are advantages to ReRAM. “Due to its unique stochastic and electrical characteristics, CrossBar’s ReRAM PUF enables more secure systems compared to incumbent PUF technologies.” Cambou discusses those other approaches in comparison to ReRAM here in detail. It’s a useful description of how PUFs are a fit for ReRAM as well.
“The approach to generate PUF on ReRAM arrays is promising based on the statistical analysis of the resistive devices,” Cambou says. He adds, “Resistive RAM is an attractive memory technology for designing secure applications, PUFs, and RNGs. It is low power, fast, compact, and less sensitive to side channel attack than flash memory.”
This has to be frustrating as hell for CrossBar. The company finds itself in yet another difficult market after ReRAM failed to materialize as the next big thing in memory innovation. There are already many technologies on the PUF case (all of which use SRAM, by the way, which is cheap). Compared to SRAM-based PUF approaches, their method allows for a higher degree of randomness, lower bit error rates, and all of this “without requiring fuzzy extractors, helper data, or heavy error correction codes,” according to Mark Davis, CrossBar president.
In terms of ReRAM more broadly, it’s worth a closer look (we’re chasing one) on why ReRAM never got off the ground. It can scale below 10nm, has outstanding flash-outpacing read latencies and can write quite a bit faster than NAND. It doesn’t require any super fancy manufacturing magic. So why did the market pass on ReRAM? Deeper dive on that coming.
CrossBar says that as for its business now, it’s becoming available at several CMOs foundry process nodes to get even higher densities and more tightly integrated devices. The company, by the way, stays afloat by licensing to SoC and memory companies as both off-the-shelf and custom IP cores. It was first to market in 2010, in the early days of next-generation memory and while the security applications angle might help them keep fighting the ReRAM fight, it’s hard to tell what the future of the technology generally looks like in the years ahead.