Fabless startup Adesto Technologies Corp. (Sunnyvale, Calif.) is developing a 1.2V-version of its conductive bridging RAM (CBRAM) non-volatile memory which it hopes will find favor with developers of wireless sensor nodes and wearable electronic systems.
CBRAM technology was originally developed at Arizona State University where it was also known as Programmable Metallization Cell technology. It involves the movement of metallic ions to form and break a conductive bridging filament. While early research focused on silver ion migration, Adesto memories are based on the movement of copper ions.
Adesto is using a 130nm CMOS process from Altis Semiconductor SA to manufacture its RM24x family of parts covering memory capacities of 32kbit to 1Mbit and offering functional and electrical compatibility with serial EEPROMs and a choice of SPI and I2C interface. Adesto has sold more than a million chips made with the process, which has a nominal Vdd of 2.7V.
Narbeh Derhacobian, CEO of Adesto, told Electronics 360 that he was eager to exploit the fact that the CBRAM technology – unlike flash memory – does not require charge pumps to achieve high on-chip voltages and can operate at lower voltages. Adesto has redesigned the RM24x range for operation at 1.2V, Derhacobian said.
One of the problems for wearable systems and Internet of Things nodes is that conventional non-volatile memory often requires a different voltage to the system-chip or microcontroller IC at 1.2V, Derhacobian said. "That means a power management IC and different voltage lines on the PCB, which all increases size and power consumption. At 1.2V you can have a single supply trace on the PCB or even a single package that combines SOC and CBRAM known-good die," he said.
Roadmap to near-threshold NVM
Derhacobian said that Adesto would be sampling 64kbit and 128kbit CBRAMs with 1.2V operation at the beginning of 2015 with volume production due in 3Q15. The redesign of the RM24x family had been modest and involved no trade offs in terms of read and write times, he added. It was also part of a potential path to near-threshold voltage operation. "We have helped build a demo chip with CBRAM as embedded memory in an SoC prototype. That was DARPA funded and to explore operation in sub-threshold CMOS. The CBRAM operated down to 0.6V," said Derhacobian. "There is a roadmap to get there but it will take a lot more work."
Part of that will be reaching agreement with a foundry on the next node to support CBRAM commercially. In July ," Derhacobian told Electronics 360 that Adesto was in discussions to put CBRAM into an Asian foundry at a sub-55nm node. Derhacobian said the discussions continue but that it is necessary to find a foundry and foundry customers who are sufficiently interested to back the definition of the insertion point for the technology. Given that planar 28nm CMOS is likely to be a long-lived node and as yet has no clear non-volatile memory option it might make sense for Adesto and others to target that node.
Embedded flash memory conventionally needs a floating-gate and oxide and it is disruptive to front-end of line production, said Derhacobian. "At 28nm HKMG there is no oxide left. All the microcontroller vendors are facing the same problem, what to do about non-volatile memory." As a back-end of line, memory technology we think CBRAM is the most likely candidate. For SoC non-volatile memory is a nice-to-have but for MCU it is a must-have."
However, Adesto is not the only company developing non-volatile memory. Micron Technology Inc. and Sony Corp. are developing a 16-Gbit ReRAM, based on copper ion movement and filament formation. A startup company Crossbar Inc. has a memory based on the movement of silver ions through a insulating amorphous silicon to form a filament.
"Micron-Sony is a validation of copper but not competition because we are not interested in high density memory. We are looking at smaller memories and embedded memories for IoT and wearable applications. It's a different market focus," said Derhacobian.
Related links and articles: