RRAM devices are strong candidates to replace Flash for both stand-alone storage applications
and embedded products, because of their high density, high speed, good endurance and
integration in the BEOL. But combining key features such as sufficient cycling and stable
retention at high temperature has proven to be a major challenge for memory makers.
Current RRAM thinking holds that a high number of write-and-erase sequences (cycles) leads to
poor temperature stability. The paper presented at IEDM, “Understanding the Trade-off in Terms
of Endurance, Retention and Window Margin of RRAM Using Experimental Results and
Simulations”, explains how these three memory characteristics are linked and how to modulate
them depending on the material and the programming conditions used.
“In this work, we demonstrated how physics rule memory features and performances,” said Luca
Perniola, head of Leti’s memory component lab. “Universal behaviors and tradeoffs are clearly
identified, putting boundaries on the best memories tailored for various specific applications.”
Investigating various classes of RRAM, such as OXRAM and CBRAM, the project determined
that best performance in each category was achieved with different RRAM, resulting in stability in
temperature up to 300°C, as well as a window margin up to 1,000 and endurance up to 109 cycles.
Exploring the different materials’ ability to allow many cycles with strong temperature stability, the
team identified three correlated parameters that influence performance: the number of cycles
RRAM can reach, the stability in temperature and the ratio between the two states of the memory.
Thus, by playing with the material stack, researchers were able to address various non-volatile
memory applications, targeting high speed, high endurance or high stability.
The team simulated four different RRAM active materials at the atomic level to extract parameters
that could explain the link between material and performance and identify the species responsible
for the switching mechanism between the two memory states. It also proposed an analytical
model to link the memory characteristics and the material parameters extracted from atomistic
simulation.
The biggest step was the fabrication on a 1T1R base wafer of one of the four RRAM, showing
high endurance and high window margin, which was rare in the research literature.
Leti, which offers the significant advantage of combining device fabrication, electrical
characterization and modeling, from ab initio calculations to device modeling, and design, led the
research. Its partners included MEP LAHC CNRS and LTM CNRS of Grenoble, and the WD San
Jose Research Center.