Overview of Embedded Nonvolatile Memories for Automotive and IoT Applications

Demands for high density and energy-efficient embedded non-volatile memories (NVMs) are rapidly spread for mobile, IoT, automotive and artificial intelligence (AI) applications. Recently, there are several types of emerging NVMs such as ReRAM, MRAM, PCM and FeRAM. Some of them are targeting for replacing legacy flash memories with advantages of cost-effective, high-density, high-endurance and so on. MRAM is an only candidate for replacing embedded SRAMs with high-speed read/write operations and high-endurances over 1012 cycles, but there are still challenges on yield ramping, or high-temperature retention as well as other emerging NVMs. Contrary, flash memories are based on legacy technology without new materials and tools, and already have maturities by silicon proven on huge volumes in mass-productions.

In the first half of this talk, embedded flash memory technologies and applications in advanced nodes are discussed. There are mainly two types of embedded flash memory technologies, one is floating-gate (FG) type and the other is charge trap type. Silicon-Oxide-Nitride-Oxide-Silicon (SONOS) is one of the later types. Technology scaling of FG flash has a limitation beyond 28-nm node due to its device structure, while SONOS flash has potentially scalable in advanced FD-SOI and Fin-FET devices below 28-nm node. Those feasibilities are shown by TCAD simulations and existing planar bulk silicon data. For example, it is shown that the latest retention characteristics of SONOS are improved by process optimizations, achieving over 10 years at 150℃ on actual devices. Whereas, the challenges for implementing flash memories on FD-SOI and Fin-FET are also discussed.

In the second half of this talk, the latest design status of ReRAM and MRAM which are fully implemented in the back-end-of-line (BEOL) based on logic baseline transistors are introduced. After that, some applications like a NV-SRAM by combining with SRAM cell and NVM cell are introduced. Analog computing-in-memory (CIM) based on NVM memories are also briefly introduced for energy efficient AI applications in the edge computing.