Progress of emerging non-volatile memory technologies in industry

Abstract

This prospective and performance summary provides a view on the state of the art of emerging non-volatile memory (eNVM) in the semiconductor industry. The overarching aim is to inform academic researchers of the status of these technologies in industry, so as to help direct the right academic research questions for future materials and device development. eNVM already have a strong foothold in the semiconductor industry with the main target of replacing embedded flash memory, and soon possibly DRAM and SRAM, i.e. replacing conventional memory. Magnetic and resistive memory are the current frontrunners among eNVM for embedded flash replacement and they are very advanced in this, which poses high demands on future academic research for eNVM for this purpose. Phase-change and ferroelectric memory are less available as commercially available products. The use of eNVM for new forms of artificial intelligence hardware is a much more open field for future academic research.

Keywords: Artificial intelligence; Electronic material; Emergent phenomena; Ferroelectric; Nanoelectronics; Neuromorphic; Spintronic; Thin film.

Figure 1

Timeline of recent technology demonstrations of eNVM with respective CMOS nodes. Sources for the different demonstrations can be found in the company details sections. Note that we only included companies with confirmed and active consumer products (notably, Intel/Micron 3D XPoint is not included, as it was discontinued), but for those, we included both products and research papers. *For Fujitsu, information about CMOS node integration could not be found. 2022 was one of their larger RRAM product launches, another was at least as early as 2016. For FRAM, 2022 was a major release, but their FRAM mass production started already in 1999. The top right inset clarifies companies where the name is not evident from the logo.

Figure 2

Schematic illustrations of the working principles of (a) FRAM, (b) PCRAM, (c) filamentary RRAM, (d) area-dependent RRAM, and (e) MRAM. Grey indicates electrodes and pale yellow indicates the memory material. For each, the top schematic represents the high resistance state, and the bottom represents the low-resistance state. The arrows in (a) indicate the ferroelectric polarisation orientation. The dots in (b) indicate the amorphous and crystalline phase of the material. The gradient in (d) indicates the height and width of a Schottky barrier. The arrows in (e) indicate the spin orientation and FL and RL stand for free layer and reference layer, respectively.

Figure 3

World Intellectual Property Organisation Patentscope search results from 2015 to July 2024 for the four included eNVM. The number of IP applications has been steadily increasing over the years. In line with Fig. 1, RRAM and MRAM are the frontrunner technologies, but the others are not far behind. Patentscope search settings were office = all, languages = en, stemming = FALSE, single family = FALSE, and including NPL = FALSE. Search terms for the individual eNVM technologies were as follows: ALLTXT:(“phase change random access memory”) or (“phase change RAM”), ALLTXT:(“ferroelectric RAM”) or (“ferroelectric random access memory”), ALLTXT:(“resistive random access memory”) or (“resistive RAM”), and ALLTXT:(“magnetoresistive RAM”) or (“magnetic RAM”) or (“magnetoresistive random access memory”) or (“magnetic random access memory”).