Embedded and stand-alone NVM: two different futures ?

“The overall emerging NVM market will grow at a CAGR_20-26 ~44%”. asserts Simone Bertolazzi, Market & Technology Analyst, Memory at Yole Intelligence, part of Yole Group. “After several years in development, embedded emerging NVM technologies have gained significant maturity and are now ready for market take off.”

The year 2020 has witnessed the introduction of the first commercial products based on eMRAM, namely Sony’s GPS SoCs manufactured by Samsung (28nm FDSOI) and used in Huawei’s smartwarches, as well as Ambiq’s low-power MCUs manufactured by TSMC on 22nm ultra-low leakage (ULL) process. A number of eMRAM-based devices could enter volume production in 2021, among which are GreenWave’s AI processors with GlobalFoundries’ eMRAM (22nm FDSOI) as well as edge-AI accelerators developed by Numen and Gyrfalcon (22nm ULL at TSMC). Products incorporating embedded RRAM (eRRAM) also hit the market in 2020 with Nuvoton-Panasonic introducing new IC devices with 40nm OxRAM for security applications.

In this dynamic context, Yole investigates disruptive technologies and related markets in depth, to point out the latest innovations and underline the business opportunities.
Released today, the Emerging Non-Volatile Memory 2021 report presents an overview of the semiconductor memory market and provides an understanding of emerging NVM technologies and applications.

Yole’s partner System Plus Consulting has carried a comprehensive analysis of Intel Optane 128GB DIMMs in its dedicated report:

For Belinda Dube, Technology & Cost Analyst at System Plus Consulting: “The 3D XPoint memory forms a vertical structure with multilevel memory cell array. This strategy is used to increase the density of the cells. The memory cells use a Phase Change Material. Instead of using transistors used by conventional memories, bit storage is based on a change of material resistance. The memory cells present a material with variable electrical resistance due to a change in the crystal structure”.

Embedded memory and stand-alone memory, two industries with different ecosystems, competitive landscapes, technology status and strategies of development… Two different futures?

What are the economic and technological challenges? How does the technology evolve? Who are the key players to watch, and on which technologies are they working on? What are the impacts of the COVID-19 pandemic and how is China advancing in the NVM business?

Yole presents today its vision of the Emerging NVM industry.

As analyzed by Yole’s memory team in its new Emerging Non-Volatile Memory 2021 report, embedded  MRAM is expected to be adopted more rapidly than RRAM. In an optimistic scenario of effective roadmap execution, Yole’s analysts forecast a ~US$1.7 billion embedded MRAM market in 2026, which corresponds to ~76% of the overall embedded emerging NVM market.

However, eRRAM will remain a strong competitor. In fact, leading players have been investing in RRAM targeting eFlash replacement at 40nm and beyond:
• TSMC has enriched its 40nm ULP process with embedded OxRAM and currently offers OxRAM at 22nm.
• GlobalFoundries licensed Adesto’s CBRAMTM from Dialog Semiconductor and is now implementing it on 22nm FDSOI for low-power consumer applications.
• UMC is pursuing 28nm OxRAM development in partnership with Nuvoton-Panasonic, and they could target the smartcard market in the coming years.
And many other companies and their strategy are deeply analyzed in Yole’s Emerging NVM report…
Finally, embedded PCM is still in the race and will target eFlash replacement in automotive MCUs. STMicroelectronics is its main promoter, having selected PCM as the best emerging NVM solution for 28nm FDSOI node in the automotive market.

According to Emilie Jolivet, Director of the Semiconductor, Memory & Computing Division at Yole: “The stand-alone emerging NVM market – comprising PCM, MRAM and RRAM – will grow from ~US$595 million in 2020 to ~US$3.3 billion in 2026. It will be driven by two key segments, namely low-latency storage (enterprise and client SCM drives) and persistent memory (NVDIMM)”.
PCM will be the leading technology thanks to the sales of 3D XPoint products – particularly PM DIMMs – that are sold by Intel in a bundle with its server CPUs. New stand-alone Optane products have been launched in 2020, one of which is the long-awaited Alder Stream SSD. This is the first product that employs the 2nd generation 3D XPoint with four stacked PCM layers. 
Yole Group’s partner, Yole SystemPlus has carried a comprehensive analysis of Intel Optane 128GB DIMM, in its dedicated report:
For Belinda Dube, Technology & Cost Analyst at Yole SystemPlus: “The 3D XPoint memory forms a vertical structure with multilevel memory cell array. This strategy is used to increase the density of the cells. The memory cells use a Phase Change Material. Instead of using transistors used by conventional memories, bit storage is based on a change of material resistance. The memory cells present a material with variable electrical resistance due to a change in the crystal structure”.
In late 2020, Intel confirmed that about 200 of the Fortune 500 companies have either directly deployed Optane PM or are in the POC stages, and the POC-to-deployment conversion rate is expected to be over 85%. However, the effort to build up a software-hardware ecosystem for PM deployment has been found to be extremely challenging and time-consuming, and is currently carried out by one key player, Intel.
In comparison to PCM, the MRAM and RRAM stand-alone markets will remain significantly smaller, holding a combined ~22% share of the stand-alone emerging NVM market in 2026. The sales ramp-up time for STT-MRAM chips that are sold to IDMs and OEMs in the low-latency storage business is taking longer than expected. The use of STT-MRAM in storage modules requires significant time and resources to engineer the system architecture and develop a suitable controller technology.
Stand-alone 3D RRAM with XPoint-like structure has also been delayed compared to previously reported roadmaps: RRAM-based SSD drives will presumably be introduced using new protocols – such as CXL or Gen-Z – that are not mature yet. More time will be necessary for the new interfaces to mature to the point of triggering the adoption of novel SCM drives as alternatives to the NVMe-based drives currently available (e.g., Samsung Z-SSD, and Intel Optane SSD).