FMC, a fabless ferro-electric memory company based in Dresden, Germany, has completed a €100 million ($116.2 million) Series C funding round, hoping to use the funds to bring its FERAM chip technology to AI data centers, replacing DRAM and SRAM, and to achieve success in the field of storage-class memory (SCM)—a field in which Intel's Optane previously failed.
DRAM and SRAM
FERAMoffers speeds similar to DRAM and SRAM, but it is non-volatile and consumes less power. Founded in 2016, FMC originated from the Nano and Microelectronics Laboratory (NaMLab gGmbH) at the Technical University of Dresden (TU Dresden). Founders included former CEO Dr. Stefan Müller and VP of Marketing Menno Mennenga. FMC previously secured €600,000 in seed funding from High-Tech Gründerfonds (HTGF) and €4 million in Series A funding from eCAPITAL/HTGF. In 2020, it raised €17.2 million in Series B funding. This latest large-scale funding brings its total funding to $141.6 million – a substantial sum.
Of this €100 million, €77 million came from an oversubscribed Series C equity financing round, supported by significant existing and new investors, making it one of the largest such financings in the semiconductor industry. The remaining €23 million comes from public funding, including the IPCEI ME/CT project and funding from the European Innovation Council (EIC).
CEO Thomas Rückes stated:
"We are developing a new generation of memory chips and system solutions that are not only more sustainable and energy-efficient, but also faster and cheaper than current industry standards. In the past, bandwidth has been the dominant metric for AI computing, but today, energy efficiency is becoming a key factor for the next generation of AI."
We initially thought FERAM technology would face a similar predicament as Optane: promising technology, but lacking a viable path to high-volume manufacturing, thus failing to reduce costs sufficiently to achieve market penetration and profitability.
However, FMC seems to be attempting to overcome this "technology + market" barrier, and the key issue is the skyrocketing memory power consumption of GPU servers in large-scale AI data centers.
Rückes insists:
"Memory chips are a major bottleneck in the AI technology stack. FMC's DRAM+ and 3D CACHE+ technologies directly address this pain point: faster and more energy-efficient than existing products. This lays the foundation for expanding AI data centers and AI edge applications. Raising such a large amount of equity funding demonstrates the importance of our technology, and we thank our top deep-tech investors for their trust in our vision."
FMC offers two product lines: DRAM+ aims to replace DRAM, providing non-volatility (reduced power consumption), memory-as-storage capabilities, and higher endurance. CACHE+ aims to replace SRAM, providing 10 times the SRAM density and 10 times the standby power reduction, also offering non-volatility.
Both are manufactured using standard CMOS processes, achieved by converting transistors and capacitors to FeFETs (ferroelectric field-effect transistors) and FeCAPs (ferroelectric capacitors), using existing semiconductor manufacturing equipment. This is certainly an advantage, but as we understand it, manufacturing feasibility isn't the biggest problem - the acceptance from upstream in the supply chain is the "elephant in the room."
We can envision DRAM+ conceptually replacing DRAM, but cost feasibility would require a huge bet from an x86 server vendor to integrate it into servers. Simultaneously, operating system vendors would have to rewrite memory management code. System applications would also need code modifications, as DRAM+ acts as both memory and storage, eliminating the need for persistent data storage.
If we're talking about GPU servers and their HBMs, an HBM manufacturer would also have to bet on replacing existing DRAM stacks with DRAM+ stacks. This isn't entirely impossible, as HBM vendor SK Hynix is an investor in FMC. But the story doesn't end there; GPU vendors would also have to rewrite the memory handling portion of their operating systems.
If Nvidia decides to take this route, things could succeed. Nvidia is powerful enough to drive the entire market in this direction.
However, if Nvidia doesn't see it that way, FMC would have to convince AMD, but AMD's push in terms of shipments is much smaller.
We believe it's more difficult for FMC to replace x86 server DRAM because power consumption issues have little impact on server manufacturers themselves, and the number of GPUs in AI data centers already exceeds that of x86 CPUs. Furthermore, no server manufacturer possesses the market momentum of Nvidia.
FMC states that this new funding will accelerate the commercialization of its DRAM+ and 3D CACHE+ chips and system solutions, and expand its global business. The company claims its technology, with its superior energy efficiency, will help expand global AI data centers and AI edge applications, setting a new industry standard in the over €100 billion memory chip market.
Investors who bet on FMC in this funding round clearly believe it's a worthwhile bet.
Are they right? Will FMC's storage-class memory succeed, while Optane fails?
We'll likely see the answer by 2030.
Source: Content from blocksandfiles
